comparison src/os/linux/vm/os_linux.cpp @ 0:a61af66fc99e jdk7-b24

Initial load
author duke
date Sat, 01 Dec 2007 00:00:00 +0000
parents
children e195fe4c40c7 5a76ab815e34
comparison
equal deleted inserted replaced
-1:000000000000 0:a61af66fc99e
1 /*
2 * Copyright 1999-2007 Sun Microsystems, Inc. All Rights Reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
24
25 // do not include precompiled header file
26 # include "incls/_os_linux.cpp.incl"
27
28 // put OS-includes here
29 # include <sys/types.h>
30 # include <sys/mman.h>
31 # include <pthread.h>
32 # include <signal.h>
33 # include <errno.h>
34 # include <dlfcn.h>
35 # include <stdio.h>
36 # include <unistd.h>
37 # include <sys/resource.h>
38 # include <pthread.h>
39 # include <sys/stat.h>
40 # include <sys/time.h>
41 # include <sys/times.h>
42 # include <sys/utsname.h>
43 # include <sys/socket.h>
44 # include <sys/wait.h>
45 # include <pwd.h>
46 # include <poll.h>
47 # include <semaphore.h>
48 # include <fcntl.h>
49 # include <string.h>
50 # include <syscall.h>
51 # include <sys/sysinfo.h>
52 # include <gnu/libc-version.h>
53 # include <sys/ipc.h>
54 # include <sys/shm.h>
55 # include <link.h>
56
57 #define MAX_PATH (2 * K)
58
59 // for timer info max values which include all bits
60 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
61 #define SEC_IN_NANOSECS 1000000000LL
62
63 ////////////////////////////////////////////////////////////////////////////////
64 // global variables
65 julong os::Linux::_physical_memory = 0;
66
67 address os::Linux::_initial_thread_stack_bottom = NULL;
68 uintptr_t os::Linux::_initial_thread_stack_size = 0;
69
70 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
71 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
72 Mutex* os::Linux::_createThread_lock = NULL;
73 pthread_t os::Linux::_main_thread;
74 int os::Linux::_page_size = -1;
75 bool os::Linux::_is_floating_stack = false;
76 bool os::Linux::_is_NPTL = false;
77 bool os::Linux::_supports_fast_thread_cpu_time = false;
78 char * os::Linux::_glibc_version = NULL;
79 char * os::Linux::_libpthread_version = NULL;
80
81 static jlong initial_time_count=0;
82
83 static int clock_tics_per_sec = 100;
84
85 // For diagnostics to print a message once. see run_periodic_checks
86 static sigset_t check_signal_done;
87 static bool check_signals = true;;
88
89 static pid_t _initial_pid = 0;
90
91 /* Signal number used to suspend/resume a thread */
92
93 /* do not use any signal number less than SIGSEGV, see 4355769 */
94 static int SR_signum = SIGUSR2;
95 sigset_t SR_sigset;
96
97 ////////////////////////////////////////////////////////////////////////////////
98 // utility functions
99
100 static int SR_initialize();
101 static int SR_finalize();
102
103 julong os::available_memory() {
104 return Linux::available_memory();
105 }
106
107 julong os::Linux::available_memory() {
108 // values in struct sysinfo are "unsigned long"
109 struct sysinfo si;
110 sysinfo(&si);
111
112 return (julong)si.freeram * si.mem_unit;
113 }
114
115 julong os::physical_memory() {
116 return Linux::physical_memory();
117 }
118
119 ////////////////////////////////////////////////////////////////////////////////
120 // environment support
121
122 bool os::getenv(const char* name, char* buf, int len) {
123 const char* val = ::getenv(name);
124 if (val != NULL && strlen(val) < (size_t)len) {
125 strcpy(buf, val);
126 return true;
127 }
128 if (len > 0) buf[0] = 0; // return a null string
129 return false;
130 }
131
132
133 // Return true if user is running as root.
134
135 bool os::have_special_privileges() {
136 static bool init = false;
137 static bool privileges = false;
138 if (!init) {
139 privileges = (getuid() != geteuid()) || (getgid() != getegid());
140 init = true;
141 }
142 return privileges;
143 }
144
145
146 #ifndef SYS_gettid
147 // i386: 224, ia64: 1105, amd64: 186, sparc 143
148 #ifdef __ia64__
149 #define SYS_gettid 1105
150 #elif __i386__
151 #define SYS_gettid 224
152 #elif __amd64__
153 #define SYS_gettid 186
154 #elif __sparc__
155 #define SYS_gettid 143
156 #else
157 #error define gettid for the arch
158 #endif
159 #endif
160
161 // Cpu architecture string
162 #if defined(IA64)
163 static char cpu_arch[] = "ia64";
164 #elif defined(IA32)
165 static char cpu_arch[] = "i386";
166 #elif defined(AMD64)
167 static char cpu_arch[] = "amd64";
168 #elif defined(SPARC)
169 # ifdef _LP64
170 static char cpu_arch[] = "sparcv9";
171 # else
172 static char cpu_arch[] = "sparc";
173 # endif
174 #else
175 #error Add appropriate cpu_arch setting
176 #endif
177
178
179 // pid_t gettid()
180 //
181 // Returns the kernel thread id of the currently running thread. Kernel
182 // thread id is used to access /proc.
183 //
184 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
185 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
186 //
187 pid_t os::Linux::gettid() {
188 int rslt = syscall(SYS_gettid);
189 if (rslt == -1) {
190 // old kernel, no NPTL support
191 return getpid();
192 } else {
193 return (pid_t)rslt;
194 }
195 }
196
197 // Most versions of linux have a bug where the number of processors are
198 // determined by looking at the /proc file system. In a chroot environment,
199 // the system call returns 1. This causes the VM to act as if it is
200 // a single processor and elide locking (see is_MP() call).
201 static bool unsafe_chroot_detected = false;
202 static char *unstable_chroot_error = "/proc file system not found.\n"
203 "Java may be unstable running multithreaded in a chroot "
204 "environment on Linux when /proc filesystem is not mounted.";
205
206 void os::Linux::initialize_system_info() {
207 _processor_count = sysconf(_SC_NPROCESSORS_CONF);
208 if (_processor_count == 1) {
209 pid_t pid = os::Linux::gettid();
210 char fname[32];
211 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
212 FILE *fp = fopen(fname, "r");
213 if (fp == NULL) {
214 unsafe_chroot_detected = true;
215 } else {
216 fclose(fp);
217 }
218 }
219 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
220 assert(_processor_count > 0, "linux error");
221 }
222
223 void os::init_system_properties_values() {
224 // char arch[12];
225 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
226
227 // The next steps are taken in the product version:
228 //
229 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
230 // This library should be located at:
231 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
232 //
233 // If "/jre/lib/" appears at the right place in the path, then we
234 // assume libjvm[_g].so is installed in a JDK and we use this path.
235 //
236 // Otherwise exit with message: "Could not create the Java virtual machine."
237 //
238 // The following extra steps are taken in the debugging version:
239 //
240 // If "/jre/lib/" does NOT appear at the right place in the path
241 // instead of exit check for $JAVA_HOME environment variable.
242 //
243 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
244 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
245 // it looks like libjvm[_g].so is installed there
246 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
247 //
248 // Otherwise exit.
249 //
250 // Important note: if the location of libjvm.so changes this
251 // code needs to be changed accordingly.
252
253 // The next few definitions allow the code to be verbatim:
254 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
255 #define getenv(n) ::getenv(n)
256
257 /*
258 * See ld(1):
259 * The linker uses the following search paths to locate required
260 * shared libraries:
261 * 1: ...
262 * ...
263 * 7: The default directories, normally /lib and /usr/lib.
264 */
265 #define DEFAULT_LIBPATH "/lib:/usr/lib"
266
267 #define EXTENSIONS_DIR "/lib/ext"
268 #define ENDORSED_DIR "/lib/endorsed"
269 #define REG_DIR "/usr/java/packages"
270
271 {
272 /* sysclasspath, java_home, dll_dir */
273 {
274 char *home_path;
275 char *dll_path;
276 char *pslash;
277 char buf[MAXPATHLEN];
278 os::jvm_path(buf, sizeof(buf));
279
280 // Found the full path to libjvm.so.
281 // Now cut the path to <java_home>/jre if we can.
282 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
283 pslash = strrchr(buf, '/');
284 if (pslash != NULL)
285 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
286 dll_path = malloc(strlen(buf) + 1);
287 if (dll_path == NULL)
288 return;
289 strcpy(dll_path, buf);
290 Arguments::set_dll_dir(dll_path);
291
292 if (pslash != NULL) {
293 pslash = strrchr(buf, '/');
294 if (pslash != NULL) {
295 *pslash = '\0'; /* get rid of /<arch> */
296 pslash = strrchr(buf, '/');
297 if (pslash != NULL)
298 *pslash = '\0'; /* get rid of /lib */
299 }
300 }
301
302 home_path = malloc(strlen(buf) + 1);
303 if (home_path == NULL)
304 return;
305 strcpy(home_path, buf);
306 Arguments::set_java_home(home_path);
307
308 if (!set_boot_path('/', ':'))
309 return;
310 }
311
312 /*
313 * Where to look for native libraries
314 *
315 * Note: Due to a legacy implementation, most of the library path
316 * is set in the launcher. This was to accomodate linking restrictions
317 * on legacy Linux implementations (which are no longer supported).
318 * Eventually, all the library path setting will be done here.
319 *
320 * However, to prevent the proliferation of improperly built native
321 * libraries, the new path component /usr/java/packages is added here.
322 * Eventually, all the library path setting will be done here.
323 */
324 {
325 char *ld_library_path;
326
327 /*
328 * Construct the invariant part of ld_library_path. Note that the
329 * space for the colon and the trailing null are provided by the
330 * nulls included by the sizeof operator (so actually we allocate
331 * a byte more than necessary).
332 */
333 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
334 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
335 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
336
337 /*
338 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
339 * should always exist (until the legacy problem cited above is
340 * addressed).
341 */
342 char *v = getenv("LD_LIBRARY_PATH");
343 if (v != NULL) {
344 char *t = ld_library_path;
345 /* That's +1 for the colon and +1 for the trailing '\0' */
346 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
347 sprintf(ld_library_path, "%s:%s", v, t);
348 }
349 Arguments::set_library_path(ld_library_path);
350 }
351
352 /*
353 * Extensions directories.
354 *
355 * Note that the space for the colon and the trailing null are provided
356 * by the nulls included by the sizeof operator (so actually one byte more
357 * than necessary is allocated).
358 */
359 {
360 char *buf = malloc(strlen(Arguments::get_java_home()) +
361 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
362 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
363 Arguments::get_java_home());
364 Arguments::set_ext_dirs(buf);
365 }
366
367 /* Endorsed standards default directory. */
368 {
369 char * buf;
370 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
371 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
372 Arguments::set_endorsed_dirs(buf);
373 }
374 }
375
376 #undef malloc
377 #undef getenv
378 #undef EXTENSIONS_DIR
379 #undef ENDORSED_DIR
380
381 // Done
382 return;
383 }
384
385 ////////////////////////////////////////////////////////////////////////////////
386 // breakpoint support
387
388 void os::breakpoint() {
389 BREAKPOINT;
390 }
391
392 extern "C" void breakpoint() {
393 // use debugger to set breakpoint here
394 }
395
396 ////////////////////////////////////////////////////////////////////////////////
397 // signal support
398
399 debug_only(static bool signal_sets_initialized = false);
400 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
401
402 bool os::Linux::is_sig_ignored(int sig) {
403 struct sigaction oact;
404 sigaction(sig, (struct sigaction*)NULL, &oact);
405 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
406 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
407 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
408 return true;
409 else
410 return false;
411 }
412
413 void os::Linux::signal_sets_init() {
414 // Should also have an assertion stating we are still single-threaded.
415 assert(!signal_sets_initialized, "Already initialized");
416 // Fill in signals that are necessarily unblocked for all threads in
417 // the VM. Currently, we unblock the following signals:
418 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
419 // by -Xrs (=ReduceSignalUsage));
420 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
421 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
422 // the dispositions or masks wrt these signals.
423 // Programs embedding the VM that want to use the above signals for their
424 // own purposes must, at this time, use the "-Xrs" option to prevent
425 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
426 // (See bug 4345157, and other related bugs).
427 // In reality, though, unblocking these signals is really a nop, since
428 // these signals are not blocked by default.
429 sigemptyset(&unblocked_sigs);
430 sigemptyset(&allowdebug_blocked_sigs);
431 sigaddset(&unblocked_sigs, SIGILL);
432 sigaddset(&unblocked_sigs, SIGSEGV);
433 sigaddset(&unblocked_sigs, SIGBUS);
434 sigaddset(&unblocked_sigs, SIGFPE);
435 sigaddset(&unblocked_sigs, SR_signum);
436
437 if (!ReduceSignalUsage) {
438 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
439 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
440 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
441 }
442 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
443 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
444 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
445 }
446 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
447 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
448 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
449 }
450 }
451 // Fill in signals that are blocked by all but the VM thread.
452 sigemptyset(&vm_sigs);
453 if (!ReduceSignalUsage)
454 sigaddset(&vm_sigs, BREAK_SIGNAL);
455 debug_only(signal_sets_initialized = true);
456
457 }
458
459 // These are signals that are unblocked while a thread is running Java.
460 // (For some reason, they get blocked by default.)
461 sigset_t* os::Linux::unblocked_signals() {
462 assert(signal_sets_initialized, "Not initialized");
463 return &unblocked_sigs;
464 }
465
466 // These are the signals that are blocked while a (non-VM) thread is
467 // running Java. Only the VM thread handles these signals.
468 sigset_t* os::Linux::vm_signals() {
469 assert(signal_sets_initialized, "Not initialized");
470 return &vm_sigs;
471 }
472
473 // These are signals that are blocked during cond_wait to allow debugger in
474 sigset_t* os::Linux::allowdebug_blocked_signals() {
475 assert(signal_sets_initialized, "Not initialized");
476 return &allowdebug_blocked_sigs;
477 }
478
479 void os::Linux::hotspot_sigmask(Thread* thread) {
480
481 //Save caller's signal mask before setting VM signal mask
482 sigset_t caller_sigmask;
483 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
484
485 OSThread* osthread = thread->osthread();
486 osthread->set_caller_sigmask(caller_sigmask);
487
488 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
489
490 if (!ReduceSignalUsage) {
491 if (thread->is_VM_thread()) {
492 // Only the VM thread handles BREAK_SIGNAL ...
493 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
494 } else {
495 // ... all other threads block BREAK_SIGNAL
496 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
497 }
498 }
499 }
500
501 //////////////////////////////////////////////////////////////////////////////
502 // detecting pthread library
503
504 void os::Linux::libpthread_init() {
505 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
506 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
507 // generic name for earlier versions.
508 // Define macros here so we can build HotSpot on old systems.
509 # ifndef _CS_GNU_LIBC_VERSION
510 # define _CS_GNU_LIBC_VERSION 2
511 # endif
512 # ifndef _CS_GNU_LIBPTHREAD_VERSION
513 # define _CS_GNU_LIBPTHREAD_VERSION 3
514 # endif
515
516 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
517 if (n > 0) {
518 char *str = (char *)malloc(n);
519 confstr(_CS_GNU_LIBC_VERSION, str, n);
520 os::Linux::set_glibc_version(str);
521 } else {
522 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
523 static char _gnu_libc_version[32];
524 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
525 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
526 os::Linux::set_glibc_version(_gnu_libc_version);
527 }
528
529 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
530 if (n > 0) {
531 char *str = (char *)malloc(n);
532 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
533
534 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
535 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
536 // is the case:
537 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
538 strstr(str, "NPTL")) {
539 // LinuxThreads has a hard limit on max number of threads. So
540 // sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
541 // On the other hand, NPTL does not have such a limit, sysconf()
542 // will return -1 and errno is not changed. Check if it is really
543 // NPTL:
544 if (sysconf(_SC_THREAD_THREADS_MAX) > 0) {
545 free(str);
546 str = "linuxthreads";
547 }
548 }
549 os::Linux::set_libpthread_version(str);
550 } else {
551 // glibc before 2.3.2 only has LinuxThreads.
552 os::Linux::set_libpthread_version("linuxthreads");
553 }
554
555 if (strstr(libpthread_version(), "NPTL")) {
556 os::Linux::set_is_NPTL();
557 } else {
558 os::Linux::set_is_LinuxThreads();
559 }
560
561 // LinuxThreads have two flavors: floating-stack mode, which allows variable
562 // stack size; and fixed-stack mode. NPTL is always floating-stack.
563 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
564 os::Linux::set_is_floating_stack();
565 }
566 }
567
568 /////////////////////////////////////////////////////////////////////////////
569 // thread stack
570
571 // Force Linux kernel to expand current thread stack. If "bottom" is close
572 // to the stack guard, caller should block all signals.
573 //
574 // MAP_GROWSDOWN:
575 // A special mmap() flag that is used to implement thread stacks. It tells
576 // kernel that the memory region should extend downwards when needed. This
577 // allows early versions of LinuxThreads to only mmap the first few pages
578 // when creating a new thread. Linux kernel will automatically expand thread
579 // stack as needed (on page faults).
580 //
581 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
582 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
583 // region, it's hard to tell if the fault is due to a legitimate stack
584 // access or because of reading/writing non-exist memory (e.g. buffer
585 // overrun). As a rule, if the fault happens below current stack pointer,
586 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
587 // application (see Linux kernel fault.c).
588 //
589 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
590 // stack overflow detection.
591 //
592 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
593 // not use this flag. However, the stack of initial thread is not created
594 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
595 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
596 // and then attach the thread to JVM.
597 //
598 // To get around the problem and allow stack banging on Linux, we need to
599 // manually expand thread stack after receiving the SIGSEGV.
