1 /*
2 * Copyright (c) 1999, 2010, 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
831 // changed with the flag -Xss
832 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
833 stack_size = JavaThread::stack_size_at_create();
834 break;
835 case os::compiler_thread:
836 if (CompilerThreadStackSize > 0) {
837 stack_size = (size_t)(CompilerThreadStackSize * K);
838 break;
839 } // else fall through:
840 // use VMThreadStackSize if CompilerThreadStackSize is not defined
841 case os::vm_thread:
842 case os::pgc_thread:
843 case os::cgc_thread:
844 case os::watcher_thread:
845 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
846 break;
847 }
848 }
849
850 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
851 pthread_attr_setstacksize(&attr, stack_size);
852 } else {
853 // let pthread_create() pick the default value.
854 }
855
856 // glibc guard page
857 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
858
859 ThreadState state;
860
861 {
862 // Serialize thread creation if we are running with fixed stack LinuxThreads
863 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
864 if (lock) {
865 os::Linux::createThread_lock()->lock_without_safepoint_check();
866 }
867
868 pthread_t tid;
869 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
870
871 pthread_attr_destroy(&attr);
872
873 if (ret != 0) {
874 if (PrintMiscellaneous && (Verbose || WizardMode)) {
875 perror("pthread_create()");
876 }
877 // Need to clean up stuff we've allocated so far
878 thread->set_osthread(NULL);
879 delete osthread;
880 if (lock) os::Linux::createThread_lock()->unlock();
881 return false;
882 }
883
884 // Store pthread info into the OSThread
885 osthread->set_pthread_id(tid);
886
887 // Wait until child thread is either initialized or aborted
888 {
889 Monitor* sync_with_child = osthread->startThread_lock();
890 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
891 while ((state = osthread->get_state()) == ALLOCATED) {
892 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
893 }
894 }
895
896 if (lock) {
897 os::Linux::createThread_lock()->unlock();
898 }
899 }
900
901 // Aborted due to thread limit being reached
902 if (state == ZOMBIE) {
903 thread->set_osthread(NULL);
904 delete osthread;
905 return false;
906 }
907
908 // The thread is returned suspended (in state INITIALIZED),
909 // and is started higher up in the call chain
910 assert(state == INITIALIZED, "race condition");
911 return true;
912 }
913
914 /////////////////////////////////////////////////////////////////////////////
915 // attach existing thread
916
917 // bootstrap the main thread
918 bool os::create_main_thread(JavaThread* thread) {
919 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
920 return create_attached_thread(thread);
921 }
922
923 bool os::create_attached_thread(JavaThread* thread) {
924 #ifdef ASSERT
925 thread->verify_not_published();
926 #endif
927
928 // Allocate the OSThread object
929 OSThread* osthread = new OSThread(NULL, NULL);
930
931 if (osthread == NULL) {
932 return false;
933 }
934
935 // Store pthread info into the OSThread
936 osthread->set_thread_id(os::Linux::gettid());
937 osthread->set_pthread_id(::pthread_self());
938
939 // initialize floating point control register
940 os::Linux::init_thread_fpu_state();
941
942 // Initial thread state is RUNNABLE
943 osthread->set_state(RUNNABLE);
944
945 thread->set_osthread(osthread);
946
947 if (UseNUMA) {
948 int lgrp_id = os::numa_get_group_id();
949 if (lgrp_id != -1) {
950 thread->set_lgrp_id(lgrp_id);
951 }
952 }
953
954 if (os::Linux::is_initial_thread()) {
955 // If current thread is initial thread, its stack is mapped on demand,
956 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
957 // the entire stack region to avoid SEGV in stack banging.
958 // It is also useful to get around the heap-stack-gap problem on SuSE
959 // kernel (see 4821821 for details). We first expand stack to the top
960 // of yellow zone, then enable stack yellow zone (order is significant,
961 // enabling yellow zone first will crash JVM on SuSE Linux), so there
962 // is no gap between the last two virtual memory regions.
963
964 JavaThread *jt = (JavaThread *)thread;
965 address addr = jt->stack_yellow_zone_base();
966 assert(addr != NULL, "initialization problem?");
967 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
968
969 osthread->set_expanding_stack();
970 os::Linux::manually_expand_stack(jt, addr);
971 osthread->clear_expanding_stack();
972 }
973
974 // initialize signal mask for this thread
975 // and save the caller's signal mask
976 os::Linux::hotspot_sigmask(thread);
977
978 return true;
979 }
980
981 void os::pd_start_thread(Thread* thread) {
982 OSThread * osthread = thread->osthread();
983 assert(osthread->get_state() != INITIALIZED, "just checking");
984 Monitor* sync_with_child = osthread->startThread_lock();
985 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
986 sync_with_child->notify();
987 }
988
989 // Free Linux resources related to the OSThread
990 void os::free_thread(OSThread* osthread) {
991 assert(osthread != NULL, "osthread not set");
992
993 if (Thread::current()->osthread() == osthread) {
994 // Restore caller's signal mask
995 sigset_t sigmask = osthread->caller_sigmask();
996 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
997 }
998
999 delete osthread;
1000 }
1001
1002 //////////////////////////////////////////////////////////////////////////////
1003 // thread local storage
1004
1005 int os::allocate_thread_local_storage() {
1006 pthread_key_t key;
1007 int rslt = pthread_key_create(&key, NULL);
1008 assert(rslt == 0, "cannot allocate thread local storage");
1009 return (int)key;
1010 }
1011
1012 // Note: This is currently not used by VM, as we don't destroy TLS key
1013 // on VM exit.
1014 void os::free_thread_local_storage(int index) {
1015 int rslt = pthread_key_delete((pthread_key_t)index);
1016 assert(rslt == 0, "invalid index");
1017 }
1018
1019 void os::thread_local_storage_at_put(int index, void* value) {
1020 int rslt = pthread_setspecific((pthread_key_t)index, value);
1021 assert(rslt == 0, "pthread_setspecific failed");
1022 }
1023
1024 extern "C" Thread* get_thread() {
1025 return ThreadLocalStorage::thread();
1026 }
1027
1028 //////////////////////////////////////////////////////////////////////////////
1029 // initial thread
1030
1031 // Check if current thread is the initial thread, similar to Solaris thr_main.
1032 bool os::Linux::is_initial_thread(void) {
1033 char dummy;
1034 // If called before init complete, thread stack bottom will be null.
1035 // Can be called if fatal error occurs before initialization.
1036 if (initial_thread_stack_bottom() == NULL) return false;
1037 assert(initial_thread_stack_bottom() != NULL &&
1038 initial_thread_stack_size() != 0,
1039 "os::init did not locate initial thread's stack region");
1040 if ((address)&dummy >= initial_thread_stack_bottom() &&
1041 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1042 return true;
1043 else return false;
1044 }
1045
1046 // Find the virtual memory area that contains addr
1047 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1048 FILE *fp = fopen("/proc/self/maps", "r");
1049 if (fp) {
1050 address low, high;
1051 while (!feof(fp)) {
1052 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1053 if (low <= addr && addr < high) {
1054 if (vma_low) *vma_low = low;
1055 if (vma_high) *vma_high = high;
1056 fclose (fp);
1057 return true;
1058 }
1059 }
1060 for (;;) {
1061 int ch = fgetc(fp);
1062 if (ch == EOF || ch == (int)'\n') break;
1063 }
1064 }
1065 fclose(fp);
1066 }
1067 return false;
1068 }
1069
1070 // Locate initial thread stack. This special handling of initial thread stack
1071 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1072 // bogus value for initial thread.
1073 void os::Linux::capture_initial_stack(size_t max_size) {
1074 // stack size is the easy part, get it from RLIMIT_STACK
1075 size_t stack_size;
1076 struct rlimit rlim;
1077 getrlimit(RLIMIT_STACK, &rlim);
1078 stack_size = rlim.rlim_cur;
1079
1080 // 6308388: a bug in ld.so will relocate its own .data section to the
1081 // lower end of primordial stack; reduce ulimit -s value a little bit
1082 // so we won't install guard page on ld.so's data section.
1083 stack_size -= 2 * page_size();
1084
1085 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1086 // 7.1, in both cases we will get 2G in return value.
1087 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1088 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1089 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1090 // in case other parts in glibc still assumes 2M max stack size.
1091 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1092 #ifndef IA64
1093 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1094 #else
1095 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1096 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1097 #endif
1098
1099 // Try to figure out where the stack base (top) is. This is harder.
1100 //
1101 // When an application is started, glibc saves the initial stack pointer in
1102 // a global variable "__libc_stack_end", which is then used by system
1103 // libraries. __libc_stack_end should be pretty close to stack top. The
1104 // variable is available since the very early days. However, because it is
1105 // a private interface, it could disappear in the future.
1106 //
1107 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1108 // to __libc_stack_end, it is very close to stack top, but isn't the real
1109 // stack top. Note that /proc may not exist if VM is running as a chroot
1110 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1111 // /proc/<pid>/stat could change in the future (though unlikely).
1112 //
1113 // We try __libc_stack_end first. If that doesn't work, look for
1114 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1115 // as a hint, which should work well in most cases.
1116
1117 uintptr_t stack_start;
1118
1119 // try __libc_stack_end first
1120 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1121 if (p && *p) {
1122 stack_start = *p;
1123 } else {
1124 // see if we can get the start_stack field from /proc/self/stat
1125 FILE *fp;
1126 int pid;
1127 char state;
1128 int ppid;
1129 int pgrp;
1130 int session;
1131 int nr;
1132 int tpgrp;
1133 unsigned long flags;
1134 unsigned long minflt;
1135 unsigned long cminflt;
1136 unsigned long majflt;
1137 unsigned long cmajflt;
1138 unsigned long utime;
1139 unsigned long stime;
1140 long cutime;
1141 long cstime;
1142 long prio;
1143 long nice;
1144 long junk;
1145 long it_real;
1146 uintptr_t start;
1147 uintptr_t vsize;
1148 intptr_t rss;
1149 uintptr_t rsslim;
1150 uintptr_t scodes;
1151 uintptr_t ecode;
1152 int i;
1153
1154 // Figure what the primordial thread stack base is. Code is inspired
1155 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1156 // followed by command name surrounded by parentheses, state, etc.
1157 char stat[2048];
1158 int statlen;
1159
1160 fp = fopen("/proc/self/stat", "r");
1161 if (fp) {
1162 statlen = fread(stat, 1, 2047, fp);
1163 stat[statlen] = '\0';
1164 fclose(fp);
1165
1166 // Skip pid and the command string. Note that we could be dealing with
1167 // weird command names, e.g. user could decide to rename java launcher
1168 // to "java 1.4.2 :)", then the stat file would look like
1169 // 1234 (java 1.4.2 :)) R ... ...
1170 // We don't really need to know the command string, just find the last
1171 // occurrence of ")" and then start parsing from there. See bug 4726580.
1172 char * s = strrchr(stat, ')');
1173
1174 i = 0;
1175 if (s) {
1176 // Skip blank chars
1177 do s++; while (isspace(*s));
1178
1179 #define _UFM UINTX_FORMAT
1180 #define _DFM INTX_FORMAT
1181
1182 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1183 /* 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 */
1184 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,
1185 &state, /* 3 %c */
1186 &ppid, /* 4 %d */
1187 &pgrp, /* 5 %d */
1188 &session, /* 6 %d */
1189 &nr, /* 7 %d */
1190 &tpgrp, /* 8 %d */
1191 &flags, /* 9 %lu */
1192 &minflt, /* 10 %lu */
1193 &cminflt, /* 11 %lu */
1194 &majflt, /* 12 %lu */
1195 &cmajflt, /* 13 %lu */
1196 &utime, /* 14 %lu */
1197 &stime, /* 15 %lu */
1198 &cutime, /* 16 %ld */
1199 &cstime, /* 17 %ld */
1200 &prio, /* 18 %ld */
1201 &nice, /* 19 %ld */
1202 &junk, /* 20 %ld */
1203 &it_real, /* 21 %ld */
1204 &start, /* 22 UINTX_FORMAT */
1205 &vsize, /* 23 UINTX_FORMAT */
1206 &rss, /* 24 INTX_FORMAT */
1207 &rsslim, /* 25 UINTX_FORMAT */
1208 &scodes, /* 26 UINTX_FORMAT */
1209 &ecode, /* 27 UINTX_FORMAT */
1210 &stack_start); /* 28 UINTX_FORMAT */
1211 }
1212
1213 #undef _UFM
1214 #undef _DFM
1215
1216 if (i != 28 - 2) {
1217 assert(false, "Bad conversion from /proc/self/stat");
1218 // product mode - assume we are the initial thread, good luck in the
1219 // embedded case.
1220 warning("Can't detect initial thread stack location - bad conversion");
1221 stack_start = (uintptr_t) &rlim;
1222 }
1223 } else {
1224 // For some reason we can't open /proc/self/stat (for example, running on
1225 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1226 // most cases, so don't abort:
1227 warning("Can't detect initial thread stack location - no /proc/self/stat");
1228 stack_start = (uintptr_t) &rlim;
1229 }
1230 }
1231
1232 // Now we have a pointer (stack_start) very close to the stack top, the
1233 // next thing to do is to figure out the exact location of stack top. We
1234 // can find out the virtual memory area that contains stack_start by
1235 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1236 // and its upper limit is the real stack top. (again, this would fail if
1237 // running inside chroot, because /proc may not exist.)
1238
1239 uintptr_t stack_top;
1240 address low, high;
1241 if (find_vma((address)stack_start, &low, &high)) {
1242 // success, "high" is the true stack top. (ignore "low", because initial
1243 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1244 stack_top = (uintptr_t)high;
1245 } else {
1246 // failed, likely because /proc/self/maps does not exist
1247 warning("Can't detect initial thread stack location - find_vma failed");
1248 // best effort: stack_start is normally within a few pages below the real
1249 // stack top, use it as stack top, and reduce stack size so we won't put
1250 // guard page outside stack.
1251 stack_top = stack_start;
1252 stack_size -= 16 * page_size();
1253 }
1254
1255 // stack_top could be partially down the page so align it
1256 stack_top = align_size_up(stack_top, page_size());
1257
1258 if (max_size && stack_size > max_size) {
1259 _initial_thread_stack_size = max_size;
1260 } else {
1261 _initial_thread_stack_size = stack_size;
1262 }
1263
1264 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1265 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1266 }
1267
1268 ////////////////////////////////////////////////////////////////////////////////
1269 // time support
1270
1271 // Time since start-up in seconds to a fine granularity.
1272 // Used by VMSelfDestructTimer and the MemProfiler.
