1 /*
2 * Copyright (c) 1999, 2009, Oracle and/or its affiliates. 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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 # define __STDC_FORMAT_MACROS
26
27 // do not include precompiled header file
28 # include "incls/_os_linux.cpp.incl"
29
30 // put OS-includes here
31 # include <sys/types.h>
32 # include <sys/mman.h>
33 # include <sys/stat.h>
34 # include <sys/select.h>
35 # include <pthread.h>
36 # include <signal.h>
37 # include <errno.h>
38 # include <dlfcn.h>
39 # include <stdio.h>
40 # include <unistd.h>
41 # include <sys/resource.h>
42 # include <pthread.h>
43 # include <sys/stat.h>
44 # include <sys/time.h>
45 # include <sys/times.h>
46 # include <sys/utsname.h>
47 # include <sys/socket.h>
48 # include <sys/wait.h>
49 # include <pwd.h>
50 # include <poll.h>
51 # include <semaphore.h>
52 # include <fcntl.h>
53 # include <string.h>
54 # include <syscall.h>
55 # include <sys/sysinfo.h>
56 # include <gnu/libc-version.h>
57 # include <sys/ipc.h>
58 # include <sys/shm.h>
59 # include <link.h>
60 # include <stdint.h>
61 # include <inttypes.h>
62
63 #define MAX_PATH (2 * K)
64
65 // for timer info max values which include all bits
66 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
67 #define SEC_IN_NANOSECS 1000000000LL
68
69 ////////////////////////////////////////////////////////////////////////////////
70 // global variables
71 julong os::Linux::_physical_memory = 0;
72
73 address os::Linux::_initial_thread_stack_bottom = NULL;
74 uintptr_t os::Linux::_initial_thread_stack_size = 0;
75
76 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
77 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
78 Mutex* os::Linux::_createThread_lock = NULL;
79 pthread_t os::Linux::_main_thread;
80 int os::Linux::_page_size = -1;
81 bool os::Linux::_is_floating_stack = false;
82 bool os::Linux::_is_NPTL = false;
83 bool os::Linux::_supports_fast_thread_cpu_time = false;
84 const char * os::Linux::_glibc_version = NULL;
85 const char * os::Linux::_libpthread_version = NULL;
86
87 static jlong initial_time_count=0;
88
89 static int clock_tics_per_sec = 100;
90
91 // For diagnostics to print a message once. see run_periodic_checks
92 static sigset_t check_signal_done;
93 static bool check_signals = true;;
94
95 static pid_t _initial_pid = 0;
96
97 /* Signal number used to suspend/resume a thread */
98
99 /* do not use any signal number less than SIGSEGV, see 4355769 */
100 static int SR_signum = SIGUSR2;
101 sigset_t SR_sigset;
102
103 /* Used to protect dlsym() calls */
104 static pthread_mutex_t dl_mutex;
105
106 ////////////////////////////////////////////////////////////////////////////////
107 // utility functions
108
109 static int SR_initialize();
110 static int SR_finalize();
111
112 julong os::available_memory() {
113 return Linux::available_memory();
114 }
115
116 julong os::Linux::available_memory() {
117 // values in struct sysinfo are "unsigned long"
118 struct sysinfo si;
119 sysinfo(&si);
120
121 return (julong)si.freeram * si.mem_unit;
122 }
123
124 julong os::physical_memory() {
125 return Linux::physical_memory();
126 }
127
128 julong os::allocatable_physical_memory(julong size) {
129 #ifdef _LP64
130 return size;
131 #else
132 julong result = MIN2(size, (julong)3800*M);
133 if (!is_allocatable(result)) {
134 // See comments under solaris for alignment considerations
135 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
136 result = MIN2(size, reasonable_size);
137 }
138 return result;
139 #endif // _LP64
140 }
141
142 ////////////////////////////////////////////////////////////////////////////////
143 // environment support
144
145 bool os::getenv(const char* name, char* buf, int len) {
146 const char* val = ::getenv(name);
147 if (val != NULL && strlen(val) < (size_t)len) {
148 strcpy(buf, val);
149 return true;
150 }
151 if (len > 0) buf[0] = 0; // return a null string
152 return false;
153 }
154
155
156 // Return true if user is running as root.
157
158 bool os::have_special_privileges() {
159 static bool init = false;
160 static bool privileges = false;
161 if (!init) {
162 privileges = (getuid() != geteuid()) || (getgid() != getegid());
163 init = true;
164 }
165 return privileges;
166 }
167
168
169 #ifndef SYS_gettid
170 // i386: 224, ia64: 1105, amd64: 186, sparc 143
171 #ifdef __ia64__
172 #define SYS_gettid 1105
173 #elif __i386__
174 #define SYS_gettid 224
175 #elif __amd64__
176 #define SYS_gettid 186
177 #elif __sparc__
178 #define SYS_gettid 143
179 #else
180 #error define gettid for the arch
181 #endif
182 #endif
183
184 // Cpu architecture string
185 #if defined(ZERO)
186 static char cpu_arch[] = ZERO_LIBARCH;
187 #elif defined(IA64)
188 static char cpu_arch[] = "ia64";
189 #elif defined(IA32)
190 static char cpu_arch[] = "i386";
191 #elif defined(AMD64)
192 static char cpu_arch[] = "amd64";
193 #elif defined(ARM)
194 static char cpu_arch[] = "arm";
195 #elif defined(PPC)
196 static char cpu_arch[] = "ppc";
197 #elif defined(SPARC)
198 # ifdef _LP64
199 static char cpu_arch[] = "sparcv9";
200 # else
201 static char cpu_arch[] = "sparc";
202 # endif
203 #else
204 #error Add appropriate cpu_arch setting
205 #endif
206
207
208 // pid_t gettid()
209 //
210 // Returns the kernel thread id of the currently running thread. Kernel
211 // thread id is used to access /proc.
212 //
213 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
214 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
215 //
216 pid_t os::Linux::gettid() {
217 int rslt = syscall(SYS_gettid);
218 if (rslt == -1) {
219 // old kernel, no NPTL support
220 return getpid();
221 } else {
222 return (pid_t)rslt;
223 }
224 }
225
226 // Most versions of linux have a bug where the number of processors are
227 // determined by looking at the /proc file system. In a chroot environment,
228 // the system call returns 1. This causes the VM to act as if it is
229 // a single processor and elide locking (see is_MP() call).
230 static bool unsafe_chroot_detected = false;
231 static const char *unstable_chroot_error = "/proc file system not found.\n"
232 "Java may be unstable running multithreaded in a chroot "
233 "environment on Linux when /proc filesystem is not mounted.";
234
235 void os::Linux::initialize_system_info() {
236 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
237 if (processor_count() == 1) {
238 pid_t pid = os::Linux::gettid();
239 char fname[32];
240 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
241 FILE *fp = fopen(fname, "r");
242 if (fp == NULL) {
243 unsafe_chroot_detected = true;
244 } else {
245 fclose(fp);
246 }
247 }
248 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
249 assert(processor_count() > 0, "linux error");
250 }
251
252 void os::init_system_properties_values() {
253 // char arch[12];
254 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
255
256 // The next steps are taken in the product version:
257 //
258 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
259 // This library should be located at:
260 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
261 //
262 // If "/jre/lib/" appears at the right place in the path, then we
263 // assume libjvm[_g].so is installed in a JDK and we use this path.
264 //
265 // Otherwise exit with message: "Could not create the Java virtual machine."
266 //
267 // The following extra steps are taken in the debugging version:
268 //
269 // If "/jre/lib/" does NOT appear at the right place in the path
270 // instead of exit check for $JAVA_HOME environment variable.
271 //
272 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
273 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
274 // it looks like libjvm[_g].so is installed there
275 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
276 //
277 // Otherwise exit.
278 //
279 // Important note: if the location of libjvm.so changes this
280 // code needs to be changed accordingly.
281
282 // The next few definitions allow the code to be verbatim:
283 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
284 #define getenv(n) ::getenv(n)
285
286 /*
287 * See ld(1):
288 * The linker uses the following search paths to locate required
289 * shared libraries:
290 * 1: ...
291 * ...
292 * 7: The default directories, normally /lib and /usr/lib.
293 */
294 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
295 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
296 #else
297 #define DEFAULT_LIBPATH "/lib:/usr/lib"
298 #endif
299
300 #define EXTENSIONS_DIR "/lib/ext"
301 #define ENDORSED_DIR "/lib/endorsed"
302 #define REG_DIR "/usr/java/packages"
303
304 {
305 /* sysclasspath, java_home, dll_dir */
306 {
307 char *home_path;
308 char *dll_path;
309 char *pslash;
310 char buf[MAXPATHLEN];
311 os::jvm_path(buf, sizeof(buf));
312
313 // Found the full path to libjvm.so.
314 // Now cut the path to <java_home>/jre if we can.
315 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
316 pslash = strrchr(buf, '/');
317 if (pslash != NULL)
318 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
319 dll_path = malloc(strlen(buf) + 1);
320 if (dll_path == NULL)
321 return;
322 strcpy(dll_path, buf);
323 Arguments::set_dll_dir(dll_path);
324
325 if (pslash != NULL) {
326 pslash = strrchr(buf, '/');
327 if (pslash != NULL) {
328 *pslash = '\0'; /* get rid of /<arch> */
329 pslash = strrchr(buf, '/');
330 if (pslash != NULL)
331 *pslash = '\0'; /* get rid of /lib */
332 }
333 }
334
335 home_path = malloc(strlen(buf) + 1);
336 if (home_path == NULL)
337 return;
338 strcpy(home_path, buf);
339 Arguments::set_java_home(home_path);
340
341 if (!set_boot_path('/', ':'))
342 return;
343 }
344
345 /*
346 * Where to look for native libraries
347 *
348 * Note: Due to a legacy implementation, most of the library path
349 * is set in the launcher. This was to accomodate linking restrictions
350 * on legacy Linux implementations (which are no longer supported).
351 * Eventually, all the library path setting will be done here.
352 *
353 * However, to prevent the proliferation of improperly built native
354 * libraries, the new path component /usr/java/packages is added here.
355 * Eventually, all the library path setting will be done here.
356 */
357 {
358 char *ld_library_path;
359
360 /*
361 * Construct the invariant part of ld_library_path. Note that the
362 * space for the colon and the trailing null are provided by the
363 * nulls included by the sizeof operator (so actually we allocate
364 * a byte more than necessary).
365 */
366 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
367 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
368 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
369
370 /*
371 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
372 * should always exist (until the legacy problem cited above is
373 * addressed).
374 */
375 char *v = getenv("LD_LIBRARY_PATH");
376 if (v != NULL) {
377 char *t = ld_library_path;
378 /* That's +1 for the colon and +1 for the trailing '\0' */
379 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
380 sprintf(ld_library_path, "%s:%s", v, t);
381 }
382 Arguments::set_library_path(ld_library_path);
383 }
384
385 /*
386 * Extensions directories.
387 *
388 * Note that the space for the colon and the trailing null are provided
389 * by the nulls included by the sizeof operator (so actually one byte more
390 * than necessary is allocated).
391 */
392 {
393 char *buf = malloc(strlen(Arguments::get_java_home()) +
394 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
395 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
396 Arguments::get_java_home());
397 Arguments::set_ext_dirs(buf);
398 }
399
400 /* Endorsed standards default directory. */
401 {
402 char * buf;
403 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
404 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
405 Arguments::set_endorsed_dirs(buf);
406 }
407 }
408
409 #undef malloc
410 #undef getenv
411 #undef EXTENSIONS_DIR
412 #undef ENDORSED_DIR
413
414 // Done
415 return;
416 }
417
418 ////////////////////////////////////////////////////////////////////////////////
419 // breakpoint support
420
421 void os::breakpoint() {
422 BREAKPOINT;
423 }
424
425 extern "C" void breakpoint() {
426 // use debugger to set breakpoint here
427 }
428
429 ////////////////////////////////////////////////////////////////////////////////
430 // signal support
431
432 debug_only(static bool signal_sets_initialized = false);
433 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
434
435 bool os::Linux::is_sig_ignored(int sig) {
436 struct sigaction oact;
437 sigaction(sig, (struct sigaction*)NULL, &oact);
438 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
439 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
440 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
441 return true;
442 else
443 return false;
444 }
445
446 void os::Linux::signal_sets_init() {
447 // Should also have an assertion stating we are still single-threaded.
448 assert(!signal_sets_initialized, "Already initialized");
449 // Fill in signals that are necessarily unblocked for all threads in
450 // the VM. Currently, we unblock the following signals:
451 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
452 // by -Xrs (=ReduceSignalUsage));
453 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
454 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
455 // the dispositions or masks wrt these signals.
456 // Programs embedding the VM that want to use the above signals for their
457 // own purposes must, at this time, use the "-Xrs" option to prevent
458 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
459 // (See bug 4345157, and other related bugs).
460 // In reality, though, unblocking these signals is really a nop, since
461 // these signals are not blocked by default.
462 sigemptyset(&unblocked_sigs);
463 sigemptyset(&allowdebug_blocked_sigs);
464 sigaddset(&unblocked_sigs, SIGILL);
465 sigaddset(&unblocked_sigs, SIGSEGV);
466 sigaddset(&unblocked_sigs, SIGBUS);
467 sigaddset(&unblocked_sigs, SIGFPE);
468 sigaddset(&unblocked_sigs, SR_signum);
469
470 if (!ReduceSignalUsage) {
471 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
472 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
473 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
474 }
475 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
476 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
477 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
478 }
479 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
480 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
481 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
482 }
483 }
484 // Fill in signals that are blocked by all but the VM thread.
485 sigemptyset(&vm_sigs);
486 if (!ReduceSignalUsage)
487 sigaddset(&vm_sigs, BREAK_SIGNAL);
488 debug_only(signal_sets_initialized = true);
489
490 }
491
492 // These are signals that are unblocked while a thread is running Java.
493 // (For some reason, they get blocked by default.)
494 sigset_t* os::Linux::unblocked_signals() {
495 assert(signal_sets_initialized, "Not initialized");
496 return &unblocked_sigs;
497 }
498
499 // These are the signals that are blocked while a (non-VM) thread is
500 // running Java. Only the VM thread handles these signals.
501 sigset_t* os::Linux::vm_signals() {
502 assert(signal_sets_initialized, "Not initialized");
503 return &vm_sigs;
504 }
505
506 // These are signals that are blocked during cond_wait to allow debugger in
507 sigset_t* os::Linux::allowdebug_blocked_signals() {
508 assert(signal_sets_initialized, "Not initialized");
509 return &allowdebug_blocked_sigs;
510 }
511
512 void os::Linux::hotspot_sigmask(Thread* thread) {
513
514 //Save caller's signal mask before setting VM signal mask
515 sigset_t caller_sigmask;
516 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
517
518 OSThread* osthread = thread->osthread();
519 osthread->set_caller_sigmask(caller_sigmask);
520
521 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
522
523 if (!ReduceSignalUsage) {
524 if (thread->is_VM_thread()) {
525 // Only the VM thread handles BREAK_SIGNAL ...
526 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
527 } else {
528 // ... all other threads block BREAK_SIGNAL
529 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
530 }
531 }
532 }
533
534 //////////////////////////////////////////////////////////////////////////////
535 // detecting pthread library
536
537 void os::Linux::libpthread_init() {
538 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
539 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
540 // generic name for earlier versions.
541 // Define macros here so we can build HotSpot on old systems.
542 # ifndef _CS_GNU_LIBC_VERSION
543 # define _CS_GNU_LIBC_VERSION 2
544 # endif
545 # ifndef _CS_GNU_LIBPTHREAD_VERSION
546 # define _CS_GNU_LIBPTHREAD_VERSION 3
547 # endif
548
549 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
550 if (n > 0) {
551 char *str = (char *)malloc(n);
552 confstr(_CS_GNU_LIBC_VERSION, str, n);
553 os::Linux::set_glibc_version(str);
554 } else {
555 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
556 static char _gnu_libc_version[32];
557 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
558 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
559 os::Linux::set_glibc_version(_gnu_libc_version);
560 }
561
562 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
563 if (n > 0) {
564 char *str = (char *)malloc(n);
565 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
566 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
567 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
568 // is the case. LinuxThreads has a hard limit on max number of threads.
569 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
570 // On the other hand, NPTL does not have such a limit, sysconf()
571 // will return -1 and errno is not changed. Check if it is really NPTL.
572 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
573 strstr(str, "NPTL") &&
574 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
575 free(str);
576 os::Linux::set_libpthread_version("linuxthreads");
577 } else {
578 os::Linux::set_libpthread_version(str);
579 }
580 } else {
581 // glibc before 2.3.2 only has LinuxThreads.
582 os::Linux::set_libpthread_version("linuxthreads");
583 }
584
585 if (strstr(libpthread_version(), "NPTL")) {
586 os::Linux::set_is_NPTL();
587 } else {
588 os::Linux::set_is_LinuxThreads();
589 }
590
591 // LinuxThreads have two flavors: floating-stack mode, which allows variable
592 // stack size; and fixed-stack mode. NPTL is always floating-stack.
593 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
594 os::Linux::set_is_floating_stack();
595 }
596 }
597
598 /////////////////////////////////////////////////////////////////////////////
599 // thread stack
600
601 // Force Linux kernel to expand current thread stack. If "bottom" is close
602 // to the stack guard, caller should block all signals.
603 //
604 // MAP_GROWSDOWN:
605 // A special mmap() flag that is used to implement thread stacks. It tells
606 // kernel that the memory region should extend downwards when needed. This
607 // allows early versions of LinuxThreads to only mmap the first few pages
608 // when creating a new thread. Linux kernel will automatically expand thread
609 // stack as needed (on page faults).
610 //
611 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
612 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
613 // region, it's hard to tell if the fault is due to a legitimate stack
614 // access or because of reading/writing non-exist memory (e.g. buffer
615 // overrun). As a rule, if the fault happens below current stack pointer,
616 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
617 // application (see Linux kernel fault.c).
618 //
619 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
620 // stack overflow detection.
621 //
622 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
623 // not use this flag. However, the stack of initial thread is not created
624 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
625 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
626 // and then attach the thread to JVM.
627 //
628 // To get around the problem and allow stack banging on Linux, we need to
629 // manually expand thread stack after receiving the SIGSEGV.
630 //
631 // There are two ways to expand thread stack to address "bottom", we used
632 // both of them in JVM before 1.5:
633 // 1. adjust stack pointer first so that it is below "bottom", and then
634 // touch "bottom"
635 // 2. mmap() the page in question
636 //
637 // Now alternate signal stack is gone, it's harder to use 2. For instance,
638 // if current sp is already near the lower end of page 101, and we need to
639 // call mmap() to map page 100, it is possible that part of the mmap() frame
640 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
641 // That will destroy the mmap() frame and cause VM to crash.
642 //
643 // The following code works by adjusting sp first, then accessing the "bottom"
644 // page to force a page fault. Linux kernel will then automatically expand the
645 // stack mapping.
646 //
647 // _expand_stack_to() assumes its frame size is less than page size, which
648 // should always be true if the function is not inlined.
649
650 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
651 #define NOINLINE
652 #else
653 #define NOINLINE __attribute__ ((noinline))
654 #endif
655
656 static void _expand_stack_to(address bottom) NOINLINE;
657
658 static void _expand_stack_to(address bottom) {
659 address sp;
660 size_t size;
661 volatile char *p;
662
663 // Adjust bottom to point to the largest address within the same page, it
664 // gives us a one-page buffer if alloca() allocates slightly more memory.
