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 // flush buffered output, finish log files
1471 ostream_abort();
1472
1473 // Check for abort hook
1474 abort_hook_t abort_hook = Arguments::abort_hook();
1475 if (abort_hook != NULL) {
1476 abort_hook();
1477 }
1478
1479 }
1480
1481 // Note: os::abort() might be called very early during initialization, or
1482 // called from signal handler. Before adding something to os::abort(), make
1483 // sure it is async-safe and can handle partially initialized VM.
1484 void os::abort(bool dump_core) {
1485 os::shutdown();
1486 if (dump_core) {
1487 #ifndef PRODUCT
1488 fdStream out(defaultStream::output_fd());
1489 out.print_raw("Current thread is ");
1490 char buf[16];
1491 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1492 out.print_raw_cr(buf);
1493 out.print_raw_cr("Dumping core ...");
1494 #endif
1495 ::abort(); // dump core
1496 }
1497
1498 ::exit(1);
1499 }
1500
1501 // Die immediately, no exit hook, no abort hook, no cleanup.
1502 void os::die() {
1503 // _exit() on LinuxThreads only kills current thread
1504 ::abort();
1505 }
1506
1507 // unused on linux for now.
1508 void os::set_error_file(const char *logfile) {}
1509
1510 intx os::current_thread_id() { return (intx)pthread_self(); }
1511 int os::current_process_id() {
1512
1513 // Under the old linux thread library, linux gives each thread
1514 // its own process id. Because of this each thread will return
1515 // a different pid if this method were to return the result
1516 // of getpid(2). Linux provides no api that returns the pid
1517 // of the launcher thread for the vm. This implementation
1518 // returns a unique pid, the pid of the launcher thread
1519 // that starts the vm 'process'.
1520
1521 // Under the NPTL, getpid() returns the same pid as the
1522 // launcher thread rather than a unique pid per thread.
1523 // Use gettid() if you want the old pre NPTL behaviour.
1524
1525 // if you are looking for the result of a call to getpid() that
1526 // returns a unique pid for the calling thread, then look at the
1527 // OSThread::thread_id() method in osThread_linux.hpp file
1528
1529 return (int)(_initial_pid ? _initial_pid : getpid());
1530 }
1531
1532 // DLL functions
1533
1534 const char* os::dll_file_extension() { return ".so"; }
1535
1536 const char* os::get_temp_directory() {
1537 const char *prop = Arguments::get_property("java.io.tmpdir");
1538 return prop == NULL ? "/tmp" : prop;
1539 }
1540
1541 static bool file_exists(const char* filename) {
1542 struct stat statbuf;
1543 if (filename == NULL || strlen(filename) == 0) {
1544 return false;
1545 }
1546 return os::stat(filename, &statbuf) == 0;
1547 }
1548
1549 void os::dll_build_name(char* buffer, size_t buflen,
1550 const char* pname, const char* fname) {
1551 // Copied from libhpi
1552 const size_t pnamelen = pname ? strlen(pname) : 0;
1553
1554 // Quietly truncate on buffer overflow. Should be an error.
1555 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1556 *buffer = '\0';
1557 return;
1558 }
1559
1560 if (pnamelen == 0) {
1561 snprintf(buffer, buflen, "lib%s.so", fname);
1562 } else if (strchr(pname, *os::path_separator()) != NULL) {
1563 int n;
1564 char** pelements = split_path(pname, &n);
1565 for (int i = 0 ; i < n ; i++) {
1566 // Really shouldn't be NULL, but check can't hurt
1567 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1568 continue; // skip the empty path values
1569 }
1570 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1571 if (file_exists(buffer)) {
1572 break;
1573 }
1574 }
1575 // release the storage
1576 for (int i = 0 ; i < n ; i++) {
1577 if (pelements[i] != NULL) {
1578 FREE_C_HEAP_ARRAY(char, pelements[i]);
1579 }
1580 }
1581 if (pelements != NULL) {
1582 FREE_C_HEAP_ARRAY(char*, pelements);
1583 }
1584 } else {
1585 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1586 }
1587 }
1588
1589 const char* os::get_current_directory(char *buf, int buflen) {
1590 return getcwd(buf, buflen);
1591 }
1592
1593 // check if addr is inside libjvm[_g].so
1594 bool os::address_is_in_vm(address addr) {
1595 static address libjvm_base_addr;
1596 Dl_info dlinfo;
1597
1598 if (libjvm_base_addr == NULL) {
1599 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1600 libjvm_base_addr = (address)dlinfo.dli_fbase;
1601 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1602 }
1603
1604 if (dladdr((void *)addr, &dlinfo)) {
1605 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1606 }
1607
1608 return false;
1609 }
1610
1611 bool os::dll_address_to_function_name(address addr, char *buf,
1612 int buflen, int *offset) {
1613 Dl_info dlinfo;
1614
1615 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1616 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1617 if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1618 return true;
1619 } else {
1620 if (buf) buf[0] = '\0';
1621 if (offset) *offset = -1;
1622 return false;
1623 }
1624 }
1625
1626 struct _address_to_library_name {
1627 address addr; // input : memory address
1628 size_t buflen; // size of fname
1629 char* fname; // output: library name
1630 address base; // library base addr
1631 };
1632
1633 static int address_to_library_name_callback(struct dl_phdr_info *info,
1634 size_t size, void *data) {
1635 int i;
1636 bool found = false;
1637 address libbase = NULL;
1638 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1639
1640 // iterate through all loadable segments
1641 for (i = 0; i < info->dlpi_phnum; i++) {
1642 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1643 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1644 // base address of a library is the lowest address of its loaded
1645 // segments.
1646 if (libbase == NULL || libbase > segbase) {
1647 libbase = segbase;
1648 }
1649 // see if 'addr' is within current segment
1650 if (segbase <= d->addr &&
1651 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1652 found = true;
1653 }
1654 }
1655 }
1656
1657 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1658 // so dll_address_to_library_name() can fall through to use dladdr() which
1659 // can figure out executable name from argv[0].
1660 if (found && info->dlpi_name && info->dlpi_name[0]) {
1661 d->base = libbase;
1662 if (d->fname) {
1663 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1664 }
1665 return 1;
1666 }
1667 return 0;
1668 }
1669
1670 bool os::dll_address_to_library_name(address addr, char* buf,
1671 int buflen, int* offset) {
1672 Dl_info dlinfo;
1673 struct _address_to_library_name data;
1674
1675 // There is a bug in old glibc dladdr() implementation that it could resolve
1676 // to wrong library name if the .so file has a base address != NULL. Here
1677 // we iterate through the program headers of all loaded libraries to find
1678 // out which library 'addr' really belongs to. This workaround can be
1679 // removed once the minimum requirement for glibc is moved to 2.3.x.
1680 data.addr = addr;
1681 data.fname = buf;
1682 data.buflen = buflen;
1683 data.base = NULL;
1684 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1685
1686 if (rslt) {
1687 // buf already contains library name
1688 if (offset) *offset = addr - data.base;
1689 return true;
1690 } else if (dladdr((void*)addr, &dlinfo)){
1691 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1692 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1693 return true;
1694 } else {
1695 if (buf) buf[0] = '\0';
1696 if (offset) *offset = -1;
1697 return false;
1698 }
1699 }
1700
1701 // Loads .dll/.so and
1702 // in case of error it checks if .dll/.so was built for the
1703 // same architecture as Hotspot is running on
1704
1705 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1706 {
1707 void * result= ::dlopen(filename, RTLD_LAZY);
1708 if (result != NULL) {
1709 // Successful loading
1710 return result;
1711 }
1712
1713 Elf32_Ehdr elf_head;
1714
1715 // Read system error message into ebuf
1716 // It may or may not be overwritten below
1717 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1718 ebuf[ebuflen-1]='\0';
1719 int diag_msg_max_length=ebuflen-strlen(ebuf);
1720 char* diag_msg_buf=ebuf+strlen(ebuf);
1721
1722 if (diag_msg_max_length==0) {
1723 // No more space in ebuf for additional diagnostics message
1724 return NULL;
1725 }
1726
1727
1728 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1729
1730 if (file_descriptor < 0) {
1731 // Can't open library, report dlerror() message
1732 return NULL;
1733 }
1734
1735 bool failed_to_read_elf_head=
1736 (sizeof(elf_head)!=
1737 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1738
1739 ::close(file_descriptor);
1740 if (failed_to_read_elf_head) {
1741 // file i/o error - report dlerror() msg
1742 return NULL;
1743 }
1744
1745 typedef struct {
1746 Elf32_Half code; // Actual value as defined in elf.h
1747 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1748 char elf_class; // 32 or 64 bit
1749 char endianess; // MSB or LSB
1750 char* name; // String representation
1751 } arch_t;
1752
1753 #ifndef EM_486
1754 #define EM_486 6 /* Intel 80486 */
1755 #endif
1756
1757 static const arch_t arch_array[]={
1758 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1759 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1760 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1761 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1762 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1763 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1764 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1765 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1766 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1767 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1768 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1769 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1770 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1771 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1772 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1773 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1774 };
1775
1776 #if (defined IA32)
1777 static Elf32_Half running_arch_code=EM_386;
1778 #elif (defined AMD64)
1779 static Elf32_Half running_arch_code=EM_X86_64;
1780 #elif (defined IA64)
1781 static Elf32_Half running_arch_code=EM_IA_64;
1782 #elif (defined __sparc) && (defined _LP64)
1783 static Elf32_Half running_arch_code=EM_SPARCV9;
1784 #elif (defined __sparc) && (!defined _LP64)
1785 static Elf32_Half running_arch_code=EM_SPARC;
1786 #elif (defined __powerpc64__)
1787 static Elf32_Half running_arch_code=EM_PPC64;
1788 #elif (defined __powerpc__)
1789 static Elf32_Half running_arch_code=EM_PPC;
1790 #elif (defined ARM)
1791 static Elf32_Half running_arch_code=EM_ARM;
1792 #elif (defined S390)
1793 static Elf32_Half running_arch_code=EM_S390;
1794 #elif (defined ALPHA)
1795 static Elf32_Half running_arch_code=EM_ALPHA;
1796 #elif (defined MIPSEL)
1797 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1798 #elif (defined PARISC)
1799 static Elf32_Half running_arch_code=EM_PARISC;
1800 #elif (defined MIPS)
1801 static Elf32_Half running_arch_code=EM_MIPS;
1802 #elif (defined M68K)
1803 static Elf32_Half running_arch_code=EM_68K;
1804 #else
1805 #error Method os::dll_load requires that one of following is defined:\
1806 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1807 #endif
1808
1809 // Identify compatability class for VM's architecture and library's architecture
1810 // Obtain string descriptions for architectures
1811
1812 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1813 int running_arch_index=-1;
1814
1815 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1816 if (running_arch_code == arch_array[i].code) {
1817 running_arch_index = i;
1818 }
1819 if (lib_arch.code == arch_array[i].code) {
1820 lib_arch.compat_class = arch_array[i].compat_class;
1821 lib_arch.name = arch_array[i].name;
1822 }
1823 }
1824
1825 assert(running_arch_index != -1,
1826 "Didn't find running architecture code (running_arch_code) in arch_array");
1827 if (running_arch_index == -1) {
1828 // Even though running architecture detection failed
1829 // we may still continue with reporting dlerror() message
1830 return NULL;
1831 }
1832
1833 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1834 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1835 return NULL;
1836 }
1837
1838 #ifndef S390
1839 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1840 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1841 return NULL;
1842 }
1843 #endif // !S390
1844
1845 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1846 if ( lib_arch.name!=NULL ) {
1847 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1848 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1849 lib_arch.name, arch_array[running_arch_index].name);
1850 } else {
1851 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1852 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1853 lib_arch.code,
1854 arch_array[running_arch_index].name);
1855 }
1856 }
1857
1858 return NULL;
1859 }
1860
1861 /*
1862 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1863 * chances are you might want to run the generated bits against glibc-2.0
1864 * libdl.so, so always use locking for any version of glibc.
1865 */
1866 void* os::dll_lookup(void* handle, const char* name) {
1867 pthread_mutex_lock(&dl_mutex);
1868 void* res = dlsym(handle, name);
1869 pthread_mutex_unlock(&dl_mutex);
1870 return res;
1871 }
1872
1873
1874 bool _print_ascii_file(const char* filename, outputStream* st) {
1875 int fd = open(filename, O_RDONLY);
1876 if (fd == -1) {
1877 return false;
1878 }
1879
1880 char buf[32];
1881 int bytes;
1882 while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1883 st->print_raw(buf, bytes);
1884 }
1885
1886 close(fd);
1887
1888 return true;
1889 }
1890
1891 void os::print_dll_info(outputStream *st) {
1892 st->print_cr("Dynamic libraries:");
1893
1894 char fname[32];
1895 pid_t pid = os::Linux::gettid();
1896
1897 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1898
1899 if (!_print_ascii_file(fname, st)) {
1900 st->print("Can not get library information for pid = %d\n", pid);
1901 }
1902 }
1903
1904
1905 void os::print_os_info(outputStream* st) {
1906 st->print("OS:");
1907
1908 // Try to identify popular distros.
1909 // Most Linux distributions have /etc/XXX-release file, which contains
1910 // the OS version string. Some have more than one /etc/XXX-release file
1911 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1912 // so the order is important.