1273 double os::elapsedTime() {
1274
1275 return (double)(os::elapsed_counter()) * 0.000001;
1276 }
1277
1278 jlong os::elapsed_counter() {
1279 timeval time;
1280 int status = gettimeofday(&time, NULL);
1281 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1282 }
1283
1284 jlong os::elapsed_frequency() {
1285 return (1000 * 1000);
1286 }
1287
1288 // For now, we say that linux does not support vtime. I have no idea
1289 // whether it can actually be made to (DLD, 9/13/05).
1290
1291 bool os::supports_vtime() { return false; }
1292 bool os::enable_vtime() { return false; }
1293 bool os::vtime_enabled() { return false; }
1294 double os::elapsedVTime() {
1295 // better than nothing, but not much
1296 return elapsedTime();
1297 }
1298
1299 jlong os::javaTimeMillis() {
1300 timeval time;
1301 int status = gettimeofday(&time, NULL);
1302 assert(status != -1, "linux error");
1303 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1304 }
1305
1306 #ifndef CLOCK_MONOTONIC
1307 #define CLOCK_MONOTONIC (1)
1308 #endif
1309
1310 void os::Linux::clock_init() {
1311 // we do dlopen's in this particular order due to bug in linux
1312 // dynamical loader (see 6348968) leading to crash on exit
1313 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1314 if (handle == NULL) {
1315 handle = dlopen("librt.so", RTLD_LAZY);
1316 }
1317
1318 if (handle) {
1319 int (*clock_getres_func)(clockid_t, struct timespec*) =
1320 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1321 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1322 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1323 if (clock_getres_func && clock_gettime_func) {
1324 // See if monotonic clock is supported by the kernel. Note that some
1325 // early implementations simply return kernel jiffies (updated every
1326 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1327 // for nano time (though the monotonic property is still nice to have).
1328 // It's fixed in newer kernels, however clock_getres() still returns
1329 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1330 // resolution for now. Hopefully as people move to new kernels, this
1331 // won't be a problem.
1332 struct timespec res;
1333 struct timespec tp;
1334 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1335 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1336 // yes, monotonic clock is supported
1337 _clock_gettime = clock_gettime_func;
1338 } else {
1339 // close librt if there is no monotonic clock
1340 dlclose(handle);
1341 }
1342 }
1343 }
1344 }
1345
1346 #ifndef SYS_clock_getres
1347
1348 #if defined(IA32) || defined(AMD64)
1349 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1350 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1351 #else
1352 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1353 #define sys_clock_getres(x,y) -1
1354 #endif
1355
1356 #else
1357 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1358 #endif
1359
1360 void os::Linux::fast_thread_clock_init() {
1361 if (!UseLinuxPosixThreadCPUClocks) {
1362 return;
1363 }
1364 clockid_t clockid;
1365 struct timespec tp;
1366 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1367 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1368
1369 // Switch to using fast clocks for thread cpu time if
1370 // the sys_clock_getres() returns 0 error code.
1371 // Note, that some kernels may support the current thread
1372 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1373 // returned by the pthread_getcpuclockid().
1374 // If the fast Posix clocks are supported then the sys_clock_getres()
1375 // must return at least tp.tv_sec == 0 which means a resolution
1376 // better than 1 sec. This is extra check for reliability.
1377
1378 if(pthread_getcpuclockid_func &&
1379 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1380 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1381
1382 _supports_fast_thread_cpu_time = true;
1383 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1384 }
1385 }
1386
1387 jlong os::javaTimeNanos() {
1388 if (Linux::supports_monotonic_clock()) {
1389 struct timespec tp;
1390 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1391 assert(status == 0, "gettime error");
1392 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1393 return result;
1394 } else {
1395 timeval time;
1396 int status = gettimeofday(&time, NULL);
1397 assert(status != -1, "linux error");
1398 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1399 return 1000 * usecs;
1400 }
1401 }
1402
1403 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1404 if (Linux::supports_monotonic_clock()) {
1405 info_ptr->max_value = ALL_64_BITS;
1406
1407 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1408 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1409 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1410 } else {
1411 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1412 info_ptr->max_value = ALL_64_BITS;
1413
1414 // gettimeofday is a real time clock so it skips
1415 info_ptr->may_skip_backward = true;
1416 info_ptr->may_skip_forward = true;
1417 }
1418
1419 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1420 }
1421
1422 // Return the real, user, and system times in seconds from an
1423 // arbitrary fixed point in the past.
1424 bool os::getTimesSecs(double* process_real_time,
1425 double* process_user_time,
1426 double* process_system_time) {
1427 struct tms ticks;
1428 clock_t real_ticks = times(&ticks);
1429
1430 if (real_ticks == (clock_t) (-1)) {
1431 return false;
1432 } else {
1433 double ticks_per_second = (double) clock_tics_per_sec;
1434 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1435 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1436 *process_real_time = ((double) real_ticks) / ticks_per_second;
1437
1438 return true;
1439 }
1440 }
1441
1442
1443 char * os::local_time_string(char *buf, size_t buflen) {
1444 struct tm t;
1445 time_t long_time;
1446 time(&long_time);
1447 localtime_r(&long_time, &t);
1448 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1449 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1450 t.tm_hour, t.tm_min, t.tm_sec);
1451 return buf;
1452 }
1453
1454 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1455 return localtime_r(clock, res);
1456 }
1457
1458 ////////////////////////////////////////////////////////////////////////////////
1459 // runtime exit support
1460
1461 // Note: os::shutdown() might be called very early during initialization, or
1462 // called from signal handler. Before adding something to os::shutdown(), make
1463 // sure it is async-safe and can handle partially initialized VM.
1464 void os::shutdown() {
1465
1466 // allow PerfMemory to attempt cleanup of any persistent resources
1467 perfMemory_exit();
1468
1469 // needs to remove object in file system
1470 AttachListener::abort();
1471
1472 // flush buffered output, finish log files
1473 ostream_abort();
1474
1475 // Check for abort hook
1476 abort_hook_t abort_hook = Arguments::abort_hook();
1477 if (abort_hook != NULL) {
1478 abort_hook();
1479 }
1480
1481 }
1482
1483 // Note: os::abort() might be called very early during initialization, or
1484 // called from signal handler. Before adding something to os::abort(), make
1485 // sure it is async-safe and can handle partially initialized VM.
1486 void os::abort(bool dump_core) {
1487 os::shutdown();
1488 if (dump_core) {
1489 #ifndef PRODUCT
1490 fdStream out(defaultStream::output_fd());
1491 out.print_raw("Current thread is ");
1492 char buf[16];
1493 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1494 out.print_raw_cr(buf);
1495 out.print_raw_cr("Dumping core ...");
1496 #endif
1497 ::abort(); // dump core
1498 }
1499
1500 ::exit(1);
1501 }
1502
1503 // Die immediately, no exit hook, no abort hook, no cleanup.
1504 void os::die() {
1505 // _exit() on LinuxThreads only kills current thread
1506 ::abort();
1507 }
1508
1509 // unused on linux for now.
1510 void os::set_error_file(const char *logfile) {}
1511
1512 intx os::current_thread_id() { return (intx)pthread_self(); }
1513 int os::current_process_id() {
1514
1515 // Under the old linux thread library, linux gives each thread
1516 // its own process id. Because of this each thread will return
1517 // a different pid if this method were to return the result
1518 // of getpid(2). Linux provides no api that returns the pid
1519 // of the launcher thread for the vm. This implementation
1520 // returns a unique pid, the pid of the launcher thread
1521 // that starts the vm 'process'.
1522
1523 // Under the NPTL, getpid() returns the same pid as the
1524 // launcher thread rather than a unique pid per thread.
1525 // Use gettid() if you want the old pre NPTL behaviour.
1526
1527 // if you are looking for the result of a call to getpid() that
1528 // returns a unique pid for the calling thread, then look at the
1529 // OSThread::thread_id() method in osThread_linux.hpp file
1530
1531 return (int)(_initial_pid ? _initial_pid : getpid());
1532 }
1533
1534 // DLL functions
1535
1536 const char* os::dll_file_extension() { return ".so"; }
1537
1538 const char* os::get_temp_directory() {
1539 const char *prop = Arguments::get_property("java.io.tmpdir");
1540 return prop == NULL ? "/tmp" : prop;
1541 }
1542
1543 static bool file_exists(const char* filename) {
1544 struct stat statbuf;
1545 if (filename == NULL || strlen(filename) == 0) {
1546 return false;
1547 }
1548 return os::stat(filename, &statbuf) == 0;
1549 }
1550
1551 void os::dll_build_name(char* buffer, size_t buflen,
1552 const char* pname, const char* fname) {
1553 // Copied from libhpi
1554 const size_t pnamelen = pname ? strlen(pname) : 0;
1555
1556 // Quietly truncate on buffer overflow. Should be an error.
1557 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1558 *buffer = '\0';
1559 return;
1560 }
1561
1562 if (pnamelen == 0) {
1563 snprintf(buffer, buflen, "lib%s.so", fname);
1564 } else if (strchr(pname, *os::path_separator()) != NULL) {
1565 int n;
1566 char** pelements = split_path(pname, &n);
1567 for (int i = 0 ; i < n ; i++) {
1568 // Really shouldn't be NULL, but check can't hurt
1569 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1570 continue; // skip the empty path values
1571 }
1572 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1573 if (file_exists(buffer)) {
1574 break;
1575 }
1576 }
1577 // release the storage
1578 for (int i = 0 ; i < n ; i++) {
1579 if (pelements[i] != NULL) {
1580 FREE_C_HEAP_ARRAY(char, pelements[i]);
1581 }
1582 }
1583 if (pelements != NULL) {
1584 FREE_C_HEAP_ARRAY(char*, pelements);
1585 }
1586 } else {
1587 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1588 }
1589 }
1590
1591 const char* os::get_current_directory(char *buf, int buflen) {
1592 return getcwd(buf, buflen);
1593 }
1594
1595 // check if addr is inside libjvm[_g].so
1596 bool os::address_is_in_vm(address addr) {
1597 static address libjvm_base_addr;
1598 Dl_info dlinfo;
1599
1600 if (libjvm_base_addr == NULL) {
1601 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1602 libjvm_base_addr = (address)dlinfo.dli_fbase;
1603 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1604 }
1605
1606 if (dladdr((void *)addr, &dlinfo)) {
1607 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1608 }
1609
1610 return false;
1611 }
1612
1613 bool os::dll_address_to_function_name(address addr, char *buf,
1614 int buflen, int *offset) {
1615 Dl_info dlinfo;
1616
1617 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1618 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1619 if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1620 return true;
1621 } else {
1622 if (buf) buf[0] = '\0';
1623 if (offset) *offset = -1;
1624 return false;
1625 }
1626 }
1627
1628 struct _address_to_library_name {
1629 address addr; // input : memory address
1630 size_t buflen; // size of fname
1631 char* fname; // output: library name
1632 address base; // library base addr
1633 };
1634
1635 static int address_to_library_name_callback(struct dl_phdr_info *info,
1636 size_t size, void *data) {
1637 int i;
1638 bool found = false;
1639 address libbase = NULL;
1640 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1641
1642 // iterate through all loadable segments
1643 for (i = 0; i < info->dlpi_phnum; i++) {
1644 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1645 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1646 // base address of a library is the lowest address of its loaded
1647 // segments.
1648 if (libbase == NULL || libbase > segbase) {
1649 libbase = segbase;
1650 }
1651 // see if 'addr' is within current segment
1652 if (segbase <= d->addr &&
1653 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1654 found = true;
1655 }
1656 }
1657 }
1658
1659 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1660 // so dll_address_to_library_name() can fall through to use dladdr() which
1661 // can figure out executable name from argv[0].
1662 if (found && info->dlpi_name && info->dlpi_name[0]) {
1663 d->base = libbase;
1664 if (d->fname) {
1665 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1666 }
1667 return 1;
1668 }
1669 return 0;
1670 }
1671
1672 bool os::dll_address_to_library_name(address addr, char* buf,
1673 int buflen, int* offset) {
1674 Dl_info dlinfo;
1675 struct _address_to_library_name data;
1676
1677 // There is a bug in old glibc dladdr() implementation that it could resolve
1678 // to wrong library name if the .so file has a base address != NULL. Here
1679 // we iterate through the program headers of all loaded libraries to find
1680 // out which library 'addr' really belongs to. This workaround can be
1681 // removed once the minimum requirement for glibc is moved to 2.3.x.