665 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
666 bottom += os::Linux::page_size() - 1;
667
668 // sp might be slightly above current stack pointer; if that's the case, we
669 // will alloca() a little more space than necessary, which is OK. Don't use
670 // os::current_stack_pointer(), as its result can be slightly below current
671 // stack pointer, causing us to not alloca enough to reach "bottom".
672 sp = (address)&sp;
673
674 if (sp > bottom) {
675 size = sp - bottom;
676 p = (volatile char *)alloca(size);
677 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
678 p[0] = '\0';
679 }
680 }
681
682 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
683 assert(t!=NULL, "just checking");
684 assert(t->osthread()->expanding_stack(), "expand should be set");
685 assert(t->stack_base() != NULL, "stack_base was not initialized");
686
687 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
688 sigset_t mask_all, old_sigset;
689 sigfillset(&mask_all);
690 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
691 _expand_stack_to(addr);
692 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
693 return true;
694 }
695 return false;
696 }
697
698 //////////////////////////////////////////////////////////////////////////////
699 // create new thread
700
701 static address highest_vm_reserved_address();
702
703 // check if it's safe to start a new thread
704 static bool _thread_safety_check(Thread* thread) {
705 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
706 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
707 // Heap is mmap'ed at lower end of memory space. Thread stacks are
708 // allocated (MAP_FIXED) from high address space. Every thread stack
709 // occupies a fixed size slot (usually 2Mbytes, but user can change
710 // it to other values if they rebuild LinuxThreads).
711 //
712 // Problem with MAP_FIXED is that mmap() can still succeed even part of
713 // the memory region has already been mmap'ed. That means if we have too
714 // many threads and/or very large heap, eventually thread stack will
715 // collide with heap.
716 //
717 // Here we try to prevent heap/stack collision by comparing current
718 // stack bottom with the highest address that has been mmap'ed by JVM
719 // plus a safety margin for memory maps created by native code.
720 //
721 // This feature can be disabled by setting ThreadSafetyMargin to 0
722 //
723 if (ThreadSafetyMargin > 0) {
724 address stack_bottom = os::current_stack_base() - os::current_stack_size();
725
726 // not safe if our stack extends below the safety margin
727 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
728 } else {
729 return true;
730 }
731 } else {
732 // Floating stack LinuxThreads or NPTL:
733 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
734 // there's not enough space left, pthread_create() will fail. If we come
735 // here, that means enough space has been reserved for stack.
736 return true;
737 }
738 }
739
740 // Thread start routine for all newly created threads
741 static void *java_start(Thread *thread) {
742 // Try to randomize the cache line index of hot stack frames.
743 // This helps when threads of the same stack traces evict each other's
744 // cache lines. The threads can be either from the same JVM instance, or
745 // from different JVM instances. The benefit is especially true for
746 // processors with hyperthreading technology.
747 static int counter = 0;
748 int pid = os::current_process_id();
749 alloca(((pid ^ counter++) & 7) * 128);
750
751 ThreadLocalStorage::set_thread(thread);
752
753 OSThread* osthread = thread->osthread();
754 Monitor* sync = osthread->startThread_lock();
755
756 // non floating stack LinuxThreads needs extra check, see above
757 if (!_thread_safety_check(thread)) {
758 // notify parent thread
759 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
760 osthread->set_state(ZOMBIE);
761 sync->notify_all();
762 return NULL;
763 }
764
765 // thread_id is kernel thread id (similar to Solaris LWP id)
766 osthread->set_thread_id(os::Linux::gettid());
767
768 if (UseNUMA) {
769 int lgrp_id = os::numa_get_group_id();
770 if (lgrp_id != -1) {
771 thread->set_lgrp_id(lgrp_id);
772 }
773 }
774 // initialize signal mask for this thread
775 os::Linux::hotspot_sigmask(thread);
776
777 // initialize floating point control register
778 os::Linux::init_thread_fpu_state();
779
780 // handshaking with parent thread
781 {
782 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
783
784 // notify parent thread
785 osthread->set_state(INITIALIZED);
786 sync->notify_all();
787
788 // wait until os::start_thread()
789 while (osthread->get_state() == INITIALIZED) {
790 sync->wait(Mutex::_no_safepoint_check_flag);
791 }
792 }
793
794 // call one more level start routine
795 thread->run();
796
797 return 0;
798 }
799
800 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
801 assert(thread->osthread() == NULL, "caller responsible");
802
803 // Allocate the OSThread object
804 OSThread* osthread = new OSThread(NULL, NULL);
805 if (osthread == NULL) {
806 return false;
807 }
808
809 // set the correct thread state
810 osthread->set_thread_type(thr_type);
811
812 // Initial state is ALLOCATED but not INITIALIZED
813 osthread->set_state(ALLOCATED);
814
815 thread->set_osthread(osthread);
816
817 // init thread attributes
818 pthread_attr_t attr;
819 pthread_attr_init(&attr);
820 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
821
822 // stack size
823 if (os::Linux::supports_variable_stack_size()) {
824 // calculate stack size if it's not specified by caller
825 if (stack_size == 0) {
826 stack_size = os::Linux::default_stack_size(thr_type);
827
828 switch (thr_type) {
829 case os::java_thread:
830 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
831 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
832 break;
833 case os::compiler_thread:
834 if (CompilerThreadStackSize > 0) {
835 stack_size = (size_t)(CompilerThreadStackSize * K);
836 break;
837 } // else fall through:
838 // use VMThreadStackSize if CompilerThreadStackSize is not defined
839 case os::vm_thread:
840 case os::pgc_thread:
841 case os::cgc_thread:
842 case os::watcher_thread:
843 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
844 break;
845 }
846 }
847
848 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
849 pthread_attr_setstacksize(&attr, stack_size);
850 } else {
851 // let pthread_create() pick the default value.
852 }
853
854 // glibc guard page
855 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
856
857 ThreadState state;
858
859 {
860 // Serialize thread creation if we are running with fixed stack LinuxThreads
861 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
862 if (lock) {
863 os::Linux::createThread_lock()->lock_without_safepoint_check();
864 }
865
866 pthread_t tid;
867 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
868
869 pthread_attr_destroy(&attr);
870
871 if (ret != 0) {
872 if (PrintMiscellaneous && (Verbose || WizardMode)) {
873 perror("pthread_create()");
874 }
875 // Need to clean up stuff we've allocated so far
876 thread->set_osthread(NULL);
877 delete osthread;
878 if (lock) os::Linux::createThread_lock()->unlock();
879 return false;
880 }
881
882 // Store pthread info into the OSThread
883 osthread->set_pthread_id(tid);
884
885 // Wait until child thread is either initialized or aborted
886 {
887 Monitor* sync_with_child = osthread->startThread_lock();
888 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
889 while ((state = osthread->get_state()) == ALLOCATED) {
890 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
891 }
892 }
893
894 if (lock) {
895 os::Linux::createThread_lock()->unlock();
896 }
897 }
898
899 // Aborted due to thread limit being reached
900 if (state == ZOMBIE) {
901 thread->set_osthread(NULL);
902 delete osthread;
903 return false;
904 }
905
906 // The thread is returned suspended (in state INITIALIZED),
907 // and is started higher up in the call chain
908 assert(state == INITIALIZED, "race condition");
909 return true;
910 }
911
912 /////////////////////////////////////////////////////////////////////////////
913 // attach existing thread
914
915 // bootstrap the main thread
916 bool os::create_main_thread(JavaThread* thread) {
917 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
918 return create_attached_thread(thread);
919 }
920
921 bool os::create_attached_thread(JavaThread* thread) {
922 #ifdef ASSERT
923 thread->verify_not_published();
924 #endif
925
926 // Allocate the OSThread object
927 OSThread* osthread = new OSThread(NULL, NULL);
928
929 if (osthread == NULL) {
930 return false;
931 }
932
933 // Store pthread info into the OSThread
934 osthread->set_thread_id(os::Linux::gettid());
935 osthread->set_pthread_id(::pthread_self());
936
937 // initialize floating point control register
938 os::Linux::init_thread_fpu_state();
939
940 // Initial thread state is RUNNABLE
941 osthread->set_state(RUNNABLE);
942
943 thread->set_osthread(osthread);
944
945 if (UseNUMA) {
946 int lgrp_id = os::numa_get_group_id();
947 if (lgrp_id != -1) {
948 thread->set_lgrp_id(lgrp_id);
949 }
950 }
951
952 if (os::Linux::is_initial_thread()) {
953 // If current thread is initial thread, its stack is mapped on demand,
954 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
955 // the entire stack region to avoid SEGV in stack banging.
956 // It is also useful to get around the heap-stack-gap problem on SuSE
957 // kernel (see 4821821 for details). We first expand stack to the top
958 // of yellow zone, then enable stack yellow zone (order is significant,
959 // enabling yellow zone first will crash JVM on SuSE Linux), so there
960 // is no gap between the last two virtual memory regions.
961
962 JavaThread *jt = (JavaThread *)thread;
963 address addr = jt->stack_yellow_zone_base();
964 assert(addr != NULL, "initialization problem?");
965 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
966
967 osthread->set_expanding_stack();
968 os::Linux::manually_expand_stack(jt, addr);
969 osthread->clear_expanding_stack();
970 }
971
972 // initialize signal mask for this thread
973 // and save the caller's signal mask
974 os::Linux::hotspot_sigmask(thread);
975
976 return true;
977 }
978
979 void os::pd_start_thread(Thread* thread) {
980 OSThread * osthread = thread->osthread();
981 assert(osthread->get_state() != INITIALIZED, "just checking");
982 Monitor* sync_with_child = osthread->startThread_lock();
983 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
984 sync_with_child->notify();
985 }
986
987 // Free Linux resources related to the OSThread
988 void os::free_thread(OSThread* osthread) {
989 assert(osthread != NULL, "osthread not set");
990
991 if (Thread::current()->osthread() == osthread) {
992 // Restore caller's signal mask
993 sigset_t sigmask = osthread->caller_sigmask();
994 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
995 }
996
997 delete osthread;
998 }
999
1000 //////////////////////////////////////////////////////////////////////////////
1001 // thread local storage
1002
1003 int os::allocate_thread_local_storage() {
1004 pthread_key_t key;
1005 int rslt = pthread_key_create(&key, NULL);
1006 assert(rslt == 0, "cannot allocate thread local storage");
1007 return (int)key;
1008 }
1009
1010 // Note: This is currently not used by VM, as we don't destroy TLS key
1011 // on VM exit.
1012 void os::free_thread_local_storage(int index) {
1013 int rslt = pthread_key_delete((pthread_key_t)index);
1014 assert(rslt == 0, "invalid index");
1015 }
1016
1017 void os::thread_local_storage_at_put(int index, void* value) {
1018 int rslt = pthread_setspecific((pthread_key_t)index, value);
1019 assert(rslt == 0, "pthread_setspecific failed");
1020 }
1021
1022 extern "C" Thread* get_thread() {
1023 return ThreadLocalStorage::thread();
1024 }
1025
1026 //////////////////////////////////////////////////////////////////////////////
1027 // initial thread
1028
1029 // Check if current thread is the initial thread, similar to Solaris thr_main.
1030 bool os::Linux::is_initial_thread(void) {
1031 char dummy;
1032 // If called before init complete, thread stack bottom will be null.
1033 // Can be called if fatal error occurs before initialization.
1034 if (initial_thread_stack_bottom() == NULL) return false;
1035 assert(initial_thread_stack_bottom() != NULL &&
1036 initial_thread_stack_size() != 0,
1037 "os::init did not locate initial thread's stack region");
1038 if ((address)&dummy >= initial_thread_stack_bottom() &&
1039 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1040 return true;
1041 else return false;
1042 }
1043
1044 // Find the virtual memory area that contains addr
1045 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1046 FILE *fp = fopen("/proc/self/maps", "r");
1047 if (fp) {
1048 address low, high;
1049 while (!feof(fp)) {
1050 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1051 if (low <= addr && addr < high) {
1052 if (vma_low) *vma_low = low;
1053 if (vma_high) *vma_high = high;
1054 fclose (fp);
1055 return true;
1056 }
1057 }
1058 for (;;) {
1059 int ch = fgetc(fp);
1060 if (ch == EOF || ch == (int)'\n') break;
1061 }
1062 }
1063 fclose(fp);
1064 }
1065 return false;
1066 }
1067
1068 // Locate initial thread stack. This special handling of initial thread stack
1069 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1070 // bogus value for initial thread.
1071 void os::Linux::capture_initial_stack(size_t max_size) {
1072 // stack size is the easy part, get it from RLIMIT_STACK
1073 size_t stack_size;
1074 struct rlimit rlim;
1075 getrlimit(RLIMIT_STACK, &rlim);
1076 stack_size = rlim.rlim_cur;
1077
1078 // 6308388: a bug in ld.so will relocate its own .data section to the
1079 // lower end of primordial stack; reduce ulimit -s value a little bit
1080 // so we won't install guard page on ld.so's data section.
1081 stack_size -= 2 * page_size();
1082
1083 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1084 // 7.1, in both cases we will get 2G in return value.
1085 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1086 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1087 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1088 // in case other parts in glibc still assumes 2M max stack size.
1089 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1090 #ifndef IA64
1091 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1092 #else
1093 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1094 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1095 #endif
1096
1097 // Try to figure out where the stack base (top) is. This is harder.
1098 //
1099 // When an application is started, glibc saves the initial stack pointer in
1100 // a global variable "__libc_stack_end", which is then used by system
1101 // libraries. __libc_stack_end should be pretty close to stack top. The
1102 // variable is available since the very early days. However, because it is
1103 // a private interface, it could disappear in the future.
1104 //
1105 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1106 // to __libc_stack_end, it is very close to stack top, but isn't the real
1107 // stack top. Note that /proc may not exist if VM is running as a chroot
1108 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1109 // /proc/<pid>/stat could change in the future (though unlikely).
1110 //
1111 // We try __libc_stack_end first. If that doesn't work, look for
1112 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1113 // as a hint, which should work well in most cases.
1114
1115 uintptr_t stack_start;
1116
1117 // try __libc_stack_end first
1118 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1119 if (p && *p) {
1120 stack_start = *p;
1121 } else {
1122 // see if we can get the start_stack field from /proc/self/stat
1123 FILE *fp;
1124 int pid;
1125 char state;
1126 int ppid;
1127 int pgrp;
1128 int session;
1129 int nr;
1130 int tpgrp;
1131 unsigned long flags;
1132 unsigned long minflt;
1133 unsigned long cminflt;
1134 unsigned long majflt;
1135 unsigned long cmajflt;
1136 unsigned long utime;
1137 unsigned long stime;
1138 long cutime;
1139 long cstime;
1140 long prio;
1141 long nice;
1142 long junk;
1143 long it_real;
1144 uintptr_t start;
1145 uintptr_t vsize;
1146 intptr_t rss;
1147 uintptr_t rsslim;
1148 uintptr_t scodes;
1149 uintptr_t ecode;
1150 int i;
1151
1152 // Figure what the primordial thread stack base is. Code is inspired
1153 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1154 // followed by command name surrounded by parentheses, state, etc.
1155 char stat[2048];
1156 int statlen;
1157
1158 fp = fopen("/proc/self/stat", "r");
1159 if (fp) {
1160 statlen = fread(stat, 1, 2047, fp);
1161 stat[statlen] = '\0';
1162 fclose(fp);
1163
1164 // Skip pid and the command string. Note that we could be dealing with
1165 // weird command names, e.g. user could decide to rename java launcher
1166 // to "java 1.4.2 :)", then the stat file would look like
1167 // 1234 (java 1.4.2 :)) R ... ...
1168 // We don't really need to know the command string, just find the last
1169 // occurrence of ")" and then start parsing from there. See bug 4726580.
1170 char * s = strrchr(stat, ')');
1171
1172 i = 0;
1173 if (s) {
1174 // Skip blank chars
1175 do s++; while (isspace(*s));
1176
1177 #define _UFM UINTX_FORMAT
1178 #define _DFM INTX_FORMAT
1179
1180 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1181 /* 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 */
1182 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1183 &state, /* 3 %c */
1184 &ppid, /* 4 %d */
1185 &pgrp, /* 5 %d */
1186 &session, /* 6 %d */
1187 &nr, /* 7 %d */
1188 &tpgrp, /* 8 %d */
1189 &flags, /* 9 %lu */
1190 &minflt, /* 10 %lu */
1191 &cminflt, /* 11 %lu */
1192 &majflt, /* 12 %lu */
1193 &cmajflt, /* 13 %lu */
1194 &utime, /* 14 %lu */
1195 &stime, /* 15 %lu */
1196 &cutime, /* 16 %ld */
1197 &cstime, /* 17 %ld */
1198 &prio, /* 18 %ld */
1199 &nice, /* 19 %ld */
1200 &junk, /* 20 %ld */
1201 &it_real, /* 21 %ld */
1202 &start, /* 22 UINTX_FORMAT */
1203 &vsize, /* 23 UINTX_FORMAT */
1204 &rss, /* 24 INTX_FORMAT */
1205 &rsslim, /* 25 UINTX_FORMAT */
1206 &scodes, /* 26 UINTX_FORMAT */
1207 &ecode, /* 27 UINTX_FORMAT */
1208 &stack_start); /* 28 UINTX_FORMAT */
1209 }
1210
1211 #undef _UFM
1212 #undef _DFM
1213
1214 if (i != 28 - 2) {
1215 assert(false, "Bad conversion from /proc/self/stat");
1216 // product mode - assume we are the initial thread, good luck in the
1217 // embedded case.
1218 warning("Can't detect initial thread stack location - bad conversion");
1219 stack_start = (uintptr_t) &rlim;
1220 }
1221 } else {
1222 // For some reason we can't open /proc/self/stat (for example, running on
1223 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1224 // most cases, so don't abort:
1225 warning("Can't detect initial thread stack location - no /proc/self/stat");
1226 stack_start = (uintptr_t) &rlim;
1227 }
1228 }
1229
1230 // Now we have a pointer (stack_start) very close to the stack top, the
1231 // next thing to do is to figure out the exact location of stack top. We
1232 // can find out the virtual memory area that contains stack_start by
1233 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1234 // and its upper limit is the real stack top. (again, this would fail if
1235 // running inside chroot, because /proc may not exist.)
1236
1237 uintptr_t stack_top;
1238 address low, high;
1239 if (find_vma((address)stack_start, &low, &high)) {
1240 // success, "high" is the true stack top. (ignore "low", because initial
1241 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1242 stack_top = (uintptr_t)high;
1243 } else {
1244 // failed, likely because /proc/self/maps does not exist
1245 warning("Can't detect initial thread stack location - find_vma failed");
1246 // best effort: stack_start is normally within a few pages below the real
1247 // stack top, use it as stack top, and reduce stack size so we won't put
1248 // guard page outside stack.
1249 stack_top = stack_start;
1250 stack_size -= 16 * page_size();
1251 }
1252
1253 // stack_top could be partially down the page so align it
1254 stack_top = align_size_up(stack_top, page_size());
1255
1256 if (max_size && stack_size > max_size) {
1257 _initial_thread_stack_size = max_size;
1258 } else {
1259 _initial_thread_stack_size = stack_size;
1260 }
1261
1262 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1263 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1264 }
1265
1266 ////////////////////////////////////////////////////////////////////////////////
1267 // time support
1268
1269 // Time since start-up in seconds to a fine granularity.