1913 if (!_print_ascii_file("/etc/mandrake-release", st) &&
1914 !_print_ascii_file("/etc/sun-release", st) &&
1915 !_print_ascii_file("/etc/redhat-release", st) &&
1916 !_print_ascii_file("/etc/SuSE-release", st) &&
1917 !_print_ascii_file("/etc/turbolinux-release", st) &&
1918 !_print_ascii_file("/etc/gentoo-release", st) &&
1919 !_print_ascii_file("/etc/debian_version", st) &&
1920 !_print_ascii_file("/etc/ltib-release", st) &&
1921 !_print_ascii_file("/etc/angstrom-version", st)) {
1922 st->print("Linux");
1923 }
1924 st->cr();
1925
1926 // kernel
1927 st->print("uname:");
1928 struct utsname name;
1929 uname(&name);
1930 st->print(name.sysname); st->print(" ");
1931 st->print(name.release); st->print(" ");
1932 st->print(name.version); st->print(" ");
1933 st->print(name.machine);
1934 st->cr();
1935
1936 // Print warning if unsafe chroot environment detected
1937 if (unsafe_chroot_detected) {
1938 st->print("WARNING!! ");
1939 st->print_cr(unstable_chroot_error);
1940 }
1941
1942 // libc, pthread
1943 st->print("libc:");
1944 st->print(os::Linux::glibc_version()); st->print(" ");
1945 st->print(os::Linux::libpthread_version()); st->print(" ");
1946 if (os::Linux::is_LinuxThreads()) {
1947 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1948 }
1949 st->cr();
1950
1951 // rlimit
1952 st->print("rlimit:");
1953 struct rlimit rlim;
1954
1955 st->print(" STACK ");
1956 getrlimit(RLIMIT_STACK, &rlim);
1957 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1958 else st->print("%uk", rlim.rlim_cur >> 10);
1959
1960 st->print(", CORE ");
1961 getrlimit(RLIMIT_CORE, &rlim);
1962 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1963 else st->print("%uk", rlim.rlim_cur >> 10);
1964
1965 st->print(", NPROC ");
1966 getrlimit(RLIMIT_NPROC, &rlim);
1967 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1968 else st->print("%d", rlim.rlim_cur);
1969
1970 st->print(", NOFILE ");
1971 getrlimit(RLIMIT_NOFILE, &rlim);
1972 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1973 else st->print("%d", rlim.rlim_cur);
1974
1975 st->print(", AS ");
1976 getrlimit(RLIMIT_AS, &rlim);
1977 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1978 else st->print("%uk", rlim.rlim_cur >> 10);
1979 st->cr();
1980
1981 // load average
1982 st->print("load average:");
1983 double loadavg[3];
1984 os::loadavg(loadavg, 3);
1985 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1986 st->cr();
1987
1988 // meminfo
1989 st->print("\n/proc/meminfo:\n");
1990 _print_ascii_file("/proc/meminfo", st);
1991 st->cr();
1992 }
1993
1994 void os::print_memory_info(outputStream* st) {
1995
1996 st->print("Memory:");
1997 st->print(" %dk page", os::vm_page_size()>>10);
1998
1999 // values in struct sysinfo are "unsigned long"
2000 struct sysinfo si;
2001 sysinfo(&si);
2002
2003 st->print(", physical " UINT64_FORMAT "k",
2004 os::physical_memory() >> 10);
2005 st->print("(" UINT64_FORMAT "k free)",
2006 os::available_memory() >> 10);
2007 st->print(", swap " UINT64_FORMAT "k",
2008 ((jlong)si.totalswap * si.mem_unit) >> 10);
2009 st->print("(" UINT64_FORMAT "k free)",
2010 ((jlong)si.freeswap * si.mem_unit) >> 10);
2011 st->cr();
2012 }
2013
2014 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2015 // but they're the same for all the linux arch that we support
2016 // and they're the same for solaris but there's no common place to put this.
2017 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2018 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2019 "ILL_COPROC", "ILL_BADSTK" };
2020
2021 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2022 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2023 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2024
2025 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2026
2027 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2028
2029 void os::print_siginfo(outputStream* st, void* siginfo) {
2030 st->print("siginfo:");
2031
2032 const int buflen = 100;
2033 char buf[buflen];
2034 siginfo_t *si = (siginfo_t*)siginfo;
2035 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2036 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2037 st->print("si_errno=%s", buf);
2038 } else {
2039 st->print("si_errno=%d", si->si_errno);
2040 }
2041 const int c = si->si_code;
2042 assert(c > 0, "unexpected si_code");
2043 switch (si->si_signo) {
2044 case SIGILL:
2045 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2046 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2047 break;
2048 case SIGFPE:
2049 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2050 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2051 break;
2052 case SIGSEGV:
2053 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2054 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2055 break;
2056 case SIGBUS:
2057 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2058 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2059 break;
2060 default:
2061 st->print(", si_code=%d", si->si_code);
2062 // no si_addr
2063 }
2064
2065 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2066 UseSharedSpaces) {
2067 FileMapInfo* mapinfo = FileMapInfo::current_info();
2068 if (mapinfo->is_in_shared_space(si->si_addr)) {
2069 st->print("\n\nError accessing class data sharing archive." \
2070 " Mapped file inaccessible during execution, " \
2071 " possible disk/network problem.");
2072 }
2073 }
2074 st->cr();
2075 }
2076
2077
2078 static void print_signal_handler(outputStream* st, int sig,
2079 char* buf, size_t buflen);
2080
2081 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2082 st->print_cr("Signal Handlers:");
2083 print_signal_handler(st, SIGSEGV, buf, buflen);
2084 print_signal_handler(st, SIGBUS , buf, buflen);
2085 print_signal_handler(st, SIGFPE , buf, buflen);
2086 print_signal_handler(st, SIGPIPE, buf, buflen);
2087 print_signal_handler(st, SIGXFSZ, buf, buflen);
2088 print_signal_handler(st, SIGILL , buf, buflen);
2089 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2090 print_signal_handler(st, SR_signum, buf, buflen);
2091 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2092 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2093 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2094 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2095 }
2096
2097 static char saved_jvm_path[MAXPATHLEN] = {0};
2098
2099 // Find the full path to the current module, libjvm.so or libjvm_g.so
2100 void os::jvm_path(char *buf, jint buflen) {
2101 // Error checking.
2102 if (buflen < MAXPATHLEN) {
2103 assert(false, "must use a large-enough buffer");
2104 buf[0] = '\0';
2105 return;
2106 }
2107 // Lazy resolve the path to current module.
2108 if (saved_jvm_path[0] != 0) {
2109 strcpy(buf, saved_jvm_path);
2110 return;
2111 }
2112
2113 char dli_fname[MAXPATHLEN];
2114 bool ret = dll_address_to_library_name(
2115 CAST_FROM_FN_PTR(address, os::jvm_path),
2116 dli_fname, sizeof(dli_fname), NULL);
2117 assert(ret != 0, "cannot locate libjvm");
2118 char *rp = realpath(dli_fname, buf);
2119 if (rp == NULL)
2120 return;
2121
2122 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2123 // Support for the gamma launcher. Typical value for buf is
2124 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2125 // the right place in the string, then assume we are installed in a JDK and
2126 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2127 // up the path so it looks like libjvm.so is installed there (append a
2128 // fake suffix hotspot/libjvm.so).
2129 const char *p = buf + strlen(buf) - 1;
2130 for (int count = 0; p > buf && count < 5; ++count) {
2131 for (--p; p > buf && *p != '/'; --p)
2132 /* empty */ ;
2133 }
2134
2135 if (strncmp(p, "/jre/lib/", 9) != 0) {
2136 // Look for JAVA_HOME in the environment.
2137 char* java_home_var = ::getenv("JAVA_HOME");
2138 if (java_home_var != NULL && java_home_var[0] != 0) {
2139 char* jrelib_p;
2140 int len;
2141
2142 // Check the current module name "libjvm.so" or "libjvm_g.so".
2143 p = strrchr(buf, '/');
2144 assert(strstr(p, "/libjvm") == p, "invalid library name");
2145 p = strstr(p, "_g") ? "_g" : "";
2146
2147 rp = realpath(java_home_var, buf);
2148 if (rp == NULL)
2149 return;
2150
2151 // determine if this is a legacy image or modules image
2152 // modules image doesn't have "jre" subdirectory
2153 len = strlen(buf);
2154 jrelib_p = buf + len;
2155 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2156 if (0 != access(buf, F_OK)) {
2157 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2158 }
2159
2160 if (0 == access(buf, F_OK)) {
2161 // Use current module name "libjvm[_g].so" instead of
2162 // "libjvm"debug_only("_g")".so" since for fastdebug version
2163 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2164 // It is used when we are choosing the HPI library's name
2165 // "libhpi[_g].so" in hpi::initialize_get_interface().
2166 len = strlen(buf);
2167 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2168 } else {
2169 // Go back to path of .so
2170 rp = realpath(dli_fname, buf);
2171 if (rp == NULL)
2172 return;
2173 }
2174 }
2175 }
2176 }
2177
2178 strcpy(saved_jvm_path, buf);
2179 }
2180
2181 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2182 // no prefix required, not even "_"
2183 }
2184
2185 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2186 // no suffix required
2187 }
2188
2189 ////////////////////////////////////////////////////////////////////////////////
2190 // sun.misc.Signal support
2191
2192 static volatile jint sigint_count = 0;
2193
2194 static void
2195 UserHandler(int sig, void *siginfo, void *context) {
2196 // 4511530 - sem_post is serialized and handled by the manager thread. When
2197 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2198 // don't want to flood the manager thread with sem_post requests.
2199 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2200 return;
2201
2202 // Ctrl-C is pressed during error reporting, likely because the error
2203 // handler fails to abort. Let VM die immediately.
2204 if (sig == SIGINT && is_error_reported()) {
2205 os::die();
2206 }
2207
2208 os::signal_notify(sig);
2209 }
2210
2211 void* os::user_handler() {
2212 return CAST_FROM_FN_PTR(void*, UserHandler);
2213 }
2214
2215 extern "C" {
2216 typedef void (*sa_handler_t)(int);
2217 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2218 }
2219
2220 void* os::signal(int signal_number, void* handler) {
2221 struct sigaction sigAct, oldSigAct;
2222
2223 sigfillset(&(sigAct.sa_mask));
2224 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2225 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2226
2227 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2228 // -1 means registration failed
2229 return (void *)-1;
2230 }
2231
2232 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2233 }
2234
2235 void os::signal_raise(int signal_number) {
2236 ::raise(signal_number);
2237 }
2238
2239 /*
2240 * The following code is moved from os.cpp for making this
2241 * code platform specific, which it is by its very nature.
2242 */
2243
2244 // Will be modified when max signal is changed to be dynamic
2245 int os::sigexitnum_pd() {
2246 return NSIG;
2247 }
2248
2249 // a counter for each possible signal value
2250 static volatile jint pending_signals[NSIG+1] = { 0 };
2251
2252 // Linux(POSIX) specific hand shaking semaphore.
2253 static sem_t sig_sem;
2254
2255 void os::signal_init_pd() {
2256 // Initialize signal structures
2257 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2258
2259 // Initialize signal semaphore
2260 ::sem_init(&sig_sem, 0, 0);
2261 }
2262
2263 void os::signal_notify(int sig) {
2264 Atomic::inc(&pending_signals[sig]);
2265 ::sem_post(&sig_sem);
2266 }
2267
2268 static int check_pending_signals(bool wait) {
2269 Atomic::store(0, &sigint_count);
2270 for (;;) {
2271 for (int i = 0; i < NSIG + 1; i++) {
2272 jint n = pending_signals[i];
2273 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2274 return i;
2275 }
2276 }
2277 if (!wait) {
2278 return -1;
2279 }
2280 JavaThread *thread = JavaThread::current();
2281 ThreadBlockInVM tbivm(thread);
2282
2283 bool threadIsSuspended;
2284 do {
2285 thread->set_suspend_equivalent();
2286 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2287 ::sem_wait(&sig_sem);
2288
2289 // were we externally suspended while we were waiting?
2290 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2291 if (threadIsSuspended) {
2292 //
2293 // The semaphore has been incremented, but while we were waiting
2294 // another thread suspended us. We don't want to continue running
2295 // while suspended because that would surprise the thread that
2296 // suspended us.
2297 //
2298 ::sem_post(&sig_sem);
2299
2300 thread->java_suspend_self();
2301 }
2302 } while (threadIsSuspended);
2303 }
2304 }
2305
2306 int os::signal_lookup() {
2307 return check_pending_signals(false);
2308 }
2309
2310 int os::signal_wait() {
2311 return check_pending_signals(true);
2312 }
2313
2314 ////////////////////////////////////////////////////////////////////////////////
2315 // Virtual Memory
2316
2317 int os::vm_page_size() {
2318 // Seems redundant as all get out
2319 assert(os::Linux::page_size() != -1, "must call os::init");
2320 return os::Linux::page_size();
2321 }
2322
2323 // Solaris allocates memory by pages.
2324 int os::vm_allocation_granularity() {
2325 assert(os::Linux::page_size() != -1, "must call os::init");
2326 return os::Linux::page_size();
2327 }
2328
2329 // Rationale behind this function:
2330 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2331 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2332 // samples for JITted code. Here we create private executable mapping over the code cache
2333 // and then we can use standard (well, almost, as mapping can change) way to provide
2334 // info for the reporting script by storing timestamp and location of symbol
2335 void linux_wrap_code(char* base, size_t size) {
2336 static volatile jint cnt = 0;
2337
2338 if (!UseOprofile) {
2339 return;
2340 }
2341
2342 char buf[PATH_MAX+1];
2343 int num = Atomic::add(1, &cnt);
2344
2345 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2346 os::get_temp_directory(), os::current_process_id(), num);
2347 unlink(buf);
2348
2349 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2350
2351 if (fd != -1) {
2352 off_t rv = lseek(fd, size-2, SEEK_SET);
2353 if (rv != (off_t)-1) {
2354 if (write(fd, "", 1) == 1) {
2355 mmap(base, size,
2356 PROT_READ|PROT_WRITE|PROT_EXEC,
2357 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2358 }
2359 }
2360 close(fd);
2361 unlink(buf);
2362 }
2363 }
2364
2365 // NOTE: Linux kernel does not really reserve the pages for us.
2366 // All it does is to check if there are enough free pages
2367 // left at the time of mmap(). This could be a potential
2368 // problem.