600 //
601 // There are two ways to expand thread stack to address "bottom", we used
602 // both of them in JVM before 1.5:
603 // 1. adjust stack pointer first so that it is below "bottom", and then
604 // touch "bottom"
605 // 2. mmap() the page in question
606 //
607 // Now alternate signal stack is gone, it's harder to use 2. For instance,
608 // if current sp is already near the lower end of page 101, and we need to
609 // call mmap() to map page 100, it is possible that part of the mmap() frame
610 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
611 // That will destroy the mmap() frame and cause VM to crash.
612 //
613 // The following code works by adjusting sp first, then accessing the "bottom"
614 // page to force a page fault. Linux kernel will then automatically expand the
615 // stack mapping.
616 //
617 // _expand_stack_to() assumes its frame size is less than page size, which
618 // should always be true if the function is not inlined.
619
620 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
621 #define NOINLINE
622 #else
623 #define NOINLINE __attribute__ ((noinline))
624 #endif
625
626 static void _expand_stack_to(address bottom) NOINLINE;
627
628 static void _expand_stack_to(address bottom) {
629 address sp;
630 size_t size;
631 volatile char *p;
632
633 // Adjust bottom to point to the largest address within the same page, it
634 // gives us a one-page buffer if alloca() allocates slightly more memory.
635 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
636 bottom += os::Linux::page_size() - 1;
637
638 // sp might be slightly above current stack pointer; if that's the case, we
639 // will alloca() a little more space than necessary, which is OK. Don't use
640 // os::current_stack_pointer(), as its result can be slightly below current
641 // stack pointer, causing us to not alloca enough to reach "bottom".
642 sp = (address)&sp;
643
644 if (sp > bottom) {
645 size = sp - bottom;
646 p = (volatile char *)alloca(size);
647 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
648 p[0] = '\0';
649 }
650 }
651
652 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
653 assert(t!=NULL, "just checking");
654 assert(t->osthread()->expanding_stack(), "expand should be set");
655 assert(t->stack_base() != NULL, "stack_base was not initialized");
656
657 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
658 sigset_t mask_all, old_sigset;
659 sigfillset(&mask_all);
660 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
661 _expand_stack_to(addr);
662 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
663 return true;
664 }
665 return false;
666 }
667
668 //////////////////////////////////////////////////////////////////////////////
669 // create new thread
670
671 static address highest_vm_reserved_address();
672
673 // check if it's safe to start a new thread
674 static bool _thread_safety_check(Thread* thread) {
675 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
676 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
677 // Heap is mmap'ed at lower end of memory space. Thread stacks are
678 // allocated (MAP_FIXED) from high address space. Every thread stack
679 // occupies a fixed size slot (usually 2Mbytes, but user can change
680 // it to other values if they rebuild LinuxThreads).
681 //
682 // Problem with MAP_FIXED is that mmap() can still succeed even part of
683 // the memory region has already been mmap'ed. That means if we have too
684 // many threads and/or very large heap, eventually thread stack will
685 // collide with heap.
686 //
687 // Here we try to prevent heap/stack collision by comparing current
688 // stack bottom with the highest address that has been mmap'ed by JVM
689 // plus a safety margin for memory maps created by native code.
690 //
691 // This feature can be disabled by setting ThreadSafetyMargin to 0
692 //
693 if (ThreadSafetyMargin > 0) {
694 address stack_bottom = os::current_stack_base() - os::current_stack_size();
695
696 // not safe if our stack extends below the safety margin
697 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
698 } else {
699 return true;
700 }
701 } else {
702 // Floating stack LinuxThreads or NPTL:
703 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
704 // there's not enough space left, pthread_create() will fail. If we come
705 // here, that means enough space has been reserved for stack.
706 return true;
707 }
708 }
709
710 // Thread start routine for all newly created threads
711 static void *java_start(Thread *thread) {
712 // Try to randomize the cache line index of hot stack frames.
713 // This helps when threads of the same stack traces evict each other's
714 // cache lines. The threads can be either from the same JVM instance, or
715 // from different JVM instances. The benefit is especially true for
716 // processors with hyperthreading technology.
717 static int counter = 0;
718 int pid = os::current_process_id();
719 alloca(((pid ^ counter++) & 7) * 128);
720
721 ThreadLocalStorage::set_thread(thread);
722
723 OSThread* osthread = thread->osthread();
724 Monitor* sync = osthread->startThread_lock();
725
726 // non floating stack LinuxThreads needs extra check, see above
727 if (!_thread_safety_check(thread)) {
728 // notify parent thread
729 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
730 osthread->set_state(ZOMBIE);
731 sync->notify_all();
732 return NULL;
733 }
734
735 // thread_id is kernel thread id (similar to Solaris LWP id)
736 osthread->set_thread_id(os::Linux::gettid());
737
738 if (UseNUMA) {
739 int lgrp_id = os::numa_get_group_id();
740 if (lgrp_id != -1) {
741 thread->set_lgrp_id(lgrp_id);
742 }
743 }
744 // initialize signal mask for this thread
745 os::Linux::hotspot_sigmask(thread);
746
747 // initialize floating point control register
748 os::Linux::init_thread_fpu_state();
749
750 // handshaking with parent thread
751 {
752 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
753
754 // notify parent thread
755 osthread->set_state(INITIALIZED);
756 sync->notify_all();
757
758 // wait until os::start_thread()
759 while (osthread->get_state() == INITIALIZED) {
760 sync->wait(Mutex::_no_safepoint_check_flag);
761 }
762 }
763
764 // call one more level start routine
765 thread->run();
766
767 return 0;
768 }
769
770 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
771 assert(thread->osthread() == NULL, "caller responsible");
772
773 // Allocate the OSThread object
774 OSThread* osthread = new OSThread(NULL, NULL);
775 if (osthread == NULL) {
776 return false;
777 }
778
779 // set the correct thread state
780 osthread->set_thread_type(thr_type);
781
782 // Initial state is ALLOCATED but not INITIALIZED
783 osthread->set_state(ALLOCATED);
784
785 thread->set_osthread(osthread);
786
787 // init thread attributes
788 pthread_attr_t attr;
789 pthread_attr_init(&attr);
790 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
791
792 // stack size
793 if (os::Linux::supports_variable_stack_size()) {
794 // calculate stack size if it's not specified by caller
795 if (stack_size == 0) {
796 stack_size = os::Linux::default_stack_size(thr_type);
797
798 switch (thr_type) {
799 case os::java_thread:
800 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
801 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
802 break;
803 case os::compiler_thread:
804 if (CompilerThreadStackSize > 0) {
805 stack_size = (size_t)(CompilerThreadStackSize * K);
806 break;
807 } // else fall through:
808 // use VMThreadStackSize if CompilerThreadStackSize is not defined
809 case os::vm_thread:
810 case os::pgc_thread:
811 case os::cgc_thread:
812 case os::watcher_thread:
813 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
814 break;
815 }
816 }
817
818 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
819 pthread_attr_setstacksize(&attr, stack_size);
820 } else {
821 // let pthread_create() pick the default value.
822 }
823
824 // glibc guard page
825 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
826
827 ThreadState state;
828
829 {
830 // Serialize thread creation if we are running with fixed stack LinuxThreads
831 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
832 if (lock) {
833 os::Linux::createThread_lock()->lock_without_safepoint_check();
834 }
835
836 pthread_t tid;
837 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
838
839 pthread_attr_destroy(&attr);
840
841 if (ret != 0) {
842 if (PrintMiscellaneous && (Verbose || WizardMode)) {
843 perror("pthread_create()");
844 }
845 // Need to clean up stuff we've allocated so far
846 thread->set_osthread(NULL);
847 delete osthread;
848 if (lock) os::Linux::createThread_lock()->unlock();
849 return false;
850 }
851
852 // Store pthread info into the OSThread
853 osthread->set_pthread_id(tid);
854
855 // Wait until child thread is either initialized or aborted
856 {
857 Monitor* sync_with_child = osthread->startThread_lock();
858 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
859 while ((state = osthread->get_state()) == ALLOCATED) {
860 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
861 }
862 }
863
864 if (lock) {
865 os::Linux::createThread_lock()->unlock();
866 }
867 }
868
869 // Aborted due to thread limit being reached
870 if (state == ZOMBIE) {
871 thread->set_osthread(NULL);
872 delete osthread;
873 return false;
874 }
875
876 // The thread is returned suspended (in state INITIALIZED),
877 // and is started higher up in the call chain
878 assert(state == INITIALIZED, "race condition");
879 return true;
880 }
881
882 /////////////////////////////////////////////////////////////////////////////
883 // attach existing thread
884
885 // bootstrap the main thread
886 bool os::create_main_thread(JavaThread* thread) {
887 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
888 return create_attached_thread(thread);
889 }
890
891 bool os::create_attached_thread(JavaThread* thread) {
892 #ifdef ASSERT
893 thread->verify_not_published();
894 #endif
895
896 // Allocate the OSThread object
897 OSThread* osthread = new OSThread(NULL, NULL);
898
899 if (osthread == NULL) {
900 return false;
901 }
902
903 // Store pthread info into the OSThread
904 osthread->set_thread_id(os::Linux::gettid());
905 osthread->set_pthread_id(::pthread_self());
906
907 // initialize floating point control register
908 os::Linux::init_thread_fpu_state();
909
910 // Initial thread state is RUNNABLE
911 osthread->set_state(RUNNABLE);
912
913 thread->set_osthread(osthread);
914
915 if (UseNUMA) {
916 int lgrp_id = os::numa_get_group_id();
917 if (lgrp_id != -1) {
918 thread->set_lgrp_id(lgrp_id);
919 }
920 }
921
922 if (os::Linux::is_initial_thread()) {
923 // If current thread is initial thread, its stack is mapped on demand,
924 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
925 // the entire stack region to avoid SEGV in stack banging.
926 // It is also useful to get around the heap-stack-gap problem on SuSE
927 // kernel (see 4821821 for details). We first expand stack to the top
928 // of yellow zone, then enable stack yellow zone (order is significant,
929 // enabling yellow zone first will crash JVM on SuSE Linux), so there
930 // is no gap between the last two virtual memory regions.
931
932 JavaThread *jt = (JavaThread *)thread;
933 address addr = jt->stack_yellow_zone_base();
934 assert(addr != NULL, "initialization problem?");
935 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
936
937 osthread->set_expanding_stack();
938 os::Linux::manually_expand_stack(jt, addr);
939 osthread->clear_expanding_stack();
940 }
941
942 // initialize signal mask for this thread
943 // and save the caller's signal mask
944 os::Linux::hotspot_sigmask(thread);
945
946 return true;
947 }
948
949 void os::pd_start_thread(Thread* thread) {
950 OSThread * osthread = thread->osthread();
951 assert(osthread->get_state() != INITIALIZED, "just checking");
952 Monitor* sync_with_child = osthread->startThread_lock();
953 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
954 sync_with_child->notify();
955 }
956
957 // Free Linux resources related to the OSThread
958 void os::free_thread(OSThread* osthread) {
959 assert(osthread != NULL, "osthread not set");
960
961 if (Thread::current()->osthread() == osthread) {
962 // Restore caller's signal mask
963 sigset_t sigmask = osthread->caller_sigmask();
964 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
965 }
966
967 delete osthread;
968 }
969
970 //////////////////////////////////////////////////////////////////////////////
971 // thread local storage
972
973 int os::allocate_thread_local_storage() {
974 pthread_key_t key;
975 int rslt = pthread_key_create(&key, NULL);
976 assert(rslt == 0, "cannot allocate thread local storage");
977 return (int)key;
978 }
979
980 // Note: This is currently not used by VM, as we don't destroy TLS key
981 // on VM exit.
982 void os::free_thread_local_storage(int index) {
983 int rslt = pthread_key_delete((pthread_key_t)index);
984 assert(rslt == 0, "invalid index");
985 }
986
987 void os::thread_local_storage_at_put(int index, void* value) {
988 int rslt = pthread_setspecific((pthread_key_t)index, value);
989 assert(rslt == 0, "pthread_setspecific failed");
990 }
991
992 extern "C" Thread* get_thread() {
993 return ThreadLocalStorage::thread();
994 }
995
996 //////////////////////////////////////////////////////////////////////////////
997 // initial thread
998
999 // Check if current thread is the initial thread, similar to Solaris thr_main.
1000 bool os::Linux::is_initial_thread(void) {
1001 char dummy;
1002 // If called before init complete, thread stack bottom will be null.
1003 // Can be called if fatal error occurs before initialization.
1004 if (initial_thread_stack_bottom() == NULL) return false;
1005 assert(initial_thread_stack_bottom() != NULL &&
1006 initial_thread_stack_size() != 0,
1007 "os::init did not locate initial thread's stack region");
1008 if ((address)&dummy >= initial_thread_stack_bottom() &&
1009 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1010 return true;
1011 else return false;
1012 }
1013
1014 // Find the virtual memory area that contains addr
1015 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1016 FILE *fp = fopen("/proc/self/maps", "r");
1017 if (fp) {
1018 address low, high;
1019 while (!feof(fp)) {
1020 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1021 if (low <= addr && addr < high) {
1022 if (vma_low) *vma_low = low;
1023 if (vma_high) *vma_high = high;
1024 fclose (fp);
1025 return true;
1026 }
1027 }
1028 for (;;) {
1029 int ch = fgetc(fp);
1030 if (ch == EOF || ch == (int)'\n') break;
1031 }
1032 }
1033 fclose(fp);
1034 }
1035 return false;
1036 }
1037
1038 // Locate initial thread stack. This special handling of initial thread stack
1039 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1040 // bogus value for initial thread.
1041 void os::Linux::capture_initial_stack(size_t max_size) {
1042 // stack size is the easy part, get it from RLIMIT_STACK
1043 size_t stack_size;
1044 struct rlimit rlim;
1045 getrlimit(RLIMIT_STACK, &rlim);
1046 stack_size = rlim.rlim_cur;
1047
1048 // 6308388: a bug in ld.so will relocate its own .data section to the
1049 // lower end of primordial stack; reduce ulimit -s value a little bit
1050 // so we won't install guard page on ld.so's data section.
1051 stack_size -= 2 * page_size();
1052
1053 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1054 // 7.1, in both cases we will get 2G in return value.
1055 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1056 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1057 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1058 // in case other parts in glibc still assumes 2M max stack size.
1059 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1060 #ifndef IA64
1061 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1062 #else
1063 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1064 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1065 #endif
1066
1067 // Try to figure out where the stack base (top) is. This is harder.
1068 //
1069 // When an application is started, glibc saves the initial stack pointer in
1070 // a global variable "__libc_stack_end", which is then used by system
1071 // libraries. __libc_stack_end should be pretty close to stack top. The
1072 // variable is available since the very early days. However, because it is
1073 // a private interface, it could disappear in the future.
1074 //
1075 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1076 // to __libc_stack_end, it is very close to stack top, but isn't the real
1077 // stack top. Note that /proc may not exist if VM is running as a chroot
1078 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1079 // /proc/<pid>/stat could change in the future (though unlikely).
1080 //
1081 // We try __libc_stack_end first. If that doesn't work, look for
1082 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1083 // as a hint, which should work well in most cases.
1084
1085 uintptr_t stack_start;
1086
1087 // try __libc_stack_end first
1088 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1089 if (p && *p) {
1090 stack_start = *p;
1091 } else {
1092 // see if we can get the start_stack field from /proc/self/stat
1093 FILE *fp;
1094 int pid;
1095 char state;
1096 int ppid;
1097 int pgrp;
1098 int session;
1099 int nr;
1100 int tpgrp;
1101 unsigned long flags;
1102 unsigned long minflt;
1103 unsigned long cminflt;
1104 unsigned long majflt;
1105 unsigned long cmajflt;
1106 unsigned long utime;
1107 unsigned long stime;
1108 long cutime;
1109 long cstime;
1110 long prio;
1111 long nice;
1112 long junk;
1113 long it_real;
1114 uintptr_t start;
1115 uintptr_t vsize;
1116 uintptr_t rss;
1117 unsigned long rsslim;
1118 uintptr_t scodes;
1119 uintptr_t ecode;
1120 int i;
1121
1122 // Figure what the primordial thread stack base is. Code is inspired
1123 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1124 // followed by command name surrounded by parentheses, state, etc.
1125 char stat[2048];
1126 int statlen;
1127
1128 fp = fopen("/proc/self/stat", "r");
1129 if (fp) {
1130 statlen = fread(stat, 1, 2047, fp);
1131 stat[statlen] = '\0';
1132 fclose(fp);
1133
1134 // Skip pid and the command string. Note that we could be dealing with
1135 // weird command names, e.g. user could decide to rename java launcher
1136 // to "java 1.4.2 :)", then the stat file would look like
1137 // 1234 (java 1.4.2 :)) R ... ...
1138 // We don't really need to know the command string, just find the last
1139 // occurrence of ")" and then start parsing from there. See bug 4726580.
1140 char * s = strrchr(stat, ')');
1141
1142 i = 0;
1143 if (s) {
1144 // Skip blank chars
1145 do s++; while (isspace(*s));
1146
1147 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1148 /* 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 */
1149 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",
1150 &state, /* 3 %c */
1151 &ppid, /* 4 %d */
1152 &pgrp, /* 5 %d */
1153 &session, /* 6 %d */
1154 &nr, /* 7 %d */
1155 &tpgrp, /* 8 %d */
1156 &flags, /* 9 %lu */
1157 &minflt, /* 10 %lu */
1158 &cminflt, /* 11 %lu */
1159 &majflt, /* 12 %lu */
1160 &cmajflt, /* 13 %lu */
1161 &utime, /* 14 %lu */
1162 &stime, /* 15 %lu */
1163 &cutime, /* 16 %ld */
1164 &cstime, /* 17 %ld */
1165 &prio, /* 18 %ld */
1166 &nice, /* 19 %ld */
1167 &junk, /* 20 %ld */
1168 &it_real, /* 21 %ld */
1169 &start, /* 22 %lu */
1170 &vsize, /* 23 %lu */
1171 &rss, /* 24 %ld */
1172 &rsslim, /* 25 %lu */
1173 &scodes, /* 26 %lu */
1174 &ecode, /* 27 %lu */
1175 &stack_start); /* 28 %lu */
1176 }
1177
1178 if (i != 28 - 2) {
1179 assert(false, "Bad conversion from /proc/self/stat");
1180 // product mode - assume we are the initial thread, good luck in the
1181 // embedded case.