1682 data.addr = addr;
1683 data.fname = buf;
1684 data.buflen = buflen;
1685 data.base = NULL;
1686 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1687
1688 if (rslt) {
1689 // buf already contains library name
1690 if (offset) *offset = addr - data.base;
1691 return true;
1692 } else if (dladdr((void*)addr, &dlinfo)){
1693 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1694 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1695 return true;
1696 } else {
1697 if (buf) buf[0] = '\0';
1698 if (offset) *offset = -1;
1699 return false;
1700 }
1701 }
1702
1703 // Loads .dll/.so and
1704 // in case of error it checks if .dll/.so was built for the
1705 // same architecture as Hotspot is running on
1706
1707 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1708 {
1709 void * result= ::dlopen(filename, RTLD_LAZY);
1710 if (result != NULL) {
1711 // Successful loading
1712 return result;
1713 }
1714
1715 Elf32_Ehdr elf_head;
1716
1717 // Read system error message into ebuf
1718 // It may or may not be overwritten below
1719 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1720 ebuf[ebuflen-1]='\0';
1721 int diag_msg_max_length=ebuflen-strlen(ebuf);
1722 char* diag_msg_buf=ebuf+strlen(ebuf);
1723
1724 if (diag_msg_max_length==0) {
1725 // No more space in ebuf for additional diagnostics message
1726 return NULL;
1727 }
1728
1729
1730 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1731
1732 if (file_descriptor < 0) {
1733 // Can't open library, report dlerror() message
1734 return NULL;
1735 }
1736
1737 bool failed_to_read_elf_head=
1738 (sizeof(elf_head)!=
1739 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1740
1741 ::close(file_descriptor);
1742 if (failed_to_read_elf_head) {
1743 // file i/o error - report dlerror() msg
1744 return NULL;
1745 }
1746
1747 typedef struct {
1748 Elf32_Half code; // Actual value as defined in elf.h
1749 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1750 char elf_class; // 32 or 64 bit
1751 char endianess; // MSB or LSB
1752 char* name; // String representation
1753 } arch_t;
1754
1755 #ifndef EM_486
1756 #define EM_486 6 /* Intel 80486 */
1757 #endif
1758
1759 static const arch_t arch_array[]={
1760 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1761 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1762 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1763 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1764 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1765 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1766 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1767 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1768 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1769 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1770 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1771 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1772 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1773 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1774 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1775 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1776 };
1777
1778 #if (defined IA32)
1779 static Elf32_Half running_arch_code=EM_386;
1780 #elif (defined AMD64)
1781 static Elf32_Half running_arch_code=EM_X86_64;
1782 #elif (defined IA64)
1783 static Elf32_Half running_arch_code=EM_IA_64;
1784 #elif (defined __sparc) && (defined _LP64)
1785 static Elf32_Half running_arch_code=EM_SPARCV9;
1786 #elif (defined __sparc) && (!defined _LP64)
1787 static Elf32_Half running_arch_code=EM_SPARC;
1788 #elif (defined __powerpc64__)
1789 static Elf32_Half running_arch_code=EM_PPC64;
1790 #elif (defined __powerpc__)
1791 static Elf32_Half running_arch_code=EM_PPC;
1792 #elif (defined ARM)
1793 static Elf32_Half running_arch_code=EM_ARM;
1794 #elif (defined S390)
1795 static Elf32_Half running_arch_code=EM_S390;
1796 #elif (defined ALPHA)
1797 static Elf32_Half running_arch_code=EM_ALPHA;
1798 #elif (defined MIPSEL)
1799 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1800 #elif (defined PARISC)
1801 static Elf32_Half running_arch_code=EM_PARISC;
1802 #elif (defined MIPS)
1803 static Elf32_Half running_arch_code=EM_MIPS;
1804 #elif (defined M68K)
1805 static Elf32_Half running_arch_code=EM_68K;
1806 #else
1807 #error Method os::dll_load requires that one of following is defined:\
1808 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1809 #endif
1810
1811 // Identify compatability class for VM's architecture and library's architecture
1812 // Obtain string descriptions for architectures
1813
1814 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1815 int running_arch_index=-1;
1816
1817 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1818 if (running_arch_code == arch_array[i].code) {
1819 running_arch_index = i;
1820 }
1821 if (lib_arch.code == arch_array[i].code) {
1822 lib_arch.compat_class = arch_array[i].compat_class;
1823 lib_arch.name = arch_array[i].name;
1824 }
1825 }
1826
1827 assert(running_arch_index != -1,
1828 "Didn't find running architecture code (running_arch_code) in arch_array");
1829 if (running_arch_index == -1) {
1830 // Even though running architecture detection failed
1831 // we may still continue with reporting dlerror() message
1832 return NULL;
1833 }
1834
1835 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1836 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1837 return NULL;
1838 }
1839
1840 #ifndef S390
1841 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1842 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1843 return NULL;
1844 }
1845 #endif // !S390
1846
1847 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1848 if ( lib_arch.name!=NULL ) {
1849 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1850 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1851 lib_arch.name, arch_array[running_arch_index].name);
1852 } else {
1853 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1854 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1855 lib_arch.code,
1856 arch_array[running_arch_index].name);
1857 }
1858 }
1859
1860 return NULL;
1861 }
1862
1863 /*
1864 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1865 * chances are you might want to run the generated bits against glibc-2.0
1866 * libdl.so, so always use locking for any version of glibc.
1867 */
1868 void* os::dll_lookup(void* handle, const char* name) {
1869 pthread_mutex_lock(&dl_mutex);
1870 void* res = dlsym(handle, name);
1871 pthread_mutex_unlock(&dl_mutex);
1872 return res;
1873 }
1874
1875
1876 bool _print_ascii_file(const char* filename, outputStream* st) {
1877 int fd = open(filename, O_RDONLY);
1878 if (fd == -1) {
1879 return false;
1880 }
1881
1882 char buf[32];
1883 int bytes;
1884 while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1885 st->print_raw(buf, bytes);
1886 }
1887
1888 close(fd);
1889
1890 return true;
1891 }
1892
1893 void os::print_dll_info(outputStream *st) {
1894 st->print_cr("Dynamic libraries:");
1895
1896 char fname[32];
1897 pid_t pid = os::Linux::gettid();
1898
1899 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1900
1901 if (!_print_ascii_file(fname, st)) {
1902 st->print("Can not get library information for pid = %d\n", pid);
1903 }
1904 }
1905
1906
1907 void os::print_os_info(outputStream* st) {
1908 st->print("OS:");
1909
1910 // Try to identify popular distros.
1911 // Most Linux distributions have /etc/XXX-release file, which contains
1912 // the OS version string. Some have more than one /etc/XXX-release file
1913 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1914 // so the order is important.
1915 if (!_print_ascii_file("/etc/mandrake-release", st) &&
1916 !_print_ascii_file("/etc/sun-release", st) &&
1917 !_print_ascii_file("/etc/redhat-release", st) &&
1918 !_print_ascii_file("/etc/SuSE-release", st) &&
1919 !_print_ascii_file("/etc/turbolinux-release", st) &&
1920 !_print_ascii_file("/etc/gentoo-release", st) &&
1921 !_print_ascii_file("/etc/debian_version", st) &&
1922 !_print_ascii_file("/etc/ltib-release", st) &&
1923 !_print_ascii_file("/etc/angstrom-version", st)) {
1924 st->print("Linux");
1925 }
1926 st->cr();
1927
1928 // kernel
1929 st->print("uname:");
1930 struct utsname name;
1931 uname(&name);
1932 st->print(name.sysname); st->print(" ");
1933 st->print(name.release); st->print(" ");
1934 st->print(name.version); st->print(" ");
1935 st->print(name.machine);
1936 st->cr();
1937
1938 // Print warning if unsafe chroot environment detected
1939 if (unsafe_chroot_detected) {
1940 st->print("WARNING!! ");
1941 st->print_cr(unstable_chroot_error);
1942 }
1943
1944 // libc, pthread
1945 st->print("libc:");
1946 st->print(os::Linux::glibc_version()); st->print(" ");
1947 st->print(os::Linux::libpthread_version()); st->print(" ");
1948 if (os::Linux::is_LinuxThreads()) {
1949 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1950 }
1951 st->cr();
1952
1953 // rlimit
1954 st->print("rlimit:");
1955 struct rlimit rlim;
1956
1957 st->print(" STACK ");
1958 getrlimit(RLIMIT_STACK, &rlim);
1959 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1960 else st->print("%uk", rlim.rlim_cur >> 10);
1961
1962 st->print(", CORE ");
1963 getrlimit(RLIMIT_CORE, &rlim);
1964 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1965 else st->print("%uk", rlim.rlim_cur >> 10);
1966
1967 st->print(", NPROC ");
1968 getrlimit(RLIMIT_NPROC, &rlim);
1969 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1970 else st->print("%d", rlim.rlim_cur);
1971
1972 st->print(", NOFILE ");
1973 getrlimit(RLIMIT_NOFILE, &rlim);
1974 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1975 else st->print("%d", rlim.rlim_cur);
1976
1977 st->print(", AS ");
1978 getrlimit(RLIMIT_AS, &rlim);
1979 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1980 else st->print("%uk", rlim.rlim_cur >> 10);
1981 st->cr();
1982
1983 // load average
1984 st->print("load average:");
1985 double loadavg[3];
1986 os::loadavg(loadavg, 3);
1987 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1988 st->cr();
1989
1990 // meminfo
1991 st->print("\n/proc/meminfo:\n");
1992 _print_ascii_file("/proc/meminfo", st);
1993 st->cr();
1994 }
1995
1996 void os::print_memory_info(outputStream* st) {
1997
1998 st->print("Memory:");
1999 st->print(" %dk page", os::vm_page_size()>>10);
2000
2001 // values in struct sysinfo are "unsigned long"
2002 struct sysinfo si;
2003 sysinfo(&si);
2004
2005 st->print(", physical " UINT64_FORMAT "k",
2006 os::physical_memory() >> 10);
2007 st->print("(" UINT64_FORMAT "k free)",
2008 os::available_memory() >> 10);
2009 st->print(", swap " UINT64_FORMAT "k",
2010 ((jlong)si.totalswap * si.mem_unit) >> 10);
2011 st->print("(" UINT64_FORMAT "k free)",
2012 ((jlong)si.freeswap * si.mem_unit) >> 10);
2013 st->cr();
2014 }
2015
2016 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2017 // but they're the same for all the linux arch that we support
2018 // and they're the same for solaris but there's no common place to put this.
2019 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2020 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2021 "ILL_COPROC", "ILL_BADSTK" };
2022
2023 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2024 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2025 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2026
2027 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2028
2029 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2030
2031 void os::print_siginfo(outputStream* st, void* siginfo) {
2032 st->print("siginfo:");
2033
2034 const int buflen = 100;
2035 char buf[buflen];
2036 siginfo_t *si = (siginfo_t*)siginfo;
2037 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2038 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2039 st->print("si_errno=%s", buf);
2040 } else {
2041 st->print("si_errno=%d", si->si_errno);
2042 }
2043 const int c = si->si_code;
2044 assert(c > 0, "unexpected si_code");
2045 switch (si->si_signo) {
2046 case SIGILL:
2047 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2048 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2049 break;
2050 case SIGFPE:
2051 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2052 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2053 break;
2054 case SIGSEGV:
2055 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2056 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2057 break;
2058 case SIGBUS:
2059 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2060 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2061 break;
2062 default:
2063 st->print(", si_code=%d", si->si_code);
2064 // no si_addr
2065 }
2066
2067 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2068 UseSharedSpaces) {
2069 FileMapInfo* mapinfo = FileMapInfo::current_info();
2070 if (mapinfo->is_in_shared_space(si->si_addr)) {
2071 st->print("\n\nError accessing class data sharing archive." \
2072 " Mapped file inaccessible during execution, " \
2073 " possible disk/network problem.");
2074 }
2075 }
2076 st->cr();
2077 }
2078
2079
2080 static void print_signal_handler(outputStream* st, int sig,
2081 char* buf, size_t buflen);
2082
2083 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2084 st->print_cr("Signal Handlers:");
2085 print_signal_handler(st, SIGSEGV, buf, buflen);
2086 print_signal_handler(st, SIGBUS , buf, buflen);
2087 print_signal_handler(st, SIGFPE , buf, buflen);
2088 print_signal_handler(st, SIGPIPE, buf, buflen);
2089 print_signal_handler(st, SIGXFSZ, buf, buflen);
2090 print_signal_handler(st, SIGILL , buf, buflen);
2091 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2092 print_signal_handler(st, SR_signum, buf, buflen);
2093 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2094 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2095 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2096 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2097 }
2098
2099 static char saved_jvm_path[MAXPATHLEN] = {0};
2100
2101 // Find the full path to the current module, libjvm.so or libjvm_g.so
2102 void os::jvm_path(char *buf, jint buflen) {
2103 // Error checking.
2104 if (buflen < MAXPATHLEN) {
2105 assert(false, "must use a large-enough buffer");
2106 buf[0] = '\0';
2107 return;
2108 }
2109 // Lazy resolve the path to current module.
2110 if (saved_jvm_path[0] != 0) {
2111 strcpy(buf, saved_jvm_path);
2112 return;
2113 }
2114
2115 char dli_fname[MAXPATHLEN];
2116 bool ret = dll_address_to_library_name(
2117 CAST_FROM_FN_PTR(address, os::jvm_path),
2118 dli_fname, sizeof(dli_fname), NULL);
2119 assert(ret != 0, "cannot locate libjvm");
2120 char *rp = realpath(dli_fname, buf);
2121 if (rp == NULL)
2122 return;
2123
2124 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2125 // Support for the gamma launcher. Typical value for buf is
2126 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2127 // the right place in the string, then assume we are installed in a JDK and
2128 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2129 // up the path so it looks like libjvm.so is installed there (append a
2130 // fake suffix hotspot/libjvm.so).
2131 const char *p = buf + strlen(buf) - 1;
2132 for (int count = 0; p > buf && count < 5; ++count) {
2133 for (--p; p > buf && *p != '/'; --p)
2134 /* empty */ ;
2135 }
2136
2137 if (strncmp(p, "/jre/lib/", 9) != 0) {
2138 // Look for JAVA_HOME in the environment.
2139 char* java_home_var = ::getenv("JAVA_HOME");
2140 if (java_home_var != NULL && java_home_var[0] != 0) {
2141 char* jrelib_p;
2142 int len;
2143
2144 // Check the current module name "libjvm.so" or "libjvm_g.so".
2145 p = strrchr(buf, '/');
2146 assert(strstr(p, "/libjvm") == p, "invalid library name");
2147 p = strstr(p, "_g") ? "_g" : "";
2148
2149 rp = realpath(java_home_var, buf);
2150 if (rp == NULL)
2151 return;
2152
2153 // determine if this is a legacy image or modules image
2154 // modules image doesn't have "jre" subdirectory
2155 len = strlen(buf);
2156 jrelib_p = buf + len;
2157 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2158 if (0 != access(buf, F_OK)) {
2159 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2160 }
2161
2162 if (0 == access(buf, F_OK)) {
2163 // Use current module name "libjvm[_g].so" instead of
2164 // "libjvm"debug_only("_g")".so" since for fastdebug version
2165 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2166 // It is used when we are choosing the HPI library's name
2167 // "libhpi[_g].so" in hpi::initialize_get_interface().
2168 len = strlen(buf);
2169 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2170 } else {
2171 // Go back to path of .so
2172 rp = realpath(dli_fname, buf);
2173 if (rp == NULL)
2174 return;
2175 }
2176 }
2177 }
2178 }
2179
2180 strcpy(saved_jvm_path, buf);
2181 }
2182
2183 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2184 // no prefix required, not even "_"
2185 }
2186
2187 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2188 // no suffix required
2189 }
2190
2191 ////////////////////////////////////////////////////////////////////////////////
2192 // sun.misc.Signal support
2193
2194 static volatile jint sigint_count = 0;
2195
2196 static void
2197 UserHandler(int sig, void *siginfo, void *context) {
2198 // 4511530 - sem_post is serialized and handled by the manager thread. When
2199 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2200 // don't want to flood the manager thread with sem_post requests.
2201 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2202 return;
2203
2204 // Ctrl-C is pressed during error reporting, likely because the error
2205 // handler fails to abort. Let VM die immediately.
2206 if (sig == SIGINT && is_error_reported()) {
2207 os::die();
2208 }
2209
2210 os::signal_notify(sig);
2211 }
2212
2213 void* os::user_handler() {
2214 return CAST_FROM_FN_PTR(void*, UserHandler);
2215 }
2216
2217 extern "C" {
2218 typedef void (*sa_handler_t)(int);
2219 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2220 }
2221
2222 void* os::signal(int signal_number, void* handler) {
2223 struct sigaction sigAct, oldSigAct;
2224
2225 sigfillset(&(sigAct.sa_mask));
2226 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2227 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2228
2229 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2230 // -1 means registration failed
2231 return (void *)-1;
2232 }
2233
2234 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2235 }
2236
2237 void os::signal_raise(int signal_number) {
2238 ::raise(signal_number);
2239 }
2240
2241 /*
2242 * The following code is moved from os.cpp for making this
2243 * code platform specific, which it is by its very nature.