2369 bool os::commit_memory(char* addr, size_t size, bool exec) {
2370 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2371 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2372 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2373 return res != (uintptr_t) MAP_FAILED;
2374 }
2375
2376 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2377 bool exec) {
2378 return commit_memory(addr, size, exec);
2379 }
2380
2381 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2382
2383 void os::free_memory(char *addr, size_t bytes) {
2384 ::mmap(addr, bytes, PROT_READ | PROT_WRITE,
2385 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2386 }
2387
2388 void os::numa_make_global(char *addr, size_t bytes) {
2389 Linux::numa_interleave_memory(addr, bytes);
2390 }
2391
2392 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2393 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2394 }
2395
2396 bool os::numa_topology_changed() { return false; }
2397
2398 size_t os::numa_get_groups_num() {
2399 int max_node = Linux::numa_max_node();
2400 return max_node > 0 ? max_node + 1 : 1;
2401 }
2402
2403 int os::numa_get_group_id() {
2404 int cpu_id = Linux::sched_getcpu();
2405 if (cpu_id != -1) {
2406 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2407 if (lgrp_id != -1) {
2408 return lgrp_id;
2409 }
2410 }
2411 return 0;
2412 }
2413
2414 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2415 for (size_t i = 0; i < size; i++) {
2416 ids[i] = i;
2417 }
2418 return size;
2419 }
2420
2421 bool os::get_page_info(char *start, page_info* info) {
2422 return false;
2423 }
2424
2425 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2426 return end;
2427 }
2428
2429 extern "C" void numa_warn(int number, char *where, ...) { }
2430 extern "C" void numa_error(char *where) { }
2431
2432
2433 // If we are running with libnuma version > 2, then we should
2434 // be trying to use symbols with versions 1.1
2435 // If we are running with earlier version, which did not have symbol versions,
2436 // we should use the base version.
2437 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2438 void *f = dlvsym(handle, name, "libnuma_1.1");
2439 if (f == NULL) {
2440 f = dlsym(handle, name);
2441 }
2442 return f;
2443 }
2444
2445 bool os::Linux::libnuma_init() {
2446 // sched_getcpu() should be in libc.
2447 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2448 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2449
2450 if (sched_getcpu() != -1) { // Does it work?
2451 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2452 if (handle != NULL) {
2453 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2454 libnuma_dlsym(handle, "numa_node_to_cpus")));
2455 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2456 libnuma_dlsym(handle, "numa_max_node")));
2457 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2458 libnuma_dlsym(handle, "numa_available")));
2459 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2460 libnuma_dlsym(handle, "numa_tonode_memory")));
2461 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2462 libnuma_dlsym(handle, "numa_interleave_memory")));
2463
2464
2465 if (numa_available() != -1) {
2466 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2467 // Create a cpu -> node mapping
2468 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2469 rebuild_cpu_to_node_map();
2470 return true;
2471 }
2472 }
2473 }
2474 return false;
2475 }
2476
2477 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2478 // The table is later used in get_node_by_cpu().
2479 void os::Linux::rebuild_cpu_to_node_map() {
2480 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2481 // in libnuma (possible values are starting from 16,
2482 // and continuing up with every other power of 2, but less
2483 // than the maximum number of CPUs supported by kernel), and
2484 // is a subject to change (in libnuma version 2 the requirements
2485 // are more reasonable) we'll just hardcode the number they use
2486 // in the library.
2487 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2488
2489 size_t cpu_num = os::active_processor_count();
2490 size_t cpu_map_size = NCPUS / BitsPerCLong;
2491 size_t cpu_map_valid_size =
2492 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2493
2494 cpu_to_node()->clear();
2495 cpu_to_node()->at_grow(cpu_num - 1);
2496 size_t node_num = numa_get_groups_num();
2497
2498 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2499 for (size_t i = 0; i < node_num; i++) {
2500 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2501 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2502 if (cpu_map[j] != 0) {
2503 for (size_t k = 0; k < BitsPerCLong; k++) {
2504 if (cpu_map[j] & (1UL << k)) {
2505 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2506 }
2507 }
2508 }
2509 }
2510 }
2511 }
2512 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2513 }
2514
2515 int os::Linux::get_node_by_cpu(int cpu_id) {
2516 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2517 return cpu_to_node()->at(cpu_id);
2518 }
2519 return -1;
2520 }
2521
2522 GrowableArray<int>* os::Linux::_cpu_to_node;
2523 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2524 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2525 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2526 os::Linux::numa_available_func_t os::Linux::_numa_available;
2527 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2528 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2529 unsigned long* os::Linux::_numa_all_nodes;
2530
2531 bool os::uncommit_memory(char* addr, size_t size) {
2532 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2533 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2534 return res != (uintptr_t) MAP_FAILED;
2535 }
2536
2537 // Linux uses a growable mapping for the stack, and if the mapping for
2538 // the stack guard pages is not removed when we detach a thread the
2539 // stack cannot grow beyond the pages where the stack guard was
2540 // mapped. If at some point later in the process the stack expands to
2541 // that point, the Linux kernel cannot expand the stack any further
2542 // because the guard pages are in the way, and a segfault occurs.
2543 //
2544 // However, it's essential not to split the stack region by unmapping
2545 // a region (leaving a hole) that's already part of the stack mapping,
2546 // so if the stack mapping has already grown beyond the guard pages at
2547 // the time we create them, we have to truncate the stack mapping.
2548 // So, we need to know the extent of the stack mapping when
2549 // create_stack_guard_pages() is called.
2550
2551 // Find the bounds of the stack mapping. Return true for success.
2552 //
2553 // We only need this for stacks that are growable: at the time of
2554 // writing thread stacks don't use growable mappings (i.e. those
2555 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2556 // only applies to the main thread.
2557 static bool
2558 get_stack_bounds(uintptr_t *bottom, uintptr_t *top)
2559 {
2560 FILE *f = fopen("/proc/self/maps", "r");
2561 if (f == NULL)
2562 return false;
2563
2564 while (!feof(f)) {
2565 size_t dummy;
2566 char *str = NULL;
2567 ssize_t len = getline(&str, &dummy, f);
2568 if (len == -1) {
2569 fclose(f);
2570 return false;
2571 }
2572
2573 if (len > 0 && str[len-1] == '\n') {
2574 str[len-1] = 0;
2575 len--;
2576 }
2577
2578 static const char *stack_str = "[stack]";
2579 if (len > (ssize_t)strlen(stack_str)
2580 && (strcmp(str + len - strlen(stack_str), stack_str) == 0)) {
2581 if (sscanf(str, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2582 uintptr_t sp = (uintptr_t)__builtin_frame_address(0);
2583 if (sp >= *bottom && sp <= *top) {
2584 free(str);
2585 fclose(f);
2586 return true;
2587 }
2588 }
2589 }
2590 free(str);
2591 }
2592 fclose(f);
2593 return false;
2594 }
2595
2596 // If the (growable) stack mapping already extends beyond the point
2597 // where we're going to put our guard pages, truncate the mapping at
2598 // that point by munmap()ping it. This ensures that when we later
2599 // munmap() the guard pages we don't leave a hole in the stack
2600 // mapping. This only affects the main/initial thread, but guard
2601 // against future OS changes
2602 bool os::create_stack_guard_pages(char* addr, size_t size) {
2603 uintptr_t stack_extent, stack_base;
2604 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2605 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2606 assert(os::Linux::is_initial_thread(),
2607 "growable stack in non-initial thread");
2608 if (stack_extent < (uintptr_t)addr)
2609 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2610 }
2611
2612 return os::commit_memory(addr, size);
2613 }
2614
2615 // If this is a growable mapping, remove the guard pages entirely by
2616 // munmap()ping them. If not, just call uncommit_memory(). This only
2617 // affects the main/initial thread, but guard against future OS changes
2618 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2619 uintptr_t stack_extent, stack_base;
2620 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2621 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2622 assert(os::Linux::is_initial_thread(),
2623 "growable stack in non-initial thread");
2624
2625 return ::munmap(addr, size) == 0;
2626 }
2627
2628 return os::uncommit_memory(addr, size);
2629 }
2630
2631 static address _highest_vm_reserved_address = NULL;
2632
2633 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2634 // at 'requested_addr'. If there are existing memory mappings at the same
2635 // location, however, they will be overwritten. If 'fixed' is false,
2636 // 'requested_addr' is only treated as a hint, the return value may or
2637 // may not start from the requested address. Unlike Linux mmap(), this
2638 // function returns NULL to indicate failure.
2639 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2640 char * addr;
2641 int flags;
2642
2643 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2644 if (fixed) {
2645 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2646 flags |= MAP_FIXED;
2647 }
2648
2649 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2650 // to PROT_EXEC if executable when we commit the page.
2651 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2652 flags, -1, 0);
2653
2654 if (addr != MAP_FAILED) {
2655 // anon_mmap() should only get called during VM initialization,
2656 // don't need lock (actually we can skip locking even it can be called
2657 // from multiple threads, because _highest_vm_reserved_address is just a
2658 // hint about the upper limit of non-stack memory regions.)
2659 if ((address)addr + bytes > _highest_vm_reserved_address) {
2660 _highest_vm_reserved_address = (address)addr + bytes;
2661 }
2662 }
2663
2664 return addr == MAP_FAILED ? NULL : addr;
2665 }
2666
2667 // Don't update _highest_vm_reserved_address, because there might be memory
2668 // regions above addr + size. If so, releasing a memory region only creates
2669 // a hole in the address space, it doesn't help prevent heap-stack collision.
2670 //
2671 static int anon_munmap(char * addr, size_t size) {
2672 return ::munmap(addr, size) == 0;
2673 }
2674
2675 char* os::reserve_memory(size_t bytes, char* requested_addr,
2676 size_t alignment_hint) {
2677 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2678 }
2679
2680 bool os::release_memory(char* addr, size_t size) {
2681 return anon_munmap(addr, size);
2682 }
2683
2684 static address highest_vm_reserved_address() {
2685 return _highest_vm_reserved_address;
2686 }
2687
2688 static bool linux_mprotect(char* addr, size_t size, int prot) {
2689 // Linux wants the mprotect address argument to be page aligned.
2690 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2691
2692 // According to SUSv3, mprotect() should only be used with mappings
2693 // established by mmap(), and mmap() always maps whole pages. Unaligned
2694 // 'addr' likely indicates problem in the VM (e.g. trying to change
2695 // protection of malloc'ed or statically allocated memory). Check the
2696 // caller if you hit this assert.
2697 assert(addr == bottom, "sanity check");
2698
2699 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2700 return ::mprotect(bottom, size, prot) == 0;
2701 }
2702
2703 // Set protections specified
2704 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2705 bool is_committed) {
2706 unsigned int p = 0;
2707 switch (prot) {
2708 case MEM_PROT_NONE: p = PROT_NONE; break;
2709 case MEM_PROT_READ: p = PROT_READ; break;
2710 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2711 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2712 default:
2713 ShouldNotReachHere();
2714 }
2715 // is_committed is unused.
2716 return linux_mprotect(addr, bytes, p);
2717 }
2718
2719 bool os::guard_memory(char* addr, size_t size) {
2720 return linux_mprotect(addr, size, PROT_NONE);
2721 }
2722
2723 bool os::unguard_memory(char* addr, size_t size) {
2724 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2725 }
2726
2727 // Large page support
2728
2729 static size_t _large_page_size = 0;
2730
2731 bool os::large_page_init() {
2732 if (!UseLargePages) return false;
2733
2734 if (LargePageSizeInBytes) {
2735 _large_page_size = LargePageSizeInBytes;
2736 } else {
2737 // large_page_size on Linux is used to round up heap size. x86 uses either
2738 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2739 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2740 // page as large as 256M.
2741 //
2742 // Here we try to figure out page size by parsing /proc/meminfo and looking
2743 // for a line with the following format:
2744 // Hugepagesize: 2048 kB
2745 //
2746 // If we can't determine the value (e.g. /proc is not mounted, or the text
2747 // format has been changed), we'll use the largest page size supported by
2748 // the processor.
2749
2750 #ifndef ZERO
2751 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
2752 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
2753 #endif // ZERO
2754
2755 FILE *fp = fopen("/proc/meminfo", "r");
2756 if (fp) {
2757 while (!feof(fp)) {
2758 int x = 0;
2759 char buf[16];
2760 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2761 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2762 _large_page_size = x * K;
2763 break;
2764 }
2765 } else {
2766 // skip to next line
2767 for (;;) {
2768 int ch = fgetc(fp);
2769 if (ch == EOF || ch == (int)'\n') break;
2770 }
2771 }
2772 }
2773 fclose(fp);
2774 }
2775 }
2776
2777 const size_t default_page_size = (size_t)Linux::page_size();
2778 if (_large_page_size > default_page_size) {
2779 _page_sizes[0] = _large_page_size;
2780 _page_sizes[1] = default_page_size;
2781 _page_sizes[2] = 0;
2782 }
2783
2784 // Large page support is available on 2.6 or newer kernel, some vendors
2785 // (e.g. Redhat) have backported it to their 2.4 based distributions.
2786 // We optimistically assume the support is available. If later it turns out
2787 // not true, VM will automatically switch to use regular page size.
2788 return true;
2789 }
2790
2791 #ifndef SHM_HUGETLB
2792 #define SHM_HUGETLB 04000
2793 #endif
2794
2795 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
2796 // "exec" is passed in but not used. Creating the shared image for
2797 // the code cache doesn't have an SHM_X executable permission to check.
2798 assert(UseLargePages, "only for large pages");
2799
2800 key_t key = IPC_PRIVATE;
2801 char *addr;
2802
2803 bool warn_on_failure = UseLargePages &&
2804 (!FLAG_IS_DEFAULT(UseLargePages) ||
2805 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2806 );
2807 char msg[128];
2808
2809 // Create a large shared memory region to attach to based on size.
2810 // Currently, size is the total size of the heap
2811 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2812 if (shmid == -1) {
2813 // Possible reasons for shmget failure:
2814 // 1. shmmax is too small for Java heap.