1182 warning("Can't detect initial thread stack location - bad conversion");
1183 stack_start = (uintptr_t) &rlim;
1184 }
1185 } else {
1186 // For some reason we can't open /proc/self/stat (for example, running on
1187 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1188 // most cases, so don't abort:
1189 warning("Can't detect initial thread stack location - no /proc/self/stat");
1190 stack_start = (uintptr_t) &rlim;
1191 }
1192 }
1193
1194 // Now we have a pointer (stack_start) very close to the stack top, the
1195 // next thing to do is to figure out the exact location of stack top. We
1196 // can find out the virtual memory area that contains stack_start by
1197 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1198 // and its upper limit is the real stack top. (again, this would fail if
1199 // running inside chroot, because /proc may not exist.)
1200
1201 uintptr_t stack_top;
1202 address low, high;
1203 if (find_vma((address)stack_start, &low, &high)) {
1204 // success, "high" is the true stack top. (ignore "low", because initial
1205 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1206 stack_top = (uintptr_t)high;
1207 } else {
1208 // failed, likely because /proc/self/maps does not exist
1209 warning("Can't detect initial thread stack location - find_vma failed");
1210 // best effort: stack_start is normally within a few pages below the real
1211 // stack top, use it as stack top, and reduce stack size so we won't put
1212 // guard page outside stack.
1213 stack_top = stack_start;
1214 stack_size -= 16 * page_size();
1215 }
1216
1217 // stack_top could be partially down the page so align it
1218 stack_top = align_size_up(stack_top, page_size());
1219
1220 if (max_size && stack_size > max_size) {
1221 _initial_thread_stack_size = max_size;
1222 } else {
1223 _initial_thread_stack_size = stack_size;
1224 }
1225
1226 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1227 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1228 }
1229
1230 ////////////////////////////////////////////////////////////////////////////////
1231 // time support
1232
1233 // Time since start-up in seconds to a fine granularity.
1234 // Used by VMSelfDestructTimer and the MemProfiler.
1235 double os::elapsedTime() {
1236
1237 return (double)(os::elapsed_counter()) * 0.000001;
1238 }
1239
1240 jlong os::elapsed_counter() {
1241 timeval time;
1242 int status = gettimeofday(&time, NULL);
1243 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1244 }
1245
1246 jlong os::elapsed_frequency() {
1247 return (1000 * 1000);
1248 }
1249
1250 jlong os::timeofday() {
1251 timeval time;
1252 int status = gettimeofday(&time, NULL);
1253 assert(status != -1, "linux error");
1254 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1255 }
1256
1257 // Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1258 // _use_global_time is only set if CacheTimeMillis is true
1259 jlong os::javaTimeMillis() {
1260 return (_use_global_time ? read_global_time() : timeofday());
1261 }
1262
1263 #ifndef CLOCK_MONOTONIC
1264 #define CLOCK_MONOTONIC (1)
1265 #endif
1266
1267 void os::Linux::clock_init() {
1268 // we do dlopen's in this particular order due to bug in linux
1269 // dynamical loader (see 6348968) leading to crash on exit
1270 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1271 if (handle == NULL) {
1272 handle = dlopen("librt.so", RTLD_LAZY);
1273 }
1274
1275 if (handle) {
1276 int (*clock_getres_func)(clockid_t, struct timespec*) =
1277 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1278 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1279 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1280 if (clock_getres_func && clock_gettime_func) {
1281 // See if monotonic clock is supported by the kernel. Note that some
1282 // early implementations simply return kernel jiffies (updated every
1283 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1284 // for nano time (though the monotonic property is still nice to have).
1285 // It's fixed in newer kernels, however clock_getres() still returns
1286 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1287 // resolution for now. Hopefully as people move to new kernels, this
1288 // won't be a problem.
1289 struct timespec res;
1290 struct timespec tp;
1291 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1292 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1293 // yes, monotonic clock is supported
1294 _clock_gettime = clock_gettime_func;
1295 } else {
1296 // close librt if there is no monotonic clock
1297 dlclose(handle);
1298 }
1299 }
1300 }
1301 }
1302
1303 #ifndef SYS_clock_getres
1304
1305 #if defined(IA32) || defined(AMD64)
1306 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1307 #else
1308 #error Value of SYS_clock_getres not known on this platform
1309 #endif
1310
1311 #endif
1312
1313 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1314
1315 void os::Linux::fast_thread_clock_init() {
1316 if (!UseLinuxPosixThreadCPUClocks) {
1317 return;
1318 }
1319 clockid_t clockid;
1320 struct timespec tp;
1321 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1322 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1323
1324 // Switch to using fast clocks for thread cpu time if
1325 // the sys_clock_getres() returns 0 error code.
1326 // Note, that some kernels may support the current thread
1327 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1328 // returned by the pthread_getcpuclockid().
1329 // If the fast Posix clocks are supported then the sys_clock_getres()
1330 // must return at least tp.tv_sec == 0 which means a resolution
1331 // better than 1 sec. This is extra check for reliability.
1332
1333 if(pthread_getcpuclockid_func &&
1334 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1335 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1336
1337 _supports_fast_thread_cpu_time = true;
1338 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1339 }
1340 }
1341
1342 jlong os::javaTimeNanos() {
1343 if (Linux::supports_monotonic_clock()) {
1344 struct timespec tp;
1345 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1346 assert(status == 0, "gettime error");
1347 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1348 return result;
1349 } else {
1350 timeval time;
1351 int status = gettimeofday(&time, NULL);
1352 assert(status != -1, "linux error");
1353 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1354 return 1000 * usecs;
1355 }
1356 }
1357
1358 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1359 if (Linux::supports_monotonic_clock()) {
1360 info_ptr->max_value = ALL_64_BITS;
1361
1362 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1363 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1364 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1365 } else {
1366 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1367 info_ptr->max_value = ALL_64_BITS;
1368
1369 // gettimeofday is a real time clock so it skips
1370 info_ptr->may_skip_backward = true;
1371 info_ptr->may_skip_forward = true;
1372 }
1373
1374 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1375 }
1376
1377 // Return the real, user, and system times in seconds from an
1378 // arbitrary fixed point in the past.
1379 bool os::getTimesSecs(double* process_real_time,
1380 double* process_user_time,
1381 double* process_system_time) {
1382 struct tms ticks;
1383 clock_t real_ticks = times(&ticks);
1384
1385 if (real_ticks == (clock_t) (-1)) {
1386 return false;
1387 } else {
1388 double ticks_per_second = (double) clock_tics_per_sec;
1389 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1390 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1391 *process_real_time = ((double) real_ticks) / ticks_per_second;
1392
1393 return true;
1394 }
1395 }
1396
1397
1398 char * os::local_time_string(char *buf, size_t buflen) {
1399 struct tm t;
1400 time_t long_time;
1401 time(&long_time);
1402 localtime_r(&long_time, &t);
1403 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1404 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1405 t.tm_hour, t.tm_min, t.tm_sec);
1406 return buf;
1407 }
1408
1409 ////////////////////////////////////////////////////////////////////////////////
1410 // runtime exit support
1411
1412 // Note: os::shutdown() might be called very early during initialization, or
1413 // called from signal handler. Before adding something to os::shutdown(), make
1414 // sure it is async-safe and can handle partially initialized VM.
1415 void os::shutdown() {
1416
1417 // allow PerfMemory to attempt cleanup of any persistent resources
1418 perfMemory_exit();
1419
1420 // needs to remove object in file system
1421 AttachListener::abort();
1422
1423 // flush buffered output, finish log files
1424 ostream_abort();
1425
1426 // Check for abort hook
1427 abort_hook_t abort_hook = Arguments::abort_hook();
1428 if (abort_hook != NULL) {
1429 abort_hook();
1430 }
1431
1432 }
1433
1434 // Note: os::abort() might be called very early during initialization, or
1435 // called from signal handler. Before adding something to os::abort(), make
1436 // sure it is async-safe and can handle partially initialized VM.
1437 void os::abort(bool dump_core) {
1438 os::shutdown();
1439 if (dump_core) {
1440 #ifndef PRODUCT
1441 fdStream out(defaultStream::output_fd());
1442 out.print_raw("Current thread is ");
1443 char buf[16];
1444 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1445 out.print_raw_cr(buf);
1446 out.print_raw_cr("Dumping core ...");
1447 #endif
1448 ::abort(); // dump core
1449 }
1450
1451 ::exit(1);
1452 }
1453
1454 // Die immediately, no exit hook, no abort hook, no cleanup.
1455 void os::die() {
1456 // _exit() on LinuxThreads only kills current thread
1457 ::abort();
1458 }
1459
1460 // unused on linux for now.
1461 void os::set_error_file(const char *logfile) {}
1462
1463 intx os::current_thread_id() { return (intx)pthread_self(); }
1464 int os::current_process_id() {
1465
1466 // Under the old linux thread library, linux gives each thread
1467 // its own process id. Because of this each thread will return
1468 // a different pid if this method were to return the result
1469 // of getpid(2). Linux provides no api that returns the pid
1470 // of the launcher thread for the vm. This implementation
1471 // returns a unique pid, the pid of the launcher thread
1472 // that starts the vm 'process'.
1473
1474 // Under the NPTL, getpid() returns the same pid as the
1475 // launcher thread rather than a unique pid per thread.
1476 // Use gettid() if you want the old pre NPTL behaviour.
1477
1478 // if you are looking for the result of a call to getpid() that
1479 // returns a unique pid for the calling thread, then look at the
1480 // OSThread::thread_id() method in osThread_linux.hpp file
1481
1482 return (int)(_initial_pid ? _initial_pid : getpid());
1483 }
1484
1485 // DLL functions
1486
1487 const char* os::dll_file_extension() { return ".so"; }
1488
1489 const char* os::get_temp_directory() { return "/tmp/"; }
1490
1491 const char* os::get_current_directory(char *buf, int buflen) {
1492 return getcwd(buf, buflen);
1493 }
1494
1495 // check if addr is inside libjvm[_g].so
1496 bool os::address_is_in_vm(address addr) {
1497 static address libjvm_base_addr;
1498 Dl_info dlinfo;
1499
1500 if (libjvm_base_addr == NULL) {
1501 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1502 libjvm_base_addr = (address)dlinfo.dli_fbase;
1503 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1504 }
1505
1506 if (dladdr((void *)addr, &dlinfo)) {
1507 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1508 }
1509
1510 return false;
1511 }
1512
1513 bool os::dll_address_to_function_name(address addr, char *buf,
1514 int buflen, int *offset) {
1515 Dl_info dlinfo;
1516
1517 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1518 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1519 if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1520 return true;
1521 } else {
1522 if (buf) buf[0] = '\0';
1523 if (offset) *offset = -1;
1524 return false;
1525 }
1526 }
1527
1528 struct _address_to_library_name {
1529 address addr; // input : memory address
1530 size_t buflen; // size of fname
1531 char* fname; // output: library name
1532 address base; // library base addr
1533 };
1534
1535 static int address_to_library_name_callback(struct dl_phdr_info *info,
1536 size_t size, void *data) {
1537 int i;
1538 bool found = false;
1539 address libbase = NULL;
1540 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1541
1542 // iterate through all loadable segments
1543 for (i = 0; i < info->dlpi_phnum; i++) {
1544 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1545 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1546 // base address of a library is the lowest address of its loaded
1547 // segments.
1548 if (libbase == NULL || libbase > segbase) {
1549 libbase = segbase;
1550 }
1551 // see if 'addr' is within current segment
1552 if (segbase <= d->addr &&
1553 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1554 found = true;
1555 }
1556 }
1557 }
1558
1559 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1560 // so dll_address_to_library_name() can fall through to use dladdr() which
1561 // can figure out executable name from argv[0].
1562 if (found && info->dlpi_name && info->dlpi_name[0]) {
1563 d->base = libbase;
1564 if (d->fname) {
1565 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1566 }
1567 return 1;
1568 }
1569 return 0;
1570 }
1571
1572 bool os::dll_address_to_library_name(address addr, char* buf,
1573 int buflen, int* offset) {
1574 Dl_info dlinfo;
1575 struct _address_to_library_name data;
1576
1577 // There is a bug in old glibc dladdr() implementation that it could resolve
1578 // to wrong library name if the .so file has a base address != NULL. Here
1579 // we iterate through the program headers of all loaded libraries to find
1580 // out which library 'addr' really belongs to. This workaround can be
1581 // removed once the minimum requirement for glibc is moved to 2.3.x.
1582 data.addr = addr;
1583 data.fname = buf;
1584 data.buflen = buflen;
1585 data.base = NULL;
1586 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1587
1588 if (rslt) {
1589 // buf already contains library name
1590 if (offset) *offset = addr - data.base;
1591 return true;
1592 } else if (dladdr((void*)addr, &dlinfo)){
1593 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1594 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1595 return true;
1596 } else {
1597 if (buf) buf[0] = '\0';
1598 if (offset) *offset = -1;
1599 return false;
1600 }
1601 }
1602
1603 // Loads .dll/.so and
1604 // in case of error it checks if .dll/.so was built for the
1605 // same architecture as Hotspot is running on
1606
1607 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1608 {
1609 void * result= ::dlopen(filename, RTLD_LAZY);
1610 if (result != NULL) {
1611 // Successful loading
1612 return result;
1613 }
1614
1615 Elf32_Ehdr elf_head;
1616
1617 // Read system error message into ebuf
1618 // It may or may not be overwritten below
1619 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1620 ebuf[ebuflen-1]='\0';
1621 int diag_msg_max_length=ebuflen-strlen(ebuf);
1622 char* diag_msg_buf=ebuf+strlen(ebuf);
1623
1624 if (diag_msg_max_length==0) {
1625 // No more space in ebuf for additional diagnostics message
1626 return NULL;
1627 }
1628
1629
1630 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1631
1632 if (file_descriptor < 0) {
1633 // Can't open library, report dlerror() message
1634 return NULL;
1635 }
1636
1637 bool failed_to_read_elf_head=
1638 (sizeof(elf_head)!=
1639 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1640
1641 ::close(file_descriptor);
1642 if (failed_to_read_elf_head) {
1643 // file i/o error - report dlerror() msg
1644 return NULL;
1645 }
1646
1647 typedef struct {
1648 Elf32_Half code; // Actual value as defined in elf.h
1649 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1650 char elf_class; // 32 or 64 bit
1651 char endianess; // MSB or LSB
1652 char* name; // String representation
1653 } arch_t;
1654
1655 #ifndef EM_486
1656 #define EM_486 6 /* Intel 80486 */
1657 #endif
1658
1659 static const arch_t arch_array[]={
1660 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1661 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1662 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1663 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1664 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1665 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1666 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1667 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1668 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}
1669 };
1670
1671 #if (defined IA32)
1672 static Elf32_Half running_arch_code=EM_386;
1673 #elif (defined AMD64)
1674 static Elf32_Half running_arch_code=EM_X86_64;
1675 #elif (defined IA64)
1676 static Elf32_Half running_arch_code=EM_IA_64;
1677 #elif (defined __sparc) && (defined _LP64)
1678 static Elf32_Half running_arch_code=EM_SPARCV9;
1679 #elif (defined __sparc) && (!defined _LP64)
1680 static Elf32_Half running_arch_code=EM_SPARC;
1681 #elif (defined __powerpc64__)
1682 static Elf32_Half running_arch_code=EM_PPC64;
1683 #elif (defined __powerpc__)
1684 static Elf32_Half running_arch_code=EM_PPC;
1685 #else
1686 #error Method os::dll_load requires that one of following is defined:\
1687 IA32, AMD64, IA64, __sparc, __powerpc__
1688 #endif
1689
1690 // Identify compatability class for VM's architecture and library's architecture
1691 // Obtain string descriptions for architectures
1692
1693 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1694 int running_arch_index=-1;
1695
1696 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1697 if (running_arch_code == arch_array[i].code) {
1698 running_arch_index = i;
1699 }
1700 if (lib_arch.code == arch_array[i].code) {
1701 lib_arch.compat_class = arch_array[i].compat_class;
1702 lib_arch.name = arch_array[i].name;
1703 }
1704 }
1705
1706 assert(running_arch_index != -1,
1707 "Didn't find running architecture code (running_arch_code) in arch_array");
1708 if (running_arch_index == -1) {
1709 // Even though running architecture detection failed
1710 // we may still continue with reporting dlerror() message
1711 return NULL;
1712 }
1713
1714 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1715 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1716 return NULL;
1717 }
1718
1719 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1720 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1721 return NULL;
1722 }
1723
1724 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1725 if ( lib_arch.name!=NULL ) {
1726 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1727 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1728 lib_arch.name, arch_array[running_arch_index].name);
1729 } else {
1730 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1731 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1732 lib_arch.code,
1733 arch_array[running_arch_index].name);
1734 }
1735 }
1736
1737 return NULL;
1738 }
1739
1740
1741
1742
1743 bool _print_ascii_file(const char* filename, outputStream* st) {
1744 int fd = open(filename, O_RDONLY);
1745 if (fd == -1) {
1746 return false;
1747 }
1748
1749 char buf[32];
1750 int bytes;
1751 while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1752 st->print_raw(buf, bytes);
1753 }
1754
1755 close(fd);
1756
1757 return true;
1758 }
1759
1760 void os::print_dll_info(outputStream *st) {
1761 st->print_cr("Dynamic libraries:");
1762
1763 char fname[32];
1764 pid_t pid = os::Linux::gettid();
1765
1766 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1767
1768 if (!_print_ascii_file(fname, st)) {
1769 st->print("Can not get library information for pid = %d\n", pid);
1770 }
1771 }
1772
1773
1774 void os::print_os_info(outputStream* st) {
1775 st->print("OS:");
1776
1777 // Try to identify popular distros.