2244 */
2245
2246 // Will be modified when max signal is changed to be dynamic
2247 int os::sigexitnum_pd() {
2248 return NSIG;
2249 }
2250
2251 // a counter for each possible signal value
2252 static volatile jint pending_signals[NSIG+1] = { 0 };
2253
2254 // Linux(POSIX) specific hand shaking semaphore.
2255 static sem_t sig_sem;
2256
2257 void os::signal_init_pd() {
2258 // Initialize signal structures
2259 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2260
2261 // Initialize signal semaphore
2262 ::sem_init(&sig_sem, 0, 0);
2263 }
2264
2265 void os::signal_notify(int sig) {
2266 Atomic::inc(&pending_signals[sig]);
2267 ::sem_post(&sig_sem);
2268 }
2269
2270 static int check_pending_signals(bool wait) {
2271 Atomic::store(0, &sigint_count);
2272 for (;;) {
2273 for (int i = 0; i < NSIG + 1; i++) {
2274 jint n = pending_signals[i];
2275 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2276 return i;
2277 }
2278 }
2279 if (!wait) {
2280 return -1;
2281 }
2282 JavaThread *thread = JavaThread::current();
2283 ThreadBlockInVM tbivm(thread);
2284
2285 bool threadIsSuspended;
2286 do {
2287 thread->set_suspend_equivalent();
2288 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2289 ::sem_wait(&sig_sem);
2290
2291 // were we externally suspended while we were waiting?
2292 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2293 if (threadIsSuspended) {
2294 //
2295 // The semaphore has been incremented, but while we were waiting
2296 // another thread suspended us. We don't want to continue running
2297 // while suspended because that would surprise the thread that
2298 // suspended us.
2299 //
2300 ::sem_post(&sig_sem);
2301
2302 thread->java_suspend_self();
2303 }
2304 } while (threadIsSuspended);
2305 }
2306 }
2307
2308 int os::signal_lookup() {
2309 return check_pending_signals(false);
2310 }
2311
2312 int os::signal_wait() {
2313 return check_pending_signals(true);
2314 }
2315
2316 ////////////////////////////////////////////////////////////////////////////////
2317 // Virtual Memory
2318
2319 int os::vm_page_size() {
2320 // Seems redundant as all get out
2321 assert(os::Linux::page_size() != -1, "must call os::init");
2322 return os::Linux::page_size();
2323 }
2324
2325 // Solaris allocates memory by pages.
2326 int os::vm_allocation_granularity() {
2327 assert(os::Linux::page_size() != -1, "must call os::init");
2328 return os::Linux::page_size();
2329 }
2330
2331 // Rationale behind this function:
2332 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2333 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2334 // samples for JITted code. Here we create private executable mapping over the code cache
2335 // and then we can use standard (well, almost, as mapping can change) way to provide
2336 // info for the reporting script by storing timestamp and location of symbol
2337 void linux_wrap_code(char* base, size_t size) {
2338 static volatile jint cnt = 0;
2339
2340 if (!UseOprofile) {
2341 return;
2342 }
2343
2344 char buf[PATH_MAX+1];
2345 int num = Atomic::add(1, &cnt);
2346
2347 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2348 os::get_temp_directory(), os::current_process_id(), num);
2349 unlink(buf);
2350
2351 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2352
2353 if (fd != -1) {
2354 off_t rv = lseek(fd, size-2, SEEK_SET);
2355 if (rv != (off_t)-1) {
2356 if (write(fd, "", 1) == 1) {
2357 mmap(base, size,
2358 PROT_READ|PROT_WRITE|PROT_EXEC,
2359 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2360 }
2361 }
2362 close(fd);
2363 unlink(buf);
2364 }
2365 }
2366
2367 // NOTE: Linux kernel does not really reserve the pages for us.
2368 // All it does is to check if there are enough free pages
2369 // left at the time of mmap(). This could be a potential
2370 // problem.
2371 bool os::commit_memory(char* addr, size_t size, bool exec) {
2372 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2373 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2374 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2375 return res != (uintptr_t) MAP_FAILED;
2376 }
2377
2378 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2379 bool exec) {
2380 return commit_memory(addr, size, exec);
2381 }
2382
2383 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2384
2385 void os::free_memory(char *addr, size_t bytes) {
2386 ::mmap(addr, bytes, PROT_READ | PROT_WRITE,
2387 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2388 }
2389
2390 void os::numa_make_global(char *addr, size_t bytes) {
2391 Linux::numa_interleave_memory(addr, bytes);
2392 }
2393
2394 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2395 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2396 }
2397
2398 bool os::numa_topology_changed() { return false; }
2399
2400 size_t os::numa_get_groups_num() {
2401 int max_node = Linux::numa_max_node();
2402 return max_node > 0 ? max_node + 1 : 1;
2403 }
2404
2405 int os::numa_get_group_id() {
2406 int cpu_id = Linux::sched_getcpu();
2407 if (cpu_id != -1) {
2408 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2409 if (lgrp_id != -1) {
2410 return lgrp_id;
2411 }
2412 }
2413 return 0;
2414 }
2415
2416 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2417 for (size_t i = 0; i < size; i++) {
2418 ids[i] = i;
2419 }
2420 return size;
2421 }
2422
2423 bool os::get_page_info(char *start, page_info* info) {
2424 return false;
2425 }
2426
2427 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2428 return end;
2429 }
2430
2431 extern "C" void numa_warn(int number, char *where, ...) { }
2432 extern "C" void numa_error(char *where) { }
2433
2434
2435 // If we are running with libnuma version > 2, then we should
2436 // be trying to use symbols with versions 1.1
2437 // If we are running with earlier version, which did not have symbol versions,
2438 // we should use the base version.
2439 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2440 void *f = dlvsym(handle, name, "libnuma_1.1");
2441 if (f == NULL) {
2442 f = dlsym(handle, name);
2443 }
2444 return f;
2445 }
2446
2447 bool os::Linux::libnuma_init() {
2448 // sched_getcpu() should be in libc.
2449 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2450 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2451
2452 if (sched_getcpu() != -1) { // Does it work?
2453 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2454 if (handle != NULL) {
2455 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2456 libnuma_dlsym(handle, "numa_node_to_cpus")));
2457 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2458 libnuma_dlsym(handle, "numa_max_node")));
2459 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2460 libnuma_dlsym(handle, "numa_available")));
2461 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2462 libnuma_dlsym(handle, "numa_tonode_memory")));
2463 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2464 libnuma_dlsym(handle, "numa_interleave_memory")));
2465
2466
2467 if (numa_available() != -1) {
2468 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2469 // Create a cpu -> node mapping
2470 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2471 rebuild_cpu_to_node_map();
2472 return true;
2473 }
2474 }
2475 }
2476 return false;
2477 }
2478
2479 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2480 // The table is later used in get_node_by_cpu().
2481 void os::Linux::rebuild_cpu_to_node_map() {
2482 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2483 // in libnuma (possible values are starting from 16,
2484 // and continuing up with every other power of 2, but less
2485 // than the maximum number of CPUs supported by kernel), and
2486 // is a subject to change (in libnuma version 2 the requirements
2487 // are more reasonable) we'll just hardcode the number they use
2488 // in the library.
2489 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2490
2491 size_t cpu_num = os::active_processor_count();
2492 size_t cpu_map_size = NCPUS / BitsPerCLong;
2493 size_t cpu_map_valid_size =
2494 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2495
2496 cpu_to_node()->clear();
2497 cpu_to_node()->at_grow(cpu_num - 1);
2498 size_t node_num = numa_get_groups_num();
2499
2500 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2501 for (size_t i = 0; i < node_num; i++) {
2502 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2503 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2504 if (cpu_map[j] != 0) {
2505 for (size_t k = 0; k < BitsPerCLong; k++) {
2506 if (cpu_map[j] & (1UL << k)) {
2507 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2508 }
2509 }
2510 }
2511 }
2512 }
2513 }
2514 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2515 }
2516
2517 int os::Linux::get_node_by_cpu(int cpu_id) {
2518 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2519 return cpu_to_node()->at(cpu_id);
2520 }
2521 return -1;
2522 }
2523
2524 GrowableArray<int>* os::Linux::_cpu_to_node;
2525 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2526 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2527 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2528 os::Linux::numa_available_func_t os::Linux::_numa_available;
2529 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2530 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2531 unsigned long* os::Linux::_numa_all_nodes;
2532
2533 bool os::uncommit_memory(char* addr, size_t size) {
2534 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2535 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2536 return res != (uintptr_t) MAP_FAILED;
2537 }
2538
2539 // Linux uses a growable mapping for the stack, and if the mapping for
2540 // the stack guard pages is not removed when we detach a thread the
2541 // stack cannot grow beyond the pages where the stack guard was
2542 // mapped. If at some point later in the process the stack expands to
2543 // that point, the Linux kernel cannot expand the stack any further
2544 // because the guard pages are in the way, and a segfault occurs.
2545 //
2546 // However, it's essential not to split the stack region by unmapping
2547 // a region (leaving a hole) that's already part of the stack mapping,
2548 // so if the stack mapping has already grown beyond the guard pages at
2549 // the time we create them, we have to truncate the stack mapping.
2550 // So, we need to know the extent of the stack mapping when
2551 // create_stack_guard_pages() is called.
2552
2553 // Find the bounds of the stack mapping. Return true for success.
2554 //
2555 // We only need this for stacks that are growable: at the time of
2556 // writing thread stacks don't use growable mappings (i.e. those
2557 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2558 // only applies to the main thread.
2559 static bool
2560 get_stack_bounds(uintptr_t *bottom, uintptr_t *top)
2561 {
2562 FILE *f = fopen("/proc/self/maps", "r");
2563 if (f == NULL)
2564 return false;
2565
2566 while (!feof(f)) {
2567 size_t dummy;
2568 char *str = NULL;
2569 ssize_t len = getline(&str, &dummy, f);
2570 if (len == -1) {
2571 fclose(f);
2572 return false;
2573 }
2574
2575 if (len > 0 && str[len-1] == '\n') {
2576 str[len-1] = 0;
2577 len--;
2578 }
2579
2580 static const char *stack_str = "[stack]";
2581 if (len > (ssize_t)strlen(stack_str)
2582 && (strcmp(str + len - strlen(stack_str), stack_str) == 0)) {
2583 if (sscanf(str, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2584 uintptr_t sp = (uintptr_t)__builtin_frame_address(0);
2585 if (sp >= *bottom && sp <= *top) {
2586 free(str);
2587 fclose(f);
2588 return true;
2589 }
2590 }
2591 }
2592 free(str);
2593 }
2594 fclose(f);
2595 return false;
2596 }
2597
2598 // If the (growable) stack mapping already extends beyond the point
2599 // where we're going to put our guard pages, truncate the mapping at
2600 // that point by munmap()ping it. This ensures that when we later
2601 // munmap() the guard pages we don't leave a hole in the stack
2602 // mapping. This only affects the main/initial thread, but guard
2603 // against future OS changes
2604 bool os::create_stack_guard_pages(char* addr, size_t size) {
2605 uintptr_t stack_extent, stack_base;
2606 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2607 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2608 assert(os::Linux::is_initial_thread(),
2609 "growable stack in non-initial thread");
2610 if (stack_extent < (uintptr_t)addr)
2611 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2612 }
2613
2614 return os::commit_memory(addr, size);
2615 }
2616
2617 // If this is a growable mapping, remove the guard pages entirely by
2618 // munmap()ping them. If not, just call uncommit_memory(). This only
2619 // affects the main/initial thread, but guard against future OS changes
2620 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2621 uintptr_t stack_extent, stack_base;
2622 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2623 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2624 assert(os::Linux::is_initial_thread(),
2625 "growable stack in non-initial thread");
2626
2627 return ::munmap(addr, size) == 0;
2628 }
2629
2630 return os::uncommit_memory(addr, size);
2631 }
2632
2633 static address _highest_vm_reserved_address = NULL;
2634
2635 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2636 // at 'requested_addr'. If there are existing memory mappings at the same
2637 // location, however, they will be overwritten. If 'fixed' is false,
2638 // 'requested_addr' is only treated as a hint, the return value may or
2639 // may not start from the requested address. Unlike Linux mmap(), this
2640 // function returns NULL to indicate failure.
2641 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2642 char * addr;
2643 int flags;
2644
2645 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2646 if (fixed) {
2647 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2648 flags |= MAP_FIXED;
2649 }
2650
2651 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2652 // to PROT_EXEC if executable when we commit the page.
2653 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2654 flags, -1, 0);
2655
2656 if (addr != MAP_FAILED) {
2657 // anon_mmap() should only get called during VM initialization,
2658 // don't need lock (actually we can skip locking even it can be called
2659 // from multiple threads, because _highest_vm_reserved_address is just a
2660 // hint about the upper limit of non-stack memory regions.)
2661 if ((address)addr + bytes > _highest_vm_reserved_address) {
2662 _highest_vm_reserved_address = (address)addr + bytes;
2663 }
2664 }
2665
2666 return addr == MAP_FAILED ? NULL : addr;
2667 }
2668
2669 // Don't update _highest_vm_reserved_address, because there might be memory
2670 // regions above addr + size. If so, releasing a memory region only creates
2671 // a hole in the address space, it doesn't help prevent heap-stack collision.
2672 //
2673 static int anon_munmap(char * addr, size_t size) {
2674 return ::munmap(addr, size) == 0;
2675 }
2676
2677 char* os::reserve_memory(size_t bytes, char* requested_addr,
2678 size_t alignment_hint) {
2679 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2680 }
2681
2682 bool os::release_memory(char* addr, size_t size) {
2683 return anon_munmap(addr, size);
2684 }
2685
2686 static address highest_vm_reserved_address() {
2687 return _highest_vm_reserved_address;
2688 }
2689
2690 static bool linux_mprotect(char* addr, size_t size, int prot) {
2691 // Linux wants the mprotect address argument to be page aligned.
2692 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2693
2694 // According to SUSv3, mprotect() should only be used with mappings
2695 // established by mmap(), and mmap() always maps whole pages. Unaligned
2696 // 'addr' likely indicates problem in the VM (e.g. trying to change
2697 // protection of malloc'ed or statically allocated memory). Check the
2698 // caller if you hit this assert.