2815 // > check shmmax value: cat /proc/sys/kernel/shmmax
2816 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2817 // 2. not enough large page memory.
2818 // > check available large pages: cat /proc/meminfo
2819 // > increase amount of large pages:
2820 // echo new_value > /proc/sys/vm/nr_hugepages
2821 // Note 1: different Linux may use different name for this property,
2822 // e.g. on Redhat AS-3 it is "hugetlb_pool".
2823 // Note 2: it's possible there's enough physical memory available but
2824 // they are so fragmented after a long run that they can't
2825 // coalesce into large pages. Try to reserve large pages when
2826 // the system is still "fresh".
2827 if (warn_on_failure) {
2828 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2829 warning(msg);
2830 }
2831 return NULL;
2832 }
2833
2834 // attach to the region
2835 addr = (char*)shmat(shmid, req_addr, 0);
2836 int err = errno;
2837
2838 // Remove shmid. If shmat() is successful, the actual shared memory segment
2839 // will be deleted when it's detached by shmdt() or when the process
2840 // terminates. If shmat() is not successful this will remove the shared
2841 // segment immediately.
2842 shmctl(shmid, IPC_RMID, NULL);
2843
2844 if ((intptr_t)addr == -1) {
2845 if (warn_on_failure) {
2846 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2847 warning(msg);
2848 }
2849 return NULL;
2850 }
2851
2852 return addr;
2853 }
2854
2855 bool os::release_memory_special(char* base, size_t bytes) {
2856 // detaching the SHM segment will also delete it, see reserve_memory_special()
2857 int rslt = shmdt(base);
2858 return rslt == 0;
2859 }
2860
2861 size_t os::large_page_size() {
2862 return _large_page_size;
2863 }
2864
2865 // Linux does not support anonymous mmap with large page memory. The only way
2866 // to reserve large page memory without file backing is through SysV shared
2867 // memory API. The entire memory region is committed and pinned upfront.
2868 // Hopefully this will change in the future...
2869 bool os::can_commit_large_page_memory() {
2870 return false;
2871 }
2872
2873 bool os::can_execute_large_page_memory() {
2874 return false;
2875 }
2876
2877 // Reserve memory at an arbitrary address, only if that area is
2878 // available (and not reserved for something else).
2879
2880 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2881 const int max_tries = 10;
2882 char* base[max_tries];
2883 size_t size[max_tries];
2884 const size_t gap = 0x000000;
2885
2886 // Assert only that the size is a multiple of the page size, since
2887 // that's all that mmap requires, and since that's all we really know
2888 // about at this low abstraction level. If we need higher alignment,
2889 // we can either pass an alignment to this method or verify alignment
2890 // in one of the methods further up the call chain. See bug 5044738.
2891 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2892
2893 // Repeatedly allocate blocks until the block is allocated at the
2894 // right spot. Give up after max_tries. Note that reserve_memory() will
2895 // automatically update _highest_vm_reserved_address if the call is
2896 // successful. The variable tracks the highest memory address every reserved
2897 // by JVM. It is used to detect heap-stack collision if running with
2898 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2899 // space than needed, it could confuse the collision detecting code. To
2900 // solve the problem, save current _highest_vm_reserved_address and
2901 // calculate the correct value before return.
2902 address old_highest = _highest_vm_reserved_address;
2903
2904 // Linux mmap allows caller to pass an address as hint; give it a try first,
2905 // if kernel honors the hint then we can return immediately.
2906 char * addr = anon_mmap(requested_addr, bytes, false);
2907 if (addr == requested_addr) {
2908 return requested_addr;
2909 }
2910
2911 if (addr != NULL) {
2912 // mmap() is successful but it fails to reserve at the requested address
2913 anon_munmap(addr, bytes);
2914 }
2915
2916 int i;
2917 for (i = 0; i < max_tries; ++i) {
2918 base[i] = reserve_memory(bytes);
2919
2920 if (base[i] != NULL) {
2921 // Is this the block we wanted?
2922 if (base[i] == requested_addr) {
2923 size[i] = bytes;
2924 break;
2925 }
2926
2927 // Does this overlap the block we wanted? Give back the overlapped
2928 // parts and try again.
2929
2930 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2931 if (top_overlap >= 0 && top_overlap < bytes) {
2932 unmap_memory(base[i], top_overlap);
2933 base[i] += top_overlap;
2934 size[i] = bytes - top_overlap;
2935 } else {
2936 size_t bottom_overlap = base[i] + bytes - requested_addr;
2937 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2938 unmap_memory(requested_addr, bottom_overlap);
2939 size[i] = bytes - bottom_overlap;
2940 } else {
2941 size[i] = bytes;
2942 }
2943 }
2944 }
2945 }
2946
2947 // Give back the unused reserved pieces.
2948
2949 for (int j = 0; j < i; ++j) {
2950 if (base[j] != NULL) {
2951 unmap_memory(base[j], size[j]);
2952 }
2953 }
2954
2955 if (i < max_tries) {
2956 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2957 return requested_addr;
2958 } else {
2959 _highest_vm_reserved_address = old_highest;
2960 return NULL;
2961 }
2962 }
2963
2964 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2965 return ::read(fd, buf, nBytes);
2966 }
2967
2968 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2969 // Solaris uses poll(), linux uses park().
2970 // Poll() is likely a better choice, assuming that Thread.interrupt()
2971 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2972 // SIGSEGV, see 4355769.
2973
2974 const int NANOSECS_PER_MILLISECS = 1000000;
2975
2976 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2977 assert(thread == Thread::current(), "thread consistency check");
2978
2979 ParkEvent * const slp = thread->_SleepEvent ;
2980 slp->reset() ;
2981 OrderAccess::fence() ;
2982
2983 if (interruptible) {
2984 jlong prevtime = javaTimeNanos();
2985
2986 for (;;) {
2987 if (os::is_interrupted(thread, true)) {
2988 return OS_INTRPT;
2989 }
2990
2991 jlong newtime = javaTimeNanos();
2992
2993 if (newtime - prevtime < 0) {
2994 // time moving backwards, should only happen if no monotonic clock
2995 // not a guarantee() because JVM should not abort on kernel/glibc bugs
2996 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2997 } else {
2998 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2999 }
3000
3001 if(millis <= 0) {
3002 return OS_OK;
3003 }
3004
3005 prevtime = newtime;
3006
3007 {
3008 assert(thread->is_Java_thread(), "sanity check");
3009 JavaThread *jt = (JavaThread *) thread;
3010 ThreadBlockInVM tbivm(jt);
3011 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3012
3013 jt->set_suspend_equivalent();
3014 // cleared by handle_special_suspend_equivalent_condition() or
3015 // java_suspend_self() via check_and_wait_while_suspended()
3016
3017 slp->park(millis);
3018
3019 // were we externally suspended while we were waiting?
3020 jt->check_and_wait_while_suspended();
3021 }
3022 }
3023 } else {
3024 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3025 jlong prevtime = javaTimeNanos();
3026
3027 for (;;) {
3028 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3029 // the 1st iteration ...
3030 jlong newtime = javaTimeNanos();
3031
3032 if (newtime - prevtime < 0) {
3033 // time moving backwards, should only happen if no monotonic clock
3034 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3035 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3036 } else {
3037 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3038 }
3039
3040 if(millis <= 0) break ;
3041
3042 prevtime = newtime;
3043 slp->park(millis);
3044 }
3045 return OS_OK ;
3046 }
3047 }
3048
3049 int os::naked_sleep() {
3050 // %% make the sleep time an integer flag. for now use 1 millisec.
3051 return os::sleep(Thread::current(), 1, false);
3052 }
3053
3054 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3055 void os::infinite_sleep() {
3056 while (true) { // sleep forever ...
3057 ::sleep(100); // ... 100 seconds at a time
3058 }
3059 }
3060
3061 // Used to convert frequent JVM_Yield() to nops
3062 bool os::dont_yield() {
3063 return DontYieldALot;
3064 }
3065
3066 void os::yield() {
3067 sched_yield();
3068 }
3069
3070 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3071
3072 void os::yield_all(int attempts) {
3073 // Yields to all threads, including threads with lower priorities
3074 // Threads on Linux are all with same priority. The Solaris style
3075 // os::yield_all() with nanosleep(1ms) is not necessary.
3076 sched_yield();
3077 }
3078
3079 // Called from the tight loops to possibly influence time-sharing heuristics
3080 void os::loop_breaker(int attempts) {
3081 os::yield_all(attempts);
3082 }
3083
3084 ////////////////////////////////////////////////////////////////////////////////
3085 // thread priority support
3086
3087 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3088 // only supports dynamic priority, static priority must be zero. For real-time
3089 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3090 // However, for large multi-threaded applications, SCHED_RR is not only slower
3091 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3092 // of 5 runs - Sep 2005).
3093 //
3094 // The following code actually changes the niceness of kernel-thread/LWP. It
3095 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3096 // not the entire user process, and user level threads are 1:1 mapped to kernel
3097 // threads. It has always been the case, but could change in the future. For
3098 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3099 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3100
3101 int os::java_to_os_priority[MaxPriority + 1] = {
3102 19, // 0 Entry should never be used
3103
3104 4, // 1 MinPriority
3105 3, // 2
3106 2, // 3
3107
3108 1, // 4
3109 0, // 5 NormPriority
3110 -1, // 6
3111
3112 -2, // 7
3113 -3, // 8
3114 -4, // 9 NearMaxPriority
3115
3116 -5 // 10 MaxPriority
3117 };
3118
3119 static int prio_init() {
3120 if (ThreadPriorityPolicy == 1) {
3121 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3122 // if effective uid is not root. Perhaps, a more elegant way of doing
3123 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3124 if (geteuid() != 0) {
3125 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3126 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3127 }
3128 ThreadPriorityPolicy = 0;
3129 }
3130 }
3131 return 0;
3132 }
3133
3134 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3135 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3136
3137 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3138 return (ret == 0) ? OS_OK : OS_ERR;
3139 }
3140
3141 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3142 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3143 *priority_ptr = java_to_os_priority[NormPriority];
3144 return OS_OK;
3145 }
3146
3147 errno = 0;
3148 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3149 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3150 }
3151
3152 // Hint to the underlying OS that a task switch would not be good.
3153 // Void return because it's a hint and can fail.
3154 void os::hint_no_preempt() {}
3155
3156 ////////////////////////////////////////////////////////////////////////////////
3157 // suspend/resume support
3158
3159 // the low-level signal-based suspend/resume support is a remnant from the
3160 // old VM-suspension that used to be for java-suspension, safepoints etc,
3161 // within hotspot. Now there is a single use-case for this:
3162 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3163 // that runs in the watcher thread.
3164 // The remaining code is greatly simplified from the more general suspension
3165 // code that used to be used.
3166 //
3167 // The protocol is quite simple:
3168 // - suspend:
3169 // - sends a signal to the target thread
3170 // - polls the suspend state of the osthread using a yield loop
3171 // - target thread signal handler (SR_handler) sets suspend state
3172 // and blocks in sigsuspend until continued
3173 // - resume:
3174 // - sets target osthread state to continue
3175 // - sends signal to end the sigsuspend loop in the SR_handler
3176 //
3177 // Note that the SR_lock plays no role in this suspend/resume protocol.
3178 //
3179
3180 static void resume_clear_context(OSThread *osthread) {
3181 osthread->set_ucontext(NULL);
3182 osthread->set_siginfo(NULL);
3183
3184 // notify the suspend action is completed, we have now resumed
3185 osthread->sr.clear_suspended();
3186 }
3187
3188 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3189 osthread->set_ucontext(context);
3190 osthread->set_siginfo(siginfo);
3191 }
3192
3193 //
3194 // Handler function invoked when a thread's execution is suspended or
3195 // resumed. We have to be careful that only async-safe functions are
3196 // called here (Note: most pthread functions are not async safe and
3197 // should be avoided.)
3198 //
3199 // Note: sigwait() is a more natural fit than sigsuspend() from an
3200 // interface point of view, but sigwait() prevents the signal hander
3201 // from being run. libpthread would get very confused by not having
3202 // its signal handlers run and prevents sigwait()'s use with the
3203 // mutex granting granting signal.
3204 //
3205 // Currently only ever called on the VMThread
3206 //
3207 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3208 // Save and restore errno to avoid confusing native code with EINTR
3209 // after sigsuspend.
3210 int old_errno = errno;
3211
3212 Thread* thread = Thread::current();
3213 OSThread* osthread = thread->osthread();
3214 assert(thread->is_VM_thread(), "Must be VMThread");
3215 // read current suspend action
3216 int action = osthread->sr.suspend_action();
3217 if (action == SR_SUSPEND) {
3218 suspend_save_context(osthread, siginfo, context);
3219
3220 // Notify the suspend action is about to be completed. do_suspend()
3221 // waits until SR_SUSPENDED is set and then returns. We will wait
3222 // here for a resume signal and that completes the suspend-other
3223 // action. do_suspend/do_resume is always called as a pair from
3224 // the same thread - so there are no races
3225
3226 // notify the caller
3227 osthread->sr.set_suspended();
3228
3229 sigset_t suspend_set; // signals for sigsuspend()
3230
3231 // get current set of blocked signals and unblock resume signal
3232 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3233 sigdelset(&suspend_set, SR_signum);
3234
3235 // wait here until we are resumed
3236 do {
3237 sigsuspend(&suspend_set);
3238 // ignore all returns until we get a resume signal
3239 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3240
3241 resume_clear_context(osthread);
3242
3243 } else {
3244 assert(action == SR_CONTINUE, "unexpected sr action");
3245 // nothing special to do - just leave the handler
3246 }
3247
3248 errno = old_errno;
3249 }
3250
3251
3252 static int SR_initialize() {
3253 struct sigaction act;
3254 char *s;
3255 /* Get signal number to use for suspend/resume */
3256 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3257 int sig = ::strtol(s, 0, 10);
3258 if (sig > 0 || sig < _NSIG) {
3259 SR_signum = sig;
3260 }
3261 }
3262
3263 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3264 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3265
3266 sigemptyset(&SR_sigset);
3267 sigaddset(&SR_sigset, SR_signum);
3268
3269 /* Set up signal handler for suspend/resume */
3270 act.sa_flags = SA_RESTART|SA_SIGINFO;
3271 act.sa_handler = (void (*)(int)) SR_handler;
3272
3273 // SR_signum is blocked by default.