1270 // Used by VMSelfDestructTimer and the MemProfiler.
1271 double os::elapsedTime() {
1272
1273 return (double)(os::elapsed_counter()) * 0.000001;
1274 }
1275
1276 jlong os::elapsed_counter() {
1277 timeval time;
1278 int status = gettimeofday(&time, NULL);
1279 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1280 }
1281
1282 jlong os::elapsed_frequency() {
1283 return (1000 * 1000);
1284 }
1285
1286 // For now, we say that linux does not support vtime. I have no idea
1287 // whether it can actually be made to (DLD, 9/13/05).
1288
1289 bool os::supports_vtime() { return false; }
1290 bool os::enable_vtime() { return false; }
1291 bool os::vtime_enabled() { return false; }
1292 double os::elapsedVTime() {
1293 // better than nothing, but not much
1294 return elapsedTime();
1295 }
1296
1297 jlong os::javaTimeMillis() {
1298 timeval time;
1299 int status = gettimeofday(&time, NULL);
1300 assert(status != -1, "linux error");
1301 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1302 }
1303
1304 #ifndef CLOCK_MONOTONIC
1305 #define CLOCK_MONOTONIC (1)
1306 #endif
1307
1308 void os::Linux::clock_init() {
1309 // we do dlopen's in this particular order due to bug in linux
1310 // dynamical loader (see 6348968) leading to crash on exit
1311 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1312 if (handle == NULL) {
1313 handle = dlopen("librt.so", RTLD_LAZY);
1314 }
1315
1316 if (handle) {
1317 int (*clock_getres_func)(clockid_t, struct timespec*) =
1318 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1319 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1320 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1321 if (clock_getres_func && clock_gettime_func) {
1322 // See if monotonic clock is supported by the kernel. Note that some
1323 // early implementations simply return kernel jiffies (updated every
1324 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1325 // for nano time (though the monotonic property is still nice to have).
1326 // It's fixed in newer kernels, however clock_getres() still returns
1327 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1328 // resolution for now. Hopefully as people move to new kernels, this
1329 // won't be a problem.
1330 struct timespec res;
1331 struct timespec tp;
1332 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1333 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1334 // yes, monotonic clock is supported
1335 _clock_gettime = clock_gettime_func;
1336 } else {
1337 // close librt if there is no monotonic clock
1338 dlclose(handle);
1339 }
1340 }
1341 }
1342 }
1343
1344 #ifndef SYS_clock_getres
1345
1346 #if defined(IA32) || defined(AMD64)
1347 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1348 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1349 #else
1350 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1351 #define sys_clock_getres(x,y) -1
1352 #endif
1353
1354 #else
1355 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1356 #endif
1357
1358 void os::Linux::fast_thread_clock_init() {
1359 if (!UseLinuxPosixThreadCPUClocks) {
1360 return;
1361 }
1362 clockid_t clockid;
1363 struct timespec tp;
1364 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1365 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1366
1367 // Switch to using fast clocks for thread cpu time if
1368 // the sys_clock_getres() returns 0 error code.
1369 // Note, that some kernels may support the current thread
1370 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1371 // returned by the pthread_getcpuclockid().
1372 // If the fast Posix clocks are supported then the sys_clock_getres()
1373 // must return at least tp.tv_sec == 0 which means a resolution
1374 // better than 1 sec. This is extra check for reliability.
1375
1376 if(pthread_getcpuclockid_func &&
1377 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1378 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1379
1380 _supports_fast_thread_cpu_time = true;
1381 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1382 }
1383 }
1384
1385 jlong os::javaTimeNanos() {
1386 if (Linux::supports_monotonic_clock()) {
1387 struct timespec tp;
1388 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1389 assert(status == 0, "gettime error");
1390 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1391 return result;
1392 } else {
1393 timeval time;
1394 int status = gettimeofday(&time, NULL);
1395 assert(status != -1, "linux error");
1396 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1397 return 1000 * usecs;
1398 }
1399 }
1400
1401 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1402 if (Linux::supports_monotonic_clock()) {
1403 info_ptr->max_value = ALL_64_BITS;
1404
1405 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1406 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1407 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1408 } else {
1409 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1410 info_ptr->max_value = ALL_64_BITS;
1411
1412 // gettimeofday is a real time clock so it skips
1413 info_ptr->may_skip_backward = true;
1414 info_ptr->may_skip_forward = true;
1415 }
1416
1417 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1418 }
1419
1420 // Return the real, user, and system times in seconds from an
1421 // arbitrary fixed point in the past.
1422 bool os::getTimesSecs(double* process_real_time,
1423 double* process_user_time,
1424 double* process_system_time) {
1425 struct tms ticks;
1426 clock_t real_ticks = times(&ticks);
1427
1428 if (real_ticks == (clock_t) (-1)) {
1429 return false;
1430 } else {
1431 double ticks_per_second = (double) clock_tics_per_sec;
1432 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1433 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1434 *process_real_time = ((double) real_ticks) / ticks_per_second;
1435
1436 return true;
1437 }
1438 }
1439
1440
1441 char * os::local_time_string(char *buf, size_t buflen) {
1442 struct tm t;
1443 time_t long_time;
1444 time(&long_time);
1445 localtime_r(&long_time, &t);
1446 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1447 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1448 t.tm_hour, t.tm_min, t.tm_sec);
1449 return buf;
1450 }
1451
1452 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1453 return localtime_r(clock, res);
1454 }
1455
1456 ////////////////////////////////////////////////////////////////////////////////
1457 // runtime exit support
1458
1459 // Note: os::shutdown() might be called very early during initialization, or
1460 // called from signal handler. Before adding something to os::shutdown(), make
1461 // sure it is async-safe and can handle partially initialized VM.
1462 void os::shutdown() {
1463
1464 // allow PerfMemory to attempt cleanup of any persistent resources
1465 perfMemory_exit();
1466
1467 // needs to remove object in file system
1468 AttachListener::abort();
1469
1470 // Check for abort hook
1471 abort_hook_t abort_hook = Arguments::abort_hook();
1472 if (abort_hook != NULL) {
1473 abort_hook();
1474 }
1475
1476 }
1477
1478 // Note: os::abort() might be called very early during initialization, or
1479 // called from signal handler. Before adding something to os::abort(), make
1480 // sure it is async-safe and can handle partially initialized VM.
1481 void os::abort(bool dump_core) {
1482 os::shutdown();
1483 if (dump_core) {
1484 #ifndef PRODUCT
1485 fdStream out(defaultStream::output_fd());
1486 out.print_raw("Current thread is ");
1487 char buf[16];
1488 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1489 out.print_raw_cr(buf);
1490 out.print_raw_cr("Dumping core ...");
1491 #endif
1492 ::abort(); // dump core
1493 }
1494
1495 ::exit(1);
1496 }
1497
1498 // Die immediately, no exit hook, no abort hook, no cleanup.
1499 void os::die() {
1500 // _exit() on LinuxThreads only kills current thread
1501 ::abort();
1502 }
1503
1504 // unused on linux for now.
1505 void os::set_error_file(const char *logfile) {}
1506
1507 intx os::current_thread_id() { return (intx)pthread_self(); }
1508 int os::current_process_id() {
1509
1510 // Under the old linux thread library, linux gives each thread
1511 // its own process id. Because of this each thread will return
1512 // a different pid if this method were to return the result
1513 // of getpid(2). Linux provides no api that returns the pid
1514 // of the launcher thread for the vm. This implementation
1515 // returns a unique pid, the pid of the launcher thread
1516 // that starts the vm 'process'.
1517
1518 // Under the NPTL, getpid() returns the same pid as the
1519 // launcher thread rather than a unique pid per thread.
1520 // Use gettid() if you want the old pre NPTL behaviour.
1521
1522 // if you are looking for the result of a call to getpid() that
1523 // returns a unique pid for the calling thread, then look at the
1524 // OSThread::thread_id() method in osThread_linux.hpp file
1525
1526 return (int)(_initial_pid ? _initial_pid : getpid());
1527 }
1528
1529 // DLL functions
1530
1531 const char* os::dll_file_extension() { return ".so"; }
1532
1533 const char* os::get_temp_directory() {
1534 const char *prop = Arguments::get_property("java.io.tmpdir");
1535 return prop == NULL ? "/tmp" : prop;
1536 }
1537
1538 static bool file_exists(const char* filename) {
1539 struct stat statbuf;
1540 if (filename == NULL || strlen(filename) == 0) {
1541 return false;
1542 }
1543 return os::stat(filename, &statbuf) == 0;
1544 }
1545
1546 void os::dll_build_name(char* buffer, size_t buflen,
1547 const char* pname, const char* fname) {
1548 // Copied from libhpi
1549 const size_t pnamelen = pname ? strlen(pname) : 0;
1550
1551 // Quietly truncate on buffer overflow. Should be an error.
1552 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1553 *buffer = '\0';
1554 return;
1555 }
1556
1557 if (pnamelen == 0) {
1558 snprintf(buffer, buflen, "lib%s.so", fname);
1559 } else if (strchr(pname, *os::path_separator()) != NULL) {
1560 int n;
1561 char** pelements = split_path(pname, &n);
1562 for (int i = 0 ; i < n ; i++) {
1563 // Really shouldn't be NULL, but check can't hurt
1564 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1565 continue; // skip the empty path values
1566 }
1567 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1568 if (file_exists(buffer)) {
1569 break;
1570 }
1571 }
1572 // release the storage
1573 for (int i = 0 ; i < n ; i++) {
1574 if (pelements[i] != NULL) {
1575 FREE_C_HEAP_ARRAY(char, pelements[i]);
1576 }
1577 }
1578 if (pelements != NULL) {
1579 FREE_C_HEAP_ARRAY(char*, pelements);
1580 }
1581 } else {
1582 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1583 }
1584 }
1585
1586 const char* os::get_current_directory(char *buf, int buflen) {
1587 return getcwd(buf, buflen);
1588 }
1589
1590 // check if addr is inside libjvm[_g].so
1591 bool os::address_is_in_vm(address addr) {
1592 static address libjvm_base_addr;
1593 Dl_info dlinfo;
1594
1595 if (libjvm_base_addr == NULL) {
1596 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1597 libjvm_base_addr = (address)dlinfo.dli_fbase;
1598 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1599 }
1600
1601 if (dladdr((void *)addr, &dlinfo)) {
1602 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1603 }
1604
1605 return false;
1606 }
1607
1608 bool os::dll_address_to_function_name(address addr, char *buf,
1609 int buflen, int *offset) {
1610 Dl_info dlinfo;
1611
1612 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1613 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1614 if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1615 return true;
1616 } else {
1617 if (buf) buf[0] = '\0';
1618 if (offset) *offset = -1;
1619 return false;
1620 }
1621 }
1622
1623 struct _address_to_library_name {
1624 address addr; // input : memory address
1625 size_t buflen; // size of fname
1626 char* fname; // output: library name
1627 address base; // library base addr
1628 };
1629
1630 static int address_to_library_name_callback(struct dl_phdr_info *info,
1631 size_t size, void *data) {
1632 int i;
1633 bool found = false;
1634 address libbase = NULL;
1635 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1636
1637 // iterate through all loadable segments
1638 for (i = 0; i < info->dlpi_phnum; i++) {
1639 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1640 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1641 // base address of a library is the lowest address of its loaded
1642 // segments.
1643 if (libbase == NULL || libbase > segbase) {
1644 libbase = segbase;
1645 }
1646 // see if 'addr' is within current segment
1647 if (segbase <= d->addr &&
1648 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1649 found = true;
1650 }
1651 }
1652 }
1653
1654 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1655 // so dll_address_to_library_name() can fall through to use dladdr() which
1656 // can figure out executable name from argv[0].
1657 if (found && info->dlpi_name && info->dlpi_name[0]) {
1658 d->base = libbase;
1659 if (d->fname) {
1660 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1661 }
1662 return 1;
1663 }
1664 return 0;
1665 }
1666
1667 bool os::dll_address_to_library_name(address addr, char* buf,
1668 int buflen, int* offset) {
1669 Dl_info dlinfo;
1670 struct _address_to_library_name data;
1671
1672 // There is a bug in old glibc dladdr() implementation that it could resolve
1673 // to wrong library name if the .so file has a base address != NULL. Here
1674 // we iterate through the program headers of all loaded libraries to find
1675 // out which library 'addr' really belongs to. This workaround can be
1676 // removed once the minimum requirement for glibc is moved to 2.3.x.
1677 data.addr = addr;
1678 data.fname = buf;
1679 data.buflen = buflen;
1680 data.base = NULL;
1681 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1682
1683 if (rslt) {
1684 // buf already contains library name
1685 if (offset) *offset = addr - data.base;
1686 return true;
1687 } else if (dladdr((void*)addr, &dlinfo)){
1688 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1689 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1690 return true;
1691 } else {
1692 if (buf) buf[0] = '\0';
1693 if (offset) *offset = -1;
1694 return false;
1695 }
1696 }
1697
1698 // Loads .dll/.so and
1699 // in case of error it checks if .dll/.so was built for the
1700 // same architecture as Hotspot is running on
1701
1702 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1703 {
1704 void * result= ::dlopen(filename, RTLD_LAZY);
1705 if (result != NULL) {
1706 // Successful loading
1707 return result;
1708 }
1709
1710 Elf32_Ehdr elf_head;
1711
1712 // Read system error message into ebuf
1713 // It may or may not be overwritten below
1714 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1715 ebuf[ebuflen-1]='\0';
1716 int diag_msg_max_length=ebuflen-strlen(ebuf);
1717 char* diag_msg_buf=ebuf+strlen(ebuf);
1718
1719 if (diag_msg_max_length==0) {
1720 // No more space in ebuf for additional diagnostics message
1721 return NULL;
1722 }
1723
1724
1725 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1726
1727 if (file_descriptor < 0) {
1728 // Can't open library, report dlerror() message
1729 return NULL;
1730 }
1731
1732 bool failed_to_read_elf_head=
1733 (sizeof(elf_head)!=
1734 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1735
1736 ::close(file_descriptor);
1737 if (failed_to_read_elf_head) {
1738 // file i/o error - report dlerror() msg
1739 return NULL;
1740 }
1741
1742 typedef struct {
1743 Elf32_Half code; // Actual value as defined in elf.h
1744 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1745 char elf_class; // 32 or 64 bit
1746 char endianess; // MSB or LSB
1747 char* name; // String representation
1748 } arch_t;
1749
1750 #ifndef EM_486
1751 #define EM_486 6 /* Intel 80486 */
1752 #endif
1753
1754 static const arch_t arch_array[]={
1755 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1756 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1757 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1758 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1759 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1760 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1761 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1762 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1763 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1764 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1765 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1766 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1767 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1768 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1769 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1770 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1771 };
1772
1773 #if (defined IA32)
1774 static Elf32_Half running_arch_code=EM_386;
1775 #elif (defined AMD64)
1776 static Elf32_Half running_arch_code=EM_X86_64;
1777 #elif (defined IA64)
1778 static Elf32_Half running_arch_code=EM_IA_64;
1779 #elif (defined __sparc) && (defined _LP64)
1780 static Elf32_Half running_arch_code=EM_SPARCV9;
1781 #elif (defined __sparc) && (!defined _LP64)
1782 static Elf32_Half running_arch_code=EM_SPARC;
1783 #elif (defined __powerpc64__)
1784 static Elf32_Half running_arch_code=EM_PPC64;
1785 #elif (defined __powerpc__)
1786 static Elf32_Half running_arch_code=EM_PPC;
1787 #elif (defined ARM)
1788 static Elf32_Half running_arch_code=EM_ARM;
1789 #elif (defined S390)
1790 static Elf32_Half running_arch_code=EM_S390;
1791 #elif (defined ALPHA)
1792 static Elf32_Half running_arch_code=EM_ALPHA;
1793 #elif (defined MIPSEL)
1794 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1795 #elif (defined PARISC)
1796 static Elf32_Half running_arch_code=EM_PARISC;
1797 #elif (defined MIPS)
1798 static Elf32_Half running_arch_code=EM_MIPS;
1799 #elif (defined M68K)
1800 static Elf32_Half running_arch_code=EM_68K;
1801 #else
1802 #error Method os::dll_load requires that one of following is defined:\
1803 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1804 #endif
1805
1806 // Identify compatability class for VM's architecture and library's architecture
1807 // Obtain string descriptions for architectures
1808
1809 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1810 int running_arch_index=-1;
1811
1812 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1813 if (running_arch_code == arch_array[i].code) {
1814 running_arch_index = i;
1815 }
1816 if (lib_arch.code == arch_array[i].code) {
1817 lib_arch.compat_class = arch_array[i].compat_class;
1818 lib_arch.name = arch_array[i].name;
1819 }
1820 }
1821
1822 assert(running_arch_index != -1,
1823 "Didn't find running architecture code (running_arch_code) in arch_array");
1824 if (running_arch_index == -1) {
1825 // Even though running architecture detection failed
1826 // we may still continue with reporting dlerror() message
1827 return NULL;
1828 }
1829
1830 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1831 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1832 return NULL;
1833 }
1834
1835 #ifndef S390
1836 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1837 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1838 return NULL;
1839 }
1840 #endif // !S390
1841
1842 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1843 if ( lib_arch.name!=NULL ) {
1844 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1845 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1846 lib_arch.name, arch_array[running_arch_index].name);
1847 } else {
1848 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1849 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1850 lib_arch.code,
1851 arch_array[running_arch_index].name);
1852 }
1853 }
1854
1855 return NULL;
1856 }
1857
1858 /*
1859 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1860 * chances are you might want to run the generated bits against glibc-2.0
1861 * libdl.so, so always use locking for any version of glibc.
1862 */
1863 void* os::dll_lookup(void* handle, const char* name) {
1864 pthread_mutex_lock(&dl_mutex);
1865 void* res = dlsym(handle, name);
1866 pthread_mutex_unlock(&dl_mutex);
1867 return res;
1868 }
1869
1870
1871 bool _print_ascii_file(const char* filename, outputStream* st) {
1872 int fd = open(filename, O_RDONLY);
1873 if (fd == -1) {
1874 return false;
1875 }
1876
1877 char buf[32];
1878 int bytes;
1879 while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1880 st->print_raw(buf, bytes);
1881 }
1882
1883 close(fd);
1884
1885 return true;
1886 }
1887
1888 void os::print_dll_info(outputStream *st) {
1889 st->print_cr("Dynamic libraries:");
1890
1891 char fname[32];
1892 pid_t pid = os::Linux::gettid();
1893
1894 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1895
1896 if (!_print_ascii_file(fname, st)) {
1897 st->print("Can not get library information for pid = %d\n", pid);
1898 }
1899 }
1900
1901
1902 void os::print_os_info(outputStream* st) {
1903 st->print("OS:");
1904
1905 // Try to identify popular distros.
1906 // Most Linux distributions have /etc/XXX-release file, which contains
1907 // the OS version string. Some have more than one /etc/XXX-release file
1908 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1909 // so the order is important.