2699 assert(addr == bottom, "sanity check");
2700
2701 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2702 return ::mprotect(bottom, size, prot) == 0;
2703 }
2704
2705 // Set protections specified
2706 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2707 bool is_committed) {
2708 unsigned int p = 0;
2709 switch (prot) {
2710 case MEM_PROT_NONE: p = PROT_NONE; break;
2711 case MEM_PROT_READ: p = PROT_READ; break;
2712 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2713 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2714 default:
2715 ShouldNotReachHere();
2716 }
2717 // is_committed is unused.
2718 return linux_mprotect(addr, bytes, p);
2719 }
2720
2721 bool os::guard_memory(char* addr, size_t size) {
2722 return linux_mprotect(addr, size, PROT_NONE);
2723 }
2724
2725 bool os::unguard_memory(char* addr, size_t size) {
2726 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2727 }
2728
2729 // Large page support
2730
2731 static size_t _large_page_size = 0;
2732
2733 bool os::large_page_init() {
2734 if (!UseLargePages) return false;
2735
2736 if (LargePageSizeInBytes) {
2737 _large_page_size = LargePageSizeInBytes;
2738 } else {
2739 // large_page_size on Linux is used to round up heap size. x86 uses either
2740 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2741 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2742 // page as large as 256M.
2743 //
2744 // Here we try to figure out page size by parsing /proc/meminfo and looking
2745 // for a line with the following format:
2746 // Hugepagesize: 2048 kB
2747 //
2748 // If we can't determine the value (e.g. /proc is not mounted, or the text
2749 // format has been changed), we'll use the largest page size supported by
2750 // the processor.
2751
2752 #ifndef ZERO
2753 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
2754 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
2755 #endif // ZERO
2756
2757 FILE *fp = fopen("/proc/meminfo", "r");
2758 if (fp) {
2759 while (!feof(fp)) {
2760 int x = 0;
2761 char buf[16];
2762 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2763 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2764 _large_page_size = x * K;
2765 break;
2766 }
2767 } else {
2768 // skip to next line
2769 for (;;) {
2770 int ch = fgetc(fp);
2771 if (ch == EOF || ch == (int)'\n') break;
2772 }
2773 }
2774 }
2775 fclose(fp);
2776 }
2777 }
2778
2779 const size_t default_page_size = (size_t)Linux::page_size();
2780 if (_large_page_size > default_page_size) {
2781 _page_sizes[0] = _large_page_size;
2782 _page_sizes[1] = default_page_size;
2783 _page_sizes[2] = 0;
2784 }
2785
2786 // Large page support is available on 2.6 or newer kernel, some vendors
2787 // (e.g. Redhat) have backported it to their 2.4 based distributions.
2788 // We optimistically assume the support is available. If later it turns out
2789 // not true, VM will automatically switch to use regular page size.
2790 return true;
2791 }
2792
2793 #ifndef SHM_HUGETLB
2794 #define SHM_HUGETLB 04000
2795 #endif
2796
2797 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
2798 // "exec" is passed in but not used. Creating the shared image for
2799 // the code cache doesn't have an SHM_X executable permission to check.
2800 assert(UseLargePages, "only for large pages");
2801
2802 key_t key = IPC_PRIVATE;
2803 char *addr;
2804
2805 bool warn_on_failure = UseLargePages &&
2806 (!FLAG_IS_DEFAULT(UseLargePages) ||
2807 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2808 );
2809 char msg[128];
2810
2811 // Create a large shared memory region to attach to based on size.
2812 // Currently, size is the total size of the heap
2813 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2814 if (shmid == -1) {
2815 // Possible reasons for shmget failure:
2816 // 1. shmmax is too small for Java heap.
2817 // > check shmmax value: cat /proc/sys/kernel/shmmax
2818 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2819 // 2. not enough large page memory.
2820 // > check available large pages: cat /proc/meminfo
2821 // > increase amount of large pages:
2822 // echo new_value > /proc/sys/vm/nr_hugepages
2823 // Note 1: different Linux may use different name for this property,
2824 // e.g. on Redhat AS-3 it is "hugetlb_pool".
2825 // Note 2: it's possible there's enough physical memory available but
2826 // they are so fragmented after a long run that they can't
2827 // coalesce into large pages. Try to reserve large pages when
2828 // the system is still "fresh".
2829 if (warn_on_failure) {
2830 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2831 warning(msg);
2832 }
2833 return NULL;
2834 }
2835
2836 // attach to the region
2837 addr = (char*)shmat(shmid, req_addr, 0);
2838 int err = errno;
2839
2840 // Remove shmid. If shmat() is successful, the actual shared memory segment
2841 // will be deleted when it's detached by shmdt() or when the process
2842 // terminates. If shmat() is not successful this will remove the shared
2843 // segment immediately.
2844 shmctl(shmid, IPC_RMID, NULL);
2845
2846 if ((intptr_t)addr == -1) {
2847 if (warn_on_failure) {
2848 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2849 warning(msg);
2850 }
2851 return NULL;
2852 }
2853
2854 return addr;
2855 }
2856
2857 bool os::release_memory_special(char* base, size_t bytes) {
2858 // detaching the SHM segment will also delete it, see reserve_memory_special()
2859 int rslt = shmdt(base);
2860 return rslt == 0;
2861 }
2862
2863 size_t os::large_page_size() {
2864 return _large_page_size;
2865 }
2866
2867 // Linux does not support anonymous mmap with large page memory. The only way
2868 // to reserve large page memory without file backing is through SysV shared
2869 // memory API. The entire memory region is committed and pinned upfront.
2870 // Hopefully this will change in the future...
2871 bool os::can_commit_large_page_memory() {
2872 return false;
2873 }
2874
2875 bool os::can_execute_large_page_memory() {
2876 return false;
2877 }
2878
2879 // Reserve memory at an arbitrary address, only if that area is
2880 // available (and not reserved for something else).
2881
2882 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2883 const int max_tries = 10;
2884 char* base[max_tries];
2885 size_t size[max_tries];
2886 const size_t gap = 0x000000;
2887
2888 // Assert only that the size is a multiple of the page size, since
2889 // that's all that mmap requires, and since that's all we really know
2890 // about at this low abstraction level. If we need higher alignment,
2891 // we can either pass an alignment to this method or verify alignment
2892 // in one of the methods further up the call chain. See bug 5044738.
2893 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2894
2895 // Repeatedly allocate blocks until the block is allocated at the
2896 // right spot. Give up after max_tries. Note that reserve_memory() will
2897 // automatically update _highest_vm_reserved_address if the call is
2898 // successful. The variable tracks the highest memory address every reserved
2899 // by JVM. It is used to detect heap-stack collision if running with
2900 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2901 // space than needed, it could confuse the collision detecting code. To
2902 // solve the problem, save current _highest_vm_reserved_address and
2903 // calculate the correct value before return.
2904 address old_highest = _highest_vm_reserved_address;
2905
2906 // Linux mmap allows caller to pass an address as hint; give it a try first,
2907 // if kernel honors the hint then we can return immediately.
2908 char * addr = anon_mmap(requested_addr, bytes, false);
2909 if (addr == requested_addr) {
2910 return requested_addr;
2911 }
2912
2913 if (addr != NULL) {
2914 // mmap() is successful but it fails to reserve at the requested address
2915 anon_munmap(addr, bytes);
2916 }
2917
2918 int i;
2919 for (i = 0; i < max_tries; ++i) {
2920 base[i] = reserve_memory(bytes);
2921
2922 if (base[i] != NULL) {
2923 // Is this the block we wanted?
2924 if (base[i] == requested_addr) {
2925 size[i] = bytes;
2926 break;
2927 }
2928
2929 // Does this overlap the block we wanted? Give back the overlapped
2930 // parts and try again.
2931
2932 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2933 if (top_overlap >= 0 && top_overlap < bytes) {
2934 unmap_memory(base[i], top_overlap);
2935 base[i] += top_overlap;
2936 size[i] = bytes - top_overlap;
2937 } else {
2938 size_t bottom_overlap = base[i] + bytes - requested_addr;
2939 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2940 unmap_memory(requested_addr, bottom_overlap);
2941 size[i] = bytes - bottom_overlap;
2942 } else {
2943 size[i] = bytes;
2944 }
2945 }
2946 }
2947 }
2948
2949 // Give back the unused reserved pieces.
2950
2951 for (int j = 0; j < i; ++j) {
2952 if (base[j] != NULL) {
2953 unmap_memory(base[j], size[j]);
2954 }
2955 }
2956
2957 if (i < max_tries) {
2958 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2959 return requested_addr;
2960 } else {
2961 _highest_vm_reserved_address = old_highest;
2962 return NULL;
2963 }
2964 }
2965
2966 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2967 return ::read(fd, buf, nBytes);
2968 }
2969
2970 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2971 // Solaris uses poll(), linux uses park().
2972 // Poll() is likely a better choice, assuming that Thread.interrupt()
2973 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2974 // SIGSEGV, see 4355769.
2975
2976 const int NANOSECS_PER_MILLISECS = 1000000;
2977
2978 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2979 assert(thread == Thread::current(), "thread consistency check");
2980
2981 ParkEvent * const slp = thread->_SleepEvent ;
2982 slp->reset() ;
2983 OrderAccess::fence() ;
2984
2985 if (interruptible) {
2986 jlong prevtime = javaTimeNanos();
2987
2988 for (;;) {
2989 if (os::is_interrupted(thread, true)) {
2990 return OS_INTRPT;
2991 }
2992
2993 jlong newtime = javaTimeNanos();
2994
2995 if (newtime - prevtime < 0) {
2996 // time moving backwards, should only happen if no monotonic clock
2997 // not a guarantee() because JVM should not abort on kernel/glibc bugs
2998 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2999 } else {
3000 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3001 }
3002
3003 if(millis <= 0) {
3004 return OS_OK;
3005 }
3006
3007 prevtime = newtime;
3008
3009 {
3010 assert(thread->is_Java_thread(), "sanity check");
3011 JavaThread *jt = (JavaThread *) thread;
3012 ThreadBlockInVM tbivm(jt);
3013 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3014
3015 jt->set_suspend_equivalent();
3016 // cleared by handle_special_suspend_equivalent_condition() or
3017 // java_suspend_self() via check_and_wait_while_suspended()
3018
3019 slp->park(millis);
3020
3021 // were we externally suspended while we were waiting?
3022 jt->check_and_wait_while_suspended();
3023 }
3024 }
3025 } else {
3026 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3027 jlong prevtime = javaTimeNanos();
3028
3029 for (;;) {
3030 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3031 // the 1st iteration ...
3032 jlong newtime = javaTimeNanos();
3033
3034 if (newtime - prevtime < 0) {
3035 // time moving backwards, should only happen if no monotonic clock
3036 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3037 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3038 } else {
3039 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3040 }
3041
3042 if(millis <= 0) break ;
3043
3044 prevtime = newtime;
3045 slp->park(millis);
3046 }
3047 return OS_OK ;
3048 }
3049 }
3050
3051 int os::naked_sleep() {
3052 // %% make the sleep time an integer flag. for now use 1 millisec.
3053 return os::sleep(Thread::current(), 1, false);
3054 }
3055
3056 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3057 void os::infinite_sleep() {
3058 while (true) { // sleep forever ...
3059 ::sleep(100); // ... 100 seconds at a time
3060 }
3061 }
3062
3063 // Used to convert frequent JVM_Yield() to nops
3064 bool os::dont_yield() {
3065 return DontYieldALot;
3066 }
3067
3068 void os::yield() {
3069 sched_yield();
3070 }
3071
3072 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3073
3074 void os::yield_all(int attempts) {
3075 // Yields to all threads, including threads with lower priorities
3076 // Threads on Linux are all with same priority. The Solaris style
3077 // os::yield_all() with nanosleep(1ms) is not necessary.
3078 sched_yield();
3079 }
3080
3081 // Called from the tight loops to possibly influence time-sharing heuristics
3082 void os::loop_breaker(int attempts) {
3083 os::yield_all(attempts);
3084 }
3085
3086 ////////////////////////////////////////////////////////////////////////////////
3087 // thread priority support
3088
3089 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3090 // only supports dynamic priority, static priority must be zero. For real-time
3091 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3092 // However, for large multi-threaded applications, SCHED_RR is not only slower
3093 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3094 // of 5 runs - Sep 2005).
3095 //
3096 // The following code actually changes the niceness of kernel-thread/LWP. It
3097 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3098 // not the entire user process, and user level threads are 1:1 mapped to kernel
3099 // threads. It has always been the case, but could change in the future. For
3100 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3101 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3102
3103 int os::java_to_os_priority[MaxPriority + 1] = {
3104 19, // 0 Entry should never be used
3105
3106 4, // 1 MinPriority
3107 3, // 2
3108 2, // 3
3109
3110 1, // 4
3111 0, // 5 NormPriority
3112 -1, // 6
3113
3114 -2, // 7
3115 -3, // 8
3116 -4, // 9 NearMaxPriority
3117
3118 -5 // 10 MaxPriority
3119 };
3120
3121 static int prio_init() {
3122 if (ThreadPriorityPolicy == 1) {
3123 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3124 // if effective uid is not root. Perhaps, a more elegant way of doing
3125 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3126 if (geteuid() != 0) {
3127 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3128 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3129 }
3130 ThreadPriorityPolicy = 0;
3131 }
3132 }
3133 return 0;
3134 }
3135
3136 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3137 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3138
3139 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3140 return (ret == 0) ? OS_OK : OS_ERR;
3141 }
3142
3143 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3144 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3145 *priority_ptr = java_to_os_priority[NormPriority];
3146 return OS_OK;
3147 }
3148
3149 errno = 0;
3150 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3151 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3152 }
3153
3154 // Hint to the underlying OS that a task switch would not be good.
3155 // Void return because it's a hint and can fail.
3156 void os::hint_no_preempt() {}
3157
3158 ////////////////////////////////////////////////////////////////////////////////
3159 // suspend/resume support
3160
3161 // the low-level signal-based suspend/resume support is a remnant from the
3162 // old VM-suspension that used to be for java-suspension, safepoints etc,
3163 // within hotspot. Now there is a single use-case for this:
3164 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3165 // that runs in the watcher thread.
3166 // The remaining code is greatly simplified from the more general suspension
3167 // code that used to be used.
3168 //
3169 // The protocol is quite simple:
3170 // - suspend:
3171 // - sends a signal to the target thread
3172 // - polls the suspend state of the osthread using a yield loop
3173 // - target thread signal handler (SR_handler) sets suspend state
3174 // and blocks in sigsuspend until continued
3175 // - resume:
3176 // - sets target osthread state to continue
3177 // - sends signal to end the sigsuspend loop in the SR_handler
3178 //
3179 // Note that the SR_lock plays no role in this suspend/resume protocol.