3274 // 4528190 - We also need to block pthread restart signal (32 on all
3275 // supported Linux platforms). Note that LinuxThreads need to block
3276 // this signal for all threads to work properly. So we don't have
3277 // to use hard-coded signal number when setting up the mask.
3278 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3279
3280 if (sigaction(SR_signum, &act, 0) == -1) {
3281 return -1;
3282 }
3283
3284 // Save signal flag
3285 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3286 return 0;
3287 }
3288
3289 static int SR_finalize() {
3290 return 0;
3291 }
3292
3293
3294 // returns true on success and false on error - really an error is fatal
3295 // but this seems the normal response to library errors
3296 static bool do_suspend(OSThread* osthread) {
3297 // mark as suspended and send signal
3298 osthread->sr.set_suspend_action(SR_SUSPEND);
3299 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3300 assert_status(status == 0, status, "pthread_kill");
3301
3302 // check status and wait until notified of suspension
3303 if (status == 0) {
3304 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3305 os::yield_all(i);
3306 }
3307 osthread->sr.set_suspend_action(SR_NONE);
3308 return true;
3309 }
3310 else {
3311 osthread->sr.set_suspend_action(SR_NONE);
3312 return false;
3313 }
3314 }
3315
3316 static void do_resume(OSThread* osthread) {
3317 assert(osthread->sr.is_suspended(), "thread should be suspended");
3318 osthread->sr.set_suspend_action(SR_CONTINUE);
3319
3320 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3321 assert_status(status == 0, status, "pthread_kill");
3322 // check status and wait unit notified of resumption
3323 if (status == 0) {
3324 for (int i = 0; osthread->sr.is_suspended(); i++) {
3325 os::yield_all(i);
3326 }
3327 }
3328 osthread->sr.set_suspend_action(SR_NONE);
3329 }
3330
3331 ////////////////////////////////////////////////////////////////////////////////
3332 // interrupt support
3333
3334 void os::interrupt(Thread* thread) {
3335 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3336 "possibility of dangling Thread pointer");
3337
3338 OSThread* osthread = thread->osthread();
3339
3340 if (!osthread->interrupted()) {
3341 osthread->set_interrupted(true);
3342 // More than one thread can get here with the same value of osthread,
3343 // resulting in multiple notifications. We do, however, want the store
3344 // to interrupted() to be visible to other threads before we execute unpark().
3345 OrderAccess::fence();
3346 ParkEvent * const slp = thread->_SleepEvent ;
3347 if (slp != NULL) slp->unpark() ;
3348 }
3349
3350 // For JSR166. Unpark even if interrupt status already was set
3351 if (thread->is_Java_thread())
3352 ((JavaThread*)thread)->parker()->unpark();
3353
3354 ParkEvent * ev = thread->_ParkEvent ;
3355 if (ev != NULL) ev->unpark() ;
3356
3357 }
3358
3359 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3360 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3361 "possibility of dangling Thread pointer");
3362
3363 OSThread* osthread = thread->osthread();
3364
3365 bool interrupted = osthread->interrupted();
3366
3367 if (interrupted && clear_interrupted) {
3368 osthread->set_interrupted(false);
3369 // consider thread->_SleepEvent->reset() ... optional optimization
3370 }
3371
3372 return interrupted;
3373 }
3374
3375 ///////////////////////////////////////////////////////////////////////////////////
3376 // signal handling (except suspend/resume)
3377
3378 // This routine may be used by user applications as a "hook" to catch signals.
3379 // The user-defined signal handler must pass unrecognized signals to this
3380 // routine, and if it returns true (non-zero), then the signal handler must
3381 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3382 // routine will never retun false (zero), but instead will execute a VM panic
3383 // routine kill the process.
3384 //
3385 // If this routine returns false, it is OK to call it again. This allows
3386 // the user-defined signal handler to perform checks either before or after
3387 // the VM performs its own checks. Naturally, the user code would be making
3388 // a serious error if it tried to handle an exception (such as a null check
3389 // or breakpoint) that the VM was generating for its own correct operation.
3390 //
3391 // This routine may recognize any of the following kinds of signals:
3392 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3393 // It should be consulted by handlers for any of those signals.
3394 //
3395 // The caller of this routine must pass in the three arguments supplied
3396 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3397 // field of the structure passed to sigaction(). This routine assumes that
3398 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3399 //
3400 // Note that the VM will print warnings if it detects conflicting signal
3401 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3402 //
3403 extern "C" int
3404 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3405 void* ucontext, int abort_if_unrecognized);
3406
3407 void signalHandler(int sig, siginfo_t* info, void* uc) {
3408 assert(info != NULL && uc != NULL, "it must be old kernel");
3409 JVM_handle_linux_signal(sig, info, uc, true);
3410 }
3411
3412
3413 // This boolean allows users to forward their own non-matching signals
3414 // to JVM_handle_linux_signal, harmlessly.
3415 bool os::Linux::signal_handlers_are_installed = false;
3416
3417 // For signal-chaining
3418 struct sigaction os::Linux::sigact[MAXSIGNUM];
3419 unsigned int os::Linux::sigs = 0;
3420 bool os::Linux::libjsig_is_loaded = false;
3421 typedef struct sigaction *(*get_signal_t)(int);
3422 get_signal_t os::Linux::get_signal_action = NULL;
3423
3424 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3425 struct sigaction *actp = NULL;
3426
3427 if (libjsig_is_loaded) {
3428 // Retrieve the old signal handler from libjsig
3429 actp = (*get_signal_action)(sig);
3430 }
3431 if (actp == NULL) {
3432 // Retrieve the preinstalled signal handler from jvm
3433 actp = get_preinstalled_handler(sig);
3434 }
3435
3436 return actp;
3437 }
3438
3439 static bool call_chained_handler(struct sigaction *actp, int sig,
3440 siginfo_t *siginfo, void *context) {
3441 // Call the old signal handler
3442 if (actp->sa_handler == SIG_DFL) {
3443 // It's more reasonable to let jvm treat it as an unexpected exception
3444 // instead of taking the default action.
3445 return false;
3446 } else if (actp->sa_handler != SIG_IGN) {
3447 if ((actp->sa_flags & SA_NODEFER) == 0) {
3448 // automaticlly block the signal
3449 sigaddset(&(actp->sa_mask), sig);
3450 }
3451
3452 sa_handler_t hand;
3453 sa_sigaction_t sa;
3454 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3455 // retrieve the chained handler
3456 if (siginfo_flag_set) {
3457 sa = actp->sa_sigaction;
3458 } else {
3459 hand = actp->sa_handler;
3460 }
3461
3462 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3463 actp->sa_handler = SIG_DFL;
3464 }
3465
3466 // try to honor the signal mask
3467 sigset_t oset;
3468 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3469
3470 // call into the chained handler
3471 if (siginfo_flag_set) {
3472 (*sa)(sig, siginfo, context);
3473 } else {
3474 (*hand)(sig);
3475 }
3476
3477 // restore the signal mask
3478 pthread_sigmask(SIG_SETMASK, &oset, 0);
3479 }
3480 // Tell jvm's signal handler the signal is taken care of.
3481 return true;
3482 }
3483
3484 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3485 bool chained = false;
3486 // signal-chaining
3487 if (UseSignalChaining) {
3488 struct sigaction *actp = get_chained_signal_action(sig);
3489 if (actp != NULL) {
3490 chained = call_chained_handler(actp, sig, siginfo, context);
3491 }
3492 }
3493 return chained;
3494 }
3495
3496 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3497 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3498 return &sigact[sig];
3499 }
3500 return NULL;
3501 }
3502
3503 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3504 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3505 sigact[sig] = oldAct;
3506 sigs |= (unsigned int)1 << sig;
3507 }
3508
3509 // for diagnostic
3510 int os::Linux::sigflags[MAXSIGNUM];
3511
3512 int os::Linux::get_our_sigflags(int sig) {
3513 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3514 return sigflags[sig];
3515 }
3516
3517 void os::Linux::set_our_sigflags(int sig, int flags) {
3518 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3519 sigflags[sig] = flags;
3520 }
3521
3522 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3523 // Check for overwrite.
3524 struct sigaction oldAct;
3525 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3526
3527 void* oldhand = oldAct.sa_sigaction
3528 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3529 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3530 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3531 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3532 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3533 if (AllowUserSignalHandlers || !set_installed) {
3534 // Do not overwrite; user takes responsibility to forward to us.
3535 return;
3536 } else if (UseSignalChaining) {
3537 // save the old handler in jvm
3538 save_preinstalled_handler(sig, oldAct);
3539 // libjsig also interposes the sigaction() call below and saves the
3540 // old sigaction on it own.
3541 } else {
3542 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3543 "%#lx for signal %d.", (long)oldhand, sig));
3544 }
3545 }
3546
3547 struct sigaction sigAct;
3548 sigfillset(&(sigAct.sa_mask));
3549 sigAct.sa_handler = SIG_DFL;
3550 if (!set_installed) {
3551 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3552 } else {
3553 sigAct.sa_sigaction = signalHandler;
3554 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3555 }
3556 // Save flags, which are set by ours
3557 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3558 sigflags[sig] = sigAct.sa_flags;
3559
3560 int ret = sigaction(sig, &sigAct, &oldAct);
3561 assert(ret == 0, "check");
3562
3563 void* oldhand2 = oldAct.sa_sigaction
3564 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3565 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3566 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3567 }
3568
3569 // install signal handlers for signals that HotSpot needs to
3570 // handle in order to support Java-level exception handling.
3571
3572 void os::Linux::install_signal_handlers() {
3573 if (!signal_handlers_are_installed) {
3574 signal_handlers_are_installed = true;
3575
3576 // signal-chaining
3577 typedef void (*signal_setting_t)();
3578 signal_setting_t begin_signal_setting = NULL;
3579 signal_setting_t end_signal_setting = NULL;
3580 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3581 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3582 if (begin_signal_setting != NULL) {
3583 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3584 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3585 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3586 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3587 libjsig_is_loaded = true;
3588 assert(UseSignalChaining, "should enable signal-chaining");
3589 }
3590 if (libjsig_is_loaded) {
3591 // Tell libjsig jvm is setting signal handlers
3592 (*begin_signal_setting)();
3593 }
3594
3595 set_signal_handler(SIGSEGV, true);
3596 set_signal_handler(SIGPIPE, true);
3597 set_signal_handler(SIGBUS, true);
3598 set_signal_handler(SIGILL, true);
3599 set_signal_handler(SIGFPE, true);
3600 set_signal_handler(SIGXFSZ, true);
3601
3602 if (libjsig_is_loaded) {
3603 // Tell libjsig jvm finishes setting signal handlers
3604 (*end_signal_setting)();
3605 }
3606
3607 // We don't activate signal checker if libjsig is in place, we trust ourselves
3608 // and if UserSignalHandler is installed all bets are off
3609 if (CheckJNICalls) {
3610 if (libjsig_is_loaded) {
3611 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3612 check_signals = false;
3613 }
3614 if (AllowUserSignalHandlers) {
3615 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3616 check_signals = false;
3617 }
3618 }
3619 }
3620 }
3621
3622 // This is the fastest way to get thread cpu time on Linux.
3623 // Returns cpu time (user+sys) for any thread, not only for current.
3624 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3625 // It might work on 2.6.10+ with a special kernel/glibc patch.
3626 // For reference, please, see IEEE Std 1003.1-2004:
3627 // http://www.unix.org/single_unix_specification
3628
3629 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3630 struct timespec tp;
3631 int rc = os::Linux::clock_gettime(clockid, &tp);
3632 assert(rc == 0, "clock_gettime is expected to return 0 code");
3633
3634 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3635 }
3636
3637 /////
3638 // glibc on Linux platform uses non-documented flag
3639 // to indicate, that some special sort of signal
3640 // trampoline is used.
3641 // We will never set this flag, and we should
3642 // ignore this flag in our diagnostic
3643 #ifdef SIGNIFICANT_SIGNAL_MASK
3644 #undef SIGNIFICANT_SIGNAL_MASK
3645 #endif
3646 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3647
3648 static const char* get_signal_handler_name(address handler,
3649 char* buf, int buflen) {
3650 int offset;
3651 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3652 if (found) {
3653 // skip directory names
3654 const char *p1, *p2;
3655 p1 = buf;
3656 size_t len = strlen(os::file_separator());
3657 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3658 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3659 } else {
3660 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3661 }
3662 return buf;
3663 }
3664
3665 static void print_signal_handler(outputStream* st, int sig,
3666 char* buf, size_t buflen) {
3667 struct sigaction sa;
3668
3669 sigaction(sig, NULL, &sa);
3670
3671 // See comment for SIGNIFICANT_SIGNAL_MASK define
3672 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3673
3674 st->print("%s: ", os::exception_name(sig, buf, buflen));
3675
3676 address handler = (sa.sa_flags & SA_SIGINFO)
3677 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3678 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3679
3680 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3681 st->print("SIG_DFL");
3682 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3683 st->print("SIG_IGN");
3684 } else {
3685 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3686 }
3687
3688 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3689
3690 address rh = VMError::get_resetted_sighandler(sig);
3691 // May be, handler was resetted by VMError?
3692 if(rh != NULL) {
3693 handler = rh;
3694 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3695 }
3696
3697 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3698
3699 // Check: is it our handler?
3700 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3701 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3702 // It is our signal handler
3703 // check for flags, reset system-used one!