1778 // Most Linux distributions have /etc/XXX-release file, which contains
1779 // the OS version string. Some have more than one /etc/XXX-release file
1780 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1781 // so the order is important.
1782 if (!_print_ascii_file("/etc/mandrake-release", st) &&
1783 !_print_ascii_file("/etc/sun-release", st) &&
1784 !_print_ascii_file("/etc/redhat-release", st) &&
1785 !_print_ascii_file("/etc/SuSE-release", st) &&
1786 !_print_ascii_file("/etc/turbolinux-release", st) &&
1787 !_print_ascii_file("/etc/gentoo-release", st) &&
1788 !_print_ascii_file("/etc/debian_version", st)) {
1789 st->print("Linux");
1790 }
1791 st->cr();
1792
1793 // kernel
1794 st->print("uname:");
1795 struct utsname name;
1796 uname(&name);
1797 st->print(name.sysname); st->print(" ");
1798 st->print(name.release); st->print(" ");
1799 st->print(name.version); st->print(" ");
1800 st->print(name.machine);
1801 st->cr();
1802
1803 // Print warning if unsafe chroot environment detected
1804 if (unsafe_chroot_detected) {
1805 st->print("WARNING!! ");
1806 st->print_cr(unstable_chroot_error);
1807 }
1808
1809 // libc, pthread
1810 st->print("libc:");
1811 st->print(os::Linux::glibc_version()); st->print(" ");
1812 st->print(os::Linux::libpthread_version()); st->print(" ");
1813 if (os::Linux::is_LinuxThreads()) {
1814 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1815 }
1816 st->cr();
1817
1818 // rlimit
1819 st->print("rlimit:");
1820 struct rlimit rlim;
1821
1822 st->print(" STACK ");
1823 getrlimit(RLIMIT_STACK, &rlim);
1824 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1825 else st->print("%uk", rlim.rlim_cur >> 10);
1826
1827 st->print(", CORE ");
1828 getrlimit(RLIMIT_CORE, &rlim);
1829 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1830 else st->print("%uk", rlim.rlim_cur >> 10);
1831
1832 st->print(", NPROC ");
1833 getrlimit(RLIMIT_NPROC, &rlim);
1834 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1835 else st->print("%d", rlim.rlim_cur);
1836
1837 st->print(", NOFILE ");
1838 getrlimit(RLIMIT_NOFILE, &rlim);
1839 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1840 else st->print("%d", rlim.rlim_cur);
1841
1842 st->print(", AS ");
1843 getrlimit(RLIMIT_AS, &rlim);
1844 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1845 else st->print("%uk", rlim.rlim_cur >> 10);
1846 st->cr();
1847
1848 // load average
1849 st->print("load average:");
1850 double loadavg[3];
1851 os::loadavg(loadavg, 3);
1852 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1853 st->cr();
1854 }
1855
1856 void os::print_memory_info(outputStream* st) {
1857
1858 st->print("Memory:");
1859 st->print(" %dk page", os::vm_page_size()>>10);
1860
1861 // values in struct sysinfo are "unsigned long"
1862 struct sysinfo si;
1863 sysinfo(&si);
1864
1865 st->print(", physical " UINT64_FORMAT "k",
1866 os::physical_memory() >> 10);
1867 st->print("(" UINT64_FORMAT "k free)",
1868 os::available_memory() >> 10);
1869 st->print(", swap " UINT64_FORMAT "k",
1870 ((jlong)si.totalswap * si.mem_unit) >> 10);
1871 st->print("(" UINT64_FORMAT "k free)",
1872 ((jlong)si.freeswap * si.mem_unit) >> 10);
1873 st->cr();
1874 }
1875
1876 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
1877 // but they're the same for all the linux arch that we support
1878 // and they're the same for solaris but there's no common place to put this.
1879 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
1880 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
1881 "ILL_COPROC", "ILL_BADSTK" };
1882
1883 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
1884 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
1885 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
1886
1887 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
1888
1889 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
1890
1891 void os::print_siginfo(outputStream* st, void* siginfo) {
1892 st->print("siginfo:");
1893
1894 const int buflen = 100;
1895 char buf[buflen];
1896 siginfo_t *si = (siginfo_t*)siginfo;
1897 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
1898 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
1899 st->print("si_errno=%s", buf);
1900 } else {
1901 st->print("si_errno=%d", si->si_errno);
1902 }
1903 const int c = si->si_code;
1904 assert(c > 0, "unexpected si_code");
1905 switch (si->si_signo) {
1906 case SIGILL:
1907 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
1908 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1909 break;
1910 case SIGFPE:
1911 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
1912 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1913 break;
1914 case SIGSEGV:
1915 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
1916 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1917 break;
1918 case SIGBUS:
1919 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
1920 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1921 break;
1922 default:
1923 st->print(", si_code=%d", si->si_code);
1924 // no si_addr
1925 }
1926
1927 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
1928 UseSharedSpaces) {
1929 FileMapInfo* mapinfo = FileMapInfo::current_info();
1930 if (mapinfo->is_in_shared_space(si->si_addr)) {
1931 st->print("\n\nError accessing class data sharing archive." \
1932 " Mapped file inaccessible during execution, " \
1933 " possible disk/network problem.");
1934 }
1935 }
1936 st->cr();
1937 }
1938
1939
1940 static void print_signal_handler(outputStream* st, int sig,
1941 char* buf, size_t buflen);
1942
1943 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1944 st->print_cr("Signal Handlers:");
1945 print_signal_handler(st, SIGSEGV, buf, buflen);
1946 print_signal_handler(st, SIGBUS , buf, buflen);
1947 print_signal_handler(st, SIGFPE , buf, buflen);
1948 print_signal_handler(st, SIGPIPE, buf, buflen);
1949 print_signal_handler(st, SIGXFSZ, buf, buflen);
1950 print_signal_handler(st, SIGILL , buf, buflen);
1951 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
1952 print_signal_handler(st, SR_signum, buf, buflen);
1953 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
1954 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
1955 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
1956 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
1957 }
1958
1959 static char saved_jvm_path[MAXPATHLEN] = {0};
1960
1961 // Find the full path to the current module, libjvm.so or libjvm_g.so
1962 void os::jvm_path(char *buf, jint len) {
1963 // Error checking.
1964 if (len < MAXPATHLEN) {
1965 assert(false, "must use a large-enough buffer");
1966 buf[0] = '\0';
1967 return;
1968 }
1969 // Lazy resolve the path to current module.
1970 if (saved_jvm_path[0] != 0) {
1971 strcpy(buf, saved_jvm_path);
1972 return;
1973 }
1974
1975 char dli_fname[MAXPATHLEN];
1976 bool ret = dll_address_to_library_name(
1977 CAST_FROM_FN_PTR(address, os::jvm_path),
1978 dli_fname, sizeof(dli_fname), NULL);
1979 assert(ret != 0, "cannot locate libjvm");
1980 realpath(dli_fname, buf);
1981
1982 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
1983 // Support for the gamma launcher. Typical value for buf is
1984 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
1985 // the right place in the string, then assume we are installed in a JDK and
1986 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
1987 // up the path so it looks like libjvm.so is installed there (append a
1988 // fake suffix hotspot/libjvm.so).
1989 const char *p = buf + strlen(buf) - 1;
1990 for (int count = 0; p > buf && count < 5; ++count) {
1991 for (--p; p > buf && *p != '/'; --p)
1992 /* empty */ ;
1993 }
1994
1995 if (strncmp(p, "/jre/lib/", 9) != 0) {
1996 // Look for JAVA_HOME in the environment.
1997 char* java_home_var = ::getenv("JAVA_HOME");
1998 if (java_home_var != NULL && java_home_var[0] != 0) {
1999 // Check the current module name "libjvm.so" or "libjvm_g.so".
2000 p = strrchr(buf, '/');
2001 assert(strstr(p, "/libjvm") == p, "invalid library name");
2002 p = strstr(p, "_g") ? "_g" : "";
2003
2004 realpath(java_home_var, buf);
2005 sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
2006 if (0 == access(buf, F_OK)) {
2007 // Use current module name "libjvm[_g].so" instead of
2008 // "libjvm"debug_only("_g")".so" since for fastdebug version
2009 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2010 // It is used when we are choosing the HPI library's name
2011 // "libhpi[_g].so" in hpi::initialize_get_interface().
2012 sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
2013 } else {
2014 // Go back to path of .so
2015 realpath(dli_fname, buf);
2016 }
2017 }
2018 }
2019 }
2020
2021 strcpy(saved_jvm_path, buf);
2022 }
2023
2024 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2025 // no prefix required, not even "_"
2026 }
2027
2028 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2029 // no suffix required
2030 }
2031
2032 ////////////////////////////////////////////////////////////////////////////////
2033 // sun.misc.Signal support
2034
2035 static volatile jint sigint_count = 0;
2036
2037 static void
2038 UserHandler(int sig, void *siginfo, void *context) {
2039 // 4511530 - sem_post is serialized and handled by the manager thread. When
2040 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2041 // don't want to flood the manager thread with sem_post requests.
2042 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2043 return;
2044
2045 // Ctrl-C is pressed during error reporting, likely because the error
2046 // handler fails to abort. Let VM die immediately.
2047 if (sig == SIGINT && is_error_reported()) {
2048 os::die();
2049 }
2050
2051 os::signal_notify(sig);
2052 }
2053
2054 void* os::user_handler() {
2055 return CAST_FROM_FN_PTR(void*, UserHandler);
2056 }
2057
2058 extern "C" {
2059 typedef void (*sa_handler_t)(int);
2060 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2061 }
2062
2063 void* os::signal(int signal_number, void* handler) {
2064 struct sigaction sigAct, oldSigAct;
2065
2066 sigfillset(&(sigAct.sa_mask));
2067 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2068 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2069
2070 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2071 // -1 means registration failed
2072 return (void *)-1;
2073 }
2074
2075 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2076 }
2077
2078 void os::signal_raise(int signal_number) {
2079 ::raise(signal_number);
2080 }
2081
2082 /*
2083 * The following code is moved from os.cpp for making this
2084 * code platform specific, which it is by its very nature.
2085 */
2086
2087 // Will be modified when max signal is changed to be dynamic
2088 int os::sigexitnum_pd() {
2089 return NSIG;
2090 }
2091
2092 // a counter for each possible signal value
2093 static volatile jint pending_signals[NSIG+1] = { 0 };
2094
2095 // Linux(POSIX) specific hand shaking semaphore.
2096 static sem_t sig_sem;
2097
2098 void os::signal_init_pd() {
2099 // Initialize signal structures
2100 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2101
2102 // Initialize signal semaphore
2103 ::sem_init(&sig_sem, 0, 0);
2104 }
2105
2106 void os::signal_notify(int sig) {
2107 Atomic::inc(&pending_signals[sig]);
2108 ::sem_post(&sig_sem);
2109 }
2110
2111 static int check_pending_signals(bool wait) {
2112 Atomic::store(0, &sigint_count);
2113 for (;;) {
2114 for (int i = 0; i < NSIG + 1; i++) {
2115 jint n = pending_signals[i];
2116 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2117 return i;
2118 }
2119 }
2120 if (!wait) {
2121 return -1;
2122 }
2123 JavaThread *thread = JavaThread::current();
2124 ThreadBlockInVM tbivm(thread);
2125
2126 bool threadIsSuspended;
2127 do {
2128 thread->set_suspend_equivalent();
2129 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2130 ::sem_wait(&sig_sem);
2131
2132 // were we externally suspended while we were waiting?
2133 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2134 if (threadIsSuspended) {
2135 //
2136 // The semaphore has been incremented, but while we were waiting
2137 // another thread suspended us. We don't want to continue running
2138 // while suspended because that would surprise the thread that
2139 // suspended us.
2140 //
2141 ::sem_post(&sig_sem);
2142
2143 thread->java_suspend_self();
2144 }
2145 } while (threadIsSuspended);
2146 }
2147 }
2148
2149 int os::signal_lookup() {
2150 return check_pending_signals(false);
2151 }
2152
2153 int os::signal_wait() {
2154 return check_pending_signals(true);
2155 }
2156
2157 ////////////////////////////////////////////////////////////////////////////////
2158 // Virtual Memory
2159
2160 int os::vm_page_size() {
2161 // Seems redundant as all get out
2162 assert(os::Linux::page_size() != -1, "must call os::init");
2163 return os::Linux::page_size();
2164 }
2165
2166 // Solaris allocates memory by pages.
2167 int os::vm_allocation_granularity() {
2168 assert(os::Linux::page_size() != -1, "must call os::init");
2169 return os::Linux::page_size();
2170 }
2171
2172 // Rationale behind this function:
2173 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2174 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2175 // samples for JITted code. Here we create private executable mapping over the code cache
2176 // and then we can use standard (well, almost, as mapping can change) way to provide
2177 // info for the reporting script by storing timestamp and location of symbol
2178 void linux_wrap_code(char* base, size_t size) {
2179 static volatile jint cnt = 0;
2180
2181 if (!UseOprofile) {
2182 return;
2183 }
2184
2185 char buf[40];
2186 int num = Atomic::add(1, &cnt);
2187
2188 sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num);
2189 unlink(buf);
2190
2191 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2192
2193 if (fd != -1) {
2194 off_t rv = lseek(fd, size-2, SEEK_SET);
2195 if (rv != (off_t)-1) {
2196 if (write(fd, "", 1) == 1) {
2197 mmap(base, size,
2198 PROT_READ|PROT_WRITE|PROT_EXEC,
2199 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2200 }
2201 }
2202 close(fd);
2203 unlink(buf);
2204 }
2205 }
2206
2207 // NOTE: Linux kernel does not really reserve the pages for us.
2208 // All it does is to check if there are enough free pages
2209 // left at the time of mmap(). This could be a potential
2210 // problem.
2211 bool os::commit_memory(char* addr, size_t size) {
2212 uintptr_t res = (uintptr_t) ::mmap(addr, size,
2213 PROT_READ|PROT_WRITE|PROT_EXEC,
2214 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2215 return res != (uintptr_t) MAP_FAILED;
2216 }
2217
2218 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint) {
2219 return commit_memory(addr, size);
2220 }
2221
2222 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2223 void os::free_memory(char *addr, size_t bytes) { }
2224 void os::numa_make_global(char *addr, size_t bytes) { }
2225 void os::numa_make_local(char *addr, size_t bytes) { }
2226 bool os::numa_topology_changed() { return false; }
2227 size_t os::numa_get_groups_num() { return 1; }
2228 int os::numa_get_group_id() { return 0; }
2229 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2230 if (size > 0) {
2231 ids[0] = 0;
2232 return 1;
2233 }
2234 return 0;
2235 }
2236
2237 bool os::get_page_info(char *start, page_info* info) {
2238 return false;
2239 }
2240
2241 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2242 return end;
2243 }
2244
2245 bool os::uncommit_memory(char* addr, size_t size) {
2246 return ::mmap(addr, size,
2247 PROT_READ|PROT_WRITE|PROT_EXEC,
2248 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0)
2249 != MAP_FAILED;
2250 }
2251
2252 static address _highest_vm_reserved_address = NULL;
2253
2254 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2255 // at 'requested_addr'. If there are existing memory mappings at the same
2256 // location, however, they will be overwritten. If 'fixed' is false,
2257 // 'requested_addr' is only treated as a hint, the return value may or
2258 // may not start from the requested address. Unlike Linux mmap(), this
2259 // function returns NULL to indicate failure.
2260 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2261 char * addr;
2262 int flags;
2263
2264 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2265 if (fixed) {
2266 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2267 flags |= MAP_FIXED;
2268 }
2269
2270 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE|PROT_EXEC,
2271 flags, -1, 0);
2272
2273 if (addr != MAP_FAILED) {
2274 // anon_mmap() should only get called during VM initialization,
2275 // don't need lock (actually we can skip locking even it can be called
2276 // from multiple threads, because _highest_vm_reserved_address is just a
2277 // hint about the upper limit of non-stack memory regions.)
2278 if ((address)addr + bytes > _highest_vm_reserved_address) {
2279 _highest_vm_reserved_address = (address)addr + bytes;
2280 }
2281 }
2282
2283 return addr == MAP_FAILED ? NULL : addr;
2284 }
2285
2286 // Don't update _highest_vm_reserved_address, because there might be memory
2287 // regions above addr + size. If so, releasing a memory region only creates
2288 // a hole in the address space, it doesn't help prevent heap-stack collision.
2289 //
2290 static int anon_munmap(char * addr, size_t size) {
2291 return ::munmap(addr, size) == 0;
2292 }
2293
2294 char* os::reserve_memory(size_t bytes, char* requested_addr,
2295 size_t alignment_hint) {
2296 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2297 }
2298
2299 bool os::release_memory(char* addr, size_t size) {
2300 return anon_munmap(addr, size);
2301 }
2302
2303 static address highest_vm_reserved_address() {
2304 return _highest_vm_reserved_address;
2305 }
2306
2307 static bool linux_mprotect(char* addr, size_t size, int prot) {
2308 // Linux wants the mprotect address argument to be page aligned.
2309 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2310
2311 // According to SUSv3, mprotect() should only be used with mappings
2312 // established by mmap(), and mmap() always maps whole pages. Unaligned
2313 // 'addr' likely indicates problem in the VM (e.g. trying to change
2314 // protection of malloc'ed or statically allocated memory). Check the
2315 // caller if you hit this assert.