1910 if (!_print_ascii_file("/etc/mandrake-release", st) &&
1911 !_print_ascii_file("/etc/sun-release", st) &&
1912 !_print_ascii_file("/etc/redhat-release", st) &&
1913 !_print_ascii_file("/etc/SuSE-release", st) &&
1914 !_print_ascii_file("/etc/turbolinux-release", st) &&
1915 !_print_ascii_file("/etc/gentoo-release", st) &&
1916 !_print_ascii_file("/etc/debian_version", st) &&
1917 !_print_ascii_file("/etc/ltib-release", st) &&
1918 !_print_ascii_file("/etc/angstrom-version", st)) {
1919 st->print("Linux");
1920 }
1921 st->cr();
1922
1923 // kernel
1924 st->print("uname:");
1925 struct utsname name;
1926 uname(&name);
1927 st->print(name.sysname); st->print(" ");
1928 st->print(name.release); st->print(" ");
1929 st->print(name.version); st->print(" ");
1930 st->print(name.machine);
1931 st->cr();
1932
1933 // Print warning if unsafe chroot environment detected
1934 if (unsafe_chroot_detected) {
1935 st->print("WARNING!! ");
1936 st->print_cr(unstable_chroot_error);
1937 }
1938
1939 // libc, pthread
1940 st->print("libc:");
1941 st->print(os::Linux::glibc_version()); st->print(" ");
1942 st->print(os::Linux::libpthread_version()); st->print(" ");
1943 if (os::Linux::is_LinuxThreads()) {
1944 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1945 }
1946 st->cr();
1947
1948 // rlimit
1949 st->print("rlimit:");
1950 struct rlimit rlim;
1951
1952 st->print(" STACK ");
1953 getrlimit(RLIMIT_STACK, &rlim);
1954 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1955 else st->print("%uk", rlim.rlim_cur >> 10);
1956
1957 st->print(", CORE ");
1958 getrlimit(RLIMIT_CORE, &rlim);
1959 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1960 else st->print("%uk", rlim.rlim_cur >> 10);
1961
1962 st->print(", NPROC ");
1963 getrlimit(RLIMIT_NPROC, &rlim);
1964 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1965 else st->print("%d", rlim.rlim_cur);
1966
1967 st->print(", NOFILE ");
1968 getrlimit(RLIMIT_NOFILE, &rlim);
1969 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1970 else st->print("%d", rlim.rlim_cur);
1971
1972 st->print(", AS ");
1973 getrlimit(RLIMIT_AS, &rlim);
1974 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1975 else st->print("%uk", rlim.rlim_cur >> 10);
1976 st->cr();
1977
1978 // load average
1979 st->print("load average:");
1980 double loadavg[3];
1981 os::loadavg(loadavg, 3);
1982 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1983 st->cr();
1984
1985 // meminfo
1986 st->print("\n/proc/meminfo:\n");
1987 _print_ascii_file("/proc/meminfo", st);
1988 st->cr();
1989 }
1990
1991 void os::print_memory_info(outputStream* st) {
1992
1993 st->print("Memory:");
1994 st->print(" %dk page", os::vm_page_size()>>10);
1995
1996 // values in struct sysinfo are "unsigned long"
1997 struct sysinfo si;
1998 sysinfo(&si);
1999
2000 st->print(", physical " UINT64_FORMAT "k",
2001 os::physical_memory() >> 10);
2002 st->print("(" UINT64_FORMAT "k free)",
2003 os::available_memory() >> 10);
2004 st->print(", swap " UINT64_FORMAT "k",
2005 ((jlong)si.totalswap * si.mem_unit) >> 10);
2006 st->print("(" UINT64_FORMAT "k free)",
2007 ((jlong)si.freeswap * si.mem_unit) >> 10);
2008 st->cr();
2009 }
2010
2011 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2012 // but they're the same for all the linux arch that we support
2013 // and they're the same for solaris but there's no common place to put this.
2014 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2015 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2016 "ILL_COPROC", "ILL_BADSTK" };
2017
2018 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2019 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2020 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2021
2022 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2023
2024 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2025
2026 void os::print_siginfo(outputStream* st, void* siginfo) {
2027 st->print("siginfo:");
2028
2029 const int buflen = 100;
2030 char buf[buflen];
2031 siginfo_t *si = (siginfo_t*)siginfo;
2032 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2033 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2034 st->print("si_errno=%s", buf);
2035 } else {
2036 st->print("si_errno=%d", si->si_errno);
2037 }
2038 const int c = si->si_code;
2039 assert(c > 0, "unexpected si_code");
2040 switch (si->si_signo) {
2041 case SIGILL:
2042 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2043 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2044 break;
2045 case SIGFPE:
2046 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2047 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2048 break;
2049 case SIGSEGV:
2050 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2051 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2052 break;
2053 case SIGBUS:
2054 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2055 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2056 break;
2057 default:
2058 st->print(", si_code=%d", si->si_code);
2059 // no si_addr
2060 }
2061
2062 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2063 UseSharedSpaces) {
2064 FileMapInfo* mapinfo = FileMapInfo::current_info();
2065 if (mapinfo->is_in_shared_space(si->si_addr)) {
2066 st->print("\n\nError accessing class data sharing archive." \
2067 " Mapped file inaccessible during execution, " \
2068 " possible disk/network problem.");
2069 }
2070 }
2071 st->cr();
2072 }
2073
2074
2075 static void print_signal_handler(outputStream* st, int sig,
2076 char* buf, size_t buflen);
2077
2078 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2079 st->print_cr("Signal Handlers:");
2080 print_signal_handler(st, SIGSEGV, buf, buflen);
2081 print_signal_handler(st, SIGBUS , buf, buflen);
2082 print_signal_handler(st, SIGFPE , buf, buflen);
2083 print_signal_handler(st, SIGPIPE, buf, buflen);
2084 print_signal_handler(st, SIGXFSZ, buf, buflen);
2085 print_signal_handler(st, SIGILL , buf, buflen);
2086 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2087 print_signal_handler(st, SR_signum, buf, buflen);
2088 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2089 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2090 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2091 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2092 }
2093
2094 static char saved_jvm_path[MAXPATHLEN] = {0};
2095
2096 // Find the full path to the current module, libjvm.so or libjvm_g.so
2097 void os::jvm_path(char *buf, jint buflen) {
2098 // Error checking.
2099 if (buflen < MAXPATHLEN) {
2100 assert(false, "must use a large-enough buffer");
2101 buf[0] = '\0';
2102 return;
2103 }
2104 // Lazy resolve the path to current module.
2105 if (saved_jvm_path[0] != 0) {
2106 strcpy(buf, saved_jvm_path);
2107 return;
2108 }
2109
2110 char dli_fname[MAXPATHLEN];
2111 bool ret = dll_address_to_library_name(
2112 CAST_FROM_FN_PTR(address, os::jvm_path),
2113 dli_fname, sizeof(dli_fname), NULL);
2114 assert(ret != 0, "cannot locate libjvm");
2115 char *rp = realpath(dli_fname, buf);
2116 if (rp == NULL)
2117 return;
2118
2119 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2120 // Support for the gamma launcher. Typical value for buf is
2121 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2122 // the right place in the string, then assume we are installed in a JDK and
2123 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2124 // up the path so it looks like libjvm.so is installed there (append a
2125 // fake suffix hotspot/libjvm.so).
2126 const char *p = buf + strlen(buf) - 1;
2127 for (int count = 0; p > buf && count < 5; ++count) {
2128 for (--p; p > buf && *p != '/'; --p)
2129 /* empty */ ;
2130 }
2131
2132 if (strncmp(p, "/jre/lib/", 9) != 0) {
2133 // Look for JAVA_HOME in the environment.
2134 char* java_home_var = ::getenv("JAVA_HOME");
2135 if (java_home_var != NULL && java_home_var[0] != 0) {
2136 char* jrelib_p;
2137 int len;
2138
2139 // Check the current module name "libjvm.so" or "libjvm_g.so".
2140 p = strrchr(buf, '/');
2141 assert(strstr(p, "/libjvm") == p, "invalid library name");
2142 p = strstr(p, "_g") ? "_g" : "";
2143
2144 rp = realpath(java_home_var, buf);
2145 if (rp == NULL)
2146 return;
2147
2148 // determine if this is a legacy image or modules image
2149 // modules image doesn't have "jre" subdirectory
2150 len = strlen(buf);
2151 jrelib_p = buf + len;
2152 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2153 if (0 != access(buf, F_OK)) {
2154 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2155 }
2156
2157 if (0 == access(buf, F_OK)) {
2158 // Use current module name "libjvm[_g].so" instead of
2159 // "libjvm"debug_only("_g")".so" since for fastdebug version
2160 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2161 // It is used when we are choosing the HPI library's name
2162 // "libhpi[_g].so" in hpi::initialize_get_interface().
2163 len = strlen(buf);
2164 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2165 } else {
2166 // Go back to path of .so
2167 rp = realpath(dli_fname, buf);
2168 if (rp == NULL)
2169 return;
2170 }
2171 }
2172 }
2173 }
2174
2175 strcpy(saved_jvm_path, buf);
2176 }
2177
2178 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2179 // no prefix required, not even "_"
2180 }
2181
2182 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2183 // no suffix required
2184 }
2185
2186 ////////////////////////////////////////////////////////////////////////////////
2187 // sun.misc.Signal support
2188
2189 static volatile jint sigint_count = 0;
2190
2191 static void
2192 UserHandler(int sig, void *siginfo, void *context) {
2193 // 4511530 - sem_post is serialized and handled by the manager thread. When
2194 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2195 // don't want to flood the manager thread with sem_post requests.
2196 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2197 return;
2198
2199 // Ctrl-C is pressed during error reporting, likely because the error
2200 // handler fails to abort. Let VM die immediately.
2201 if (sig == SIGINT && is_error_reported()) {
2202 os::die();
2203 }
2204
2205 os::signal_notify(sig);
2206 }
2207
2208 void* os::user_handler() {
2209 return CAST_FROM_FN_PTR(void*, UserHandler);
2210 }
2211
2212 extern "C" {
2213 typedef void (*sa_handler_t)(int);
2214 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2215 }
2216
2217 void* os::signal(int signal_number, void* handler) {
2218 struct sigaction sigAct, oldSigAct;
2219
2220 sigfillset(&(sigAct.sa_mask));
2221 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2222 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2223
2224 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2225 // -1 means registration failed
2226 return (void *)-1;
2227 }
2228
2229 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2230 }
2231
2232 void os::signal_raise(int signal_number) {
2233 ::raise(signal_number);
2234 }
2235
2236 /*
2237 * The following code is moved from os.cpp for making this
2238 * code platform specific, which it is by its very nature.
2239 */
2240
2241 // Will be modified when max signal is changed to be dynamic
2242 int os::sigexitnum_pd() {
2243 return NSIG;
2244 }
2245
2246 // a counter for each possible signal value
2247 static volatile jint pending_signals[NSIG+1] = { 0 };
2248
2249 // Linux(POSIX) specific hand shaking semaphore.
2250 static sem_t sig_sem;
2251
2252 void os::signal_init_pd() {
2253 // Initialize signal structures
2254 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2255
2256 // Initialize signal semaphore
2257 ::sem_init(&sig_sem, 0, 0);
2258 }
2259
2260 void os::signal_notify(int sig) {
2261 Atomic::inc(&pending_signals[sig]);
2262 ::sem_post(&sig_sem);
2263 }
2264
2265 static int check_pending_signals(bool wait) {
2266 Atomic::store(0, &sigint_count);
2267 for (;;) {
2268 for (int i = 0; i < NSIG + 1; i++) {
2269 jint n = pending_signals[i];
2270 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2271 return i;
2272 }
2273 }
2274 if (!wait) {
2275 return -1;
2276 }
2277 JavaThread *thread = JavaThread::current();
2278 ThreadBlockInVM tbivm(thread);
2279
2280 bool threadIsSuspended;
2281 do {
2282 thread->set_suspend_equivalent();
2283 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2284 ::sem_wait(&sig_sem);
2285
2286 // were we externally suspended while we were waiting?
2287 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2288 if (threadIsSuspended) {
2289 //
2290 // The semaphore has been incremented, but while we were waiting
2291 // another thread suspended us. We don't want to continue running
2292 // while suspended because that would surprise the thread that
2293 // suspended us.
2294 //
2295 ::sem_post(&sig_sem);
2296
2297 thread->java_suspend_self();
2298 }
2299 } while (threadIsSuspended);
2300 }
2301 }
2302
2303 int os::signal_lookup() {
2304 return check_pending_signals(false);
2305 }
2306
2307 int os::signal_wait() {
2308 return check_pending_signals(true);
2309 }
2310
2311 ////////////////////////////////////////////////////////////////////////////////
2312 // Virtual Memory
2313
2314 int os::vm_page_size() {
2315 // Seems redundant as all get out
2316 assert(os::Linux::page_size() != -1, "must call os::init");
2317 return os::Linux::page_size();
2318 }
2319
2320 // Solaris allocates memory by pages.
2321 int os::vm_allocation_granularity() {
2322 assert(os::Linux::page_size() != -1, "must call os::init");
2323 return os::Linux::page_size();
2324 }
2325
2326 // Rationale behind this function:
2327 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2328 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2329 // samples for JITted code. Here we create private executable mapping over the code cache
2330 // and then we can use standard (well, almost, as mapping can change) way to provide
2331 // info for the reporting script by storing timestamp and location of symbol
2332 void linux_wrap_code(char* base, size_t size) {
2333 static volatile jint cnt = 0;
2334
2335 if (!UseOprofile) {
2336 return;
2337 }
2338
2339 char buf[PATH_MAX+1];
2340 int num = Atomic::add(1, &cnt);
2341
2342 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2343 os::get_temp_directory(), os::current_process_id(), num);
2344 unlink(buf);
2345
2346 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2347
2348 if (fd != -1) {
2349 off_t rv = lseek(fd, size-2, SEEK_SET);
2350 if (rv != (off_t)-1) {
2351 if (write(fd, "", 1) == 1) {
2352 mmap(base, size,
2353 PROT_READ|PROT_WRITE|PROT_EXEC,
2354 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2355 }
2356 }
2357 close(fd);
2358 unlink(buf);
2359 }
2360 }
2361
2362 // NOTE: Linux kernel does not really reserve the pages for us.
2363 // All it does is to check if there are enough free pages
2364 // left at the time of mmap(). This could be a potential
2365 // problem.
2366 bool os::commit_memory(char* addr, size_t size, bool exec) {
2367 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2368 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2369 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2370 return res != (uintptr_t) MAP_FAILED;
2371 }
2372
2373 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2374 bool exec) {
2375 return commit_memory(addr, size, exec);
2376 }
2377
2378 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2379
2380 void os::free_memory(char *addr, size_t bytes) {
2381 ::mmap(addr, bytes, PROT_READ | PROT_WRITE,
2382 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2383 }
2384
2385 void os::numa_make_global(char *addr, size_t bytes) {
2386 Linux::numa_interleave_memory(addr, bytes);
2387 }
2388
2389 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2390 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2391 }
2392
2393 bool os::numa_topology_changed() { return false; }
2394
2395 size_t os::numa_get_groups_num() {
2396 int max_node = Linux::numa_max_node();
2397 return max_node > 0 ? max_node + 1 : 1;
2398 }
2399
2400 int os::numa_get_group_id() {
2401 int cpu_id = Linux::sched_getcpu();
2402 if (cpu_id != -1) {
2403 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2404 if (lgrp_id != -1) {
2405 return lgrp_id;
2406 }
2407 }
2408 return 0;
2409 }
2410
2411 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2412 for (size_t i = 0; i < size; i++) {
2413 ids[i] = i;
2414 }
2415 return size;
2416 }
2417
2418 bool os::get_page_info(char *start, page_info* info) {
2419 return false;
2420 }
2421
2422 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2423 return end;
2424 }
2425
2426 extern "C" void numa_warn(int number, char *where, ...) { }
2427 extern "C" void numa_error(char *where) { }
2428
2429
2430 // If we are running with libnuma version > 2, then we should
2431 // be trying to use symbols with versions 1.1
2432 // If we are running with earlier version, which did not have symbol versions,
2433 // we should use the base version.
2434 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2435 void *f = dlvsym(handle, name, "libnuma_1.1");
2436 if (f == NULL) {
2437 f = dlsym(handle, name);
2438 }
2439 return f;
2440 }
2441
2442 bool os::Linux::libnuma_init() {
2443 // sched_getcpu() should be in libc.
2444 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2445 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2446
2447 if (sched_getcpu() != -1) { // Does it work?
2448 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2449 if (handle != NULL) {
2450 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2451 libnuma_dlsym(handle, "numa_node_to_cpus")));
2452 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2453 libnuma_dlsym(handle, "numa_max_node")));
2454 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2455 libnuma_dlsym(handle, "numa_available")));
2456 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2457 libnuma_dlsym(handle, "numa_tonode_memory")));
2458 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2459 libnuma_dlsym(handle, "numa_interleave_memory")));
2460
2461
2462 if (numa_available() != -1) {
2463 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2464 // Create a cpu -> node mapping
2465 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2466 rebuild_cpu_to_node_map();
2467 return true;
2468 }
2469 }
2470 }
2471 return false;
2472 }
2473
2474 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2475 // The table is later used in get_node_by_cpu().
2476 void os::Linux::rebuild_cpu_to_node_map() {
2477 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2478 // in libnuma (possible values are starting from 16,
2479 // and continuing up with every other power of 2, but less
2480 // than the maximum number of CPUs supported by kernel), and
2481 // is a subject to change (in libnuma version 2 the requirements
2482 // are more reasonable) we'll just hardcode the number they use
2483 // in the library.
2484 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2485
2486 size_t cpu_num = os::active_processor_count();
2487 size_t cpu_map_size = NCPUS / BitsPerCLong;
2488 size_t cpu_map_valid_size =
2489 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2490
2491 cpu_to_node()->clear();
2492 cpu_to_node()->at_grow(cpu_num - 1);
2493 size_t node_num = numa_get_groups_num();
2494
2495 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2496 for (size_t i = 0; i < node_num; i++) {
2497 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2498 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2499 if (cpu_map[j] != 0) {
2500 for (size_t k = 0; k < BitsPerCLong; k++) {
2501 if (cpu_map[j] & (1UL << k)) {
2502 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2503 }
2504 }
2505 }
2506 }
2507 }
2508 }
2509 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2510 }
2511
2512 int os::Linux::get_node_by_cpu(int cpu_id) {
2513 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2514 return cpu_to_node()->at(cpu_id);
2515 }
2516 return -1;
2517 }
2518
2519 GrowableArray<int>* os::Linux::_cpu_to_node;
2520 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2521 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2522 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2523 os::Linux::numa_available_func_t os::Linux::_numa_available;
2524 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2525 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2526 unsigned long* os::Linux::_numa_all_nodes;
2527
2528 bool os::uncommit_memory(char* addr, size_t size) {
2529 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2530 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2531 return res != (uintptr_t) MAP_FAILED;
2532 }
2533
2534 // Linux uses a growable mapping for the stack, and if the mapping for
2535 // the stack guard pages is not removed when we detach a thread the
2536 // stack cannot grow beyond the pages where the stack guard was
2537 // mapped. If at some point later in the process the stack expands to
2538 // that point, the Linux kernel cannot expand the stack any further
2539 // because the guard pages are in the way, and a segfault occurs.
2540 //
2541 // However, it's essential not to split the stack region by unmapping
2542 // a region (leaving a hole) that's already part of the stack mapping,
2543 // so if the stack mapping has already grown beyond the guard pages at
2544 // the time we create them, we have to truncate the stack mapping.
2545 // So, we need to know the extent of the stack mapping when
2546 // create_stack_guard_pages() is called.
2547
2548 // Find the bounds of the stack mapping. Return true for success.