3704 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3705 st->print(
3706 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3707 os::Linux::get_our_sigflags(sig));
3708 }
3709 }
3710 st->cr();
3711 }
3712
3713
3714 #define DO_SIGNAL_CHECK(sig) \
3715 if (!sigismember(&check_signal_done, sig)) \
3716 os::Linux::check_signal_handler(sig)
3717
3718 // This method is a periodic task to check for misbehaving JNI applications
3719 // under CheckJNI, we can add any periodic checks here
3720
3721 void os::run_periodic_checks() {
3722
3723 if (check_signals == false) return;
3724
3725 // SEGV and BUS if overridden could potentially prevent
3726 // generation of hs*.log in the event of a crash, debugging
3727 // such a case can be very challenging, so we absolutely
3728 // check the following for a good measure:
3729 DO_SIGNAL_CHECK(SIGSEGV);
3730 DO_SIGNAL_CHECK(SIGILL);
3731 DO_SIGNAL_CHECK(SIGFPE);
3732 DO_SIGNAL_CHECK(SIGBUS);
3733 DO_SIGNAL_CHECK(SIGPIPE);
3734 DO_SIGNAL_CHECK(SIGXFSZ);
3735
3736
3737 // ReduceSignalUsage allows the user to override these handlers
3738 // see comments at the very top and jvm_solaris.h
3739 if (!ReduceSignalUsage) {
3740 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3741 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3742 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3743 DO_SIGNAL_CHECK(BREAK_SIGNAL);
3744 }
3745
3746 DO_SIGNAL_CHECK(SR_signum);
3747 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3748 }
3749
3750 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3751
3752 static os_sigaction_t os_sigaction = NULL;
3753
3754 void os::Linux::check_signal_handler(int sig) {
3755 char buf[O_BUFLEN];
3756 address jvmHandler = NULL;
3757
3758
3759 struct sigaction act;
3760 if (os_sigaction == NULL) {
3761 // only trust the default sigaction, in case it has been interposed
3762 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3763 if (os_sigaction == NULL) return;
3764 }
3765
3766 os_sigaction(sig, (struct sigaction*)NULL, &act);
3767
3768
3769 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3770
3771 address thisHandler = (act.sa_flags & SA_SIGINFO)
3772 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3773 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3774
3775
3776 switch(sig) {
3777 case SIGSEGV:
3778 case SIGBUS:
3779 case SIGFPE:
3780 case SIGPIPE:
3781 case SIGILL:
3782 case SIGXFSZ:
3783 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3784 break;
3785
3786 case SHUTDOWN1_SIGNAL:
3787 case SHUTDOWN2_SIGNAL:
3788 case SHUTDOWN3_SIGNAL:
3789 case BREAK_SIGNAL:
3790 jvmHandler = (address)user_handler();
3791 break;
3792
3793 case INTERRUPT_SIGNAL:
3794 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3795 break;
3796
3797 default:
3798 if (sig == SR_signum) {
3799 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3800 } else {
3801 return;
3802 }
3803 break;
3804 }
3805
3806 if (thisHandler != jvmHandler) {
3807 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3808 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3809 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3810 // No need to check this sig any longer
3811 sigaddset(&check_signal_done, sig);
3812 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3813 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3814 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3815 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
3816 // No need to check this sig any longer
3817 sigaddset(&check_signal_done, sig);
3818 }
3819
3820 // Dump all the signal
3821 if (sigismember(&check_signal_done, sig)) {
3822 print_signal_handlers(tty, buf, O_BUFLEN);
3823 }
3824 }
3825
3826 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3827
3828 extern bool signal_name(int signo, char* buf, size_t len);
3829
3830 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3831 if (0 < exception_code && exception_code <= SIGRTMAX) {
3832 // signal
3833 if (!signal_name(exception_code, buf, size)) {
3834 jio_snprintf(buf, size, "SIG%d", exception_code);
3835 }
3836 return buf;
3837 } else {
3838 return NULL;
3839 }
3840 }
3841
3842 // this is called _before_ the most of global arguments have been parsed
3843 void os::init(void) {
3844 char dummy; /* used to get a guess on initial stack address */
3845 // first_hrtime = gethrtime();
3846
3847 // With LinuxThreads the JavaMain thread pid (primordial thread)
3848 // is different than the pid of the java launcher thread.
3849 // So, on Linux, the launcher thread pid is passed to the VM
3850 // via the sun.java.launcher.pid property.
3851 // Use this property instead of getpid() if it was correctly passed.
3852 // See bug 6351349.
3853 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3854
3855 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3856
3857 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3858
3859 init_random(1234567);
3860
3861 ThreadCritical::initialize();
3862
3863 Linux::set_page_size(sysconf(_SC_PAGESIZE));
3864 if (Linux::page_size() == -1) {
3865 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
3866 strerror(errno)));
3867 }
3868 init_page_sizes((size_t) Linux::page_size());
3869
3870 Linux::initialize_system_info();
3871
3872 // main_thread points to the aboriginal thread
3873 Linux::_main_thread = pthread_self();
3874
3875 Linux::clock_init();
3876 initial_time_count = os::elapsed_counter();
3877 pthread_mutex_init(&dl_mutex, NULL);
3878 }
3879
3880 // To install functions for atexit system call
3881 extern "C" {
3882 static void perfMemory_exit_helper() {
3883 perfMemory_exit();
3884 }
3885 }
3886
3887 // this is called _after_ the global arguments have been parsed
3888 jint os::init_2(void)
3889 {
3890 Linux::fast_thread_clock_init();
3891
3892 // Allocate a single page and mark it as readable for safepoint polling
3893 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3894 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3895
3896 os::set_polling_page( polling_page );
3897
3898 #ifndef PRODUCT
3899 if(Verbose && PrintMiscellaneous)
3900 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3901 #endif
3902
3903 if (!UseMembar) {
3904 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3905 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3906 os::set_memory_serialize_page( mem_serialize_page );
3907
3908 #ifndef PRODUCT
3909 if(Verbose && PrintMiscellaneous)
3910 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3911 #endif
3912 }
3913
3914 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3915
3916 // initialize suspend/resume support - must do this before signal_sets_init()
3917 if (SR_initialize() != 0) {
3918 perror("SR_initialize failed");
3919 return JNI_ERR;
3920 }
3921
3922 Linux::signal_sets_init();
3923 Linux::install_signal_handlers();
3924
3925 size_t threadStackSizeInBytes = ThreadStackSize * K;
3926 if (threadStackSizeInBytes != 0 &&
3927 threadStackSizeInBytes < Linux::min_stack_allowed) {
3928 tty->print_cr("\nThe stack size specified is too small, "
3929 "Specify at least %dk",
3930 Linux::min_stack_allowed / K);
3931 return JNI_ERR;
3932 }
3933
3934 // Make the stack size a multiple of the page size so that
3935 // the yellow/red zones can be guarded.
3936 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3937 vm_page_size()));
3938
3939 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3940
3941 Linux::libpthread_init();
3942 if (PrintMiscellaneous && (Verbose || WizardMode)) {
3943 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3944 Linux::glibc_version(), Linux::libpthread_version(),
3945 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3946 }
3947
3948 if (UseNUMA) {
3949 if (!Linux::libnuma_init()) {
3950 UseNUMA = false;
3951 } else {
3952 if ((Linux::numa_max_node() < 1)) {
3953 // There's only one node(they start from 0), disable NUMA.
3954 UseNUMA = false;
3955 }
3956 }
3957 if (!UseNUMA && ForceNUMA) {
3958 UseNUMA = true;
3959 }
3960 }
3961
3962 if (MaxFDLimit) {
3963 // set the number of file descriptors to max. print out error
3964 // if getrlimit/setrlimit fails but continue regardless.
3965 struct rlimit nbr_files;
3966 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3967 if (status != 0) {
3968 if (PrintMiscellaneous && (Verbose || WizardMode))
3969 perror("os::init_2 getrlimit failed");
3970 } else {
3971 nbr_files.rlim_cur = nbr_files.rlim_max;
3972 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3973 if (status != 0) {
3974 if (PrintMiscellaneous && (Verbose || WizardMode))
3975 perror("os::init_2 setrlimit failed");
3976 }
3977 }
3978 }
3979
3980 // Initialize lock used to serialize thread creation (see os::create_thread)
3981 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3982
3983 // Initialize HPI.
3984 jint hpi_result = hpi::initialize();
3985 if (hpi_result != JNI_OK) {
3986 tty->print_cr("There was an error trying to initialize the HPI library.");
3987 return hpi_result;
3988 }
3989
3990 // at-exit methods are called in the reverse order of their registration.
3991 // atexit functions are called on return from main or as a result of a
3992 // call to exit(3C). There can be only 32 of these functions registered
3993 // and atexit() does not set errno.
3994
3995 if (PerfAllowAtExitRegistration) {
3996 // only register atexit functions if PerfAllowAtExitRegistration is set.
3997 // atexit functions can be delayed until process exit time, which
3998 // can be problematic for embedded VM situations. Embedded VMs should
3999 // call DestroyJavaVM() to assure that VM resources are released.
4000
4001 // note: perfMemory_exit_helper atexit function may be removed in
4002 // the future if the appropriate cleanup code can be added to the
4003 // VM_Exit VMOperation's doit method.
4004 if (atexit(perfMemory_exit_helper) != 0) {
4005 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4006 }
4007 }
4008
4009 // initialize thread priority policy
4010 prio_init();
4011
4012 return JNI_OK;
4013 }
4014
4015 // this is called at the end of vm_initialization
4016 void os::init_3(void) { }
4017
4018 // Mark the polling page as unreadable
4019 void os::make_polling_page_unreadable(void) {
4020 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4021 fatal("Could not disable polling page");
4022 };
4023
4024 // Mark the polling page as readable
4025 void os::make_polling_page_readable(void) {
4026 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4027 fatal("Could not enable polling page");
4028 }
4029 };
4030
4031 int os::active_processor_count() {
4032 // Linux doesn't yet have a (official) notion of processor sets,
4033 // so just return the number of online processors.
4034 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4035 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4036 return online_cpus;
4037 }
4038
4039 bool os::distribute_processes(uint length, uint* distribution) {
4040 // Not yet implemented.
4041 return false;
4042 }
4043
4044 bool os::bind_to_processor(uint processor_id) {
4045 // Not yet implemented.
4046 return false;
4047 }
4048
4049 ///
4050
4051 // Suspends the target using the signal mechanism and then grabs the PC before
4052 // resuming the target. Used by the flat-profiler only
4053 ExtendedPC os::get_thread_pc(Thread* thread) {
4054 // Make sure that it is called by the watcher for the VMThread
4055 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4056 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4057
4058 ExtendedPC epc;
4059
4060 OSThread* osthread = thread->osthread();
4061 if (do_suspend(osthread)) {
4062 if (osthread->ucontext() != NULL) {
4063 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4064 } else {
4065 // NULL context is unexpected, double-check this is the VMThread
4066 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4067 }
4068 do_resume(osthread);
4069 }
4070 // failure means pthread_kill failed for some reason - arguably this is
4071 // a fatal problem, but such problems are ignored elsewhere
4072
4073 return epc;
4074 }
4075
4076 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4077 {
4078 if (is_NPTL()) {
4079 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4080 } else {
4081 #ifndef IA64
4082 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4083 // word back to default 64bit precision if condvar is signaled. Java
4084 // wants 53bit precision. Save and restore current value.
4085 int fpu = get_fpu_control_word();
4086 #endif // IA64
4087 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4088 #ifndef IA64
4089 set_fpu_control_word(fpu);
4090 #endif // IA64
4091 return status;
4092 }
4093 }
4094
4095 ////////////////////////////////////////////////////////////////////////////////
4096 // debug support
4097
4098 static address same_page(address x, address y) {
4099 int page_bits = -os::vm_page_size();
4100 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4101 return x;
4102 else if (x > y)
4103 return (address)(intptr_t(y) | ~page_bits) + 1;
4104 else
4105 return (address)(intptr_t(y) & page_bits);
4106 }
4107
4108 bool os::find(address addr, outputStream* st) {
4109 Dl_info dlinfo;
4110 memset(&dlinfo, 0, sizeof(dlinfo));
4111 if (dladdr(addr, &dlinfo)) {
4112 st->print(PTR_FORMAT ": ", addr);
4113 if (dlinfo.dli_sname != NULL) {
4114 st->print("%s+%#x", dlinfo.dli_sname,
4115 addr - (intptr_t)dlinfo.dli_saddr);
4116 } else if (dlinfo.dli_fname) {
4117 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4118 } else {
4119 st->print("<absolute address>");
4120 }
4121 if (dlinfo.dli_fname) {
4122 st->print(" in %s", dlinfo.dli_fname);
4123 }
4124 if (dlinfo.dli_fbase) {
4125 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4126 }
4127 st->cr();
4128
4129 if (Verbose) {
4130 // decode some bytes around the PC
4131 address begin = same_page(addr-40, addr);
4132 address end = same_page(addr+40, addr);
4133 address lowest = (address) dlinfo.dli_sname;
4134 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4135 if (begin < lowest) begin = lowest;
4136 Dl_info dlinfo2;
4137 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4138 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4139 end = (address) dlinfo2.dli_saddr;
4140 Disassembler::decode(begin, end, st);
4141 }
4142 return true;
4143 }
4144 return false;
4145 }
4146
4147 ////////////////////////////////////////////////////////////////////////////////
4148 // misc
4149
4150 // This does not do anything on Linux. This is basically a hook for being
4151 // able to use structured exception handling (thread-local exception filters)
4152 // on, e.g., Win32.