2316 assert(addr == bottom, "sanity check");
2317
2318 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2319 return ::mprotect(bottom, size, prot) == 0;
2320 }
2321
2322 bool os::protect_memory(char* addr, size_t size) {
2323 return linux_mprotect(addr, size, PROT_READ);
2324 }
2325
2326 bool os::guard_memory(char* addr, size_t size) {
2327 return linux_mprotect(addr, size, PROT_NONE);
2328 }
2329
2330 bool os::unguard_memory(char* addr, size_t size) {
2331 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE|PROT_EXEC);
2332 }
2333
2334 // Large page support
2335
2336 static size_t _large_page_size = 0;
2337
2338 bool os::large_page_init() {
2339 if (!UseLargePages) return false;
2340
2341 if (LargePageSizeInBytes) {
2342 _large_page_size = LargePageSizeInBytes;
2343 } else {
2344 // large_page_size on Linux is used to round up heap size. x86 uses either
2345 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2346 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2347 // page as large as 256M.
2348 //
2349 // Here we try to figure out page size by parsing /proc/meminfo and looking
2350 // for a line with the following format:
2351 // Hugepagesize: 2048 kB
2352 //
2353 // If we can't determine the value (e.g. /proc is not mounted, or the text
2354 // format has been changed), we'll use the largest page size supported by
2355 // the processor.
2356
2357 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M);
2358
2359 FILE *fp = fopen("/proc/meminfo", "r");
2360 if (fp) {
2361 while (!feof(fp)) {
2362 int x = 0;
2363 char buf[16];
2364 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2365 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2366 _large_page_size = x * K;
2367 break;
2368 }
2369 } else {
2370 // skip to next line
2371 for (;;) {
2372 int ch = fgetc(fp);
2373 if (ch == EOF || ch == (int)'\n') break;
2374 }
2375 }
2376 }
2377 fclose(fp);
2378 }
2379 }
2380
2381 const size_t default_page_size = (size_t)Linux::page_size();
2382 if (_large_page_size > default_page_size) {
2383 _page_sizes[0] = _large_page_size;
2384 _page_sizes[1] = default_page_size;
2385 _page_sizes[2] = 0;
2386 }
2387
2388 // Large page support is available on 2.6 or newer kernel, some vendors
2389 // (e.g. Redhat) have backported it to their 2.4 based distributions.
2390 // We optimistically assume the support is available. If later it turns out
2391 // not true, VM will automatically switch to use regular page size.
2392 return true;
2393 }
2394
2395 #ifndef SHM_HUGETLB
2396 #define SHM_HUGETLB 04000
2397 #endif
2398
2399 char* os::reserve_memory_special(size_t bytes) {
2400 assert(UseLargePages, "only for large pages");
2401
2402 key_t key = IPC_PRIVATE;
2403 char *addr;
2404
2405 bool warn_on_failure = UseLargePages &&
2406 (!FLAG_IS_DEFAULT(UseLargePages) ||
2407 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2408 );
2409 char msg[128];
2410
2411 // Create a large shared memory region to attach to based on size.
2412 // Currently, size is the total size of the heap
2413 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2414 if (shmid == -1) {
2415 // Possible reasons for shmget failure:
2416 // 1. shmmax is too small for Java heap.
2417 // > check shmmax value: cat /proc/sys/kernel/shmmax
2418 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2419 // 2. not enough large page memory.
2420 // > check available large pages: cat /proc/meminfo
2421 // > increase amount of large pages:
2422 // echo new_value > /proc/sys/vm/nr_hugepages
2423 // Note 1: different Linux may use different name for this property,
2424 // e.g. on Redhat AS-3 it is "hugetlb_pool".
2425 // Note 2: it's possible there's enough physical memory available but
2426 // they are so fragmented after a long run that they can't
2427 // coalesce into large pages. Try to reserve large pages when
2428 // the system is still "fresh".
2429 if (warn_on_failure) {
2430 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2431 warning(msg);
2432 }
2433 return NULL;
2434 }
2435
2436 // attach to the region
2437 addr = (char*)shmat(shmid, NULL, 0);
2438 int err = errno;
2439
2440 // Remove shmid. If shmat() is successful, the actual shared memory segment
2441 // will be deleted when it's detached by shmdt() or when the process
2442 // terminates. If shmat() is not successful this will remove the shared
2443 // segment immediately.
2444 shmctl(shmid, IPC_RMID, NULL);
2445
2446 if ((intptr_t)addr == -1) {
2447 if (warn_on_failure) {
2448 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2449 warning(msg);
2450 }
2451 return NULL;
2452 }
2453
2454 return addr;
2455 }
2456
2457 bool os::release_memory_special(char* base, size_t bytes) {
2458 // detaching the SHM segment will also delete it, see reserve_memory_special()
2459 int rslt = shmdt(base);
2460 return rslt == 0;
2461 }
2462
2463 size_t os::large_page_size() {
2464 return _large_page_size;
2465 }
2466
2467 // Linux does not support anonymous mmap with large page memory. The only way
2468 // to reserve large page memory without file backing is through SysV shared
2469 // memory API. The entire memory region is committed and pinned upfront.
2470 // Hopefully this will change in the future...
2471 bool os::can_commit_large_page_memory() {
2472 return false;
2473 }
2474
2475 // Reserve memory at an arbitrary address, only if that area is
2476 // available (and not reserved for something else).
2477
2478 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2479 const int max_tries = 10;
2480 char* base[max_tries];
2481 size_t size[max_tries];
2482 const size_t gap = 0x000000;
2483
2484 // Assert only that the size is a multiple of the page size, since
2485 // that's all that mmap requires, and since that's all we really know
2486 // about at this low abstraction level. If we need higher alignment,
2487 // we can either pass an alignment to this method or verify alignment
2488 // in one of the methods further up the call chain. See bug 5044738.
2489 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2490
2491 // Repeatedly allocate blocks until the block is allocated at the
2492 // right spot. Give up after max_tries. Note that reserve_memory() will
2493 // automatically update _highest_vm_reserved_address if the call is
2494 // successful. The variable tracks the highest memory address every reserved
2495 // by JVM. It is used to detect heap-stack collision if running with
2496 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2497 // space than needed, it could confuse the collision detecting code. To
2498 // solve the problem, save current _highest_vm_reserved_address and
2499 // calculate the correct value before return.
2500 address old_highest = _highest_vm_reserved_address;
2501
2502 // Linux mmap allows caller to pass an address as hint; give it a try first,
2503 // if kernel honors the hint then we can return immediately.
2504 char * addr = anon_mmap(requested_addr, bytes, false);
2505 if (addr == requested_addr) {
2506 return requested_addr;
2507 }
2508
2509 if (addr != NULL) {
2510 // mmap() is successful but it fails to reserve at the requested address
2511 anon_munmap(addr, bytes);
2512 }
2513
2514 int i;
2515 for (i = 0; i < max_tries; ++i) {
2516 base[i] = reserve_memory(bytes);
2517
2518 if (base[i] != NULL) {
2519 // Is this the block we wanted?
2520 if (base[i] == requested_addr) {
2521 size[i] = bytes;
2522 break;
2523 }
2524
2525 // Does this overlap the block we wanted? Give back the overlapped
2526 // parts and try again.
2527
2528 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2529 if (top_overlap >= 0 && top_overlap < bytes) {
2530 unmap_memory(base[i], top_overlap);
2531 base[i] += top_overlap;
2532 size[i] = bytes - top_overlap;
2533 } else {
2534 size_t bottom_overlap = base[i] + bytes - requested_addr;
2535 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2536 unmap_memory(requested_addr, bottom_overlap);
2537 size[i] = bytes - bottom_overlap;
2538 } else {
2539 size[i] = bytes;
2540 }
2541 }
2542 }
2543 }
2544
2545 // Give back the unused reserved pieces.
2546
2547 for (int j = 0; j < i; ++j) {
2548 if (base[j] != NULL) {
2549 unmap_memory(base[j], size[j]);
2550 }
2551 }
2552
2553 if (i < max_tries) {
2554 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2555 return requested_addr;
2556 } else {
2557 _highest_vm_reserved_address = old_highest;
2558 return NULL;
2559 }
2560 }
2561
2562 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2563 return ::read(fd, buf, nBytes);
2564 }
2565
2566 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2567 // Solaris uses poll(), linux uses park().
2568 // Poll() is likely a better choice, assuming that Thread.interrupt()
2569 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2570 // SIGSEGV, see 4355769.
2571
2572 const int NANOSECS_PER_MILLISECS = 1000000;
2573
2574 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2575 assert(thread == Thread::current(), "thread consistency check");
2576
2577 ParkEvent * const slp = thread->_SleepEvent ;
2578 slp->reset() ;
2579 OrderAccess::fence() ;
2580
2581 if (interruptible) {
2582 jlong prevtime = javaTimeNanos();
2583
2584 for (;;) {
2585 if (os::is_interrupted(thread, true)) {
2586 return OS_INTRPT;
2587 }
2588
2589 jlong newtime = javaTimeNanos();
2590
2591 if (newtime - prevtime < 0) {
2592 // time moving backwards, should only happen if no monotonic clock
2593 // not a guarantee() because JVM should not abort on kernel/glibc bugs
2594 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2595 } else {
2596 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2597 }
2598
2599 if(millis <= 0) {
2600 return OS_OK;
2601 }
2602
2603 prevtime = newtime;
2604
2605 {
2606 assert(thread->is_Java_thread(), "sanity check");
2607 JavaThread *jt = (JavaThread *) thread;
2608 ThreadBlockInVM tbivm(jt);
2609 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
2610
2611 jt->set_suspend_equivalent();
2612 // cleared by handle_special_suspend_equivalent_condition() or
2613 // java_suspend_self() via check_and_wait_while_suspended()
2614
2615 slp->park(millis);
2616
2617 // were we externally suspended while we were waiting?
2618 jt->check_and_wait_while_suspended();
2619 }
2620 }
2621 } else {
2622 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2623 jlong prevtime = javaTimeNanos();
2624
2625 for (;;) {
2626 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
2627 // the 1st iteration ...
2628 jlong newtime = javaTimeNanos();
2629
2630 if (newtime - prevtime < 0) {
2631 // time moving backwards, should only happen if no monotonic clock
2632 // not a guarantee() because JVM should not abort on kernel/glibc bugs
2633 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2634 } else {
2635 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2636 }
2637
2638 if(millis <= 0) break ;
2639
2640 prevtime = newtime;
2641 slp->park(millis);
2642 }
2643 return OS_OK ;
2644 }
2645 }
2646
2647 int os::naked_sleep() {
2648 // %% make the sleep time an integer flag. for now use 1 millisec.
2649 return os::sleep(Thread::current(), 1, false);
2650 }
2651
2652 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2653 void os::infinite_sleep() {
2654 while (true) { // sleep forever ...
2655 ::sleep(100); // ... 100 seconds at a time
2656 }
2657 }
2658
2659 // Used to convert frequent JVM_Yield() to nops
2660 bool os::dont_yield() {
2661 return DontYieldALot;
2662 }
2663
2664 void os::yield() {
2665 sched_yield();
2666 }
2667
2668 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
2669
2670 void os::yield_all(int attempts) {
2671 // Yields to all threads, including threads with lower priorities
2672 // Threads on Linux are all with same priority. The Solaris style
2673 // os::yield_all() with nanosleep(1ms) is not necessary.
2674 sched_yield();
2675 }
2676
2677 // Called from the tight loops to possibly influence time-sharing heuristics
2678 void os::loop_breaker(int attempts) {
2679 os::yield_all(attempts);
2680 }
2681
2682 ////////////////////////////////////////////////////////////////////////////////
2683 // thread priority support
2684
2685 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
2686 // only supports dynamic priority, static priority must be zero. For real-time
2687 // applications, Linux supports SCHED_RR which allows static priority (1-99).
2688 // However, for large multi-threaded applications, SCHED_RR is not only slower
2689 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
2690 // of 5 runs - Sep 2005).
2691 //
2692 // The following code actually changes the niceness of kernel-thread/LWP. It
2693 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
2694 // not the entire user process, and user level threads are 1:1 mapped to kernel
2695 // threads. It has always been the case, but could change in the future. For
2696 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
2697 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
2698
2699 int os::java_to_os_priority[MaxPriority + 1] = {
2700 19, // 0 Entry should never be used
2701
2702 4, // 1 MinPriority
2703 3, // 2
2704 2, // 3
2705
2706 1, // 4
2707 0, // 5 NormPriority
2708 -1, // 6
2709
2710 -2, // 7
2711 -3, // 8
2712 -4, // 9 NearMaxPriority
2713
2714 -5 // 10 MaxPriority
2715 };
2716
2717 static int prio_init() {
2718 if (ThreadPriorityPolicy == 1) {
2719 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
2720 // if effective uid is not root. Perhaps, a more elegant way of doing
2721 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
2722 if (geteuid() != 0) {
2723 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
2724 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
2725 }
2726 ThreadPriorityPolicy = 0;
2727 }
2728 }
2729 return 0;
2730 }
2731
2732 OSReturn os::set_native_priority(Thread* thread, int newpri) {
2733 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
2734
2735 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
2736 return (ret == 0) ? OS_OK : OS_ERR;
2737 }
2738
2739 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
2740 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
2741 *priority_ptr = java_to_os_priority[NormPriority];
2742 return OS_OK;
2743 }
2744
2745 errno = 0;
2746 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
2747 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
2748 }
2749
2750 // Hint to the underlying OS that a task switch would not be good.
2751 // Void return because it's a hint and can fail.
2752 void os::hint_no_preempt() {}
2753
2754 ////////////////////////////////////////////////////////////////////////////////
2755 // suspend/resume support
2756
2757 // the low-level signal-based suspend/resume support is a remnant from the
2758 // old VM-suspension that used to be for java-suspension, safepoints etc,
2759 // within hotspot. Now there is a single use-case for this:
2760 // - calling get_thread_pc() on the VMThread by the flat-profiler task
2761 // that runs in the watcher thread.
2762 // The remaining code is greatly simplified from the more general suspension
2763 // code that used to be used.
2764 //
2765 // The protocol is quite simple:
2766 // - suspend:
2767 // - sends a signal to the target thread
2768 // - polls the suspend state of the osthread using a yield loop
2769 // - target thread signal handler (SR_handler) sets suspend state
2770 // and blocks in sigsuspend until continued
2771 // - resume:
2772 // - sets target osthread state to continue
2773 // - sends signal to end the sigsuspend loop in the SR_handler
2774 //
2775 // Note that the SR_lock plays no role in this suspend/resume protocol.
2776 //
2777
2778 static void resume_clear_context(OSThread *osthread) {
2779 osthread->set_ucontext(NULL);
2780 osthread->set_siginfo(NULL);
2781
2782 // notify the suspend action is completed, we have now resumed
2783 osthread->sr.clear_suspended();
2784 }
2785
2786 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
2787 osthread->set_ucontext(context);
2788 osthread->set_siginfo(siginfo);
2789 }
2790
2791 //
2792 // Handler function invoked when a thread's execution is suspended or
2793 // resumed. We have to be careful that only async-safe functions are
2794 // called here (Note: most pthread functions are not async safe and
2795 // should be avoided.)
2796 //
2797 // Note: sigwait() is a more natural fit than sigsuspend() from an
2798 // interface point of view, but sigwait() prevents the signal hander
2799 // from being run. libpthread would get very confused by not having
2800 // its signal handlers run and prevents sigwait()'s use with the
2801 // mutex granting granting signal.
2802 //
2803 // Currently only ever called on the VMThread
2804 //
2805 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
2806 // Save and restore errno to avoid confusing native code with EINTR
2807 // after sigsuspend.
2808 int old_errno = errno;
2809
2810 Thread* thread = Thread::current();
2811 OSThread* osthread = thread->osthread();
2812 assert(thread->is_VM_thread(), "Must be VMThread");
2813 // read current suspend action
2814 int action = osthread->sr.suspend_action();
2815 if (action == SR_SUSPEND) {
2816 suspend_save_context(osthread, siginfo, context);
2817
2818 // Notify the suspend action is about to be completed. do_suspend()
2819 // waits until SR_SUSPENDED is set and then returns. We will wait
2820 // here for a resume signal and that completes the suspend-other
2821 // action. do_suspend/do_resume is always called as a pair from
2822 // the same thread - so there are no races
2823
2824 // notify the caller
2825 osthread->sr.set_suspended();
2826
2827 sigset_t suspend_set; // signals for sigsuspend()
2828
2829 // get current set of blocked signals and unblock resume signal
2830 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
2831 sigdelset(&suspend_set, SR_signum);
2832
2833 // wait here until we are resumed
2834 do {
2835 sigsuspend(&suspend_set);
2836 // ignore all returns until we get a resume signal
2837 } while (osthread->sr.suspend_action() != SR_CONTINUE);
2838
2839 resume_clear_context(osthread);
2840
2841 } else {
2842 assert(action == SR_CONTINUE, "unexpected sr action");
2843 // nothing special to do - just leave the handler
2844 }
2845
2846 errno = old_errno;
2847 }
2848
2849
2850 static int SR_initialize() {
2851 struct sigaction act;
2852 char *s;
2853 /* Get signal number to use for suspend/resume */
2854 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
2855 int sig = ::strtol(s, 0, 10);
2856 if (sig > 0 || sig < _NSIG) {
2857 SR_signum = sig;
2858 }
2859 }
2860
2861 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
2862 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
2863
2864 sigemptyset(&SR_sigset);
2865 sigaddset(&SR_sigset, SR_signum);
2866
2867 /* Set up signal handler for suspend/resume */
2868 act.sa_flags = SA_RESTART|SA_SIGINFO;
2869 act.sa_handler = (void (*)(int)) SR_handler;
2870
2871 // SR_signum is blocked by default.
2872 // 4528190 - We also need to block pthread restart signal (32 on all
2873 // supported Linux platforms). Note that LinuxThreads need to block
2874 // this signal for all threads to work properly. So we don't have
2875 // to use hard-coded signal number when setting up the mask.