2549 //
2550 // We only need this for stacks that are growable: at the time of
2551 // writing thread stacks don't use growable mappings (i.e. those
2552 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2553 // only applies to the main thread.
2554 static bool
2555 get_stack_bounds(uintptr_t *bottom, uintptr_t *top)
2556 {
2557 FILE *f = fopen("/proc/self/maps", "r");
2558 if (f == NULL)
2559 return false;
2560
2561 while (!feof(f)) {
2562 size_t dummy;
2563 char *str = NULL;
2564 ssize_t len = getline(&str, &dummy, f);
2565 if (len == -1) {
2566 fclose(f);
2567 return false;
2568 }
2569
2570 if (len > 0 && str[len-1] == '\n') {
2571 str[len-1] = 0;
2572 len--;
2573 }
2574
2575 static const char *stack_str = "[stack]";
2576 if (len > (ssize_t)strlen(stack_str)
2577 && (strcmp(str + len - strlen(stack_str), stack_str) == 0)) {
2578 if (sscanf(str, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2579 uintptr_t sp = (uintptr_t)__builtin_frame_address(0);
2580 if (sp >= *bottom && sp <= *top) {
2581 free(str);
2582 fclose(f);
2583 return true;
2584 }
2585 }
2586 }
2587 free(str);
2588 }
2589 fclose(f);
2590 return false;
2591 }
2592
2593 // If the (growable) stack mapping already extends beyond the point
2594 // where we're going to put our guard pages, truncate the mapping at
2595 // that point by munmap()ping it. This ensures that when we later
2596 // munmap() the guard pages we don't leave a hole in the stack
2597 // mapping. This only affects the main/initial thread, but guard
2598 // against future OS changes
2599 bool os::create_stack_guard_pages(char* addr, size_t size) {
2600 uintptr_t stack_extent, stack_base;
2601 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2602 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2603 assert(os::Linux::is_initial_thread(),
2604 "growable stack in non-initial thread");
2605 if (stack_extent < (uintptr_t)addr)
2606 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2607 }
2608
2609 return os::commit_memory(addr, size);
2610 }
2611
2612 // If this is a growable mapping, remove the guard pages entirely by
2613 // munmap()ping them. If not, just call uncommit_memory(). This only
2614 // affects the main/initial thread, but guard against future OS changes
2615 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2616 uintptr_t stack_extent, stack_base;
2617 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2618 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2619 assert(os::Linux::is_initial_thread(),
2620 "growable stack in non-initial thread");
2621
2622 return ::munmap(addr, size) == 0;
2623 }
2624
2625 return os::uncommit_memory(addr, size);
2626 }
2627
2628 static address _highest_vm_reserved_address = NULL;
2629
2630 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2631 // at 'requested_addr'. If there are existing memory mappings at the same
2632 // location, however, they will be overwritten. If 'fixed' is false,
2633 // 'requested_addr' is only treated as a hint, the return value may or
2634 // may not start from the requested address. Unlike Linux mmap(), this
2635 // function returns NULL to indicate failure.
2636 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2637 char * addr;
2638 int flags;
2639
2640 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2641 if (fixed) {
2642 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2643 flags |= MAP_FIXED;
2644 }
2645
2646 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2647 // to PROT_EXEC if executable when we commit the page.
2648 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2649 flags, -1, 0);
2650
2651 if (addr != MAP_FAILED) {
2652 // anon_mmap() should only get called during VM initialization,
2653 // don't need lock (actually we can skip locking even it can be called
2654 // from multiple threads, because _highest_vm_reserved_address is just a
2655 // hint about the upper limit of non-stack memory regions.)
2656 if ((address)addr + bytes > _highest_vm_reserved_address) {
2657 _highest_vm_reserved_address = (address)addr + bytes;
2658 }
2659 }
2660
2661 return addr == MAP_FAILED ? NULL : addr;
2662 }
2663
2664 // Don't update _highest_vm_reserved_address, because there might be memory
2665 // regions above addr + size. If so, releasing a memory region only creates
2666 // a hole in the address space, it doesn't help prevent heap-stack collision.
2667 //
2668 static int anon_munmap(char * addr, size_t size) {
2669 return ::munmap(addr, size) == 0;
2670 }
2671
2672 char* os::reserve_memory(size_t bytes, char* requested_addr,
2673 size_t alignment_hint) {
2674 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2675 }
2676
2677 bool os::release_memory(char* addr, size_t size) {
2678 return anon_munmap(addr, size);
2679 }
2680
2681 static address highest_vm_reserved_address() {
2682 return _highest_vm_reserved_address;
2683 }
2684
2685 static bool linux_mprotect(char* addr, size_t size, int prot) {
2686 // Linux wants the mprotect address argument to be page aligned.
2687 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2688
2689 // According to SUSv3, mprotect() should only be used with mappings
2690 // established by mmap(), and mmap() always maps whole pages. Unaligned
2691 // 'addr' likely indicates problem in the VM (e.g. trying to change
2692 // protection of malloc'ed or statically allocated memory). Check the
2693 // caller if you hit this assert.
2694 assert(addr == bottom, "sanity check");
2695
2696 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2697 return ::mprotect(bottom, size, prot) == 0;
2698 }
2699
2700 // Set protections specified
2701 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2702 bool is_committed) {
2703 unsigned int p = 0;
2704 switch (prot) {
2705 case MEM_PROT_NONE: p = PROT_NONE; break;
2706 case MEM_PROT_READ: p = PROT_READ; break;
2707 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2708 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2709 default:
2710 ShouldNotReachHere();
2711 }
2712 // is_committed is unused.
2713 return linux_mprotect(addr, bytes, p);
2714 }
2715
2716 bool os::guard_memory(char* addr, size_t size) {
2717 return linux_mprotect(addr, size, PROT_NONE);
2718 }
2719
2720 bool os::unguard_memory(char* addr, size_t size) {
2721 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2722 }
2723
2724 // Large page support
2725
2726 static size_t _large_page_size = 0;
2727
2728 bool os::large_page_init() {
2729 if (!UseLargePages) return false;
2730
2731 if (LargePageSizeInBytes) {
2732 _large_page_size = LargePageSizeInBytes;
2733 } else {
2734 // large_page_size on Linux is used to round up heap size. x86 uses either
2735 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2736 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2737 // page as large as 256M.
2738 //
2739 // Here we try to figure out page size by parsing /proc/meminfo and looking
2740 // for a line with the following format:
2741 // Hugepagesize: 2048 kB
2742 //
2743 // If we can't determine the value (e.g. /proc is not mounted, or the text
2744 // format has been changed), we'll use the largest page size supported by
2745 // the processor.
2746
2747 #ifndef ZERO
2748 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
2749 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
2750 #endif // ZERO
2751
2752 FILE *fp = fopen("/proc/meminfo", "r");
2753 if (fp) {
2754 while (!feof(fp)) {
2755 int x = 0;
2756 char buf[16];
2757 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2758 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2759 _large_page_size = x * K;
2760 break;
2761 }
2762 } else {
2763 // skip to next line
2764 for (;;) {
2765 int ch = fgetc(fp);
2766 if (ch == EOF || ch == (int)'\n') break;
2767 }
2768 }
2769 }
2770 fclose(fp);
2771 }
2772 }
2773
2774 const size_t default_page_size = (size_t)Linux::page_size();
2775 if (_large_page_size > default_page_size) {
2776 _page_sizes[0] = _large_page_size;
2777 _page_sizes[1] = default_page_size;
2778 _page_sizes[2] = 0;
2779 }
2780
2781 // Large page support is available on 2.6 or newer kernel, some vendors
2782 // (e.g. Redhat) have backported it to their 2.4 based distributions.
2783 // We optimistically assume the support is available. If later it turns out
2784 // not true, VM will automatically switch to use regular page size.
2785 return true;
2786 }
2787
2788 #ifndef SHM_HUGETLB
2789 #define SHM_HUGETLB 04000
2790 #endif
2791
2792 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
2793 // "exec" is passed in but not used. Creating the shared image for
2794 // the code cache doesn't have an SHM_X executable permission to check.
2795 assert(UseLargePages, "only for large pages");
2796
2797 key_t key = IPC_PRIVATE;
2798 char *addr;
2799
2800 bool warn_on_failure = UseLargePages &&
2801 (!FLAG_IS_DEFAULT(UseLargePages) ||
2802 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2803 );
2804 char msg[128];
2805
2806 // Create a large shared memory region to attach to based on size.
2807 // Currently, size is the total size of the heap
2808 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2809 if (shmid == -1) {
2810 // Possible reasons for shmget failure:
2811 // 1. shmmax is too small for Java heap.
2812 // > check shmmax value: cat /proc/sys/kernel/shmmax
2813 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2814 // 2. not enough large page memory.
2815 // > check available large pages: cat /proc/meminfo
2816 // > increase amount of large pages:
2817 // echo new_value > /proc/sys/vm/nr_hugepages
2818 // Note 1: different Linux may use different name for this property,
2819 // e.g. on Redhat AS-3 it is "hugetlb_pool".
2820 // Note 2: it's possible there's enough physical memory available but
2821 // they are so fragmented after a long run that they can't
2822 // coalesce into large pages. Try to reserve large pages when
2823 // the system is still "fresh".
2824 if (warn_on_failure) {
2825 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2826 warning(msg);
2827 }
2828 return NULL;
2829 }
2830
2831 // attach to the region
2832 addr = (char*)shmat(shmid, req_addr, 0);
2833 int err = errno;
2834
2835 // Remove shmid. If shmat() is successful, the actual shared memory segment
2836 // will be deleted when it's detached by shmdt() or when the process
2837 // terminates. If shmat() is not successful this will remove the shared
2838 // segment immediately.
2839 shmctl(shmid, IPC_RMID, NULL);
2840
2841 if ((intptr_t)addr == -1) {
2842 if (warn_on_failure) {
2843 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2844 warning(msg);
2845 }
2846 return NULL;
2847 }
2848
2849 return addr;
2850 }
2851
2852 bool os::release_memory_special(char* base, size_t bytes) {
2853 // detaching the SHM segment will also delete it, see reserve_memory_special()
2854 int rslt = shmdt(base);
2855 return rslt == 0;
2856 }
2857
2858 size_t os::large_page_size() {
2859 return _large_page_size;
2860 }
2861
2862 // Linux does not support anonymous mmap with large page memory. The only way
2863 // to reserve large page memory without file backing is through SysV shared
2864 // memory API. The entire memory region is committed and pinned upfront.
2865 // Hopefully this will change in the future...
2866 bool os::can_commit_large_page_memory() {
2867 return false;
2868 }
2869
2870 bool os::can_execute_large_page_memory() {
2871 return false;
2872 }
2873
2874 // Reserve memory at an arbitrary address, only if that area is
2875 // available (and not reserved for something else).
2876
2877 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2878 const int max_tries = 10;
2879 char* base[max_tries];
2880 size_t size[max_tries];
2881 const size_t gap = 0x000000;
2882
2883 // Assert only that the size is a multiple of the page size, since
2884 // that's all that mmap requires, and since that's all we really know
2885 // about at this low abstraction level. If we need higher alignment,
2886 // we can either pass an alignment to this method or verify alignment
2887 // in one of the methods further up the call chain. See bug 5044738.
2888 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2889
2890 // Repeatedly allocate blocks until the block is allocated at the
2891 // right spot. Give up after max_tries. Note that reserve_memory() will
2892 // automatically update _highest_vm_reserved_address if the call is
2893 // successful. The variable tracks the highest memory address every reserved
2894 // by JVM. It is used to detect heap-stack collision if running with
2895 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2896 // space than needed, it could confuse the collision detecting code. To
2897 // solve the problem, save current _highest_vm_reserved_address and
2898 // calculate the correct value before return.
2899 address old_highest = _highest_vm_reserved_address;
2900
2901 // Linux mmap allows caller to pass an address as hint; give it a try first,
2902 // if kernel honors the hint then we can return immediately.
2903 char * addr = anon_mmap(requested_addr, bytes, false);
2904 if (addr == requested_addr) {
2905 return requested_addr;
2906 }
2907
2908 if (addr != NULL) {
2909 // mmap() is successful but it fails to reserve at the requested address
2910 anon_munmap(addr, bytes);
2911 }
2912
2913 int i;
2914 for (i = 0; i < max_tries; ++i) {
2915 base[i] = reserve_memory(bytes);
2916
2917 if (base[i] != NULL) {
2918 // Is this the block we wanted?
2919 if (base[i] == requested_addr) {
2920 size[i] = bytes;
2921 break;
2922 }
2923
2924 // Does this overlap the block we wanted? Give back the overlapped
2925 // parts and try again.
2926
2927 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2928 if (top_overlap >= 0 && top_overlap < bytes) {
2929 unmap_memory(base[i], top_overlap);
2930 base[i] += top_overlap;
2931 size[i] = bytes - top_overlap;
2932 } else {
2933 size_t bottom_overlap = base[i] + bytes - requested_addr;
2934 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2935 unmap_memory(requested_addr, bottom_overlap);
2936 size[i] = bytes - bottom_overlap;
2937 } else {
2938 size[i] = bytes;
2939 }
2940 }
2941 }
2942 }
2943
2944 // Give back the unused reserved pieces.
2945
2946 for (int j = 0; j < i; ++j) {
2947 if (base[j] != NULL) {
2948 unmap_memory(base[j], size[j]);
2949 }
2950 }
2951
2952 if (i < max_tries) {
2953 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2954 return requested_addr;
2955 } else {
2956 _highest_vm_reserved_address = old_highest;
2957 return NULL;
2958 }
2959 }
2960
2961 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2962 return ::read(fd, buf, nBytes);
2963 }
2964
2965 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2966 // Solaris uses poll(), linux uses park().
2967 // Poll() is likely a better choice, assuming that Thread.interrupt()
2968 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2969 // SIGSEGV, see 4355769.
2970
2971 const int NANOSECS_PER_MILLISECS = 1000000;
2972
2973 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2974 assert(thread == Thread::current(), "thread consistency check");
2975
2976 ParkEvent * const slp = thread->_SleepEvent ;
2977 slp->reset() ;
2978 OrderAccess::fence() ;
2979
2980 if (interruptible) {
2981 jlong prevtime = javaTimeNanos();
2982
2983 for (;;) {
2984 if (os::is_interrupted(thread, true)) {
2985 return OS_INTRPT;
2986 }
2987
2988 jlong newtime = javaTimeNanos();
2989
2990 if (newtime - prevtime < 0) {
2991 // time moving backwards, should only happen if no monotonic clock
2992 // not a guarantee() because JVM should not abort on kernel/glibc bugs
2993 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2994 } else {
2995 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2996 }
2997
2998 if(millis <= 0) {
2999 return OS_OK;
3000 }
3001
3002 prevtime = newtime;
3003
3004 {
3005 assert(thread->is_Java_thread(), "sanity check");
3006 JavaThread *jt = (JavaThread *) thread;
3007 ThreadBlockInVM tbivm(jt);
3008 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3009
3010 jt->set_suspend_equivalent();
3011 // cleared by handle_special_suspend_equivalent_condition() or
3012 // java_suspend_self() via check_and_wait_while_suspended()
3013
3014 slp->park(millis);
3015
3016 // were we externally suspended while we were waiting?
3017 jt->check_and_wait_while_suspended();
3018 }
3019 }
3020 } else {
3021 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3022 jlong prevtime = javaTimeNanos();
3023
3024 for (;;) {
3025 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3026 // the 1st iteration ...
3027 jlong newtime = javaTimeNanos();
3028
3029 if (newtime - prevtime < 0) {
3030 // time moving backwards, should only happen if no monotonic clock
3031 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3032 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3033 } else {
3034 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3035 }
3036
3037 if(millis <= 0) break ;
3038
3039 prevtime = newtime;
3040 slp->park(millis);
3041 }
3042 return OS_OK ;
3043 }
3044 }
3045
3046 int os::naked_sleep() {
3047 // %% make the sleep time an integer flag. for now use 1 millisec.
3048 return os::sleep(Thread::current(), 1, false);
3049 }
3050
3051 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3052 void os::infinite_sleep() {
3053 while (true) { // sleep forever ...
3054 ::sleep(100); // ... 100 seconds at a time
3055 }
3056 }
3057
3058 // Used to convert frequent JVM_Yield() to nops
3059 bool os::dont_yield() {
3060 return DontYieldALot;
3061 }
3062
3063 void os::yield() {
3064 sched_yield();
3065 }
3066
3067 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3068
3069 void os::yield_all(int attempts) {
3070 // Yields to all threads, including threads with lower priorities
3071 // Threads on Linux are all with same priority. The Solaris style
3072 // os::yield_all() with nanosleep(1ms) is not necessary.
3073 sched_yield();
3074 }
3075
3076 // Called from the tight loops to possibly influence time-sharing heuristics
3077 void os::loop_breaker(int attempts) {
3078 os::yield_all(attempts);
3079 }
3080
3081 ////////////////////////////////////////////////////////////////////////////////
3082 // thread priority support
3083
3084 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3085 // only supports dynamic priority, static priority must be zero. For real-time
3086 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3087 // However, for large multi-threaded applications, SCHED_RR is not only slower
3088 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3089 // of 5 runs - Sep 2005).
3090 //
3091 // The following code actually changes the niceness of kernel-thread/LWP. It
3092 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3093 // not the entire user process, and user level threads are 1:1 mapped to kernel
3094 // threads. It has always been the case, but could change in the future. For
3095 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3096 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3097
3098 int os::java_to_os_priority[MaxPriority + 1] = {
3099 19, // 0 Entry should never be used
3100
3101 4, // 1 MinPriority
3102 3, // 2
3103 2, // 3
3104
3105 1, // 4
3106 0, // 5 NormPriority
3107 -1, // 6
3108
3109 -2, // 7
3110 -3, // 8
3111 -4, // 9 NearMaxPriority
3112
3113 -5 // 10 MaxPriority
3114 };
3115
3116 static int prio_init() {
3117 if (ThreadPriorityPolicy == 1) {
3118 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3119 // if effective uid is not root. Perhaps, a more elegant way of doing
3120 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3121 if (geteuid() != 0) {
3122 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3123 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3124 }
3125 ThreadPriorityPolicy = 0;
3126 }
3127 }
3128 return 0;
3129 }
3130
3131 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3132 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3133
3134 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3135 return (ret == 0) ? OS_OK : OS_ERR;
3136 }
3137
3138 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3139 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3140 *priority_ptr = java_to_os_priority[NormPriority];
3141 return OS_OK;
3142 }
3143
3144 errno = 0;
3145 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3146 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3147 }
3148
3149 // Hint to the underlying OS that a task switch would not be good.
3150 // Void return because it's a hint and can fail.
3151 void os::hint_no_preempt() {}
3152
3153 ////////////////////////////////////////////////////////////////////////////////
3154 // suspend/resume support
3155
3156 // the low-level signal-based suspend/resume support is a remnant from the
3157 // old VM-suspension that used to be for java-suspension, safepoints etc,
3158 // within hotspot. Now there is a single use-case for this:
3159 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3160 // that runs in the watcher thread.
3161 // The remaining code is greatly simplified from the more general suspension
3162 // code that used to be used.
3163 //
3164 // The protocol is quite simple:
3165 // - suspend:
3166 // - sends a signal to the target thread
3167 // - polls the suspend state of the osthread using a yield loop
3168 // - target thread signal handler (SR_handler) sets suspend state
3169 // and blocks in sigsuspend until continued
3170 // - resume:
3171 // - sets target osthread state to continue
3172 // - sends signal to end the sigsuspend loop in the SR_handler
3173 //
3174 // Note that the SR_lock plays no role in this suspend/resume protocol.