4153 void
4154 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4155 JavaCallArguments* args, Thread* thread) {
4156 f(value, method, args, thread);
4157 }
4158
4159 void os::print_statistics() {
4160 }
4161
4162 int os::message_box(const char* title, const char* message) {
4163 int i;
4164 fdStream err(defaultStream::error_fd());
4165 for (i = 0; i < 78; i++) err.print_raw("=");
4166 err.cr();
4167 err.print_raw_cr(title);
4168 for (i = 0; i < 78; i++) err.print_raw("-");
4169 err.cr();
4170 err.print_raw_cr(message);
4171 for (i = 0; i < 78; i++) err.print_raw("=");
4172 err.cr();
4173
4174 char buf[16];
4175 // Prevent process from exiting upon "read error" without consuming all CPU
4176 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4177
4178 return buf[0] == 'y' || buf[0] == 'Y';
4179 }
4180
4181 int os::stat(const char *path, struct stat *sbuf) {
4182 char pathbuf[MAX_PATH];
4183 if (strlen(path) > MAX_PATH - 1) {
4184 errno = ENAMETOOLONG;
4185 return -1;
4186 }
4187 hpi::native_path(strcpy(pathbuf, path));
4188 return ::stat(pathbuf, sbuf);
4189 }
4190
4191 bool os::check_heap(bool force) {
4192 return true;
4193 }
4194
4195 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4196 return ::vsnprintf(buf, count, format, args);
4197 }
4198
4199 // Is a (classpath) directory empty?
4200 bool os::dir_is_empty(const char* path) {
4201 DIR *dir = NULL;
4202 struct dirent *ptr;
4203
4204 dir = opendir(path);
4205 if (dir == NULL) return true;
4206
4207 /* Scan the directory */
4208 bool result = true;
4209 char buf[sizeof(struct dirent) + MAX_PATH];
4210 while (result && (ptr = ::readdir(dir)) != NULL) {
4211 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4212 result = false;
4213 }
4214 }
4215 closedir(dir);
4216 return result;
4217 }
4218
4219 // create binary file, rewriting existing file if required
4220 int os::create_binary_file(const char* path, bool rewrite_existing) {
4221 int oflags = O_WRONLY | O_CREAT;
4222 if (!rewrite_existing) {
4223 oflags |= O_EXCL;
4224 }
4225 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4226 }
4227
4228 // return current position of file pointer
4229 jlong os::current_file_offset(int fd) {
4230 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4231 }
4232
4233 // move file pointer to the specified offset
4234 jlong os::seek_to_file_offset(int fd, jlong offset) {
4235 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4236 }
4237
4238 // Map a block of memory.
4239 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4240 char *addr, size_t bytes, bool read_only,
4241 bool allow_exec) {
4242 int prot;
4243 int flags;
4244
4245 if (read_only) {
4246 prot = PROT_READ;
4247 flags = MAP_SHARED;
4248 } else {
4249 prot = PROT_READ | PROT_WRITE;
4250 flags = MAP_PRIVATE;
4251 }
4252
4253 if (allow_exec) {
4254 prot |= PROT_EXEC;
4255 }
4256
4257 if (addr != NULL) {
4258 flags |= MAP_FIXED;
4259 }
4260
4261 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4262 fd, file_offset);
4263 if (mapped_address == MAP_FAILED) {
4264 return NULL;
4265 }
4266 return mapped_address;
4267 }
4268
4269
4270 // Remap a block of memory.
4271 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4272 char *addr, size_t bytes, bool read_only,
4273 bool allow_exec) {
4274 // same as map_memory() on this OS
4275 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4276 allow_exec);
4277 }
4278
4279
4280 // Unmap a block of memory.
4281 bool os::unmap_memory(char* addr, size_t bytes) {
4282 return munmap(addr, bytes) == 0;
4283 }
4284
4285 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4286
4287 static clockid_t thread_cpu_clockid(Thread* thread) {
4288 pthread_t tid = thread->osthread()->pthread_id();
4289 clockid_t clockid;
4290
4291 // Get thread clockid
4292 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4293 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4294 return clockid;
4295 }
4296
4297 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4298 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4299 // of a thread.
4300 //
4301 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4302 // the fast estimate available on the platform.
4303
4304 jlong os::current_thread_cpu_time() {
4305 if (os::Linux::supports_fast_thread_cpu_time()) {
4306 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4307 } else {
4308 // return user + sys since the cost is the same
4309 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4310 }
4311 }
4312
4313 jlong os::thread_cpu_time(Thread* thread) {
4314 // consistent with what current_thread_cpu_time() returns
4315 if (os::Linux::supports_fast_thread_cpu_time()) {
4316 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4317 } else {
4318 return slow_thread_cpu_time(thread, true /* user + sys */);
4319 }
4320 }
4321
4322 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4323 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4324 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4325 } else {
4326 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4327 }
4328 }
4329
4330 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4331 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4332 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4333 } else {
4334 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4335 }
4336 }
4337
4338 //
4339 // -1 on error.
4340 //
4341
4342 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4343 static bool proc_pid_cpu_avail = true;
4344 static bool proc_task_unchecked = true;
4345 static const char *proc_stat_path = "/proc/%d/stat";
4346 pid_t tid = thread->osthread()->thread_id();
4347 int i;
4348 char *s;
4349 char stat[2048];
4350 int statlen;
4351 char proc_name[64];
4352 int count;
4353 long sys_time, user_time;
4354 char string[64];
4355 char cdummy;
4356 int idummy;
4357 long ldummy;
4358 FILE *fp;
4359
4360 // We first try accessing /proc/<pid>/cpu since this is faster to
4361 // process. If this file is not present (linux kernels 2.5 and above)
4362 // then we open /proc/<pid>/stat.
4363 if ( proc_pid_cpu_avail ) {
4364 sprintf(proc_name, "/proc/%d/cpu", tid);
4365 fp = fopen(proc_name, "r");
4366 if ( fp != NULL ) {
4367 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4368 fclose(fp);
4369 if ( count != 3 ) return -1;
4370
4371 if (user_sys_cpu_time) {
4372 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4373 } else {
4374 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4375 }
4376 }
4377 else proc_pid_cpu_avail = false;
4378 }
4379
4380 // The /proc/<tid>/stat aggregates per-process usage on
4381 // new Linux kernels 2.6+ where NPTL is supported.
4382 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4383 // See bug 6328462.
4384 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4385 // and possibly in some other cases, so we check its availability.
4386 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4387 // This is executed only once
4388 proc_task_unchecked = false;
4389 fp = fopen("/proc/self/task", "r");
4390 if (fp != NULL) {
4391 proc_stat_path = "/proc/self/task/%d/stat";
4392 fclose(fp);
4393 }
4394 }
4395
4396 sprintf(proc_name, proc_stat_path, tid);
4397 fp = fopen(proc_name, "r");
4398 if ( fp == NULL ) return -1;
4399 statlen = fread(stat, 1, 2047, fp);
4400 stat[statlen] = '\0';
4401 fclose(fp);
4402
4403 // Skip pid and the command string. Note that we could be dealing with
4404 // weird command names, e.g. user could decide to rename java launcher
4405 // to "java 1.4.2 :)", then the stat file would look like
4406 // 1234 (java 1.4.2 :)) R ... ...
4407 // We don't really need to know the command string, just find the last
4408 // occurrence of ")" and then start parsing from there. See bug 4726580.
4409 s = strrchr(stat, ')');
4410 i = 0;
4411 if (s == NULL ) return -1;
4412
4413 // Skip blank chars
4414 do s++; while (isspace(*s));
4415
4416 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4417 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4418 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4419 &user_time, &sys_time);
4420 if ( count != 13 ) return -1;
4421 if (user_sys_cpu_time) {
4422 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4423 } else {
4424 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4425 }
4426 }
4427
4428 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4429 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4430 info_ptr->may_skip_backward = false; // elapsed time not wall time
4431 info_ptr->may_skip_forward = false; // elapsed time not wall time
4432 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4433 }
4434
4435 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4436 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4437 info_ptr->may_skip_backward = false; // elapsed time not wall time
4438 info_ptr->may_skip_forward = false; // elapsed time not wall time
4439 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4440 }
4441
4442 bool os::is_thread_cpu_time_supported() {
4443 return true;
4444 }
4445
4446 // System loadavg support. Returns -1 if load average cannot be obtained.
4447 // Linux doesn't yet have a (official) notion of processor sets,
4448 // so just return the system wide load average.
4449 int os::loadavg(double loadavg[], int nelem) {
4450 return ::getloadavg(loadavg, nelem);
4451 }
4452
4453 void os::pause() {
4454 char filename[MAX_PATH];
4455 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4456 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4457 } else {
4458 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4459 }
4460
4461 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4462 if (fd != -1) {
4463 struct stat buf;
4464 close(fd);
4465 while (::stat(filename, &buf) == 0) {
4466 (void)::poll(NULL, 0, 100);
4467 }
4468 } else {
4469 jio_fprintf(stderr,
4470 "Could not open pause file '%s', continuing immediately.\n", filename);
4471 }
4472 }
4473
4474 extern "C" {
4475
4476 /**
4477 * NOTE: the following code is to keep the green threads code
4478 * in the libjava.so happy. Once the green threads is removed,
4479 * these code will no longer be needed.
4480 */
4481 int
4482 jdk_waitpid(pid_t pid, int* status, int options) {
4483 return waitpid(pid, status, options);
4484 }
4485
4486 int
4487 fork1() {
4488 return fork();
4489 }
4490
4491 int
4492 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4493 return sem_init(sem, pshared, value);
4494 }
4495
4496 int
4497 jdk_sem_post(sem_t *sem) {
4498 return sem_post(sem);
4499 }
4500
4501 int
4502 jdk_sem_wait(sem_t *sem) {
4503 return sem_wait(sem);
4504 }
4505
4506 int
4507 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4508 return pthread_sigmask(how , newmask, oldmask);
4509 }
4510
4511 }
4512
4513 // Refer to the comments in os_solaris.cpp park-unpark.
4514 //
4515 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4516 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4517 // For specifics regarding the bug see GLIBC BUGID 261237 :
4518 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4519 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4520 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4521 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4522 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4523 // and monitorenter when we're using 1-0 locking. All those operations may result in
4524 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4525 // of libpthread avoids the problem, but isn't practical.
4526 //
4527 // Possible remedies:
4528 //
4529 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4530 // This is palliative and probabilistic, however. If the thread is preempted
4531 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4532 // than the minimum period may have passed, and the abstime may be stale (in the
4533 // past) resultin in a hang. Using this technique reduces the odds of a hang
4534 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4535 //
4536 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4537 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4538 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4539 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4540 // thread.
4541 //
4542 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4543 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4544 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4545 // This also works well. In fact it avoids kernel-level scalability impediments
4546 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4547 // timers in a graceful fashion.
4548 //
4549 // 4. When the abstime value is in the past it appears that control returns
4550 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4551 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4552 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4553 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4554 // It may be possible to avoid reinitialization by checking the return
4555 // value from pthread_cond_timedwait(). In addition to reinitializing the
4556 // condvar we must establish the invariant that cond_signal() is only called
4557 // within critical sections protected by the adjunct mutex. This prevents
4558 // cond_signal() from "seeing" a condvar that's in the midst of being
4559 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4560 // desirable signal-after-unlock optimization that avoids futile context switching.
4561 //
4562 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4563 // structure when a condvar is used or initialized. cond_destroy() would
4564 // release the helper structure. Our reinitialize-after-timedwait fix
4565 // put excessive stress on malloc/free and locks protecting the c-heap.
4566 //
4567 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4568 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4569 // and only enabling the work-around for vulnerable environments.
4570
4571 // utility to compute the abstime argument to timedwait:
4572 // millis is the relative timeout time
4573 // abstime will be the absolute timeout time
4574 // TODO: replace compute_abstime() with unpackTime()
4575
4576 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4577 if (millis < 0) millis = 0;
4578 struct timeval now;
4579 int status = gettimeofday(&now, NULL);
4580 assert(status == 0, "gettimeofday");
4581 jlong seconds = millis / 1000;
4582 millis %= 1000;
4583 if (seconds > 50000000) { // see man cond_timedwait(3T)
4584 seconds = 50000000;
4585 }
4586 abstime->tv_sec = now.tv_sec + seconds;
4587 long usec = now.tv_usec + millis * 1000;
4588 if (usec >= 1000000) {
4589 abstime->tv_sec += 1;
4590 usec -= 1000000;
4591 }
4592 abstime->tv_nsec = usec * 1000;
4593 return abstime;
4594 }
4595
4596
4597 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4598 // Conceptually TryPark() should be equivalent to park(0).
4599
4600 int os::PlatformEvent::TryPark() {
4601 for (;;) {
4602 const int v = _Event ;
4603 guarantee ((v == 0) || (v == 1), "invariant") ;
4604 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4605 }
4606 }
4607
4608 void os::PlatformEvent::park() { // AKA "down()"
4609 // Invariant: Only the thread associated with the Event/PlatformEvent
4610 // may call park().
4611 // TODO: assert that _Assoc != NULL or _Assoc == Self
4612 int v ;
4613 for (;;) {
4614 v = _Event ;
4615 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4616 }
4617 guarantee (v >= 0, "invariant") ;
4618 if (v == 0) {
4619 // Do this the hard way by blocking ...
4620 int status = pthread_mutex_lock(_mutex);
4621 assert_status(status == 0, status, "mutex_lock");
4622 guarantee (_nParked == 0, "invariant") ;
4623 ++ _nParked ;
4624 while (_Event < 0) {
4625 status = pthread_cond_wait(_cond, _mutex);
4626 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4627 // Treat this the same as if the wait was interrupted
4628 if (status == ETIME) { status = EINTR; }
4629 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4630 }
4631 -- _nParked ;
4632
4633 // In theory we could move the ST of 0 into _Event past the unlock(),
4634 // but then we'd need a MEMBAR after the ST.
4635 _Event = 0 ;
4636 status = pthread_mutex_unlock(_mutex);
4637 assert_status(status == 0, status, "mutex_unlock");
4638 }
4639 guarantee (_Event >= 0, "invariant") ;
4640 }
4641
4642 int os::PlatformEvent::park(jlong millis) {
4643 guarantee (_nParked == 0, "invariant") ;
4644
4645 int v ;
4646 for (;;) {
4647 v = _Event ;
4648 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4649 }
4650 guarantee (v >= 0, "invariant") ;
4651 if (v != 0) return OS_OK ;
4652
4653 // We do this the hard way, by blocking the thread.