3180 //
3181
3182 static void resume_clear_context(OSThread *osthread) {
3183 osthread->set_ucontext(NULL);
3184 osthread->set_siginfo(NULL);
3185
3186 // notify the suspend action is completed, we have now resumed
3187 osthread->sr.clear_suspended();
3188 }
3189
3190 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3191 osthread->set_ucontext(context);
3192 osthread->set_siginfo(siginfo);
3193 }
3194
3195 //
3196 // Handler function invoked when a thread's execution is suspended or
3197 // resumed. We have to be careful that only async-safe functions are
3198 // called here (Note: most pthread functions are not async safe and
3199 // should be avoided.)
3200 //
3201 // Note: sigwait() is a more natural fit than sigsuspend() from an
3202 // interface point of view, but sigwait() prevents the signal hander
3203 // from being run. libpthread would get very confused by not having
3204 // its signal handlers run and prevents sigwait()'s use with the
3205 // mutex granting granting signal.
3206 //
3207 // Currently only ever called on the VMThread
3208 //
3209 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3210 // Save and restore errno to avoid confusing native code with EINTR
3211 // after sigsuspend.
3212 int old_errno = errno;
3213
3214 Thread* thread = Thread::current();
3215 OSThread* osthread = thread->osthread();
3216 assert(thread->is_VM_thread(), "Must be VMThread");
3217 // read current suspend action
3218 int action = osthread->sr.suspend_action();
3219 if (action == SR_SUSPEND) {
3220 suspend_save_context(osthread, siginfo, context);
3221
3222 // Notify the suspend action is about to be completed. do_suspend()
3223 // waits until SR_SUSPENDED is set and then returns. We will wait
3224 // here for a resume signal and that completes the suspend-other
3225 // action. do_suspend/do_resume is always called as a pair from
3226 // the same thread - so there are no races
3227
3228 // notify the caller
3229 osthread->sr.set_suspended();
3230
3231 sigset_t suspend_set; // signals for sigsuspend()
3232
3233 // get current set of blocked signals and unblock resume signal
3234 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3235 sigdelset(&suspend_set, SR_signum);
3236
3237 // wait here until we are resumed
3238 do {
3239 sigsuspend(&suspend_set);
3240 // ignore all returns until we get a resume signal
3241 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3242
3243 resume_clear_context(osthread);
3244
3245 } else {
3246 assert(action == SR_CONTINUE, "unexpected sr action");
3247 // nothing special to do - just leave the handler
3248 }
3249
3250 errno = old_errno;
3251 }
3252
3253
3254 static int SR_initialize() {
3255 struct sigaction act;
3256 char *s;
3257 /* Get signal number to use for suspend/resume */
3258 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3259 int sig = ::strtol(s, 0, 10);
3260 if (sig > 0 || sig < _NSIG) {
3261 SR_signum = sig;
3262 }
3263 }
3264
3265 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3266 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3267
3268 sigemptyset(&SR_sigset);
3269 sigaddset(&SR_sigset, SR_signum);
3270
3271 /* Set up signal handler for suspend/resume */
3272 act.sa_flags = SA_RESTART|SA_SIGINFO;
3273 act.sa_handler = (void (*)(int)) SR_handler;
3274
3275 // SR_signum is blocked by default.
3276 // 4528190 - We also need to block pthread restart signal (32 on all
3277 // supported Linux platforms). Note that LinuxThreads need to block
3278 // this signal for all threads to work properly. So we don't have
3279 // to use hard-coded signal number when setting up the mask.
3280 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3281
3282 if (sigaction(SR_signum, &act, 0) == -1) {
3283 return -1;
3284 }
3285
3286 // Save signal flag
3287 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3288 return 0;
3289 }
3290
3291 static int SR_finalize() {
3292 return 0;
3293 }
3294
3295
3296 // returns true on success and false on error - really an error is fatal
3297 // but this seems the normal response to library errors
3298 static bool do_suspend(OSThread* osthread) {
3299 // mark as suspended and send signal
3300 osthread->sr.set_suspend_action(SR_SUSPEND);
3301 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3302 assert_status(status == 0, status, "pthread_kill");
3303
3304 // check status and wait until notified of suspension
3305 if (status == 0) {
3306 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3307 os::yield_all(i);
3308 }
3309 osthread->sr.set_suspend_action(SR_NONE);
3310 return true;
3311 }
3312 else {
3313 osthread->sr.set_suspend_action(SR_NONE);
3314 return false;
3315 }
3316 }
3317
3318 static void do_resume(OSThread* osthread) {
3319 assert(osthread->sr.is_suspended(), "thread should be suspended");
3320 osthread->sr.set_suspend_action(SR_CONTINUE);
3321
3322 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3323 assert_status(status == 0, status, "pthread_kill");
3324 // check status and wait unit notified of resumption
3325 if (status == 0) {
3326 for (int i = 0; osthread->sr.is_suspended(); i++) {
3327 os::yield_all(i);
3328 }
3329 }
3330 osthread->sr.set_suspend_action(SR_NONE);
3331 }
3332
3333 ////////////////////////////////////////////////////////////////////////////////
3334 // interrupt support
3335
3336 void os::interrupt(Thread* thread) {
3337 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3338 "possibility of dangling Thread pointer");
3339
3340 OSThread* osthread = thread->osthread();
3341
3342 if (!osthread->interrupted()) {
3343 osthread->set_interrupted(true);
3344 // More than one thread can get here with the same value of osthread,
3345 // resulting in multiple notifications. We do, however, want the store
3346 // to interrupted() to be visible to other threads before we execute unpark().
3347 OrderAccess::fence();
3348 ParkEvent * const slp = thread->_SleepEvent ;
3349 if (slp != NULL) slp->unpark() ;
3350 }
3351
3352 // For JSR166. Unpark even if interrupt status already was set
3353 if (thread->is_Java_thread())
3354 ((JavaThread*)thread)->parker()->unpark();
3355
3356 ParkEvent * ev = thread->_ParkEvent ;
3357 if (ev != NULL) ev->unpark() ;
3358
3359 }
3360
3361 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3362 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3363 "possibility of dangling Thread pointer");
3364
3365 OSThread* osthread = thread->osthread();
3366
3367 bool interrupted = osthread->interrupted();
3368
3369 if (interrupted && clear_interrupted) {
3370 osthread->set_interrupted(false);
3371 // consider thread->_SleepEvent->reset() ... optional optimization
3372 }
3373
3374 return interrupted;
3375 }
3376
3377 ///////////////////////////////////////////////////////////////////////////////////
3378 // signal handling (except suspend/resume)
3379
3380 // This routine may be used by user applications as a "hook" to catch signals.
3381 // The user-defined signal handler must pass unrecognized signals to this
3382 // routine, and if it returns true (non-zero), then the signal handler must
3383 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3384 // routine will never retun false (zero), but instead will execute a VM panic
3385 // routine kill the process.
3386 //
3387 // If this routine returns false, it is OK to call it again. This allows
3388 // the user-defined signal handler to perform checks either before or after
3389 // the VM performs its own checks. Naturally, the user code would be making
3390 // a serious error if it tried to handle an exception (such as a null check
3391 // or breakpoint) that the VM was generating for its own correct operation.
3392 //
3393 // This routine may recognize any of the following kinds of signals:
3394 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3395 // It should be consulted by handlers for any of those signals.
3396 //
3397 // The caller of this routine must pass in the three arguments supplied
3398 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3399 // field of the structure passed to sigaction(). This routine assumes that
3400 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3401 //
3402 // Note that the VM will print warnings if it detects conflicting signal
3403 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3404 //
3405 extern "C" int
3406 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3407 void* ucontext, int abort_if_unrecognized);
3408
3409 void signalHandler(int sig, siginfo_t* info, void* uc) {
3410 assert(info != NULL && uc != NULL, "it must be old kernel");
3411 JVM_handle_linux_signal(sig, info, uc, true);
3412 }
3413
3414
3415 // This boolean allows users to forward their own non-matching signals
3416 // to JVM_handle_linux_signal, harmlessly.
3417 bool os::Linux::signal_handlers_are_installed = false;
3418
3419 // For signal-chaining
3420 struct sigaction os::Linux::sigact[MAXSIGNUM];
3421 unsigned int os::Linux::sigs = 0;
3422 bool os::Linux::libjsig_is_loaded = false;
3423 typedef struct sigaction *(*get_signal_t)(int);
3424 get_signal_t os::Linux::get_signal_action = NULL;
3425
3426 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3427 struct sigaction *actp = NULL;
3428
3429 if (libjsig_is_loaded) {
3430 // Retrieve the old signal handler from libjsig
3431 actp = (*get_signal_action)(sig);
3432 }
3433 if (actp == NULL) {
3434 // Retrieve the preinstalled signal handler from jvm
3435 actp = get_preinstalled_handler(sig);
3436 }
3437
3438 return actp;
3439 }
3440
3441 static bool call_chained_handler(struct sigaction *actp, int sig,
3442 siginfo_t *siginfo, void *context) {
3443 // Call the old signal handler
3444 if (actp->sa_handler == SIG_DFL) {
3445 // It's more reasonable to let jvm treat it as an unexpected exception
3446 // instead of taking the default action.
3447 return false;
3448 } else if (actp->sa_handler != SIG_IGN) {
3449 if ((actp->sa_flags & SA_NODEFER) == 0) {
3450 // automaticlly block the signal
3451 sigaddset(&(actp->sa_mask), sig);
3452 }
3453
3454 sa_handler_t hand;
3455 sa_sigaction_t sa;
3456 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3457 // retrieve the chained handler
3458 if (siginfo_flag_set) {
3459 sa = actp->sa_sigaction;
3460 } else {
3461 hand = actp->sa_handler;
3462 }
3463
3464 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3465 actp->sa_handler = SIG_DFL;
3466 }
3467
3468 // try to honor the signal mask
3469 sigset_t oset;
3470 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3471
3472 // call into the chained handler
3473 if (siginfo_flag_set) {
3474 (*sa)(sig, siginfo, context);
3475 } else {
3476 (*hand)(sig);
3477 }
3478
3479 // restore the signal mask
3480 pthread_sigmask(SIG_SETMASK, &oset, 0);
3481 }
3482 // Tell jvm's signal handler the signal is taken care of.
3483 return true;
3484 }
3485
3486 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3487 bool chained = false;
3488 // signal-chaining
3489 if (UseSignalChaining) {
3490 struct sigaction *actp = get_chained_signal_action(sig);
3491 if (actp != NULL) {
3492 chained = call_chained_handler(actp, sig, siginfo, context);
3493 }
3494 }
3495 return chained;
3496 }
3497
3498 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3499 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3500 return &sigact[sig];
3501 }
3502 return NULL;
3503 }
3504
3505 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3506 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3507 sigact[sig] = oldAct;
3508 sigs |= (unsigned int)1 << sig;
3509 }
3510
3511 // for diagnostic
3512 int os::Linux::sigflags[MAXSIGNUM];
3513
3514 int os::Linux::get_our_sigflags(int sig) {
3515 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3516 return sigflags[sig];
3517 }
3518
3519 void os::Linux::set_our_sigflags(int sig, int flags) {
3520 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3521 sigflags[sig] = flags;
3522 }
3523
3524 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3525 // Check for overwrite.
3526 struct sigaction oldAct;
3527 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3528
3529 void* oldhand = oldAct.sa_sigaction
3530 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3531 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3532 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3533 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3534 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3535 if (AllowUserSignalHandlers || !set_installed) {
3536 // Do not overwrite; user takes responsibility to forward to us.
3537 return;
3538 } else if (UseSignalChaining) {
3539 // save the old handler in jvm
3540 save_preinstalled_handler(sig, oldAct);
3541 // libjsig also interposes the sigaction() call below and saves the
3542 // old sigaction on it own.
3543 } else {
3544 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3545 "%#lx for signal %d.", (long)oldhand, sig));
3546 }
3547 }
3548
3549 struct sigaction sigAct;
3550 sigfillset(&(sigAct.sa_mask));
3551 sigAct.sa_handler = SIG_DFL;
3552 if (!set_installed) {
3553 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3554 } else {
3555 sigAct.sa_sigaction = signalHandler;
3556 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3557 }
3558 // Save flags, which are set by ours
3559 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3560 sigflags[sig] = sigAct.sa_flags;
3561
3562 int ret = sigaction(sig, &sigAct, &oldAct);
3563 assert(ret == 0, "check");
3564
3565 void* oldhand2 = oldAct.sa_sigaction
3566 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3567 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3568 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3569 }
3570
3571 // install signal handlers for signals that HotSpot needs to
3572 // handle in order to support Java-level exception handling.
3573
3574 void os::Linux::install_signal_handlers() {
3575 if (!signal_handlers_are_installed) {
3576 signal_handlers_are_installed = true;
3577
3578 // signal-chaining
3579 typedef void (*signal_setting_t)();
3580 signal_setting_t begin_signal_setting = NULL;
3581 signal_setting_t end_signal_setting = NULL;
3582 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3583 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3584 if (begin_signal_setting != NULL) {
3585 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3586 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3587 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3588 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3589 libjsig_is_loaded = true;
3590 assert(UseSignalChaining, "should enable signal-chaining");
3591 }
3592 if (libjsig_is_loaded) {
3593 // Tell libjsig jvm is setting signal handlers
3594 (*begin_signal_setting)();
3595 }
3596
3597 set_signal_handler(SIGSEGV, true);
3598 set_signal_handler(SIGPIPE, true);
3599 set_signal_handler(SIGBUS, true);
3600 set_signal_handler(SIGILL, true);
3601 set_signal_handler(SIGFPE, true);
3602 set_signal_handler(SIGXFSZ, true);
3603
3604 if (libjsig_is_loaded) {
3605 // Tell libjsig jvm finishes setting signal handlers
3606 (*end_signal_setting)();
3607 }
3608
3609 // We don't activate signal checker if libjsig is in place, we trust ourselves
3610 // and if UserSignalHandler is installed all bets are off
3611 if (CheckJNICalls) {
3612 if (libjsig_is_loaded) {
3613 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3614 check_signals = false;
3615 }
3616 if (AllowUserSignalHandlers) {
3617 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3618 check_signals = false;
3619 }
3620 }
3621 }
3622 }
3623
3624 // This is the fastest way to get thread cpu time on Linux.
3625 // Returns cpu time (user+sys) for any thread, not only for current.
3626 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3627 // It might work on 2.6.10+ with a special kernel/glibc patch.