2876 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
2877
2878 if (sigaction(SR_signum, &act, 0) == -1) {
2879 return -1;
2880 }
2881
2882 // Save signal flag
2883 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
2884 return 0;
2885 }
2886
2887 static int SR_finalize() {
2888 return 0;
2889 }
2890
2891
2892 // returns true on success and false on error - really an error is fatal
2893 // but this seems the normal response to library errors
2894 static bool do_suspend(OSThread* osthread) {
2895 // mark as suspended and send signal
2896 osthread->sr.set_suspend_action(SR_SUSPEND);
2897 int status = pthread_kill(osthread->pthread_id(), SR_signum);
2898 assert_status(status == 0, status, "pthread_kill");
2899
2900 // check status and wait until notified of suspension
2901 if (status == 0) {
2902 for (int i = 0; !osthread->sr.is_suspended(); i++) {
2903 os::yield_all(i);
2904 }
2905 osthread->sr.set_suspend_action(SR_NONE);
2906 return true;
2907 }
2908 else {
2909 osthread->sr.set_suspend_action(SR_NONE);
2910 return false;
2911 }
2912 }
2913
2914 static void do_resume(OSThread* osthread) {
2915 assert(osthread->sr.is_suspended(), "thread should be suspended");
2916 osthread->sr.set_suspend_action(SR_CONTINUE);
2917
2918 int status = pthread_kill(osthread->pthread_id(), SR_signum);
2919 assert_status(status == 0, status, "pthread_kill");
2920 // check status and wait unit notified of resumption
2921 if (status == 0) {
2922 for (int i = 0; osthread->sr.is_suspended(); i++) {
2923 os::yield_all(i);
2924 }
2925 }
2926 osthread->sr.set_suspend_action(SR_NONE);
2927 }
2928
2929 ////////////////////////////////////////////////////////////////////////////////
2930 // interrupt support
2931
2932 void os::interrupt(Thread* thread) {
2933 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
2934 "possibility of dangling Thread pointer");
2935
2936 OSThread* osthread = thread->osthread();
2937
2938 if (!osthread->interrupted()) {
2939 osthread->set_interrupted(true);
2940 // More than one thread can get here with the same value of osthread,
2941 // resulting in multiple notifications. We do, however, want the store
2942 // to interrupted() to be visible to other threads before we execute unpark().
2943 OrderAccess::fence();
2944 ParkEvent * const slp = thread->_SleepEvent ;
2945 if (slp != NULL) slp->unpark() ;
2946 }
2947
2948 // For JSR166. Unpark even if interrupt status already was set
2949 if (thread->is_Java_thread())
2950 ((JavaThread*)thread)->parker()->unpark();
2951
2952 ParkEvent * ev = thread->_ParkEvent ;
2953 if (ev != NULL) ev->unpark() ;
2954
2955 }
2956
2957 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
2958 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
2959 "possibility of dangling Thread pointer");
2960
2961 OSThread* osthread = thread->osthread();
2962
2963 bool interrupted = osthread->interrupted();
2964
2965 if (interrupted && clear_interrupted) {
2966 osthread->set_interrupted(false);
2967 // consider thread->_SleepEvent->reset() ... optional optimization
2968 }
2969
2970 return interrupted;
2971 }
2972
2973 ///////////////////////////////////////////////////////////////////////////////////
2974 // signal handling (except suspend/resume)
2975
2976 // This routine may be used by user applications as a "hook" to catch signals.
2977 // The user-defined signal handler must pass unrecognized signals to this
2978 // routine, and if it returns true (non-zero), then the signal handler must
2979 // return immediately. If the flag "abort_if_unrecognized" is true, then this
2980 // routine will never retun false (zero), but instead will execute a VM panic
2981 // routine kill the process.
2982 //
2983 // If this routine returns false, it is OK to call it again. This allows
2984 // the user-defined signal handler to perform checks either before or after
2985 // the VM performs its own checks. Naturally, the user code would be making
2986 // a serious error if it tried to handle an exception (such as a null check
2987 // or breakpoint) that the VM was generating for its own correct operation.
2988 //
2989 // This routine may recognize any of the following kinds of signals:
2990 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
2991 // It should be consulted by handlers for any of those signals.
2992 //
2993 // The caller of this routine must pass in the three arguments supplied
2994 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
2995 // field of the structure passed to sigaction(). This routine assumes that
2996 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
2997 //
2998 // Note that the VM will print warnings if it detects conflicting signal
2999 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3000 //
3001 extern "C" int
3002 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3003 void* ucontext, int abort_if_unrecognized);
3004
3005 void signalHandler(int sig, siginfo_t* info, void* uc) {
3006 assert(info != NULL && uc != NULL, "it must be old kernel");
3007 JVM_handle_linux_signal(sig, info, uc, true);
3008 }
3009
3010
3011 // This boolean allows users to forward their own non-matching signals
3012 // to JVM_handle_linux_signal, harmlessly.
3013 bool os::Linux::signal_handlers_are_installed = false;
3014
3015 // For signal-chaining
3016 struct sigaction os::Linux::sigact[MAXSIGNUM];
3017 unsigned int os::Linux::sigs = 0;
3018 bool os::Linux::libjsig_is_loaded = false;
3019 typedef struct sigaction *(*get_signal_t)(int);
3020 get_signal_t os::Linux::get_signal_action = NULL;
3021
3022 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3023 struct sigaction *actp = NULL;
3024
3025 if (libjsig_is_loaded) {
3026 // Retrieve the old signal handler from libjsig
3027 actp = (*get_signal_action)(sig);
3028 }
3029 if (actp == NULL) {
3030 // Retrieve the preinstalled signal handler from jvm
3031 actp = get_preinstalled_handler(sig);
3032 }
3033
3034 return actp;
3035 }
3036
3037 static bool call_chained_handler(struct sigaction *actp, int sig,
3038 siginfo_t *siginfo, void *context) {
3039 // Call the old signal handler
3040 if (actp->sa_handler == SIG_DFL) {
3041 // It's more reasonable to let jvm treat it as an unexpected exception
3042 // instead of taking the default action.
3043 return false;
3044 } else if (actp->sa_handler != SIG_IGN) {
3045 if ((actp->sa_flags & SA_NODEFER) == 0) {
3046 // automaticlly block the signal
3047 sigaddset(&(actp->sa_mask), sig);
3048 }
3049
3050 sa_handler_t hand;
3051 sa_sigaction_t sa;
3052 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3053 // retrieve the chained handler
3054 if (siginfo_flag_set) {
3055 sa = actp->sa_sigaction;
3056 } else {
3057 hand = actp->sa_handler;
3058 }
3059
3060 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3061 actp->sa_handler = SIG_DFL;
3062 }
3063
3064 // try to honor the signal mask
3065 sigset_t oset;
3066 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3067
3068 // call into the chained handler
3069 if (siginfo_flag_set) {
3070 (*sa)(sig, siginfo, context);
3071 } else {
3072 (*hand)(sig);
3073 }
3074
3075 // restore the signal mask
3076 pthread_sigmask(SIG_SETMASK, &oset, 0);
3077 }
3078 // Tell jvm's signal handler the signal is taken care of.
3079 return true;
3080 }
3081
3082 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3083 bool chained = false;
3084 // signal-chaining
3085 if (UseSignalChaining) {
3086 struct sigaction *actp = get_chained_signal_action(sig);
3087 if (actp != NULL) {
3088 chained = call_chained_handler(actp, sig, siginfo, context);
3089 }
3090 }
3091 return chained;
3092 }
3093
3094 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3095 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3096 return &sigact[sig];
3097 }
3098 return NULL;
3099 }
3100
3101 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3102 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3103 sigact[sig] = oldAct;
3104 sigs |= (unsigned int)1 << sig;
3105 }
3106
3107 // for diagnostic
3108 int os::Linux::sigflags[MAXSIGNUM];
3109
3110 int os::Linux::get_our_sigflags(int sig) {
3111 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3112 return sigflags[sig];
3113 }
3114
3115 void os::Linux::set_our_sigflags(int sig, int flags) {
3116 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3117 sigflags[sig] = flags;
3118 }
3119
3120 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3121 // Check for overwrite.
3122 struct sigaction oldAct;
3123 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3124
3125 void* oldhand = oldAct.sa_sigaction
3126 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3127 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3128 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3129 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3130 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3131 if (AllowUserSignalHandlers || !set_installed) {
3132 // Do not overwrite; user takes responsibility to forward to us.
3133 return;
3134 } else if (UseSignalChaining) {
3135 // save the old handler in jvm
3136 save_preinstalled_handler(sig, oldAct);
3137 // libjsig also interposes the sigaction() call below and saves the
3138 // old sigaction on it own.
3139 } else {
3140 fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig);
3141 }
3142 }
3143
3144 struct sigaction sigAct;
3145 sigfillset(&(sigAct.sa_mask));
3146 sigAct.sa_handler = SIG_DFL;
3147 if (!set_installed) {
3148 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3149 } else {
3150 sigAct.sa_sigaction = signalHandler;
3151 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3152 }
3153 // Save flags, which are set by ours
3154 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3155 sigflags[sig] = sigAct.sa_flags;
3156
3157 int ret = sigaction(sig, &sigAct, &oldAct);
3158 assert(ret == 0, "check");
3159
3160 void* oldhand2 = oldAct.sa_sigaction
3161 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3162 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3163 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3164 }
3165
3166 // install signal handlers for signals that HotSpot needs to
3167 // handle in order to support Java-level exception handling.
3168
3169 void os::Linux::install_signal_handlers() {
3170 if (!signal_handlers_are_installed) {
3171 signal_handlers_are_installed = true;
3172
3173 // signal-chaining
3174 typedef void (*signal_setting_t)();
3175 signal_setting_t begin_signal_setting = NULL;
3176 signal_setting_t end_signal_setting = NULL;
3177 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3178 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3179 if (begin_signal_setting != NULL) {
3180 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3181 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3182 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3183 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3184 libjsig_is_loaded = true;
3185 assert(UseSignalChaining, "should enable signal-chaining");
3186 }
3187 if (libjsig_is_loaded) {
3188 // Tell libjsig jvm is setting signal handlers
3189 (*begin_signal_setting)();
3190 }
3191
3192 set_signal_handler(SIGSEGV, true);
3193 set_signal_handler(SIGPIPE, true);
3194 set_signal_handler(SIGBUS, true);
3195 set_signal_handler(SIGILL, true);
3196 set_signal_handler(SIGFPE, true);
3197 set_signal_handler(SIGXFSZ, true);
3198
3199 if (libjsig_is_loaded) {
3200 // Tell libjsig jvm finishes setting signal handlers
3201 (*end_signal_setting)();
3202 }
3203
3204 // We don't activate signal checker if libjsig is in place, we trust ourselves
3205 // and if UserSignalHandler is installed all bets are off
3206 if (CheckJNICalls) {
3207 if (libjsig_is_loaded) {
3208 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3209 check_signals = false;
3210 }
3211 if (AllowUserSignalHandlers) {
3212 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3213 check_signals = false;
3214 }
3215 }
3216 }
3217 }
3218
3219 // This is the fastest way to get thread cpu time on Linux.
3220 // Returns cpu time (user+sys) for any thread, not only for current.
3221 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3222 // It might work on 2.6.10+ with a special kernel/glibc patch.
3223 // For reference, please, see IEEE Std 1003.1-2004:
3224 // http://www.unix.org/single_unix_specification
3225
3226 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3227 struct timespec tp;
3228 int rc = os::Linux::clock_gettime(clockid, &tp);
3229 assert(rc == 0, "clock_gettime is expected to return 0 code");
3230
3231 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3232 }
3233
3234 /////
3235 // glibc on Linux platform uses non-documented flag
3236 // to indicate, that some special sort of signal
3237 // trampoline is used.
3238 // We will never set this flag, and we should
3239 // ignore this flag in our diagnostic
3240 #ifdef SIGNIFICANT_SIGNAL_MASK
3241 #undef SIGNIFICANT_SIGNAL_MASK
3242 #endif
3243 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3244
3245 static const char* get_signal_handler_name(address handler,
3246 char* buf, int buflen) {
3247 int offset;
3248 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3249 if (found) {
3250 // skip directory names
3251 const char *p1, *p2;
3252 p1 = buf;
3253 size_t len = strlen(os::file_separator());
3254 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3255 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3256 } else {
3257 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3258 }
3259 return buf;
3260 }
3261
3262 static void print_signal_handler(outputStream* st, int sig,
3263 char* buf, size_t buflen) {
3264 struct sigaction sa;
3265
3266 sigaction(sig, NULL, &sa);
3267
3268 // See comment for SIGNIFICANT_SIGNAL_MASK define
3269 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3270
3271 st->print("%s: ", os::exception_name(sig, buf, buflen));
3272
3273 address handler = (sa.sa_flags & SA_SIGINFO)
3274 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3275 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3276
3277 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3278 st->print("SIG_DFL");
3279 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3280 st->print("SIG_IGN");
3281 } else {
3282 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3283 }
3284
3285 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3286
3287 address rh = VMError::get_resetted_sighandler(sig);
3288 // May be, handler was resetted by VMError?
3289 if(rh != NULL) {
3290 handler = rh;
3291 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3292 }
3293
3294 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3295
3296 // Check: is it our handler?
3297 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3298 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3299 // It is our signal handler
3300 // check for flags, reset system-used one!
3301 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3302 st->print(
3303 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3304 os::Linux::get_our_sigflags(sig));
3305 }
3306 }
3307 st->cr();
3308 }
3309
3310
3311 #define DO_SIGNAL_CHECK(sig) \
3312 if (!sigismember(&check_signal_done, sig)) \
3313 os::Linux::check_signal_handler(sig)
3314
3315 // This method is a periodic task to check for misbehaving JNI applications
3316 // under CheckJNI, we can add any periodic checks here
3317
3318 void os::run_periodic_checks() {
3319
3320 if (check_signals == false) return;
3321
3322 // SEGV and BUS if overridden could potentially prevent
3323 // generation of hs*.log in the event of a crash, debugging
3324 // such a case can be very challenging, so we absolutely
3325 // check the following for a good measure:
3326 DO_SIGNAL_CHECK(SIGSEGV);
3327 DO_SIGNAL_CHECK(SIGILL);
3328 DO_SIGNAL_CHECK(SIGFPE);
3329 DO_SIGNAL_CHECK(SIGBUS);
3330 DO_SIGNAL_CHECK(SIGPIPE);
3331 DO_SIGNAL_CHECK(SIGXFSZ);
3332
3333
3334 // ReduceSignalUsage allows the user to override these handlers
3335 // see comments at the very top and jvm_solaris.h
3336 if (!ReduceSignalUsage) {
3337 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3338 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3339 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3340 DO_SIGNAL_CHECK(BREAK_SIGNAL);
3341 }
3342
3343 DO_SIGNAL_CHECK(SR_signum);
3344 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3345 }
3346
3347 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3348
3349 static os_sigaction_t os_sigaction = NULL;
3350
3351 void os::Linux::check_signal_handler(int sig) {
3352 char buf[O_BUFLEN];
3353 address jvmHandler = NULL;
3354
3355
3356 struct sigaction act;
3357 if (os_sigaction == NULL) {
3358 // only trust the default sigaction, in case it has been interposed
3359 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3360 if (os_sigaction == NULL) return;
3361 }
3362
3363 os_sigaction(sig, (struct sigaction*)NULL, &act);
3364
3365
3366 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3367
3368 address thisHandler = (act.sa_flags & SA_SIGINFO)
3369 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3370 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3371
3372
3373 switch(sig) {
3374 case SIGSEGV:
3375 case SIGBUS:
3376 case SIGFPE:
3377 case SIGPIPE:
3378 case SIGILL:
3379 case SIGXFSZ:
3380 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3381 break;
3382
3383 case SHUTDOWN1_SIGNAL:
3384 case SHUTDOWN2_SIGNAL:
3385 case SHUTDOWN3_SIGNAL:
3386 case BREAK_SIGNAL:
3387 jvmHandler = (address)user_handler();
3388 break;
3389
3390 case INTERRUPT_SIGNAL:
3391 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3392 break;
3393
3394 default:
3395 if (sig == SR_signum) {
3396 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3397 } else {
3398 return;
3399 }
3400 break;
3401 }
3402
3403 if (thisHandler != jvmHandler) {
3404 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3405 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3406 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3407 // No need to check this sig any longer
3408 sigaddset(&check_signal_done, sig);
3409 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3410 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3411 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3412 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
3413 // No need to check this sig any longer
3414 sigaddset(&check_signal_done, sig);
3415 }
3416
3417 // Dump all the signal
3418 if (sigismember(&check_signal_done, sig)) {
3419 print_signal_handlers(tty, buf, O_BUFLEN);
3420 }
3421 }
3422
3423 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3424
3425 extern bool signal_name(int signo, char* buf, size_t len);
3426
3427 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3428 if (0 < exception_code && exception_code <= SIGRTMAX) {
3429 // signal
3430 if (!signal_name(exception_code, buf, size)) {
3431 jio_snprintf(buf, size, "SIG%d", exception_code);
3432 }
3433 return buf;
3434 } else {
3435 return NULL;
3436 }
3437 }
3438
3439 // this is called _before_ the most of global arguments have been parsed
3440 void os::init(void) {
3441 char dummy; /* used to get a guess on initial stack address */
3442 // first_hrtime = gethrtime();
3443
3444 // With LinuxThreads the JavaMain thread pid (primordial thread)
3445 // is different than the pid of the java launcher thread.
3446 // So, on Linux, the launcher thread pid is passed to the VM
3447 // via the sun.java.launcher.pid property.
3448 // Use this property instead of getpid() if it was correctly passed.
3449 // See bug 6351349.