4654 // Consider enforcing a minimum timeout value.
4655 struct timespec abst;
4656 compute_abstime(&abst, millis);
4657
4658 int ret = OS_TIMEOUT;
4659 int status = pthread_mutex_lock(_mutex);
4660 assert_status(status == 0, status, "mutex_lock");
4661 guarantee (_nParked == 0, "invariant") ;
4662 ++_nParked ;
4663
4664 // Object.wait(timo) will return because of
4665 // (a) notification
4666 // (b) timeout
4667 // (c) thread.interrupt
4668 //
4669 // Thread.interrupt and object.notify{All} both call Event::set.
4670 // That is, we treat thread.interrupt as a special case of notification.
4671 // The underlying Solaris implementation, cond_timedwait, admits
4672 // spurious/premature wakeups, but the JLS/JVM spec prevents the
4673 // JVM from making those visible to Java code. As such, we must
4674 // filter out spurious wakeups. We assume all ETIME returns are valid.
4675 //
4676 // TODO: properly differentiate simultaneous notify+interrupt.
4677 // In that case, we should propagate the notify to another waiter.
4678
4679 while (_Event < 0) {
4680 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4681 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4682 pthread_cond_destroy (_cond);
4683 pthread_cond_init (_cond, NULL) ;
4684 }
4685 assert_status(status == 0 || status == EINTR ||
4686 status == ETIME || status == ETIMEDOUT,
4687 status, "cond_timedwait");
4688 if (!FilterSpuriousWakeups) break ; // previous semantics
4689 if (status == ETIME || status == ETIMEDOUT) break ;
4690 // We consume and ignore EINTR and spurious wakeups.
4691 }
4692 --_nParked ;
4693 if (_Event >= 0) {
4694 ret = OS_OK;
4695 }
4696 _Event = 0 ;
4697 status = pthread_mutex_unlock(_mutex);
4698 assert_status(status == 0, status, "mutex_unlock");
4699 assert (_nParked == 0, "invariant") ;
4700 return ret;
4701 }
4702
4703 void os::PlatformEvent::unpark() {
4704 int v, AnyWaiters ;
4705 for (;;) {
4706 v = _Event ;
4707 if (v > 0) {
4708 // The LD of _Event could have reordered or be satisfied
4709 // by a read-aside from this processor's write buffer.
4710 // To avoid problems execute a barrier and then
4711 // ratify the value.
4712 OrderAccess::fence() ;
4713 if (_Event == v) return ;
4714 continue ;
4715 }
4716 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4717 }
4718 if (v < 0) {
4719 // Wait for the thread associated with the event to vacate
4720 int status = pthread_mutex_lock(_mutex);
4721 assert_status(status == 0, status, "mutex_lock");
4722 AnyWaiters = _nParked ;
4723 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4724 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4725 AnyWaiters = 0 ;
4726 pthread_cond_signal (_cond);
4727 }
4728 status = pthread_mutex_unlock(_mutex);
4729 assert_status(status == 0, status, "mutex_unlock");
4730 if (AnyWaiters != 0) {
4731 status = pthread_cond_signal(_cond);
4732 assert_status(status == 0, status, "cond_signal");
4733 }
4734 }
4735
4736 // Note that we signal() _after dropping the lock for "immortal" Events.
4737 // This is safe and avoids a common class of futile wakeups. In rare
4738 // circumstances this can cause a thread to return prematurely from
4739 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4740 // simply re-test the condition and re-park itself.
4741 }
4742
4743
4744 // JSR166
4745 // -------------------------------------------------------
4746
4747 /*
4748 * The solaris and linux implementations of park/unpark are fairly
4749 * conservative for now, but can be improved. They currently use a
4750 * mutex/condvar pair, plus a a count.
4751 * Park decrements count if > 0, else does a condvar wait. Unpark
4752 * sets count to 1 and signals condvar. Only one thread ever waits
4753 * on the condvar. Contention seen when trying to park implies that someone
4754 * is unparking you, so don't wait. And spurious returns are fine, so there
4755 * is no need to track notifications.
4756 */
4757
4758
4759 #define NANOSECS_PER_SEC 1000000000
4760 #define NANOSECS_PER_MILLISEC 1000000
4761 #define MAX_SECS 100000000
4762 /*
4763 * This code is common to linux and solaris and will be moved to a
4764 * common place in dolphin.
4765 *
4766 * The passed in time value is either a relative time in nanoseconds
4767 * or an absolute time in milliseconds. Either way it has to be unpacked
4768 * into suitable seconds and nanoseconds components and stored in the
4769 * given timespec structure.
4770 * Given time is a 64-bit value and the time_t used in the timespec is only
4771 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4772 * overflow if times way in the future are given. Further on Solaris versions
4773 * prior to 10 there is a restriction (see cond_timedwait) that the specified
4774 * number of seconds, in abstime, is less than current_time + 100,000,000.
4775 * As it will be 28 years before "now + 100000000" will overflow we can
4776 * ignore overflow and just impose a hard-limit on seconds using the value
4777 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4778 * years from "now".
4779 */
4780
4781 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4782 assert (time > 0, "convertTime");
4783
4784 struct timeval now;
4785 int status = gettimeofday(&now, NULL);
4786 assert(status == 0, "gettimeofday");
4787
4788 time_t max_secs = now.tv_sec + MAX_SECS;
4789
4790 if (isAbsolute) {
4791 jlong secs = time / 1000;
4792 if (secs > max_secs) {
4793 absTime->tv_sec = max_secs;
4794 }
4795 else {
4796 absTime->tv_sec = secs;
4797 }
4798 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4799 }
4800 else {
4801 jlong secs = time / NANOSECS_PER_SEC;
4802 if (secs >= MAX_SECS) {
4803 absTime->tv_sec = max_secs;
4804 absTime->tv_nsec = 0;
4805 }
4806 else {
4807 absTime->tv_sec = now.tv_sec + secs;
4808 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4809 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4810 absTime->tv_nsec -= NANOSECS_PER_SEC;
4811 ++absTime->tv_sec; // note: this must be <= max_secs
4812 }
4813 }
4814 }
4815 assert(absTime->tv_sec >= 0, "tv_sec < 0");
4816 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4817 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4818 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4819 }
4820
4821 void Parker::park(bool isAbsolute, jlong time) {
4822 // Optional fast-path check:
4823 // Return immediately if a permit is available.
4824 if (_counter > 0) {
4825 _counter = 0 ;
4826 OrderAccess::fence();
4827 return ;
4828 }
4829
4830 Thread* thread = Thread::current();
4831 assert(thread->is_Java_thread(), "Must be JavaThread");
4832 JavaThread *jt = (JavaThread *)thread;
4833
4834 // Optional optimization -- avoid state transitions if there's an interrupt pending.
4835 // Check interrupt before trying to wait
4836 if (Thread::is_interrupted(thread, false)) {
4837 return;
4838 }
4839
4840 // Next, demultiplex/decode time arguments
4841 timespec absTime;
4842 if (time < 0) { // don't wait at all
4843 return;
4844 }
4845 if (time > 0) {
4846 unpackTime(&absTime, isAbsolute, time);
4847 }
4848
4849
4850 // Enter safepoint region
4851 // Beware of deadlocks such as 6317397.
4852 // The per-thread Parker:: mutex is a classic leaf-lock.
4853 // In particular a thread must never block on the Threads_lock while
4854 // holding the Parker:: mutex. If safepoints are pending both the
4855 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4856 ThreadBlockInVM tbivm(jt);
4857
4858 // Don't wait if cannot get lock since interference arises from
4859 // unblocking. Also. check interrupt before trying wait
4860 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4861 return;
4862 }
4863
4864 int status ;
4865 if (_counter > 0) { // no wait needed
4866 _counter = 0;
4867 status = pthread_mutex_unlock(_mutex);
4868 assert (status == 0, "invariant") ;
4869 OrderAccess::fence();
4870 return;
4871 }
4872
4873 #ifdef ASSERT
4874 // Don't catch signals while blocked; let the running threads have the signals.
4875 // (This allows a debugger to break into the running thread.)
4876 sigset_t oldsigs;
4877 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4878 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4879 #endif
4880
4881 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4882 jt->set_suspend_equivalent();
4883 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4884
4885 if (time == 0) {
4886 status = pthread_cond_wait (_cond, _mutex) ;
4887 } else {
4888 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4889 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4890 pthread_cond_destroy (_cond) ;
4891 pthread_cond_init (_cond, NULL);
4892 }
4893 }
4894 assert_status(status == 0 || status == EINTR ||
4895 status == ETIME || status == ETIMEDOUT,
4896 status, "cond_timedwait");
4897
4898 #ifdef ASSERT
4899 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4900 #endif
4901
4902 _counter = 0 ;
4903 status = pthread_mutex_unlock(_mutex) ;
4904 assert_status(status == 0, status, "invariant") ;
4905 // If externally suspended while waiting, re-suspend
4906 if (jt->handle_special_suspend_equivalent_condition()) {
4907 jt->java_suspend_self();
4908 }
4909
4910 OrderAccess::fence();
4911 }
4912
4913 void Parker::unpark() {
4914 int s, status ;
4915 status = pthread_mutex_lock(_mutex);
4916 assert (status == 0, "invariant") ;
4917 s = _counter;
4918 _counter = 1;
4919 if (s < 1) {
4920 if (WorkAroundNPTLTimedWaitHang) {
4921 status = pthread_cond_signal (_cond) ;
4922 assert (status == 0, "invariant") ;
4923 status = pthread_mutex_unlock(_mutex);
4924 assert (status == 0, "invariant") ;
4925 } else {
4926 status = pthread_mutex_unlock(_mutex);
4927 assert (status == 0, "invariant") ;
4928 status = pthread_cond_signal (_cond) ;
4929 assert (status == 0, "invariant") ;
4930 }
4931 } else {
4932 pthread_mutex_unlock(_mutex);
4933 assert (status == 0, "invariant") ;
4934 }
4935 }
4936
4937
4938 extern char** environ;
4939
4940 #ifndef __NR_fork
4941 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4942 #endif
4943
4944 #ifndef __NR_execve
4945 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4946 #endif
4947
4948 // Run the specified command in a separate process. Return its exit value,
4949 // or -1 on failure (e.g. can't fork a new process).
4950 // Unlike system(), this function can be called from signal handler. It
4951 // doesn't block SIGINT et al.
4952 int os::fork_and_exec(char* cmd) {
4953 const char * argv[4] = {"sh", "-c", cmd, NULL};
4954
4955 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4956 // pthread_atfork handlers and reset pthread library. All we need is a
4957 // separate process to execve. Make a direct syscall to fork process.
4958 // On IA64 there's no fork syscall, we have to use fork() and hope for
4959 // the best...
4960 pid_t pid = NOT_IA64(syscall(__NR_fork);)
4961 IA64_ONLY(fork();)
4962
4963 if (pid < 0) {
4964 // fork failed
4965 return -1;
4966
4967 } else if (pid == 0) {
4968 // child process
4969
4970 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4971 // first to kill every thread on the thread list. Because this list is
4972 // not reset by fork() (see notes above), execve() will instead kill
4973 // every thread in the parent process. We know this is the only thread
4974 // in the new process, so make a system call directly.
4975 // IA64 should use normal execve() from glibc to match the glibc fork()
4976 // above.
4977 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4978 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
4979
4980 // execve failed
4981 _exit(-1);
4982
4983 } else {
4984 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4985 // care about the actual exit code, for now.
4986
4987 int status;
4988
4989 // Wait for the child process to exit. This returns immediately if
4990 // the child has already exited. */
4991 while (waitpid(pid, &status, 0) < 0) {
4992 switch (errno) {
4993 case ECHILD: return 0;
4994 case EINTR: break;
4995 default: return -1;
4996 }
4997 }
4998
4999 if (WIFEXITED(status)) {
5000 // The child exited normally; get its exit code.
5001 return WEXITSTATUS(status);
5002 } else if (WIFSIGNALED(status)) {
5003 // The child exited because of a signal
5004 // The best value to return is 0x80 + signal number,
5005 // because that is what all Unix shells do, and because
5006 // it allows callers to distinguish between process exit and
5007 // process death by signal.
5008 return 0x80 + WTERMSIG(status);
5009 } else {
5010 // Unknown exit code; pass it through
5011 return status;
5012 }
5013 }
5014 }
5015
5016 // is_headless_jre()
5017 //
5018 // Test for the existence of libmawt in motif21 or xawt directories
5019 // in order to report if we are running in a headless jre
5020 //
5021 bool os::is_headless_jre() {
5022 struct stat statbuf;
5023 char buf[MAXPATHLEN];
5024 char libmawtpath[MAXPATHLEN];
5025 const char *xawtstr = "/xawt/libmawt.so";
5026 const char *motifstr = "/motif21/libmawt.so";
5027 char *p;
5028
5029 // Get path to libjvm.so
5030 os::jvm_path(buf, sizeof(buf));
5031
5032 // Get rid of libjvm.so
5033 p = strrchr(buf, '/');
5034 if (p == NULL) return false;
5035 else *p = '\0';
5036
5037 // Get rid of client or server
5038 p = strrchr(buf, '/');
5039 if (p == NULL) return false;
5040 else *p = '\0';
5041
5042 // check xawt/libmawt.so
5043 strcpy(libmawtpath, buf);
5044 strcat(libmawtpath, xawtstr);
5045 if (::stat(libmawtpath, &statbuf) == 0) return false;
5046
5047 // check motif21/libmawt.so
5048 strcpy(libmawtpath, buf);
5049 strcat(libmawtpath, motifstr);
5050 if (::stat(libmawtpath, &statbuf) == 0) return false;
5051
5052 return true;
5053 }
5054