3628 // For reference, please, see IEEE Std 1003.1-2004:
3629 // http://www.unix.org/single_unix_specification
3630
3631 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3632 struct timespec tp;
3633 int rc = os::Linux::clock_gettime(clockid, &tp);
3634 assert(rc == 0, "clock_gettime is expected to return 0 code");
3635
3636 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3637 }
3638
3639 /////
3640 // glibc on Linux platform uses non-documented flag
3641 // to indicate, that some special sort of signal
3642 // trampoline is used.
3643 // We will never set this flag, and we should
3644 // ignore this flag in our diagnostic
3645 #ifdef SIGNIFICANT_SIGNAL_MASK
3646 #undef SIGNIFICANT_SIGNAL_MASK
3647 #endif
3648 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3649
3650 static const char* get_signal_handler_name(address handler,
3651 char* buf, int buflen) {
3652 int offset;
3653 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3654 if (found) {
3655 // skip directory names
3656 const char *p1, *p2;
3657 p1 = buf;
3658 size_t len = strlen(os::file_separator());
3659 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3660 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3661 } else {
3662 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3663 }
3664 return buf;
3665 }
3666
3667 static void print_signal_handler(outputStream* st, int sig,
3668 char* buf, size_t buflen) {
3669 struct sigaction sa;
3670
3671 sigaction(sig, NULL, &sa);
3672
3673 // See comment for SIGNIFICANT_SIGNAL_MASK define
3674 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3675
3676 st->print("%s: ", os::exception_name(sig, buf, buflen));
3677
3678 address handler = (sa.sa_flags & SA_SIGINFO)
3679 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3680 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3681
3682 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3683 st->print("SIG_DFL");
3684 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3685 st->print("SIG_IGN");
3686 } else {
3687 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3688 }
3689
3690 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3691
3692 address rh = VMError::get_resetted_sighandler(sig);
3693 // May be, handler was resetted by VMError?
3694 if(rh != NULL) {
3695 handler = rh;
3696 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3697 }
3698
3699 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3700
3701 // Check: is it our handler?
3702 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3703 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3704 // It is our signal handler
3705 // check for flags, reset system-used one!
3706 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3707 st->print(
3708 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3709 os::Linux::get_our_sigflags(sig));
3710 }
3711 }
3712 st->cr();
3713 }
3714
3715
3716 #define DO_SIGNAL_CHECK(sig) \
3717 if (!sigismember(&check_signal_done, sig)) \
3718 os::Linux::check_signal_handler(sig)
3719
3720 // This method is a periodic task to check for misbehaving JNI applications
3721 // under CheckJNI, we can add any periodic checks here
3722
3723 void os::run_periodic_checks() {
3724
3725 if (check_signals == false) return;
3726
3727 // SEGV and BUS if overridden could potentially prevent
3728 // generation of hs*.log in the event of a crash, debugging
3729 // such a case can be very challenging, so we absolutely
3730 // check the following for a good measure:
3731 DO_SIGNAL_CHECK(SIGSEGV);
3732 DO_SIGNAL_CHECK(SIGILL);
3733 DO_SIGNAL_CHECK(SIGFPE);
3734 DO_SIGNAL_CHECK(SIGBUS);
3735 DO_SIGNAL_CHECK(SIGPIPE);
3736 DO_SIGNAL_CHECK(SIGXFSZ);
3737
3738
3739 // ReduceSignalUsage allows the user to override these handlers
3740 // see comments at the very top and jvm_solaris.h
3741 if (!ReduceSignalUsage) {
3742 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3743 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3744 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3745 DO_SIGNAL_CHECK(BREAK_SIGNAL);
3746 }
3747
3748 DO_SIGNAL_CHECK(SR_signum);
3749 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3750 }
3751
3752 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3753
3754 static os_sigaction_t os_sigaction = NULL;
3755
3756 void os::Linux::check_signal_handler(int sig) {
3757 char buf[O_BUFLEN];
3758 address jvmHandler = NULL;
3759
3760
3761 struct sigaction act;
3762 if (os_sigaction == NULL) {
3763 // only trust the default sigaction, in case it has been interposed
3764 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3765 if (os_sigaction == NULL) return;
3766 }
3767
3768 os_sigaction(sig, (struct sigaction*)NULL, &act);
3769
3770
3771 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3772
3773 address thisHandler = (act.sa_flags & SA_SIGINFO)
3774 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3775 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3776
3777
3778 switch(sig) {
3779 case SIGSEGV:
3780 case SIGBUS:
3781 case SIGFPE:
3782 case SIGPIPE:
3783 case SIGILL:
3784 case SIGXFSZ:
3785 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3786 break;
3787
3788 case SHUTDOWN1_SIGNAL:
3789 case SHUTDOWN2_SIGNAL:
3790 case SHUTDOWN3_SIGNAL:
3791 case BREAK_SIGNAL:
3792 jvmHandler = (address)user_handler();
3793 break;
3794
3795 case INTERRUPT_SIGNAL:
3796 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3797 break;
3798
3799 default:
3800 if (sig == SR_signum) {
3801 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3802 } else {
3803 return;
3804 }
3805 break;
3806 }
3807
3808 if (thisHandler != jvmHandler) {
3809 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3810 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3811 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3812 // No need to check this sig any longer
3813 sigaddset(&check_signal_done, sig);
3814 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3815 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3816 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3817 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
3818 // No need to check this sig any longer
3819 sigaddset(&check_signal_done, sig);
3820 }
3821
3822 // Dump all the signal
3823 if (sigismember(&check_signal_done, sig)) {
3824 print_signal_handlers(tty, buf, O_BUFLEN);
3825 }
3826 }
3827
3828 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3829
3830 extern bool signal_name(int signo, char* buf, size_t len);
3831
3832 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3833 if (0 < exception_code && exception_code <= SIGRTMAX) {
3834 // signal
3835 if (!signal_name(exception_code, buf, size)) {
3836 jio_snprintf(buf, size, "SIG%d", exception_code);
3837 }
3838 return buf;
3839 } else {
3840 return NULL;
3841 }
3842 }
3843
3844 // this is called _before_ the most of global arguments have been parsed
3845 void os::init(void) {
3846 char dummy; /* used to get a guess on initial stack address */
3847 // first_hrtime = gethrtime();
3848
3849 // With LinuxThreads the JavaMain thread pid (primordial thread)
3850 // is different than the pid of the java launcher thread.
3851 // So, on Linux, the launcher thread pid is passed to the VM
3852 // via the sun.java.launcher.pid property.
3853 // Use this property instead of getpid() if it was correctly passed.
3854 // See bug 6351349.
3855 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3856
3857 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3858
3859 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3860
3861 init_random(1234567);
3862
3863 ThreadCritical::initialize();
3864
3865 Linux::set_page_size(sysconf(_SC_PAGESIZE));
3866 if (Linux::page_size() == -1) {
3867 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
3868 strerror(errno)));
3869 }
3870 init_page_sizes((size_t) Linux::page_size());
3871
3872 Linux::initialize_system_info();
3873
3874 // main_thread points to the aboriginal thread
3875 Linux::_main_thread = pthread_self();
3876
3877 Linux::clock_init();
3878 initial_time_count = os::elapsed_counter();
3879 pthread_mutex_init(&dl_mutex, NULL);
3880 }
3881
3882 // To install functions for atexit system call
3883 extern "C" {
3884 static void perfMemory_exit_helper() {
3885 perfMemory_exit();
3886 }
3887 }
3888
3889 // this is called _after_ the global arguments have been parsed
3890 jint os::init_2(void)
3891 {
3892 Linux::fast_thread_clock_init();
3893
3894 // Allocate a single page and mark it as readable for safepoint polling
3895 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3896 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3897
3898 os::set_polling_page( polling_page );
3899
3900 #ifndef PRODUCT
3901 if(Verbose && PrintMiscellaneous)
3902 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3903 #endif
3904
3905 if (!UseMembar) {
3906 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3907 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3908 os::set_memory_serialize_page( mem_serialize_page );
3909
3910 #ifndef PRODUCT
3911 if(Verbose && PrintMiscellaneous)
3912 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3913 #endif
3914 }
3915
3916 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3917
3918 // initialize suspend/resume support - must do this before signal_sets_init()
3919 if (SR_initialize() != 0) {
3920 perror("SR_initialize failed");
3921 return JNI_ERR;
3922 }
3923
3924 Linux::signal_sets_init();
3925 Linux::install_signal_handlers();
3926
3927 // Check minimum allowable stack size for thread creation and to initialize
3928 // the java system classes, including StackOverflowError - depends on page
3929 // size. Add a page for compiler2 recursion in main thread.
3930 // Add in 2*BytesPerWord times page size to account for VM stack during
3931 // class initialization depending on 32 or 64 bit VM.
3932 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
3933 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
3934 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
3935
3936 size_t threadStackSizeInBytes = ThreadStackSize * K;
3937 if (threadStackSizeInBytes != 0 &&
3938 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
3939 tty->print_cr("\nThe stack size specified is too small, "
3940 "Specify at least %dk",
3941 os::Linux::min_stack_allowed/ K);
3942 return JNI_ERR;
3943 }
3944
3945 // Make the stack size a multiple of the page size so that
3946 // the yellow/red zones can be guarded.
3947 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3948 vm_page_size()));
3949
3950 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3951
3952 Linux::libpthread_init();
3953 if (PrintMiscellaneous && (Verbose || WizardMode)) {
3954 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3955 Linux::glibc_version(), Linux::libpthread_version(),
3956 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3957 }
3958
3959 if (UseNUMA) {
3960 if (!Linux::libnuma_init()) {
3961 UseNUMA = false;
3962 } else {
3963 if ((Linux::numa_max_node() < 1)) {
3964 // There's only one node(they start from 0), disable NUMA.
3965 UseNUMA = false;
3966 }
3967 }
3968 if (!UseNUMA && ForceNUMA) {
3969 UseNUMA = true;
3970 }
3971 }
3972
3973 if (MaxFDLimit) {
3974 // set the number of file descriptors to max. print out error
3975 // if getrlimit/setrlimit fails but continue regardless.
3976 struct rlimit nbr_files;
3977 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3978 if (status != 0) {
3979 if (PrintMiscellaneous && (Verbose || WizardMode))
3980 perror("os::init_2 getrlimit failed");
3981 } else {
3982 nbr_files.rlim_cur = nbr_files.rlim_max;
3983 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3984 if (status != 0) {
3985 if (PrintMiscellaneous && (Verbose || WizardMode))
3986 perror("os::init_2 setrlimit failed");
3987 }
3988 }
3989 }
3990
3991 // Initialize lock used to serialize thread creation (see os::create_thread)
3992 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3993
3994 // Initialize HPI.
3995 jint hpi_result = hpi::initialize();
3996 if (hpi_result != JNI_OK) {
3997 tty->print_cr("There was an error trying to initialize the HPI library.");
3998 return hpi_result;
3999 }
4000
4001 // at-exit methods are called in the reverse order of their registration.
4002 // atexit functions are called on return from main or as a result of a
4003 // call to exit(3C). There can be only 32 of these functions registered
4004 // and atexit() does not set errno.
4005
4006 if (PerfAllowAtExitRegistration) {
4007 // only register atexit functions if PerfAllowAtExitRegistration is set.
4008 // atexit functions can be delayed until process exit time, which
4009 // can be problematic for embedded VM situations. Embedded VMs should
4010 // call DestroyJavaVM() to assure that VM resources are released.
4011
4012 // note: perfMemory_exit_helper atexit function may be removed in
4013 // the future if the appropriate cleanup code can be added to the
4014 // VM_Exit VMOperation's doit method.
4015 if (atexit(perfMemory_exit_helper) != 0) {
4016 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4017 }
4018 }
4019
4020 // initialize thread priority policy
4021 prio_init();
4022
4023 return JNI_OK;
4024 }
4025
4026 // this is called at the end of vm_initialization
4027 void os::init_3(void) { }
4028
4029 // Mark the polling page as unreadable
4030 void os::make_polling_page_unreadable(void) {
4031 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4032 fatal("Could not disable polling page");
4033 };
4034
4035 // Mark the polling page as readable
4036 void os::make_polling_page_readable(void) {
4037 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4038 fatal("Could not enable polling page");
4039 }
4040 };
4041
4042 int os::active_processor_count() {
4043 // Linux doesn't yet have a (official) notion of processor sets,
4044 // so just return the number of online processors.
4045 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4046 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4047 return online_cpus;
4048 }
4049
4050 bool os::distribute_processes(uint length, uint* distribution) {
4051 // Not yet implemented.
4052 return false;
4053 }
4054
4055 bool os::bind_to_processor(uint processor_id) {
4056 // Not yet implemented.
4057 return false;
4058 }
4059
4060 ///
4061
4062 // Suspends the target using the signal mechanism and then grabs the PC before
4063 // resuming the target. Used by the flat-profiler only
4064 ExtendedPC os::get_thread_pc(Thread* thread) {
4065 // Make sure that it is called by the watcher for the VMThread
4066 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4067 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4068
4069 ExtendedPC epc;
4070
4071 OSThread* osthread = thread->osthread();
4072 if (do_suspend(osthread)) {
4073 if (osthread->ucontext() != NULL) {
4074 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4075 } else {
4076 // NULL context is unexpected, double-check this is the VMThread
4077 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4078 }
4079 do_resume(osthread);
4080 }
4081 // failure means pthread_kill failed for some reason - arguably this is
4082 // a fatal problem, but such problems are ignored elsewhere
4083
4084 return epc;
4085 }
4086
4087 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4088 {
4089 if (is_NPTL()) {
4090 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4091 } else {
4092 #ifndef IA64
4093 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4094 // word back to default 64bit precision if condvar is signaled. Java
4095 // wants 53bit precision. Save and restore current value.