3175 //
3176
3177 static void resume_clear_context(OSThread *osthread) {
3178 osthread->set_ucontext(NULL);
3179 osthread->set_siginfo(NULL);
3180
3181 // notify the suspend action is completed, we have now resumed
3182 osthread->sr.clear_suspended();
3183 }
3184
3185 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3186 osthread->set_ucontext(context);
3187 osthread->set_siginfo(siginfo);
3188 }
3189
3190 //
3191 // Handler function invoked when a thread's execution is suspended or
3192 // resumed. We have to be careful that only async-safe functions are
3193 // called here (Note: most pthread functions are not async safe and
3194 // should be avoided.)
3195 //
3196 // Note: sigwait() is a more natural fit than sigsuspend() from an
3197 // interface point of view, but sigwait() prevents the signal hander
3198 // from being run. libpthread would get very confused by not having
3199 // its signal handlers run and prevents sigwait()'s use with the
3200 // mutex granting granting signal.
3201 //
3202 // Currently only ever called on the VMThread
3203 //
3204 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3205 // Save and restore errno to avoid confusing native code with EINTR
3206 // after sigsuspend.
3207 int old_errno = errno;
3208
3209 Thread* thread = Thread::current();
3210 OSThread* osthread = thread->osthread();
3211 assert(thread->is_VM_thread(), "Must be VMThread");
3212 // read current suspend action
3213 int action = osthread->sr.suspend_action();
3214 if (action == SR_SUSPEND) {
3215 suspend_save_context(osthread, siginfo, context);
3216
3217 // Notify the suspend action is about to be completed. do_suspend()
3218 // waits until SR_SUSPENDED is set and then returns. We will wait
3219 // here for a resume signal and that completes the suspend-other
3220 // action. do_suspend/do_resume is always called as a pair from
3221 // the same thread - so there are no races
3222
3223 // notify the caller
3224 osthread->sr.set_suspended();
3225
3226 sigset_t suspend_set; // signals for sigsuspend()
3227
3228 // get current set of blocked signals and unblock resume signal
3229 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3230 sigdelset(&suspend_set, SR_signum);
3231
3232 // wait here until we are resumed
3233 do {
3234 sigsuspend(&suspend_set);
3235 // ignore all returns until we get a resume signal
3236 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3237
3238 resume_clear_context(osthread);
3239
3240 } else {
3241 assert(action == SR_CONTINUE, "unexpected sr action");
3242 // nothing special to do - just leave the handler
3243 }
3244
3245 errno = old_errno;
3246 }
3247
3248
3249 static int SR_initialize() {
3250 struct sigaction act;
3251 char *s;
3252 /* Get signal number to use for suspend/resume */
3253 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3254 int sig = ::strtol(s, 0, 10);
3255 if (sig > 0 || sig < _NSIG) {
3256 SR_signum = sig;
3257 }
3258 }
3259
3260 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3261 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3262
3263 sigemptyset(&SR_sigset);
3264 sigaddset(&SR_sigset, SR_signum);
3265
3266 /* Set up signal handler for suspend/resume */
3267 act.sa_flags = SA_RESTART|SA_SIGINFO;
3268 act.sa_handler = (void (*)(int)) SR_handler;
3269
3270 // SR_signum is blocked by default.
3271 // 4528190 - We also need to block pthread restart signal (32 on all
3272 // supported Linux platforms). Note that LinuxThreads need to block
3273 // this signal for all threads to work properly. So we don't have
3274 // to use hard-coded signal number when setting up the mask.
3275 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3276
3277 if (sigaction(SR_signum, &act, 0) == -1) {
3278 return -1;
3279 }
3280
3281 // Save signal flag
3282 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3283 return 0;
3284 }
3285
3286 static int SR_finalize() {
3287 return 0;
3288 }
3289
3290
3291 // returns true on success and false on error - really an error is fatal
3292 // but this seems the normal response to library errors
3293 static bool do_suspend(OSThread* osthread) {
3294 // mark as suspended and send signal
3295 osthread->sr.set_suspend_action(SR_SUSPEND);
3296 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3297 assert_status(status == 0, status, "pthread_kill");
3298
3299 // check status and wait until notified of suspension
3300 if (status == 0) {
3301 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3302 os::yield_all(i);
3303 }
3304 osthread->sr.set_suspend_action(SR_NONE);
3305 return true;
3306 }
3307 else {
3308 osthread->sr.set_suspend_action(SR_NONE);
3309 return false;
3310 }
3311 }
3312
3313 static void do_resume(OSThread* osthread) {
3314 assert(osthread->sr.is_suspended(), "thread should be suspended");
3315 osthread->sr.set_suspend_action(SR_CONTINUE);
3316
3317 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3318 assert_status(status == 0, status, "pthread_kill");
3319 // check status and wait unit notified of resumption
3320 if (status == 0) {
3321 for (int i = 0; osthread->sr.is_suspended(); i++) {
3322 os::yield_all(i);
3323 }
3324 }
3325 osthread->sr.set_suspend_action(SR_NONE);
3326 }
3327
3328 ////////////////////////////////////////////////////////////////////////////////
3329 // interrupt support
3330
3331 void os::interrupt(Thread* thread) {
3332 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3333 "possibility of dangling Thread pointer");
3334
3335 OSThread* osthread = thread->osthread();
3336
3337 if (!osthread->interrupted()) {
3338 osthread->set_interrupted(true);
3339 // More than one thread can get here with the same value of osthread,
3340 // resulting in multiple notifications. We do, however, want the store
3341 // to interrupted() to be visible to other threads before we execute unpark().
3342 OrderAccess::fence();
3343 ParkEvent * const slp = thread->_SleepEvent ;
3344 if (slp != NULL) slp->unpark() ;
3345 }
3346
3347 // For JSR166. Unpark even if interrupt status already was set
3348 if (thread->is_Java_thread())
3349 ((JavaThread*)thread)->parker()->unpark();
3350
3351 ParkEvent * ev = thread->_ParkEvent ;
3352 if (ev != NULL) ev->unpark() ;
3353
3354 }
3355
3356 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3357 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3358 "possibility of dangling Thread pointer");
3359
3360 OSThread* osthread = thread->osthread();
3361
3362 bool interrupted = osthread->interrupted();
3363
3364 if (interrupted && clear_interrupted) {
3365 osthread->set_interrupted(false);
3366 // consider thread->_SleepEvent->reset() ... optional optimization
3367 }
3368
3369 return interrupted;
3370 }
3371
3372 ///////////////////////////////////////////////////////////////////////////////////
3373 // signal handling (except suspend/resume)
3374
3375 // This routine may be used by user applications as a "hook" to catch signals.
3376 // The user-defined signal handler must pass unrecognized signals to this
3377 // routine, and if it returns true (non-zero), then the signal handler must
3378 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3379 // routine will never retun false (zero), but instead will execute a VM panic
3380 // routine kill the process.
3381 //
3382 // If this routine returns false, it is OK to call it again. This allows
3383 // the user-defined signal handler to perform checks either before or after
3384 // the VM performs its own checks. Naturally, the user code would be making
3385 // a serious error if it tried to handle an exception (such as a null check
3386 // or breakpoint) that the VM was generating for its own correct operation.
3387 //
3388 // This routine may recognize any of the following kinds of signals:
3389 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3390 // It should be consulted by handlers for any of those signals.
3391 //
3392 // The caller of this routine must pass in the three arguments supplied
3393 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3394 // field of the structure passed to sigaction(). This routine assumes that
3395 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3396 //
3397 // Note that the VM will print warnings if it detects conflicting signal
3398 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3399 //
3400 extern "C" int
3401 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3402 void* ucontext, int abort_if_unrecognized);
3403
3404 void signalHandler(int sig, siginfo_t* info, void* uc) {
3405 assert(info != NULL && uc != NULL, "it must be old kernel");
3406 JVM_handle_linux_signal(sig, info, uc, true);
3407 }
3408
3409
3410 // This boolean allows users to forward their own non-matching signals
3411 // to JVM_handle_linux_signal, harmlessly.
3412 bool os::Linux::signal_handlers_are_installed = false;
3413
3414 // For signal-chaining
3415 struct sigaction os::Linux::sigact[MAXSIGNUM];
3416 unsigned int os::Linux::sigs = 0;
3417 bool os::Linux::libjsig_is_loaded = false;
3418 typedef struct sigaction *(*get_signal_t)(int);
3419 get_signal_t os::Linux::get_signal_action = NULL;
3420
3421 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3422 struct sigaction *actp = NULL;
3423
3424 if (libjsig_is_loaded) {
3425 // Retrieve the old signal handler from libjsig
3426 actp = (*get_signal_action)(sig);
3427 }
3428 if (actp == NULL) {
3429 // Retrieve the preinstalled signal handler from jvm
3430 actp = get_preinstalled_handler(sig);
3431 }
3432
3433 return actp;
3434 }
3435
3436 static bool call_chained_handler(struct sigaction *actp, int sig,
3437 siginfo_t *siginfo, void *context) {
3438 // Call the old signal handler
3439 if (actp->sa_handler == SIG_DFL) {
3440 // It's more reasonable to let jvm treat it as an unexpected exception
3441 // instead of taking the default action.
3442 return false;
3443 } else if (actp->sa_handler != SIG_IGN) {
3444 if ((actp->sa_flags & SA_NODEFER) == 0) {
3445 // automaticlly block the signal
3446 sigaddset(&(actp->sa_mask), sig);
3447 }
3448
3449 sa_handler_t hand;
3450 sa_sigaction_t sa;
3451 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3452 // retrieve the chained handler
3453 if (siginfo_flag_set) {
3454 sa = actp->sa_sigaction;
3455 } else {
3456 hand = actp->sa_handler;
3457 }
3458
3459 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3460 actp->sa_handler = SIG_DFL;
3461 }
3462
3463 // try to honor the signal mask
3464 sigset_t oset;
3465 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3466
3467 // call into the chained handler
3468 if (siginfo_flag_set) {
3469 (*sa)(sig, siginfo, context);
3470 } else {
3471 (*hand)(sig);
3472 }
3473
3474 // restore the signal mask
3475 pthread_sigmask(SIG_SETMASK, &oset, 0);
3476 }
3477 // Tell jvm's signal handler the signal is taken care of.
3478 return true;
3479 }
3480
3481 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3482 bool chained = false;
3483 // signal-chaining
3484 if (UseSignalChaining) {
3485 struct sigaction *actp = get_chained_signal_action(sig);
3486 if (actp != NULL) {
3487 chained = call_chained_handler(actp, sig, siginfo, context);
3488 }
3489 }
3490 return chained;
3491 }
3492
3493 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3494 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3495 return &sigact[sig];
3496 }
3497 return NULL;
3498 }
3499
3500 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3501 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3502 sigact[sig] = oldAct;
3503 sigs |= (unsigned int)1 << sig;
3504 }
3505
3506 // for diagnostic
3507 int os::Linux::sigflags[MAXSIGNUM];
3508
3509 int os::Linux::get_our_sigflags(int sig) {
3510 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3511 return sigflags[sig];
3512 }
3513
3514 void os::Linux::set_our_sigflags(int sig, int flags) {
3515 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3516 sigflags[sig] = flags;
3517 }
3518
3519 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3520 // Check for overwrite.
3521 struct sigaction oldAct;
3522 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3523
3524 void* oldhand = oldAct.sa_sigaction
3525 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3526 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3527 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3528 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3529 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3530 if (AllowUserSignalHandlers || !set_installed) {
3531 // Do not overwrite; user takes responsibility to forward to us.
3532 return;
3533 } else if (UseSignalChaining) {
3534 // save the old handler in jvm
3535 save_preinstalled_handler(sig, oldAct);
3536 // libjsig also interposes the sigaction() call below and saves the
3537 // old sigaction on it own.
3538 } else {
3539 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3540 "%#lx for signal %d.", (long)oldhand, sig));
3541 }
3542 }
3543
3544 struct sigaction sigAct;
3545 sigfillset(&(sigAct.sa_mask));
3546 sigAct.sa_handler = SIG_DFL;
3547 if (!set_installed) {
3548 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3549 } else {
3550 sigAct.sa_sigaction = signalHandler;
3551 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3552 }
3553 // Save flags, which are set by ours
3554 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3555 sigflags[sig] = sigAct.sa_flags;
3556
3557 int ret = sigaction(sig, &sigAct, &oldAct);
3558 assert(ret == 0, "check");
3559
3560 void* oldhand2 = oldAct.sa_sigaction
3561 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3562 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3563 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3564 }
3565
3566 // install signal handlers for signals that HotSpot needs to
3567 // handle in order to support Java-level exception handling.
3568
3569 void os::Linux::install_signal_handlers() {
3570 if (!signal_handlers_are_installed) {
3571 signal_handlers_are_installed = true;
3572
3573 // signal-chaining
3574 typedef void (*signal_setting_t)();
3575 signal_setting_t begin_signal_setting = NULL;
3576 signal_setting_t end_signal_setting = NULL;
3577 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3578 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3579 if (begin_signal_setting != NULL) {
3580 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3581 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3582 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3583 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3584 libjsig_is_loaded = true;
3585 assert(UseSignalChaining, "should enable signal-chaining");
3586 }
3587 if (libjsig_is_loaded) {
3588 // Tell libjsig jvm is setting signal handlers
3589 (*begin_signal_setting)();
3590 }
3591
3592 set_signal_handler(SIGSEGV, true);
3593 set_signal_handler(SIGPIPE, true);
3594 set_signal_handler(SIGBUS, true);
3595 set_signal_handler(SIGILL, true);
3596 set_signal_handler(SIGFPE, true);
3597 set_signal_handler(SIGXFSZ, true);
3598
3599 if (libjsig_is_loaded) {
3600 // Tell libjsig jvm finishes setting signal handlers
3601 (*end_signal_setting)();
3602 }
3603
3604 // We don't activate signal checker if libjsig is in place, we trust ourselves
3605 // and if UserSignalHandler is installed all bets are off
3606 if (CheckJNICalls) {
3607 if (libjsig_is_loaded) {
3608 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3609 check_signals = false;
3610 }
3611 if (AllowUserSignalHandlers) {
3612 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3613 check_signals = false;
3614 }
3615 }
3616 }
3617 }
3618
3619 // This is the fastest way to get thread cpu time on Linux.
3620 // Returns cpu time (user+sys) for any thread, not only for current.
3621 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3622 // It might work on 2.6.10+ with a special kernel/glibc patch.
3623 // For reference, please, see IEEE Std 1003.1-2004:
3624 // http://www.unix.org/single_unix_specification
3625
3626 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3627 struct timespec tp;
3628 int rc = os::Linux::clock_gettime(clockid, &tp);
3629 assert(rc == 0, "clock_gettime is expected to return 0 code");
3630
3631 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3632 }
3633
3634 /////
3635 // glibc on Linux platform uses non-documented flag
3636 // to indicate, that some special sort of signal
3637 // trampoline is used.
3638 // We will never set this flag, and we should
3639 // ignore this flag in our diagnostic
3640 #ifdef SIGNIFICANT_SIGNAL_MASK
3641 #undef SIGNIFICANT_SIGNAL_MASK
3642 #endif
3643 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3644
3645 static const char* get_signal_handler_name(address handler,
3646 char* buf, int buflen) {
3647 int offset;
3648 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3649 if (found) {
3650 // skip directory names
3651 const char *p1, *p2;
3652 p1 = buf;
3653 size_t len = strlen(os::file_separator());
3654 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3655 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3656 } else {
3657 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3658 }
3659 return buf;
3660 }
3661
3662 static void print_signal_handler(outputStream* st, int sig,
3663 char* buf, size_t buflen) {
3664 struct sigaction sa;
3665
3666 sigaction(sig, NULL, &sa);
3667
3668 // See comment for SIGNIFICANT_SIGNAL_MASK define
3669 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3670
3671 st->print("%s: ", os::exception_name(sig, buf, buflen));
3672
3673 address handler = (sa.sa_flags & SA_SIGINFO)
3674 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3675 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3676
3677 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3678 st->print("SIG_DFL");
3679 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3680 st->print("SIG_IGN");
3681 } else {
3682 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3683 }
3684
3685 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3686
3687 address rh = VMError::get_resetted_sighandler(sig);
3688 // May be, handler was resetted by VMError?
3689 if(rh != NULL) {
3690 handler = rh;
3691 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3692 }
3693
3694 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3695
3696 // Check: is it our handler?
3697 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3698 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3699 // It is our signal handler
3700 // check for flags, reset system-used one!
3701 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3702 st->print(
3703 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3704 os::Linux::get_our_sigflags(sig));
3705 }
3706 }
3707 st->cr();
3708 }
3709
3710
3711 #define DO_SIGNAL_CHECK(sig) \
3712 if (!sigismember(&check_signal_done, sig)) \
3713 os::Linux::check_signal_handler(sig)
3714
3715 // This method is a periodic task to check for misbehaving JNI applications
3716 // under CheckJNI, we can add any periodic checks here
3717
3718 void os::run_periodic_checks() {
3719
3720 if (check_signals == false) return;
3721
3722 // SEGV and BUS if overridden could potentially prevent
3723 // generation of hs*.log in the event of a crash, debugging
3724 // such a case can be very challenging, so we absolutely
3725 // check the following for a good measure:
3726 DO_SIGNAL_CHECK(SIGSEGV);
3727 DO_SIGNAL_CHECK(SIGILL);
3728 DO_SIGNAL_CHECK(SIGFPE);
3729 DO_SIGNAL_CHECK(SIGBUS);
3730 DO_SIGNAL_CHECK(SIGPIPE);
3731 DO_SIGNAL_CHECK(SIGXFSZ);
3732
3733
3734 // ReduceSignalUsage allows the user to override these handlers
3735 // see comments at the very top and jvm_solaris.h
3736 if (!ReduceSignalUsage) {
3737 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3738 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3739 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3740 DO_SIGNAL_CHECK(BREAK_SIGNAL);
3741 }
3742
3743 DO_SIGNAL_CHECK(SR_signum);
3744 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3745 }
3746
3747 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3748
3749 static os_sigaction_t os_sigaction = NULL;
3750
3751 void os::Linux::check_signal_handler(int sig) {
3752 char buf[O_BUFLEN];
3753 address jvmHandler = NULL;
3754
3755
3756 struct sigaction act;
3757 if (os_sigaction == NULL) {
3758 // only trust the default sigaction, in case it has been interposed
3759 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3760 if (os_sigaction == NULL) return;
3761 }
3762
3763 os_sigaction(sig, (struct sigaction*)NULL, &act);
3764
3765
3766 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3767
3768 address thisHandler = (act.sa_flags & SA_SIGINFO)
3769 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3770 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3771
3772
3773 switch(sig) {
3774 case SIGSEGV:
3775 case SIGBUS:
3776 case SIGFPE:
3777 case SIGPIPE:
3778 case SIGILL:
3779 case SIGXFSZ:
3780 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3781 break;
3782
3783 case SHUTDOWN1_SIGNAL:
3784 case SHUTDOWN2_SIGNAL:
3785 case SHUTDOWN3_SIGNAL:
3786 case BREAK_SIGNAL:
3787 jvmHandler = (address)user_handler();
3788 break;
3789
3790 case INTERRUPT_SIGNAL:
3791 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3792 break;
3793
3794 default:
3795 if (sig == SR_signum) {
3796 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3797 } else {
3798 return;
3799 }
3800 break;
3801 }
3802
3803 if (thisHandler != jvmHandler) {
3804 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3805 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3806 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3807 // No need to check this sig any longer
3808 sigaddset(&check_signal_done, sig);
3809 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3810 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3811 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3812 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
3813 // No need to check this sig any longer
3814 sigaddset(&check_signal_done, sig);
3815 }
3816
3817 // Dump all the signal
3818 if (sigismember(&check_signal_done, sig)) {
3819 print_signal_handlers(tty, buf, O_BUFLEN);
3820 }
3821 }
3822
3823 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3824
3825 extern bool signal_name(int signo, char* buf, size_t len);
3826
3827 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3828 if (0 < exception_code && exception_code <= SIGRTMAX) {
3829 // signal
3830 if (!signal_name(exception_code, buf, size)) {
3831 jio_snprintf(buf, size, "SIG%d", exception_code);
3832 }
3833 return buf;
3834 } else {
3835 return NULL;
3836 }
3837 }
3838
3839 // this is called _before_ the most of global arguments have been parsed
3840 void os::init(void) {
3841 char dummy; /* used to get a guess on initial stack address */
3842 // first_hrtime = gethrtime();
3843
3844 // With LinuxThreads the JavaMain thread pid (primordial thread)
3845 // is different than the pid of the java launcher thread.