3450 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3451
3452 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3453
3454 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3455
3456 init_random(1234567);
3457
3458 ThreadCritical::initialize();
3459
3460 Linux::set_page_size(sysconf(_SC_PAGESIZE));
3461 if (Linux::page_size() == -1) {
3462 fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno));
3463 }
3464 init_page_sizes((size_t) Linux::page_size());
3465
3466 Linux::initialize_system_info();
3467
3468 // main_thread points to the aboriginal thread
3469 Linux::_main_thread = pthread_self();
3470
3471 Linux::clock_init();
3472 initial_time_count = os::elapsed_counter();
3473 }
3474
3475 // To install functions for atexit system call
3476 extern "C" {
3477 static void perfMemory_exit_helper() {
3478 perfMemory_exit();
3479 }
3480 }
3481
3482 // this is called _after_ the global arguments have been parsed
3483 jint os::init_2(void)
3484 {
3485 Linux::fast_thread_clock_init();
3486
3487 // Allocate a single page and mark it as readable for safepoint polling
3488 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3489 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3490
3491 os::set_polling_page( polling_page );
3492
3493 #ifndef PRODUCT
3494 if(Verbose && PrintMiscellaneous)
3495 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3496 #endif
3497
3498 if (!UseMembar) {
3499 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3500 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3501 os::set_memory_serialize_page( mem_serialize_page );
3502
3503 #ifndef PRODUCT
3504 if(Verbose && PrintMiscellaneous)
3505 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3506 #endif
3507 }
3508
3509 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3510
3511 // initialize suspend/resume support - must do this before signal_sets_init()
3512 if (SR_initialize() != 0) {
3513 perror("SR_initialize failed");
3514 return JNI_ERR;
3515 }
3516
3517 Linux::signal_sets_init();
3518 Linux::install_signal_handlers();
3519
3520 size_t threadStackSizeInBytes = ThreadStackSize * K;
3521 if (threadStackSizeInBytes != 0 &&
3522 threadStackSizeInBytes < Linux::min_stack_allowed) {
3523 tty->print_cr("\nThe stack size specified is too small, "
3524 "Specify at least %dk",
3525 Linux::min_stack_allowed / K);
3526 return JNI_ERR;
3527 }
3528
3529 // Make the stack size a multiple of the page size so that
3530 // the yellow/red zones can be guarded.
3531 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3532 vm_page_size()));
3533
3534 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3535
3536 Linux::libpthread_init();
3537 if (PrintMiscellaneous && (Verbose || WizardMode)) {
3538 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3539 Linux::glibc_version(), Linux::libpthread_version(),
3540 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3541 }
3542
3543 if (MaxFDLimit) {
3544 // set the number of file descriptors to max. print out error
3545 // if getrlimit/setrlimit fails but continue regardless.
3546 struct rlimit nbr_files;
3547 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3548 if (status != 0) {
3549 if (PrintMiscellaneous && (Verbose || WizardMode))
3550 perror("os::init_2 getrlimit failed");
3551 } else {
3552 nbr_files.rlim_cur = nbr_files.rlim_max;
3553 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3554 if (status != 0) {
3555 if (PrintMiscellaneous && (Verbose || WizardMode))
3556 perror("os::init_2 setrlimit failed");
3557 }
3558 }
3559 }
3560
3561 // Initialize lock used to serialize thread creation (see os::create_thread)
3562 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3563
3564 // Initialize HPI.
3565 jint hpi_result = hpi::initialize();
3566 if (hpi_result != JNI_OK) {
3567 tty->print_cr("There was an error trying to initialize the HPI library.");
3568 return hpi_result;
3569 }
3570
3571 // at-exit methods are called in the reverse order of their registration.
3572 // atexit functions are called on return from main or as a result of a
3573 // call to exit(3C). There can be only 32 of these functions registered
3574 // and atexit() does not set errno.
3575
3576 if (PerfAllowAtExitRegistration) {
3577 // only register atexit functions if PerfAllowAtExitRegistration is set.
3578 // atexit functions can be delayed until process exit time, which
3579 // can be problematic for embedded VM situations. Embedded VMs should
3580 // call DestroyJavaVM() to assure that VM resources are released.
3581
3582 // note: perfMemory_exit_helper atexit function may be removed in
3583 // the future if the appropriate cleanup code can be added to the
3584 // VM_Exit VMOperation's doit method.
3585 if (atexit(perfMemory_exit_helper) != 0) {
3586 warning("os::init2 atexit(perfMemory_exit_helper) failed");
3587 }
3588 }
3589
3590 // initialize thread priority policy
3591 prio_init();
3592
3593 return JNI_OK;
3594 }
3595
3596 // Mark the polling page as unreadable
3597 void os::make_polling_page_unreadable(void) {
3598 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
3599 fatal("Could not disable polling page");
3600 };
3601
3602 // Mark the polling page as readable
3603 void os::make_polling_page_readable(void) {
3604 if( !protect_memory((char *)_polling_page, Linux::page_size()) )
3605 fatal("Could not enable polling page");
3606 };
3607
3608 int os::active_processor_count() {
3609 // Linux doesn't yet have a (official) notion of processor sets,
3610 // so just return the number of online processors.
3611 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
3612 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
3613 return online_cpus;
3614 }
3615
3616 bool os::distribute_processes(uint length, uint* distribution) {
3617 // Not yet implemented.
3618 return false;
3619 }
3620
3621 bool os::bind_to_processor(uint processor_id) {
3622 // Not yet implemented.
3623 return false;
3624 }
3625
3626 ///
3627
3628 // Suspends the target using the signal mechanism and then grabs the PC before
3629 // resuming the target. Used by the flat-profiler only
3630 ExtendedPC os::get_thread_pc(Thread* thread) {
3631 // Make sure that it is called by the watcher for the VMThread
3632 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
3633 assert(thread->is_VM_thread(), "Can only be called for VMThread");
3634
3635 ExtendedPC epc;
3636
3637 OSThread* osthread = thread->osthread();
3638 if (do_suspend(osthread)) {
3639 if (osthread->ucontext() != NULL) {
3640 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
3641 } else {
3642 // NULL context is unexpected, double-check this is the VMThread
3643 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3644 }
3645 do_resume(osthread);
3646 }
3647 // failure means pthread_kill failed for some reason - arguably this is
3648 // a fatal problem, but such problems are ignored elsewhere
3649
3650 return epc;
3651 }
3652
3653 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
3654 {
3655 if (is_NPTL()) {
3656 return pthread_cond_timedwait(_cond, _mutex, _abstime);
3657 } else {
3658 #ifndef IA64
3659 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
3660 // word back to default 64bit precision if condvar is signaled. Java
3661 // wants 53bit precision. Save and restore current value.
3662 int fpu = get_fpu_control_word();
3663 #endif // IA64
3664 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
3665 #ifndef IA64
3666 set_fpu_control_word(fpu);
3667 #endif // IA64
3668 return status;
3669 }
3670 }
3671
3672 ////////////////////////////////////////////////////////////////////////////////
3673 // debug support
3674
3675 #ifndef PRODUCT
3676 static address same_page(address x, address y) {
3677 int page_bits = -os::vm_page_size();
3678 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
3679 return x;
3680 else if (x > y)
3681 return (address)(intptr_t(y) | ~page_bits) + 1;
3682 else
3683 return (address)(intptr_t(y) & page_bits);
3684 }
3685
3686 bool os::find(address addr) {
3687 Dl_info dlinfo;
3688 memset(&dlinfo, 0, sizeof(dlinfo));
3689 if (dladdr(addr, &dlinfo)) {
3690 tty->print(PTR_FORMAT ": ", addr);
3691 if (dlinfo.dli_sname != NULL) {
3692 tty->print("%s+%#x", dlinfo.dli_sname,
3693 addr - (intptr_t)dlinfo.dli_saddr);
3694 } else if (dlinfo.dli_fname) {
3695 tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
3696 } else {
3697 tty->print("<absolute address>");
3698 }
3699 if (dlinfo.dli_fname) {
3700 tty->print(" in %s", dlinfo.dli_fname);
3701 }
3702 if (dlinfo.dli_fbase) {
3703 tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
3704 }
3705 tty->cr();
3706
3707 if (Verbose) {
3708 // decode some bytes around the PC
3709 address begin = same_page(addr-40, addr);
3710 address end = same_page(addr+40, addr);
3711 address lowest = (address) dlinfo.dli_sname;
3712 if (!lowest) lowest = (address) dlinfo.dli_fbase;
3713 if (begin < lowest) begin = lowest;
3714 Dl_info dlinfo2;
3715 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
3716 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
3717 end = (address) dlinfo2.dli_saddr;
3718 Disassembler::decode(begin, end);
3719 }
3720 return true;
3721 }
3722 return false;
3723 }
3724
3725 #endif
3726
3727 ////////////////////////////////////////////////////////////////////////////////
3728 // misc
3729
3730 // This does not do anything on Linux. This is basically a hook for being
3731 // able to use structured exception handling (thread-local exception filters)
3732 // on, e.g., Win32.
3733 void
3734 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
3735 JavaCallArguments* args, Thread* thread) {
3736 f(value, method, args, thread);
3737 }
3738
3739 void os::print_statistics() {
3740 }
3741
3742 int os::message_box(const char* title, const char* message) {
3743 int i;
3744 fdStream err(defaultStream::error_fd());
3745 for (i = 0; i < 78; i++) err.print_raw("=");
3746 err.cr();
3747 err.print_raw_cr(title);
3748 for (i = 0; i < 78; i++) err.print_raw("-");
3749 err.cr();
3750 err.print_raw_cr(message);
3751 for (i = 0; i < 78; i++) err.print_raw("=");
3752 err.cr();
3753
3754 char buf[16];
3755 // Prevent process from exiting upon "read error" without consuming all CPU
3756 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3757
3758 return buf[0] == 'y' || buf[0] == 'Y';
3759 }
3760
3761 int os::stat(const char *path, struct stat *sbuf) {
3762 char pathbuf[MAX_PATH];
3763 if (strlen(path) > MAX_PATH - 1) {
3764 errno = ENAMETOOLONG;
3765 return -1;
3766 }
3767 hpi::native_path(strcpy(pathbuf, path));
3768 return ::stat(pathbuf, sbuf);
3769 }
3770
3771 bool os::check_heap(bool force) {
3772 return true;
3773 }
3774
3775 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
3776 return ::vsnprintf(buf, count, format, args);
3777 }
3778
3779 // Is a (classpath) directory empty?
3780 bool os::dir_is_empty(const char* path) {
3781 DIR *dir = NULL;
3782 struct dirent *ptr;
3783
3784 dir = opendir(path);
3785 if (dir == NULL) return true;
3786
3787 /* Scan the directory */
3788 bool result = true;
3789 char buf[sizeof(struct dirent) + MAX_PATH];
3790 while (result && (ptr = ::readdir(dir)) != NULL) {
3791 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
3792 result = false;
3793 }
3794 }
3795 closedir(dir);
3796 return result;
3797 }
3798
3799 // create binary file, rewriting existing file if required
3800 int os::create_binary_file(const char* path, bool rewrite_existing) {
3801 int oflags = O_WRONLY | O_CREAT;
3802 if (!rewrite_existing) {
3803 oflags |= O_EXCL;
3804 }
3805 return ::open64(path, oflags, S_IREAD | S_IWRITE);
3806 }
3807
3808 // return current position of file pointer
3809 jlong os::current_file_offset(int fd) {
3810 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
3811 }
3812
3813 // move file pointer to the specified offset
3814 jlong os::seek_to_file_offset(int fd, jlong offset) {
3815 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
3816 }
3817
3818 // Map a block of memory.
3819 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
3820 char *addr, size_t bytes, bool read_only,
3821 bool allow_exec) {
3822 int prot;
3823 int flags;
3824
3825 if (read_only) {
3826 prot = PROT_READ;
3827 flags = MAP_SHARED;
3828 } else {
3829 prot = PROT_READ | PROT_WRITE;
3830 flags = MAP_PRIVATE;
3831 }
3832
3833 if (allow_exec) {
3834 prot |= PROT_EXEC;
3835 }
3836
3837 if (addr != NULL) {
3838 flags |= MAP_FIXED;
3839 }
3840
3841 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
3842 fd, file_offset);
3843 if (mapped_address == MAP_FAILED) {
3844 return NULL;
3845 }
3846 return mapped_address;
3847 }
3848
3849
3850 // Remap a block of memory.
3851 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
3852 char *addr, size_t bytes, bool read_only,
3853 bool allow_exec) {
3854 // same as map_memory() on this OS
3855 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
3856 allow_exec);
3857 }
3858
3859
3860 // Unmap a block of memory.
3861 bool os::unmap_memory(char* addr, size_t bytes) {
3862 return munmap(addr, bytes) == 0;
3863 }
3864
3865 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
3866
3867 static clockid_t thread_cpu_clockid(Thread* thread) {
3868 pthread_t tid = thread->osthread()->pthread_id();
3869 clockid_t clockid;
3870
3871 // Get thread clockid
3872 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
3873 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
3874 return clockid;
3875 }
3876
3877 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
3878 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
3879 // of a thread.
3880 //
3881 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
3882 // the fast estimate available on the platform.
3883
3884 jlong os::current_thread_cpu_time() {
3885 if (os::Linux::supports_fast_thread_cpu_time()) {
3886 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
3887 } else {
3888 // return user + sys since the cost is the same
3889 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
3890 }
3891 }
3892
3893 jlong os::thread_cpu_time(Thread* thread) {
3894 // consistent with what current_thread_cpu_time() returns
3895 if (os::Linux::supports_fast_thread_cpu_time()) {
3896 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
3897 } else {
3898 return slow_thread_cpu_time(thread, true /* user + sys */);
3899 }
3900 }
3901
3902 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
3903 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
3904 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
3905 } else {
3906 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
3907 }
3908 }
3909
3910 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
3911 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
3912 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
3913 } else {
3914 return slow_thread_cpu_time(thread, user_sys_cpu_time);
3915 }
3916 }
3917
3918 //
3919 // -1 on error.
3920 //
3921
3922 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
3923 static bool proc_pid_cpu_avail = true;
3924 static bool proc_task_unchecked = true;
3925 static const char *proc_stat_path = "/proc/%d/stat";
3926 pid_t tid = thread->osthread()->thread_id();
3927 int i;
3928 char *s;
3929 char stat[2048];
3930 int statlen;
3931 char proc_name[64];
3932 int count;
3933 long sys_time, user_time;
3934 char string[64];
3935 int idummy;
3936 long ldummy;
3937 FILE *fp;
3938
3939 // We first try accessing /proc/<pid>/cpu since this is faster to
3940 // process. If this file is not present (linux kernels 2.5 and above)
3941 // then we open /proc/<pid>/stat.
3942 if ( proc_pid_cpu_avail ) {
3943 sprintf(proc_name, "/proc/%d/cpu", tid);
3944 fp = fopen(proc_name, "r");
3945 if ( fp != NULL ) {
3946 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
3947 fclose(fp);
3948 if ( count != 3 ) return -1;
3949
3950 if (user_sys_cpu_time) {
3951 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
3952 } else {
3953 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
3954 }
3955 }
3956 else proc_pid_cpu_avail = false;
3957 }
3958
3959 // The /proc/<tid>/stat aggregates per-process usage on
3960 // new Linux kernels 2.6+ where NPTL is supported.
3961 // The /proc/self/task/<tid>/stat still has the per-thread usage.
3962 // See bug 6328462.
3963 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
3964 // and possibly in some other cases, so we check its availability.
3965 if (proc_task_unchecked && os::Linux::is_NPTL()) {
3966 // This is executed only once
3967 proc_task_unchecked = false;
3968 fp = fopen("/proc/self/task", "r");
3969 if (fp != NULL) {
3970 proc_stat_path = "/proc/self/task/%d/stat";
3971 fclose(fp);
3972 }
3973 }
3974
3975 sprintf(proc_name, proc_stat_path, tid);
3976 fp = fopen(proc_name, "r");
3977 if ( fp == NULL ) return -1;
3978 statlen = fread(stat, 1, 2047, fp);
3979 stat[statlen] = '\0';
3980 fclose(fp);
3981
3982 // Skip pid and the command string. Note that we could be dealing with
3983 // weird command names, e.g. user could decide to rename java launcher
3984 // to "java 1.4.2 :)", then the stat file would look like
3985 // 1234 (java 1.4.2 :)) R ... ...
3986 // We don't really need to know the command string, just find the last
3987 // occurrence of ")" and then start parsing from there. See bug 4726580.
3988 s = strrchr(stat, ')');
3989 i = 0;
3990 if (s == NULL ) return -1;
3991
3992 // Skip blank chars
3993 do s++; while (isspace(*s));
3994
3995 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
3996 &idummy, &idummy, &idummy, &idummy, &idummy, &idummy,
3997 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
3998 &user_time, &sys_time);
3999 if ( count != 13 ) return -1;
4000 if (user_sys_cpu_time) {
4001 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4002 } else {
4003 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4004 }
4005 }
4006
4007 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4008 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4009 info_ptr->may_skip_backward = false; // elapsed time not wall time
4010 info_ptr->may_skip_forward = false; // elapsed time not wall time
4011 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4012 }
4013
4014 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4015 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4016 info_ptr->may_skip_backward = false; // elapsed time not wall time
4017 info_ptr->may_skip_forward = false; // elapsed time not wall time
4018 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4019 }
4020
4021 bool os::is_thread_cpu_time_supported() {
4022 return true;
4023 }
4024
4025 // System loadavg support. Returns -1 if load average cannot be obtained.
4026 // Linux doesn't yet have a (official) notion of processor sets,
4027 // so just return the system wide load average.
4028 int os::loadavg(double loadavg[], int nelem) {
4029 return ::getloadavg(loadavg, nelem);
4030 }
4031
4032 void os::pause() {
4033 char filename[MAX_PATH];
4034 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4035 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4036 } else {
4037 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4038 }
4039
4040 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4041 if (fd != -1) {
4042 struct stat buf;
4043 close(fd);
4044 while (::stat(filename, &buf) == 0) {
4045 (void)::poll(NULL, 0, 100);
4046 }
4047 } else {
4048 jio_fprintf(stderr,
4049 "Could not open pause file '%s', continuing immediately.\n", filename);
4050 }
4051 }
4052
4053 extern "C" {
4054
4055 /**
4056 * NOTE: the following code is to keep the green threads code
4057 * in the libjava.so happy. Once the green threads is removed,
4058 * these code will no longer be needed.