3846 // So, on Linux, the launcher thread pid is passed to the VM
3847 // via the sun.java.launcher.pid property.
3848 // Use this property instead of getpid() if it was correctly passed.
3849 // See bug 6351349.
3850 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3851
3852 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3853
3854 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3855
3856 init_random(1234567);
3857
3858 ThreadCritical::initialize();
3859
3860 Linux::set_page_size(sysconf(_SC_PAGESIZE));
3861 if (Linux::page_size() == -1) {
3862 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
3863 strerror(errno)));
3864 }
3865 init_page_sizes((size_t) Linux::page_size());
3866
3867 Linux::initialize_system_info();
3868
3869 // main_thread points to the aboriginal thread
3870 Linux::_main_thread = pthread_self();
3871
3872 Linux::clock_init();
3873 initial_time_count = os::elapsed_counter();
3874 pthread_mutex_init(&dl_mutex, NULL);
3875 }
3876
3877 // To install functions for atexit system call
3878 extern "C" {
3879 static void perfMemory_exit_helper() {
3880 perfMemory_exit();
3881 }
3882 }
3883
3884 // this is called _after_ the global arguments have been parsed
3885 jint os::init_2(void)
3886 {
3887 Linux::fast_thread_clock_init();
3888
3889 // Allocate a single page and mark it as readable for safepoint polling
3890 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3891 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3892
3893 os::set_polling_page( polling_page );
3894
3895 #ifndef PRODUCT
3896 if(Verbose && PrintMiscellaneous)
3897 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3898 #endif
3899
3900 if (!UseMembar) {
3901 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3902 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3903 os::set_memory_serialize_page( mem_serialize_page );
3904
3905 #ifndef PRODUCT
3906 if(Verbose && PrintMiscellaneous)
3907 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3908 #endif
3909 }
3910
3911 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3912
3913 // initialize suspend/resume support - must do this before signal_sets_init()
3914 if (SR_initialize() != 0) {
3915 perror("SR_initialize failed");
3916 return JNI_ERR;
3917 }
3918
3919 Linux::signal_sets_init();
3920 Linux::install_signal_handlers();
3921
3922 size_t threadStackSizeInBytes = ThreadStackSize * K;
3923 if (threadStackSizeInBytes != 0 &&
3924 threadStackSizeInBytes < Linux::min_stack_allowed) {
3925 tty->print_cr("\nThe stack size specified is too small, "
3926 "Specify at least %dk",
3927 Linux::min_stack_allowed / K);
3928 return JNI_ERR;
3929 }
3930
3931 // Make the stack size a multiple of the page size so that
3932 // the yellow/red zones can be guarded.
3933 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3934 vm_page_size()));
3935
3936 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3937
3938 Linux::libpthread_init();
3939 if (PrintMiscellaneous && (Verbose || WizardMode)) {
3940 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3941 Linux::glibc_version(), Linux::libpthread_version(),
3942 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3943 }
3944
3945 if (UseNUMA) {
3946 if (!Linux::libnuma_init()) {
3947 UseNUMA = false;
3948 } else {
3949 if ((Linux::numa_max_node() < 1)) {
3950 // There's only one node(they start from 0), disable NUMA.
3951 UseNUMA = false;
3952 }
3953 }
3954 if (!UseNUMA && ForceNUMA) {
3955 UseNUMA = true;
3956 }
3957 }
3958
3959 if (MaxFDLimit) {
3960 // set the number of file descriptors to max. print out error
3961 // if getrlimit/setrlimit fails but continue regardless.
3962 struct rlimit nbr_files;
3963 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3964 if (status != 0) {
3965 if (PrintMiscellaneous && (Verbose || WizardMode))
3966 perror("os::init_2 getrlimit failed");
3967 } else {
3968 nbr_files.rlim_cur = nbr_files.rlim_max;
3969 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3970 if (status != 0) {
3971 if (PrintMiscellaneous && (Verbose || WizardMode))
3972 perror("os::init_2 setrlimit failed");
3973 }
3974 }
3975 }
3976
3977 // Initialize lock used to serialize thread creation (see os::create_thread)
3978 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3979
3980 // Initialize HPI.
3981 jint hpi_result = hpi::initialize();
3982 if (hpi_result != JNI_OK) {
3983 tty->print_cr("There was an error trying to initialize the HPI library.");
3984 return hpi_result;
3985 }
3986
3987 // at-exit methods are called in the reverse order of their registration.
3988 // atexit functions are called on return from main or as a result of a
3989 // call to exit(3C). There can be only 32 of these functions registered
3990 // and atexit() does not set errno.
3991
3992 if (PerfAllowAtExitRegistration) {
3993 // only register atexit functions if PerfAllowAtExitRegistration is set.
3994 // atexit functions can be delayed until process exit time, which
3995 // can be problematic for embedded VM situations. Embedded VMs should
3996 // call DestroyJavaVM() to assure that VM resources are released.
3997
3998 // note: perfMemory_exit_helper atexit function may be removed in
3999 // the future if the appropriate cleanup code can be added to the
4000 // VM_Exit VMOperation's doit method.
4001 if (atexit(perfMemory_exit_helper) != 0) {
4002 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4003 }
4004 }
4005
4006 // initialize thread priority policy
4007 prio_init();
4008
4009 return JNI_OK;
4010 }
4011
4012 // this is called at the end of vm_initialization
4013 void os::init_3(void) { }
4014
4015 // Mark the polling page as unreadable
4016 void os::make_polling_page_unreadable(void) {
4017 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4018 fatal("Could not disable polling page");
4019 };
4020
4021 // Mark the polling page as readable
4022 void os::make_polling_page_readable(void) {
4023 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4024 fatal("Could not enable polling page");
4025 }
4026 };
4027
4028 int os::active_processor_count() {
4029 // Linux doesn't yet have a (official) notion of processor sets,
4030 // so just return the number of online processors.
4031 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4032 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4033 return online_cpus;
4034 }
4035
4036 bool os::distribute_processes(uint length, uint* distribution) {
4037 // Not yet implemented.
4038 return false;
4039 }
4040
4041 bool os::bind_to_processor(uint processor_id) {
4042 // Not yet implemented.
4043 return false;
4044 }
4045
4046 ///
4047
4048 // Suspends the target using the signal mechanism and then grabs the PC before
4049 // resuming the target. Used by the flat-profiler only
4050 ExtendedPC os::get_thread_pc(Thread* thread) {
4051 // Make sure that it is called by the watcher for the VMThread
4052 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4053 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4054
4055 ExtendedPC epc;
4056
4057 OSThread* osthread = thread->osthread();
4058 if (do_suspend(osthread)) {
4059 if (osthread->ucontext() != NULL) {
4060 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4061 } else {
4062 // NULL context is unexpected, double-check this is the VMThread
4063 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4064 }
4065 do_resume(osthread);
4066 }
4067 // failure means pthread_kill failed for some reason - arguably this is
4068 // a fatal problem, but such problems are ignored elsewhere
4069
4070 return epc;
4071 }
4072
4073 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4074 {
4075 if (is_NPTL()) {
4076 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4077 } else {
4078 #ifndef IA64
4079 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4080 // word back to default 64bit precision if condvar is signaled. Java
4081 // wants 53bit precision. Save and restore current value.
4082 int fpu = get_fpu_control_word();
4083 #endif // IA64
4084 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4085 #ifndef IA64
4086 set_fpu_control_word(fpu);
4087 #endif // IA64
4088 return status;
4089 }
4090 }
4091
4092 ////////////////////////////////////////////////////////////////////////////////
4093 // debug support
4094
4095 static address same_page(address x, address y) {
4096 int page_bits = -os::vm_page_size();
4097 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4098 return x;
4099 else if (x > y)
4100 return (address)(intptr_t(y) | ~page_bits) + 1;
4101 else
4102 return (address)(intptr_t(y) & page_bits);
4103 }
4104
4105 bool os::find(address addr, outputStream* st) {
4106 Dl_info dlinfo;
4107 memset(&dlinfo, 0, sizeof(dlinfo));
4108 if (dladdr(addr, &dlinfo)) {
4109 st->print(PTR_FORMAT ": ", addr);
4110 if (dlinfo.dli_sname != NULL) {
4111 st->print("%s+%#x", dlinfo.dli_sname,
4112 addr - (intptr_t)dlinfo.dli_saddr);
4113 } else if (dlinfo.dli_fname) {
4114 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4115 } else {
4116 st->print("<absolute address>");
4117 }
4118 if (dlinfo.dli_fname) {
4119 st->print(" in %s", dlinfo.dli_fname);
4120 }
4121 if (dlinfo.dli_fbase) {
4122 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4123 }
4124 st->cr();
4125
4126 if (Verbose) {
4127 // decode some bytes around the PC
4128 address begin = same_page(addr-40, addr);
4129 address end = same_page(addr+40, addr);
4130 address lowest = (address) dlinfo.dli_sname;
4131 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4132 if (begin < lowest) begin = lowest;
4133 Dl_info dlinfo2;
4134 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4135 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4136 end = (address) dlinfo2.dli_saddr;
4137 Disassembler::decode(begin, end, st);
4138 }
4139 return true;
4140 }
4141 return false;
4142 }
4143
4144 ////////////////////////////////////////////////////////////////////////////////
4145 // misc
4146
4147 // This does not do anything on Linux. This is basically a hook for being
4148 // able to use structured exception handling (thread-local exception filters)
4149 // on, e.g., Win32.
4150 void
4151 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4152 JavaCallArguments* args, Thread* thread) {
4153 f(value, method, args, thread);
4154 }
4155
4156 void os::print_statistics() {
4157 }
4158
4159 int os::message_box(const char* title, const char* message) {
4160 int i;
4161 fdStream err(defaultStream::error_fd());
4162 for (i = 0; i < 78; i++) err.print_raw("=");
4163 err.cr();
4164 err.print_raw_cr(title);
4165 for (i = 0; i < 78; i++) err.print_raw("-");
4166 err.cr();
4167 err.print_raw_cr(message);
4168 for (i = 0; i < 78; i++) err.print_raw("=");
4169 err.cr();
4170
4171 char buf[16];
4172 // Prevent process from exiting upon "read error" without consuming all CPU
4173 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4174
4175 return buf[0] == 'y' || buf[0] == 'Y';
4176 }
4177
4178 int os::stat(const char *path, struct stat *sbuf) {
4179 char pathbuf[MAX_PATH];
4180 if (strlen(path) > MAX_PATH - 1) {
4181 errno = ENAMETOOLONG;
4182 return -1;
4183 }
4184 hpi::native_path(strcpy(pathbuf, path));
4185 return ::stat(pathbuf, sbuf);
4186 }
4187
4188 bool os::check_heap(bool force) {
4189 return true;
4190 }
4191
4192 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4193 return ::vsnprintf(buf, count, format, args);
4194 }
4195
4196 // Is a (classpath) directory empty?
4197 bool os::dir_is_empty(const char* path) {
4198 DIR *dir = NULL;
4199 struct dirent *ptr;
4200
4201 dir = opendir(path);
4202 if (dir == NULL) return true;
4203
4204 /* Scan the directory */
4205 bool result = true;
4206 char buf[sizeof(struct dirent) + MAX_PATH];
4207 while (result && (ptr = ::readdir(dir)) != NULL) {
4208 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4209 result = false;
4210 }
4211 }
4212 closedir(dir);
4213 return result;
4214 }
4215
4216 // create binary file, rewriting existing file if required
4217 int os::create_binary_file(const char* path, bool rewrite_existing) {
4218 int oflags = O_WRONLY | O_CREAT;
4219 if (!rewrite_existing) {
4220 oflags |= O_EXCL;
4221 }
4222 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4223 }
4224
4225 // return current position of file pointer
4226 jlong os::current_file_offset(int fd) {
4227 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4228 }
4229
4230 // move file pointer to the specified offset
4231 jlong os::seek_to_file_offset(int fd, jlong offset) {
4232 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4233 }
4234
4235 // Map a block of memory.
4236 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4237 char *addr, size_t bytes, bool read_only,
4238 bool allow_exec) {
4239 int prot;
4240 int flags;
4241
4242 if (read_only) {
4243 prot = PROT_READ;
4244 flags = MAP_SHARED;
4245 } else {
4246 prot = PROT_READ | PROT_WRITE;
4247 flags = MAP_PRIVATE;
4248 }
4249
4250 if (allow_exec) {
4251 prot |= PROT_EXEC;
4252 }
4253
4254 if (addr != NULL) {
4255 flags |= MAP_FIXED;
4256 }
4257
4258 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4259 fd, file_offset);
4260 if (mapped_address == MAP_FAILED) {
4261 return NULL;
4262 }
4263 return mapped_address;
4264 }
4265
4266
4267 // Remap a block of memory.
4268 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4269 char *addr, size_t bytes, bool read_only,
4270 bool allow_exec) {
4271 // same as map_memory() on this OS
4272 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4273 allow_exec);
4274 }
4275
4276
4277 // Unmap a block of memory.
4278 bool os::unmap_memory(char* addr, size_t bytes) {
4279 return munmap(addr, bytes) == 0;
4280 }
4281
4282 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4283
4284 static clockid_t thread_cpu_clockid(Thread* thread) {
4285 pthread_t tid = thread->osthread()->pthread_id();
4286 clockid_t clockid;
4287
4288 // Get thread clockid
4289 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4290 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4291 return clockid;
4292 }
4293
4294 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4295 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4296 // of a thread.
4297 //
4298 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4299 // the fast estimate available on the platform.
4300
4301 jlong os::current_thread_cpu_time() {
4302 if (os::Linux::supports_fast_thread_cpu_time()) {
4303 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4304 } else {
4305 // return user + sys since the cost is the same
4306 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4307 }
4308 }
4309
4310 jlong os::thread_cpu_time(Thread* thread) {
4311 // consistent with what current_thread_cpu_time() returns
4312 if (os::Linux::supports_fast_thread_cpu_time()) {
4313 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4314 } else {
4315 return slow_thread_cpu_time(thread, true /* user + sys */);
4316 }
4317 }
4318
4319 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4320 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4321 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4322 } else {
4323 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4324 }
4325 }
4326
4327 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4328 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4329 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4330 } else {
4331 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4332 }
4333 }
4334
4335 //
4336 // -1 on error.
4337 //
4338
4339 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4340 static bool proc_pid_cpu_avail = true;
4341 static bool proc_task_unchecked = true;
4342 static const char *proc_stat_path = "/proc/%d/stat";
4343 pid_t tid = thread->osthread()->thread_id();
4344 int i;
4345 char *s;
4346 char stat[2048];
4347 int statlen;
4348 char proc_name[64];
4349 int count;
4350 long sys_time, user_time;
4351 char string[64];
4352 char cdummy;
4353 int idummy;
4354 long ldummy;
4355 FILE *fp;
4356
4357 // We first try accessing /proc/<pid>/cpu since this is faster to
4358 // process. If this file is not present (linux kernels 2.5 and above)
4359 // then we open /proc/<pid>/stat.
4360 if ( proc_pid_cpu_avail ) {
4361 sprintf(proc_name, "/proc/%d/cpu", tid);
4362 fp = fopen(proc_name, "r");
4363 if ( fp != NULL ) {
4364 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4365 fclose(fp);
4366 if ( count != 3 ) return -1;
4367
4368 if (user_sys_cpu_time) {
4369 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4370 } else {
4371 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4372 }
4373 }
4374 else proc_pid_cpu_avail = false;
4375 }
4376
4377 // The /proc/<tid>/stat aggregates per-process usage on
4378 // new Linux kernels 2.6+ where NPTL is supported.
4379 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4380 // See bug 6328462.
4381 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4382 // and possibly in some other cases, so we check its availability.
4383 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4384 // This is executed only once
4385 proc_task_unchecked = false;
4386 fp = fopen("/proc/self/task", "r");
4387 if (fp != NULL) {
4388 proc_stat_path = "/proc/self/task/%d/stat";
4389 fclose(fp);
4390 }
4391 }
4392
4393 sprintf(proc_name, proc_stat_path, tid);
4394 fp = fopen(proc_name, "r");
4395 if ( fp == NULL ) return -1;
4396 statlen = fread(stat, 1, 2047, fp);
4397 stat[statlen] = '\0';
4398 fclose(fp);
4399
4400 // Skip pid and the command string. Note that we could be dealing with
4401 // weird command names, e.g. user could decide to rename java launcher
4402 // to "java 1.4.2 :)", then the stat file would look like
4403 // 1234 (java 1.4.2 :)) R ... ...
4404 // We don't really need to know the command string, just find the last
4405 // occurrence of ")" and then start parsing from there. See bug 4726580.
4406 s = strrchr(stat, ')');
4407 i = 0;
4408 if (s == NULL ) return -1;
4409
4410 // Skip blank chars
4411 do s++; while (isspace(*s));
4412
4413 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4414 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4415 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4416 &user_time, &sys_time);
4417 if ( count != 13 ) return -1;
4418 if (user_sys_cpu_time) {
4419 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4420 } else {
4421 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4422 }
4423 }
4424
4425 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4426 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4427 info_ptr->may_skip_backward = false; // elapsed time not wall time
4428 info_ptr->may_skip_forward = false; // elapsed time not wall time
4429 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4430 }
4431
4432 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4433 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4434 info_ptr->may_skip_backward = false; // elapsed time not wall time
4435 info_ptr->may_skip_forward = false; // elapsed time not wall time
4436 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4437 }
4438
4439 bool os::is_thread_cpu_time_supported() {
4440 return true;
4441 }
4442
4443 // System loadavg support. Returns -1 if load average cannot be obtained.
4444 // Linux doesn't yet have a (official) notion of processor sets,
4445 // so just return the system wide load average.
4446 int os::loadavg(double loadavg[], int nelem) {
4447 return ::getloadavg(loadavg, nelem);
4448 }
4449
4450 void os::pause() {
4451 char filename[MAX_PATH];
4452 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4453 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4454 } else {
4455 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4456 }
4457
4458 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4459 if (fd != -1) {
4460 struct stat buf;
4461 close(fd);
4462 while (::stat(filename, &buf) == 0) {
4463 (void)::poll(NULL, 0, 100);
4464 }
4465 } else {
4466 jio_fprintf(stderr,
4467 "Could not open pause file '%s', continuing immediately.\n", filename);
4468 }
4469 }
4470
4471 extern "C" {
4472
4473 /**
4474 * NOTE: the following code is to keep the green threads code
4475 * in the libjava.so happy. Once the green threads is removed,
4476 * these code will no longer be needed.
4477 */
4478 int
4479 jdk_waitpid(pid_t pid, int* status, int options) {
4480 return waitpid(pid, status, options);
4481 }
4482
4483 int
4484 fork1() {
4485 return fork();
4486 }
4487
4488 int
4489 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4490 return sem_init(sem, pshared, value);
4491 }
4492
4493 int
4494 jdk_sem_post(sem_t *sem) {
4495 return sem_post(sem);
4496 }
4497
4498 int
4499 jdk_sem_wait(sem_t *sem) {
4500 return sem_wait(sem);
4501 }
4502
4503 int
4504 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4505 return pthread_sigmask(how , newmask, oldmask);
4506 }
4507
4508 }
4509
4510 // Refer to the comments in os_solaris.cpp park-unpark.
4511 //
4512 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4513 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4514 // For specifics regarding the bug see GLIBC BUGID 261237 :
4515 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4516 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4517 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4518 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4519 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4520 // and monitorenter when we're using 1-0 locking. All those operations may result in
4521 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4522 // of libpthread avoids the problem, but isn't practical.
4523 //
4524 // Possible remedies:
4525 //
4526 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4527 // This is palliative and probabilistic, however. If the thread is preempted
4528 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4529 // than the minimum period may have passed, and the abstime may be stale (in the
4530 // past) resultin in a hang. Using this technique reduces the odds of a hang
4531 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4532 //
4533 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4534 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4535 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4536 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4537 // thread.
4538 //
4539 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4540 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4541 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4542 // This also works well. In fact it avoids kernel-level scalability impediments
4543 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4544 // timers in a graceful fashion.
4545 //
4546 // 4. When the abstime value is in the past it appears that control returns
4547 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4548 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4549 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4550 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4551 // It may be possible to avoid reinitialization by checking the return
4552 // value from pthread_cond_timedwait(). In addition to reinitializing the
4553 // condvar we must establish the invariant that cond_signal() is only called
4554 // within critical sections protected by the adjunct mutex. This prevents
4555 // cond_signal() from "seeing" a condvar that's in the midst of being
4556 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4557 // desirable signal-after-unlock optimization that avoids futile context switching.
4558 //
4559 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4560 // structure when a condvar is used or initialized. cond_destroy() would
4561 // release the helper structure. Our reinitialize-after-timedwait fix
4562 // put excessive stress on malloc/free and locks protecting the c-heap.
4563 //
4564 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4565 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4566 // and only enabling the work-around for vulnerable environments.
4567
4568 // utility to compute the abstime argument to timedwait:
4569 // millis is the relative timeout time
4570 // abstime will be the absolute timeout time
4571 // TODO: replace compute_abstime() with unpackTime()
4572
4573 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4574 if (millis < 0) millis = 0;
4575 struct timeval now;
4576 int status = gettimeofday(&now, NULL);
4577 assert(status == 0, "gettimeofday");
4578 jlong seconds = millis / 1000;
4579 millis %= 1000;
4580 if (seconds > 50000000) { // see man cond_timedwait(3T)
4581 seconds = 50000000;
4582 }
4583 abstime->tv_sec = now.tv_sec + seconds;
4584 long usec = now.tv_usec + millis * 1000;
4585 if (usec >= 1000000) {
4586 abstime->tv_sec += 1;
4587 usec -= 1000000;
4588 }
4589 abstime->tv_nsec = usec * 1000;
4590 return abstime;
4591 }
4592
4593
4594 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4595 // Conceptually TryPark() should be equivalent to park(0).
4596
4597 int os::PlatformEvent::TryPark() {
4598 for (;;) {
4599 const int v = _Event ;
4600 guarantee ((v == 0) || (v == 1), "invariant") ;
4601 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4602 }
4603 }
4604
4605 void os::PlatformEvent::park() { // AKA "down()"
4606 // Invariant: Only the thread associated with the Event/PlatformEvent
4607 // may call park().
4608 // TODO: assert that _Assoc != NULL or _Assoc == Self
4609 int v ;
4610 for (;;) {
4611 v = _Event ;
4612 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4613 }
4614 guarantee (v >= 0, "invariant") ;
4615 if (v == 0) {
4616 // Do this the hard way by blocking ...
4617 int status = pthread_mutex_lock(_mutex);
4618 assert_status(status == 0, status, "mutex_lock");
4619 guarantee (_nParked == 0, "invariant") ;
4620 ++ _nParked ;
4621 while (_Event < 0) {
4622 status = pthread_cond_wait(_cond, _mutex);
4623 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4624 // Treat this the same as if the wait was interrupted
4625 if (status == ETIME) { status = EINTR; }
4626 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4627 }
4628 -- _nParked ;
4629
4630 // In theory we could move the ST of 0 into _Event past the unlock(),
4631 // but then we'd need a MEMBAR after the ST.
4632 _Event = 0 ;
4633 status = pthread_mutex_unlock(_mutex);
4634 assert_status(status == 0, status, "mutex_unlock");
4635 }
4636 guarantee (_Event >= 0, "invariant") ;
4637 }
4638
4639 int os::PlatformEvent::park(jlong millis) {
4640 guarantee (_nParked == 0, "invariant") ;
4641
4642 int v ;
4643 for (;;) {
4644 v = _Event ;
4645 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4646 }
4647 guarantee (v >= 0, "invariant") ;
4648 if (v != 0) return OS_OK ;
4649
4650 // We do this the hard way, by blocking the thread.
4651 // Consider enforcing a minimum timeout value.
4652 struct timespec abst;
4653 compute_abstime(&abst, millis);
4654
4655 int ret = OS_TIMEOUT;
4656 int status = pthread_mutex_lock(_mutex);
4657 assert_status(status == 0, status, "mutex_lock");
4658 guarantee (_nParked == 0, "invariant") ;
4659 ++_nParked ;
4660
4661 // Object.wait(timo) will return because of
4662 // (a) notification
4663 // (b) timeout
4664 // (c) thread.interrupt
4665 //
4666 // Thread.interrupt and object.notify{All} both call Event::set.
4667 // That is, we treat thread.interrupt as a special case of notification.
4668 // The underlying Solaris implementation, cond_timedwait, admits
4669 // spurious/premature wakeups, but the JLS/JVM spec prevents the
4670 // JVM from making those visible to Java code. As such, we must
4671 // filter out spurious wakeups. We assume all ETIME returns are valid.
4672 //
4673 // TODO: properly differentiate simultaneous notify+interrupt.
4674 // In that case, we should propagate the notify to another waiter.
4675
4676 while (_Event < 0) {
4677 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4678 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4679 pthread_cond_destroy (_cond);
4680 pthread_cond_init (_cond, NULL) ;
4681 }
4682 assert_status(status == 0 || status == EINTR ||
4683 status == ETIME || status == ETIMEDOUT,
4684 status, "cond_timedwait");
4685 if (!FilterSpuriousWakeups) break ; // previous semantics
4686 if (status == ETIME || status == ETIMEDOUT) break ;
4687 // We consume and ignore EINTR and spurious wakeups.
4688 }
4689 --_nParked ;
4690 if (_Event >= 0) {
4691 ret = OS_OK;
4692 }
4693 _Event = 0 ;
4694 status = pthread_mutex_unlock(_mutex);
4695 assert_status(status == 0, status, "mutex_unlock");
4696 assert (_nParked == 0, "invariant") ;
4697 return ret;
4698 }
4699
4700 void os::PlatformEvent::unpark() {
4701 int v, AnyWaiters ;
4702 for (;;) {
4703 v = _Event ;
4704 if (v > 0) {
4705 // The LD of _Event could have reordered or be satisfied
4706 // by a read-aside from this processor's write buffer.
4707 // To avoid problems execute a barrier and then
4708 // ratify the value.
4709 OrderAccess::fence() ;
4710 if (_Event == v) return ;
4711 continue ;
4712 }
4713 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4714 }
4715 if (v < 0) {
4716 // Wait for the thread associated with the event to vacate
4717 int status = pthread_mutex_lock(_mutex);
4718 assert_status(status == 0, status, "mutex_lock");
4719 AnyWaiters = _nParked ;
4720 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4721 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4722 AnyWaiters = 0 ;
4723 pthread_cond_signal (_cond);
4724 }
4725 status = pthread_mutex_unlock(_mutex);
4726 assert_status(status == 0, status, "mutex_unlock");
4727 if (AnyWaiters != 0) {
4728 status = pthread_cond_signal(_cond);
4729 assert_status(status == 0, status, "cond_signal");
4730 }
4731 }
4732
4733 // Note that we signal() _after dropping the lock for "immortal" Events.
4734 // This is safe and avoids a common class of futile wakeups. In rare
4735 // circumstances this can cause a thread to return prematurely from
4736 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4737 // simply re-test the condition and re-park itself.
4738 }
4739
4740
4741 // JSR166
4742 // -------------------------------------------------------
4743
4744 /*
4745 * The solaris and linux implementations of park/unpark are fairly
4746 * conservative for now, but can be improved. They currently use a
4747 * mutex/condvar pair, plus a a count.
4748 * Park decrements count if > 0, else does a condvar wait. Unpark
4749 * sets count to 1 and signals condvar. Only one thread ever waits
4750 * on the condvar. Contention seen when trying to park implies that someone
4751 * is unparking you, so don't wait. And spurious returns are fine, so there
4752 * is no need to track notifications.
4753 */
4754
4755
4756 #define NANOSECS_PER_SEC 1000000000
4757 #define NANOSECS_PER_MILLISEC 1000000
4758 #define MAX_SECS 100000000
4759 /*
4760 * This code is common to linux and solaris and will be moved to a
4761 * common place in dolphin.
4762 *
4763 * The passed in time value is either a relative time in nanoseconds
4764 * or an absolute time in milliseconds. Either way it has to be unpacked
4765 * into suitable seconds and nanoseconds components and stored in the
4766 * given timespec structure.
4767 * Given time is a 64-bit value and the time_t used in the timespec is only
4768 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4769 * overflow if times way in the future are given. Further on Solaris versions
4770 * prior to 10 there is a restriction (see cond_timedwait) that the specified
4771 * number of seconds, in abstime, is less than current_time + 100,000,000.
4772 * As it will be 28 years before "now + 100000000" will overflow we can
4773 * ignore overflow and just impose a hard-limit on seconds using the value
4774 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4775 * years from "now".
4776 */
4777
4778 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4779 assert (time > 0, "convertTime");
4780
4781 struct timeval now;
4782 int status = gettimeofday(&now, NULL);
4783 assert(status == 0, "gettimeofday");
4784
4785 time_t max_secs = now.tv_sec + MAX_SECS;
4786
4787 if (isAbsolute) {
4788 jlong secs = time / 1000;
4789 if (secs > max_secs) {
4790 absTime->tv_sec = max_secs;
4791 }
4792 else {
4793 absTime->tv_sec = secs;
4794 }
4795 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4796 }
4797 else {
4798 jlong secs = time / NANOSECS_PER_SEC;
4799 if (secs >= MAX_SECS) {
4800 absTime->tv_sec = max_secs;
4801 absTime->tv_nsec = 0;
4802 }
4803 else {
4804 absTime->tv_sec = now.tv_sec + secs;
4805 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4806 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4807 absTime->tv_nsec -= NANOSECS_PER_SEC;
4808 ++absTime->tv_sec; // note: this must be <= max_secs
4809 }
4810 }
4811 }
4812 assert(absTime->tv_sec >= 0, "tv_sec < 0");
4813 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4814 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4815 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4816 }
4817
4818 void Parker::park(bool isAbsolute, jlong time) {
4819 // Optional fast-path check:
4820 // Return immediately if a permit is available.
4821 if (_counter > 0) {
4822 _counter = 0 ;
4823 OrderAccess::fence();
4824 return ;
4825 }
4826
4827 Thread* thread = Thread::current();
4828 assert(thread->is_Java_thread(), "Must be JavaThread");
4829 JavaThread *jt = (JavaThread *)thread;
4830
4831 // Optional optimization -- avoid state transitions if there's an interrupt pending.
4832 // Check interrupt before trying to wait
4833 if (Thread::is_interrupted(thread, false)) {
4834 return;
4835 }
4836
4837 // Next, demultiplex/decode time arguments
4838 timespec absTime;
4839 if (time < 0) { // don't wait at all
4840 return;
4841 }
4842 if (time > 0) {
4843 unpackTime(&absTime, isAbsolute, time);
4844 }
4845
4846
4847 // Enter safepoint region
4848 // Beware of deadlocks such as 6317397.
4849 // The per-thread Parker:: mutex is a classic leaf-lock.
4850 // In particular a thread must never block on the Threads_lock while
4851 // holding the Parker:: mutex. If safepoints are pending both the
4852 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4853 ThreadBlockInVM tbivm(jt);
4854
4855 // Don't wait if cannot get lock since interference arises from
4856 // unblocking. Also. check interrupt before trying wait
4857 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4858 return;
4859 }
4860
4861 int status ;
4862 if (_counter > 0) { // no wait needed
4863 _counter = 0;
4864 status = pthread_mutex_unlock(_mutex);
4865 assert (status == 0, "invariant") ;
4866 OrderAccess::fence();
4867 return;
4868 }
4869
4870 #ifdef ASSERT
4871 // Don't catch signals while blocked; let the running threads have the signals.
4872 // (This allows a debugger to break into the running thread.)
4873 sigset_t oldsigs;
4874 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4875 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4876 #endif
4877
4878 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4879 jt->set_suspend_equivalent();
4880 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4881
4882 if (time == 0) {
4883 status = pthread_cond_wait (_cond, _mutex) ;
4884 } else {
4885 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4886 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4887 pthread_cond_destroy (_cond) ;
4888 pthread_cond_init (_cond, NULL);
4889 }
4890 }
4891 assert_status(status == 0 || status == EINTR ||
4892 status == ETIME || status == ETIMEDOUT,
4893 status, "cond_timedwait");
4894
4895 #ifdef ASSERT
4896 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4897 #endif
4898
4899 _counter = 0 ;
4900 status = pthread_mutex_unlock(_mutex) ;
4901 assert_status(status == 0, status, "invariant") ;
4902 // If externally suspended while waiting, re-suspend
4903 if (jt->handle_special_suspend_equivalent_condition()) {
4904 jt->java_suspend_self();
4905 }
4906
4907 OrderAccess::fence();
4908 }
4909
4910 void Parker::unpark() {
4911 int s, status ;
4912 status = pthread_mutex_lock(_mutex);
4913 assert (status == 0, "invariant") ;
4914 s = _counter;
4915 _counter = 1;
4916 if (s < 1) {
4917 if (WorkAroundNPTLTimedWaitHang) {
4918 status = pthread_cond_signal (_cond) ;
4919 assert (status == 0, "invariant") ;
4920 status = pthread_mutex_unlock(_mutex);
4921 assert (status == 0, "invariant") ;
4922 } else {
4923 status = pthread_mutex_unlock(_mutex);
4924 assert (status == 0, "invariant") ;
4925 status = pthread_cond_signal (_cond) ;
4926 assert (status == 0, "invariant") ;
4927 }
4928 } else {
4929 pthread_mutex_unlock(_mutex);
4930 assert (status == 0, "invariant") ;
4931 }
4932 }
4933
4934
4935 extern char** environ;
4936
4937 #ifndef __NR_fork
4938 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4939 #endif
4940
4941 #ifndef __NR_execve
4942 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4943 #endif
4944
4945 // Run the specified command in a separate process. Return its exit value,
4946 // or -1 on failure (e.g. can't fork a new process).
4947 // Unlike system(), this function can be called from signal handler. It
4948 // doesn't block SIGINT et al.
4949 int os::fork_and_exec(char* cmd) {
4950 const char * argv[4] = {"sh", "-c", cmd, NULL};
4951
4952 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4953 // pthread_atfork handlers and reset pthread library. All we need is a
4954 // separate process to execve. Make a direct syscall to fork process.
4955 // On IA64 there's no fork syscall, we have to use fork() and hope for
4956 // the best...
4957 pid_t pid = NOT_IA64(syscall(__NR_fork);)
4958 IA64_ONLY(fork();)
4959
4960 if (pid < 0) {
4961 // fork failed
4962 return -1;
4963
4964 } else if (pid == 0) {
4965 // child process
4966
4967 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4968 // first to kill every thread on the thread list. Because this list is
4969 // not reset by fork() (see notes above), execve() will instead kill
4970 // every thread in the parent process. We know this is the only thread
4971 // in the new process, so make a system call directly.
4972 // IA64 should use normal execve() from glibc to match the glibc fork()
4973 // above.
4974 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4975 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
4976
4977 // execve failed
4978 _exit(-1);
4979
4980 } else {
4981 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4982 // care about the actual exit code, for now.
4983
4984 int status;
4985
4986 // Wait for the child process to exit. This returns immediately if
4987 // the child has already exited. */
4988 while (waitpid(pid, &status, 0) < 0) {
4989 switch (errno) {
4990 case ECHILD: return 0;
4991 case EINTR: break;
4992 default: return -1;
4993 }
4994 }
4995
4996 if (WIFEXITED(status)) {
4997 // The child exited normally; get its exit code.
4998 return WEXITSTATUS(status);
4999 } else if (WIFSIGNALED(status)) {
5000 // The child exited because of a signal
5001 // The best value to return is 0x80 + signal number,
5002 // because that is what all Unix shells do, and because
5003 // it allows callers to distinguish between process exit and
5004 // process death by signal.
5005 return 0x80 + WTERMSIG(status);
5006 } else {
5007 // Unknown exit code; pass it through
5008 return status;
5009 }
5010 }
5011 }
5012
5013 // is_headless_jre()
5014 //
5015 // Test for the existence of libmawt in motif21 or xawt directories
5016 // in order to report if we are running in a headless jre
5017 //
5018 bool os::is_headless_jre() {
5019 struct stat statbuf;
5020 char buf[MAXPATHLEN];
5021 char libmawtpath[MAXPATHLEN];
5022 const char *xawtstr = "/xawt/libmawt.so";
5023 const char *motifstr = "/motif21/libmawt.so";
5024 char *p;
5025
5026 // Get path to libjvm.so
5027 os::jvm_path(buf, sizeof(buf));
5028
5029 // Get rid of libjvm.so
5030 p = strrchr(buf, '/');
5031 if (p == NULL) return false;
5032 else *p = '\0';
5033
5034 // Get rid of client or server
5035 p = strrchr(buf, '/');
5036 if (p == NULL) return false;
5037 else *p = '\0';
5038
5039 // check xawt/libmawt.so
5040 strcpy(libmawtpath, buf);
5041 strcat(libmawtpath, xawtstr);
5042 if (::stat(libmawtpath, &statbuf) == 0) return false;
5043
5044 // check motif21/libmawt.so
5045 strcpy(libmawtpath, buf);
5046 strcat(libmawtpath, motifstr);
5047 if (::stat(libmawtpath, &statbuf) == 0) return false;
5048
5049 return true;
5050 }
5051