4059 */
4060 int
4061 jdk_waitpid(pid_t pid, int* status, int options) {
4062 return waitpid(pid, status, options);
4063 }
4064
4065 int
4066 fork1() {
4067 return fork();
4068 }
4069
4070 int
4071 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4072 return sem_init(sem, pshared, value);
4073 }
4074
4075 int
4076 jdk_sem_post(sem_t *sem) {
4077 return sem_post(sem);
4078 }
4079
4080 int
4081 jdk_sem_wait(sem_t *sem) {
4082 return sem_wait(sem);
4083 }
4084
4085 int
4086 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4087 return pthread_sigmask(how , newmask, oldmask);
4088 }
4089
4090 }
4091
4092 // Refer to the comments in os_solaris.cpp park-unpark.
4093 //
4094 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4095 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4096 // For specifics regarding the bug see GLIBC BUGID 261237 :
4097 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4098 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4099 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4100 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4101 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4102 // and monitorenter when we're using 1-0 locking. All those operations may result in
4103 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4104 // of libpthread avoids the problem, but isn't practical.
4105 //
4106 // Possible remedies:
4107 //
4108 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4109 // This is palliative and probabilistic, however. If the thread is preempted
4110 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4111 // than the minimum period may have passed, and the abstime may be stale (in the
4112 // past) resultin in a hang. Using this technique reduces the odds of a hang
4113 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4114 //
4115 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4116 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4117 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4118 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4119 // thread.
4120 //
4121 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4122 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4123 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4124 // This also works well. In fact it avoids kernel-level scalability impediments
4125 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4126 // timers in a graceful fashion.
4127 //
4128 // 4. When the abstime value is in the past it appears that control returns
4129 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4130 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4131 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4132 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4133 // It may be possible to avoid reinitialization by checking the return
4134 // value from pthread_cond_timedwait(). In addition to reinitializing the
4135 // condvar we must establish the invariant that cond_signal() is only called
4136 // within critical sections protected by the adjunct mutex. This prevents
4137 // cond_signal() from "seeing" a condvar that's in the midst of being
4138 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4139 // desirable signal-after-unlock optimization that avoids futile context switching.
4140 //
4141 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4142 // structure when a condvar is used or initialized. cond_destroy() would
4143 // release the helper structure. Our reinitialize-after-timedwait fix
4144 // put excessive stress on malloc/free and locks protecting the c-heap.
4145 //
4146 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4147 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4148 // and only enabling the work-around for vulnerable environments.
4149
4150 // utility to compute the abstime argument to timedwait:
4151 // millis is the relative timeout time
4152 // abstime will be the absolute timeout time
4153 // TODO: replace compute_abstime() with unpackTime()
4154
4155 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4156 if (millis < 0) millis = 0;
4157 struct timeval now;
4158 int status = gettimeofday(&now, NULL);
4159 assert(status == 0, "gettimeofday");
4160 jlong seconds = millis / 1000;
4161 millis %= 1000;
4162 if (seconds > 50000000) { // see man cond_timedwait(3T)
4163 seconds = 50000000;
4164 }
4165 abstime->tv_sec = now.tv_sec + seconds;
4166 long usec = now.tv_usec + millis * 1000;
4167 if (usec >= 1000000) {
4168 abstime->tv_sec += 1;
4169 usec -= 1000000;
4170 }
4171 abstime->tv_nsec = usec * 1000;
4172 return abstime;
4173 }
4174
4175
4176 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4177 // Conceptually TryPark() should be equivalent to park(0).
4178
4179 int os::PlatformEvent::TryPark() {
4180 for (;;) {
4181 const int v = _Event ;
4182 guarantee ((v == 0) || (v == 1), "invariant") ;
4183 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4184 }
4185 }
4186
4187 void os::PlatformEvent::park() { // AKA "down()"
4188 // Invariant: Only the thread associated with the Event/PlatformEvent
4189 // may call park().
4190 // TODO: assert that _Assoc != NULL or _Assoc == Self
4191 int v ;
4192 for (;;) {
4193 v = _Event ;
4194 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4195 }
4196 guarantee (v >= 0, "invariant") ;
4197 if (v == 0) {
4198 // Do this the hard way by blocking ...
4199 int status = pthread_mutex_lock(_mutex);
4200 assert_status(status == 0, status, "mutex_lock");
4201 guarantee (_nParked == 0, "invariant") ;
4202 ++ _nParked ;
4203 while (_Event < 0) {
4204 status = pthread_cond_wait(_cond, _mutex);
4205 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4206 // Treat this the same as if the wait was interrupted
4207 if (status == ETIME) { status = EINTR; }
4208 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4209 }
4210 -- _nParked ;
4211
4212 // In theory we could move the ST of 0 into _Event past the unlock(),
4213 // but then we'd need a MEMBAR after the ST.
4214 _Event = 0 ;
4215 status = pthread_mutex_unlock(_mutex);
4216 assert_status(status == 0, status, "mutex_unlock");
4217 }
4218 guarantee (_Event >= 0, "invariant") ;
4219 }
4220
4221 int os::PlatformEvent::park(jlong millis) {
4222 guarantee (_nParked == 0, "invariant") ;
4223
4224 int v ;
4225 for (;;) {
4226 v = _Event ;
4227 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4228 }
4229 guarantee (v >= 0, "invariant") ;
4230 if (v != 0) return OS_OK ;
4231
4232 // We do this the hard way, by blocking the thread.
4233 // Consider enforcing a minimum timeout value.
4234 struct timespec abst;
4235 compute_abstime(&abst, millis);
4236
4237 int ret = OS_TIMEOUT;
4238 int status = pthread_mutex_lock(_mutex);
4239 assert_status(status == 0, status, "mutex_lock");
4240 guarantee (_nParked == 0, "invariant") ;
4241 ++_nParked ;
4242
4243 // Object.wait(timo) will return because of
4244 // (a) notification
4245 // (b) timeout
4246 // (c) thread.interrupt
4247 //
4248 // Thread.interrupt and object.notify{All} both call Event::set.
4249 // That is, we treat thread.interrupt as a special case of notification.
4250 // The underlying Solaris implementation, cond_timedwait, admits
4251 // spurious/premature wakeups, but the JLS/JVM spec prevents the
4252 // JVM from making those visible to Java code. As such, we must
4253 // filter out spurious wakeups. We assume all ETIME returns are valid.
4254 //
4255 // TODO: properly differentiate simultaneous notify+interrupt.
4256 // In that case, we should propagate the notify to another waiter.
4257
4258 while (_Event < 0) {
4259 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4260 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4261 pthread_cond_destroy (_cond);
4262 pthread_cond_init (_cond, NULL) ;
4263 }
4264 assert_status(status == 0 || status == EINTR ||
4265 status == ETIME || status == ETIMEDOUT,
4266 status, "cond_timedwait");
4267 if (!FilterSpuriousWakeups) break ; // previous semantics
4268 if (status == ETIME || status == ETIMEDOUT) break ;
4269 // We consume and ignore EINTR and spurious wakeups.
4270 }
4271 --_nParked ;
4272 if (_Event >= 0) {
4273 ret = OS_OK;
4274 }
4275 _Event = 0 ;
4276 status = pthread_mutex_unlock(_mutex);
4277 assert_status(status == 0, status, "mutex_unlock");
4278 assert (_nParked == 0, "invariant") ;
4279 return ret;
4280 }
4281
4282 void os::PlatformEvent::unpark() {
4283 int v, AnyWaiters ;
4284 for (;;) {
4285 v = _Event ;
4286 if (v > 0) {
4287 // The LD of _Event could have reordered or be satisfied
4288 // by a read-aside from this processor's write buffer.
4289 // To avoid problems execute a barrier and then
4290 // ratify the value.
4291 OrderAccess::fence() ;
4292 if (_Event == v) return ;
4293 continue ;
4294 }
4295 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4296 }
4297 if (v < 0) {
4298 // Wait for the thread associated with the event to vacate
4299 int status = pthread_mutex_lock(_mutex);
4300 assert_status(status == 0, status, "mutex_lock");
4301 AnyWaiters = _nParked ;
4302 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4303 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4304 AnyWaiters = 0 ;
4305 pthread_cond_signal (_cond);
4306 }
4307 status = pthread_mutex_unlock(_mutex);
4308 assert_status(status == 0, status, "mutex_unlock");
4309 if (AnyWaiters != 0) {
4310 status = pthread_cond_signal(_cond);
4311 assert_status(status == 0, status, "cond_signal");
4312 }
4313 }
4314
4315 // Note that we signal() _after dropping the lock for "immortal" Events.
4316 // This is safe and avoids a common class of futile wakeups. In rare
4317 // circumstances this can cause a thread to return prematurely from
4318 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4319 // simply re-test the condition and re-park itself.
4320 }
4321
4322
4323 // JSR166
4324 // -------------------------------------------------------
4325
4326 /*
4327 * The solaris and linux implementations of park/unpark are fairly
4328 * conservative for now, but can be improved. They currently use a
4329 * mutex/condvar pair, plus a a count.
4330 * Park decrements count if > 0, else does a condvar wait. Unpark
4331 * sets count to 1 and signals condvar. Only one thread ever waits
4332 * on the condvar. Contention seen when trying to park implies that someone
4333 * is unparking you, so don't wait. And spurious returns are fine, so there
4334 * is no need to track notifications.
4335 */
4336
4337
4338 #define NANOSECS_PER_SEC 1000000000
4339 #define NANOSECS_PER_MILLISEC 1000000
4340 #define MAX_SECS 100000000
4341 /*
4342 * This code is common to linux and solaris and will be moved to a
4343 * common place in dolphin.
4344 *
4345 * The passed in time value is either a relative time in nanoseconds
4346 * or an absolute time in milliseconds. Either way it has to be unpacked
4347 * into suitable seconds and nanoseconds components and stored in the
4348 * given timespec structure.
4349 * Given time is a 64-bit value and the time_t used in the timespec is only
4350 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4351 * overflow if times way in the future are given. Further on Solaris versions
4352 * prior to 10 there is a restriction (see cond_timedwait) that the specified
4353 * number of seconds, in abstime, is less than current_time + 100,000,000.
4354 * As it will be 28 years before "now + 100000000" will overflow we can
4355 * ignore overflow and just impose a hard-limit on seconds using the value
4356 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4357 * years from "now".
4358 */
4359
4360 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4361 assert (time > 0, "convertTime");
4362
4363 struct timeval now;
4364 int status = gettimeofday(&now, NULL);
4365 assert(status == 0, "gettimeofday");
4366
4367 time_t max_secs = now.tv_sec + MAX_SECS;
4368
4369 if (isAbsolute) {
4370 jlong secs = time / 1000;
4371 if (secs > max_secs) {
4372 absTime->tv_sec = max_secs;
4373 }
4374 else {
4375 absTime->tv_sec = secs;
4376 }
4377 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4378 }
4379 else {
4380 jlong secs = time / NANOSECS_PER_SEC;
4381 if (secs >= MAX_SECS) {
4382 absTime->tv_sec = max_secs;
4383 absTime->tv_nsec = 0;
4384 }
4385 else {
4386 absTime->tv_sec = now.tv_sec + secs;
4387 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4388 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4389 absTime->tv_nsec -= NANOSECS_PER_SEC;
4390 ++absTime->tv_sec; // note: this must be <= max_secs
4391 }
4392 }
4393 }
4394 assert(absTime->tv_sec >= 0, "tv_sec < 0");
4395 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4396 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4397 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4398 }
4399
4400 void Parker::park(bool isAbsolute, jlong time) {
4401 // Optional fast-path check:
4402 // Return immediately if a permit is available.
4403 if (_counter > 0) {
4404 _counter = 0 ;
4405 return ;
4406 }
4407
4408 Thread* thread = Thread::current();
4409 assert(thread->is_Java_thread(), "Must be JavaThread");
4410 JavaThread *jt = (JavaThread *)thread;
4411
4412 // Optional optimization -- avoid state transitions if there's an interrupt pending.
4413 // Check interrupt before trying to wait
4414 if (Thread::is_interrupted(thread, false)) {
4415 return;
4416 }
4417
4418 // Next, demultiplex/decode time arguments
4419 timespec absTime;
4420 if (time < 0) { // don't wait at all
4421 return;
4422 }
4423 if (time > 0) {
4424 unpackTime(&absTime, isAbsolute, time);
4425 }
4426
4427
4428 // Enter safepoint region
4429 // Beware of deadlocks such as 6317397.
4430 // The per-thread Parker:: mutex is a classic leaf-lock.
4431 // In particular a thread must never block on the Threads_lock while
4432 // holding the Parker:: mutex. If safepoints are pending both the
4433 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4434 ThreadBlockInVM tbivm(jt);
4435
4436 // Don't wait if cannot get lock since interference arises from
4437 // unblocking. Also. check interrupt before trying wait
4438 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4439 return;
4440 }
4441
4442 int status ;
4443 if (_counter > 0) { // no wait needed
4444 _counter = 0;
4445 status = pthread_mutex_unlock(_mutex);
4446 assert (status == 0, "invariant") ;
4447 return;
4448 }
4449
4450 #ifdef ASSERT
4451 // Don't catch signals while blocked; let the running threads have the signals.
4452 // (This allows a debugger to break into the running thread.)
4453 sigset_t oldsigs;
4454 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4455 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4456 #endif
4457
4458 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4459 jt->set_suspend_equivalent();
4460 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4461
4462 if (time == 0) {
4463 status = pthread_cond_wait (_cond, _mutex) ;
4464 } else {
4465 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4466 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4467 pthread_cond_destroy (_cond) ;
4468 pthread_cond_init (_cond, NULL);
4469 }
4470 }
4471 assert_status(status == 0 || status == EINTR ||
4472 status == ETIME || status == ETIMEDOUT,
4473 status, "cond_timedwait");
4474
4475 #ifdef ASSERT
4476 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4477 #endif
4478
4479 _counter = 0 ;
4480 status = pthread_mutex_unlock(_mutex) ;
4481 assert_status(status == 0, status, "invariant") ;
4482 // If externally suspended while waiting, re-suspend
4483 if (jt->handle_special_suspend_equivalent_condition()) {
4484 jt->java_suspend_self();
4485 }
4486
4487 }
4488
4489 void Parker::unpark() {
4490 int s, status ;
4491 status = pthread_mutex_lock(_mutex);
4492 assert (status == 0, "invariant") ;
4493 s = _counter;
4494 _counter = 1;
4495 if (s < 1) {
4496 if (WorkAroundNPTLTimedWaitHang) {
4497 status = pthread_cond_signal (_cond) ;
4498 assert (status == 0, "invariant") ;
4499 status = pthread_mutex_unlock(_mutex);
4500 assert (status == 0, "invariant") ;
4501 } else {
4502 status = pthread_mutex_unlock(_mutex);
4503 assert (status == 0, "invariant") ;
4504 status = pthread_cond_signal (_cond) ;
4505 assert (status == 0, "invariant") ;
4506 }
4507 } else {
4508 pthread_mutex_unlock(_mutex);
4509 assert (status == 0, "invariant") ;
4510 }
4511 }
4512
4513
4514 extern char** environ;
4515
4516 #ifndef __NR_fork
4517 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4518 #endif
4519
4520 #ifndef __NR_execve
4521 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4522 #endif
4523
4524 // Run the specified command in a separate process. Return its exit value,
4525 // or -1 on failure (e.g. can't fork a new process).
4526 // Unlike system(), this function can be called from signal handler. It
4527 // doesn't block SIGINT et al.
4528 int os::fork_and_exec(char* cmd) {
4529 char * argv[4];
4530 argv[0] = "sh";
4531 argv[1] = "-c";
4532 argv[2] = cmd;
4533 argv[3] = NULL;
4534
4535 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4536 // pthread_atfork handlers and reset pthread library. All we need is a
4537 // separate process to execve. Make a direct syscall to fork process.
4538 // On IA64 there's no fork syscall, we have to use fork() and hope for
4539 // the best...
4540 pid_t pid = NOT_IA64(syscall(__NR_fork);)
4541 IA64_ONLY(fork();)
4542
4543 if (pid < 0) {
4544 // fork failed
4545 return -1;
4546
4547 } else if (pid == 0) {
4548 // child process
4549
4550 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4551 // first to kill every thread on the thread list. Because this list is
4552 // not reset by fork() (see notes above), execve() will instead kill
4553 // every thread in the parent process. We know this is the only thread
4554 // in the new process, so make a system call directly.
4555 // IA64 should use normal execve() from glibc to match the glibc fork()
4556 // above.
4557 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4558 IA64_ONLY(execve("/bin/sh", argv, environ);)
4559
4560 // execve failed
4561 _exit(-1);
4562
4563 } else {
4564 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4565 // care about the actual exit code, for now.
4566
4567 int status;
4568
4569 // Wait for the child process to exit. This returns immediately if
4570 // the child has already exited. */
4571 while (waitpid(pid, &status, 0) < 0) {
4572 switch (errno) {
4573 case ECHILD: return 0;
4574 case EINTR: break;
4575 default: return -1;
4576 }
4577 }
4578
4579 if (WIFEXITED(status)) {
4580 // The child exited normally; get its exit code.
4581 return WEXITSTATUS(status);
4582 } else if (WIFSIGNALED(status)) {
4583 // The child exited because of a signal
4584 // The best value to return is 0x80 + signal number,
4585 // because that is what all Unix shells do, and because
4586 // it allows callers to distinguish between process exit and
4587 // process death by signal.
4588 return 0x80 + WTERMSIG(status);
4589 } else {
4590 // Unknown exit code; pass it through
4591 return status;
4592 }
4593 }
4594 }