4096 int fpu = get_fpu_control_word();
4097 #endif // IA64
4098 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4099 #ifndef IA64
4100 set_fpu_control_word(fpu);
4101 #endif // IA64
4102 return status;
4103 }
4104 }
4105
4106 ////////////////////////////////////////////////////////////////////////////////
4107 // debug support
4108
4109 static address same_page(address x, address y) {
4110 int page_bits = -os::vm_page_size();
4111 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4112 return x;
4113 else if (x > y)
4114 return (address)(intptr_t(y) | ~page_bits) + 1;
4115 else
4116 return (address)(intptr_t(y) & page_bits);
4117 }
4118
4119 bool os::find(address addr, outputStream* st) {
4120 Dl_info dlinfo;
4121 memset(&dlinfo, 0, sizeof(dlinfo));
4122 if (dladdr(addr, &dlinfo)) {
4123 st->print(PTR_FORMAT ": ", addr);
4124 if (dlinfo.dli_sname != NULL) {
4125 st->print("%s+%#x", dlinfo.dli_sname,
4126 addr - (intptr_t)dlinfo.dli_saddr);
4127 } else if (dlinfo.dli_fname) {
4128 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4129 } else {
4130 st->print("<absolute address>");
4131 }
4132 if (dlinfo.dli_fname) {
4133 st->print(" in %s", dlinfo.dli_fname);
4134 }
4135 if (dlinfo.dli_fbase) {
4136 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4137 }
4138 st->cr();
4139
4140 if (Verbose) {
4141 // decode some bytes around the PC
4142 address begin = same_page(addr-40, addr);
4143 address end = same_page(addr+40, addr);
4144 address lowest = (address) dlinfo.dli_sname;
4145 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4146 if (begin < lowest) begin = lowest;
4147 Dl_info dlinfo2;
4148 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4149 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4150 end = (address) dlinfo2.dli_saddr;
4151 Disassembler::decode(begin, end, st);
4152 }
4153 return true;
4154 }
4155 return false;
4156 }
4157
4158 ////////////////////////////////////////////////////////////////////////////////
4159 // misc
4160
4161 // This does not do anything on Linux. This is basically a hook for being
4162 // able to use structured exception handling (thread-local exception filters)
4163 // on, e.g., Win32.
4164 void
4165 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4166 JavaCallArguments* args, Thread* thread) {
4167 f(value, method, args, thread);
4168 }
4169
4170 void os::print_statistics() {
4171 }
4172
4173 int os::message_box(const char* title, const char* message) {
4174 int i;
4175 fdStream err(defaultStream::error_fd());
4176 for (i = 0; i < 78; i++) err.print_raw("=");
4177 err.cr();
4178 err.print_raw_cr(title);
4179 for (i = 0; i < 78; i++) err.print_raw("-");
4180 err.cr();
4181 err.print_raw_cr(message);
4182 for (i = 0; i < 78; i++) err.print_raw("=");
4183 err.cr();
4184
4185 char buf[16];
4186 // Prevent process from exiting upon "read error" without consuming all CPU
4187 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4188
4189 return buf[0] == 'y' || buf[0] == 'Y';
4190 }
4191
4192 int os::stat(const char *path, struct stat *sbuf) {
4193 char pathbuf[MAX_PATH];
4194 if (strlen(path) > MAX_PATH - 1) {
4195 errno = ENAMETOOLONG;
4196 return -1;
4197 }
4198 hpi::native_path(strcpy(pathbuf, path));
4199 return ::stat(pathbuf, sbuf);
4200 }
4201
4202 bool os::check_heap(bool force) {
4203 return true;
4204 }
4205
4206 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4207 return ::vsnprintf(buf, count, format, args);
4208 }
4209
4210 // Is a (classpath) directory empty?
4211 bool os::dir_is_empty(const char* path) {
4212 DIR *dir = NULL;
4213 struct dirent *ptr;
4214
4215 dir = opendir(path);
4216 if (dir == NULL) return true;
4217
4218 /* Scan the directory */
4219 bool result = true;
4220 char buf[sizeof(struct dirent) + MAX_PATH];
4221 while (result && (ptr = ::readdir(dir)) != NULL) {
4222 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4223 result = false;
4224 }
4225 }
4226 closedir(dir);
4227 return result;
4228 }
4229
4230 // create binary file, rewriting existing file if required
4231 int os::create_binary_file(const char* path, bool rewrite_existing) {
4232 int oflags = O_WRONLY | O_CREAT;
4233 if (!rewrite_existing) {
4234 oflags |= O_EXCL;
4235 }
4236 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4237 }
4238
4239 // return current position of file pointer
4240 jlong os::current_file_offset(int fd) {
4241 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4242 }
4243
4244 // move file pointer to the specified offset
4245 jlong os::seek_to_file_offset(int fd, jlong offset) {
4246 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4247 }
4248
4249 // Map a block of memory.
4250 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4251 char *addr, size_t bytes, bool read_only,
4252 bool allow_exec) {
4253 int prot;
4254 int flags;
4255
4256 if (read_only) {
4257 prot = PROT_READ;
4258 flags = MAP_SHARED;
4259 } else {
4260 prot = PROT_READ | PROT_WRITE;
4261 flags = MAP_PRIVATE;
4262 }
4263
4264 if (allow_exec) {
4265 prot |= PROT_EXEC;
4266 }
4267
4268 if (addr != NULL) {
4269 flags |= MAP_FIXED;
4270 }
4271
4272 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4273 fd, file_offset);
4274 if (mapped_address == MAP_FAILED) {
4275 return NULL;
4276 }
4277 return mapped_address;
4278 }
4279
4280
4281 // Remap a block of memory.
4282 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4283 char *addr, size_t bytes, bool read_only,
4284 bool allow_exec) {
4285 // same as map_memory() on this OS
4286 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4287 allow_exec);
4288 }
4289
4290
4291 // Unmap a block of memory.
4292 bool os::unmap_memory(char* addr, size_t bytes) {
4293 return munmap(addr, bytes) == 0;
4294 }
4295
4296 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4297
4298 static clockid_t thread_cpu_clockid(Thread* thread) {
4299 pthread_t tid = thread->osthread()->pthread_id();
4300 clockid_t clockid;
4301
4302 // Get thread clockid
4303 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4304 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4305 return clockid;
4306 }
4307
4308 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4309 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4310 // of a thread.
4311 //
4312 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4313 // the fast estimate available on the platform.
4314
4315 jlong os::current_thread_cpu_time() {
4316 if (os::Linux::supports_fast_thread_cpu_time()) {
4317 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4318 } else {
4319 // return user + sys since the cost is the same
4320 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4321 }
4322 }
4323
4324 jlong os::thread_cpu_time(Thread* thread) {
4325 // consistent with what current_thread_cpu_time() returns
4326 if (os::Linux::supports_fast_thread_cpu_time()) {
4327 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4328 } else {
4329 return slow_thread_cpu_time(thread, true /* user + sys */);
4330 }
4331 }
4332
4333 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4334 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4335 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4336 } else {
4337 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4338 }
4339 }
4340
4341 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4342 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4343 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4344 } else {
4345 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4346 }
4347 }
4348
4349 //
4350 // -1 on error.
4351 //
4352
4353 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4354 static bool proc_pid_cpu_avail = true;
4355 static bool proc_task_unchecked = true;
4356 static const char *proc_stat_path = "/proc/%d/stat";
4357 pid_t tid = thread->osthread()->thread_id();
4358 int i;
4359 char *s;
4360 char stat[2048];
4361 int statlen;
4362 char proc_name[64];
4363 int count;
4364 long sys_time, user_time;
4365 char string[64];
4366 char cdummy;
4367 int idummy;
4368 long ldummy;
4369 FILE *fp;
4370
4371 // We first try accessing /proc/<pid>/cpu since this is faster to
4372 // process. If this file is not present (linux kernels 2.5 and above)
4373 // then we open /proc/<pid>/stat.
4374 if ( proc_pid_cpu_avail ) {
4375 sprintf(proc_name, "/proc/%d/cpu", tid);
4376 fp = fopen(proc_name, "r");
4377 if ( fp != NULL ) {
4378 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4379 fclose(fp);
4380 if ( count != 3 ) return -1;
4381
4382 if (user_sys_cpu_time) {
4383 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4384 } else {
4385 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4386 }
4387 }
4388 else proc_pid_cpu_avail = false;
4389 }
4390
4391 // The /proc/<tid>/stat aggregates per-process usage on
4392 // new Linux kernels 2.6+ where NPTL is supported.
4393 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4394 // See bug 6328462.
4395 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4396 // and possibly in some other cases, so we check its availability.
4397 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4398 // This is executed only once
4399 proc_task_unchecked = false;
4400 fp = fopen("/proc/self/task", "r");
4401 if (fp != NULL) {
4402 proc_stat_path = "/proc/self/task/%d/stat";
4403 fclose(fp);
4404 }
4405 }
4406
4407 sprintf(proc_name, proc_stat_path, tid);
4408 fp = fopen(proc_name, "r");
4409 if ( fp == NULL ) return -1;
4410 statlen = fread(stat, 1, 2047, fp);
4411 stat[statlen] = '\0';
4412 fclose(fp);
4413
4414 // Skip pid and the command string. Note that we could be dealing with
4415 // weird command names, e.g. user could decide to rename java launcher
4416 // to "java 1.4.2 :)", then the stat file would look like
4417 // 1234 (java 1.4.2 :)) R ... ...
4418 // We don't really need to know the command string, just find the last
4419 // occurrence of ")" and then start parsing from there. See bug 4726580.
4420 s = strrchr(stat, ')');
4421 i = 0;
4422 if (s == NULL ) return -1;
4423
4424 // Skip blank chars
4425 do s++; while (isspace(*s));
4426
4427 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4428 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4429 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4430 &user_time, &sys_time);
4431 if ( count != 13 ) return -1;
4432 if (user_sys_cpu_time) {
4433 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4434 } else {
4435 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4436 }
4437 }
4438
4439 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4440 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4441 info_ptr->may_skip_backward = false; // elapsed time not wall time
4442 info_ptr->may_skip_forward = false; // elapsed time not wall time
4443 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4444 }
4445
4446 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4447 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4448 info_ptr->may_skip_backward = false; // elapsed time not wall time
4449 info_ptr->may_skip_forward = false; // elapsed time not wall time
4450 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4451 }
4452
4453 bool os::is_thread_cpu_time_supported() {
4454 return true;
4455 }
4456
4457 // System loadavg support. Returns -1 if load average cannot be obtained.
4458 // Linux doesn't yet have a (official) notion of processor sets,
4459 // so just return the system wide load average.
4460 int os::loadavg(double loadavg[], int nelem) {
4461 return ::getloadavg(loadavg, nelem);
4462 }
4463
4464 void os::pause() {
4465 char filename[MAX_PATH];
4466 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4467 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4468 } else {
4469 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4470 }
4471
4472 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4473 if (fd != -1) {
4474 struct stat buf;
4475 close(fd);
4476 while (::stat(filename, &buf) == 0) {
4477 (void)::poll(NULL, 0, 100);
4478 }
4479 } else {
4480 jio_fprintf(stderr,
4481 "Could not open pause file '%s', continuing immediately.\n", filename);
4482 }
4483 }
4484
4485 extern "C" {
4486
4487 /**
4488 * NOTE: the following code is to keep the green threads code
4489 * in the libjava.so happy. Once the green threads is removed,
4490 * these code will no longer be needed.
4491 */
4492 int
4493 jdk_waitpid(pid_t pid, int* status, int options) {
4494 return waitpid(pid, status, options);
4495 }
4496
4497 int
4498 fork1() {
4499 return fork();
4500 }
4501
4502 int
4503 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4504 return sem_init(sem, pshared, value);
4505 }
4506
4507 int
4508 jdk_sem_post(sem_t *sem) {
4509 return sem_post(sem);
4510 }
4511
4512 int
4513 jdk_sem_wait(sem_t *sem) {
4514 return sem_wait(sem);
4515 }
4516
4517 int
4518 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4519 return pthread_sigmask(how , newmask, oldmask);
4520 }
4521
4522 }
4523
4524 // Refer to the comments in os_solaris.cpp park-unpark.
4525 //
4526 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4527 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4528 // For specifics regarding the bug see GLIBC BUGID 261237 :
4529 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4530 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4531 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4532 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4533 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4534 // and monitorenter when we're using 1-0 locking. All those operations may result in
4535 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4536 // of libpthread avoids the problem, but isn't practical.
4537 //
4538 // Possible remedies:
4539 //
4540 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4541 // This is palliative and probabilistic, however. If the thread is preempted
4542 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4543 // than the minimum period may have passed, and the abstime may be stale (in the
4544 // past) resultin in a hang. Using this technique reduces the odds of a hang
4545 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4546 //
4547 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4548 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4549 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4550 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4551 // thread.
4552 //
4553 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4554 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4555 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4556 // This also works well. In fact it avoids kernel-level scalability impediments
4557 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4558 // timers in a graceful fashion.
4559 //
4560 // 4. When the abstime value is in the past it appears that control returns
4561 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4562 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4563 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4564 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4565 // It may be possible to avoid reinitialization by checking the return
4566 // value from pthread_cond_timedwait(). In addition to reinitializing the
4567 // condvar we must establish the invariant that cond_signal() is only called
4568 // within critical sections protected by the adjunct mutex. This prevents
4569 // cond_signal() from "seeing" a condvar that's in the midst of being
4570 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4571 // desirable signal-after-unlock optimization that avoids futile context switching.
4572 //
4573 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4574 // structure when a condvar is used or initialized. cond_destroy() would
4575 // release the helper structure. Our reinitialize-after-timedwait fix
4576 // put excessive stress on malloc/free and locks protecting the c-heap.
4577 //
4578 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4579 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4580 // and only enabling the work-around for vulnerable environments.
4581
4582 // utility to compute the abstime argument to timedwait:
4583 // millis is the relative timeout time
4584 // abstime will be the absolute timeout time
4585 // TODO: replace compute_abstime() with unpackTime()
4586
4587 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4588 if (millis < 0) millis = 0;
4589 struct timeval now;
4590 int status = gettimeofday(&now, NULL);
4591 assert(status == 0, "gettimeofday");
4592 jlong seconds = millis / 1000;
4593 millis %= 1000;
4594 if (seconds > 50000000) { // see man cond_timedwait(3T)
4595 seconds = 50000000;
4596 }
4597 abstime->tv_sec = now.tv_sec + seconds;
4598 long usec = now.tv_usec + millis * 1000;
4599 if (usec >= 1000000) {
4600 abstime->tv_sec += 1;
4601 usec -= 1000000;
4602 }
4603 abstime->tv_nsec = usec * 1000;
4604 return abstime;
4605 }
4606
4607
4608 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4609 // Conceptually TryPark() should be equivalent to park(0).
4610
4611 int os::PlatformEvent::TryPark() {
4612 for (;;) {
4613 const int v = _Event ;
4614 guarantee ((v == 0) || (v == 1), "invariant") ;
4615 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4616 }
4617 }
4618
4619 void os::PlatformEvent::park() { // AKA "down()"
4620 // Invariant: Only the thread associated with the Event/PlatformEvent
4621 // may call park().
4622 // TODO: assert that _Assoc != NULL or _Assoc == Self
4623 int v ;
4624 for (;;) {
4625 v = _Event ;
4626 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4627 }
4628 guarantee (v >= 0, "invariant") ;
4629 if (v == 0) {
4630 // Do this the hard way by blocking ...
4631 int status = pthread_mutex_lock(_mutex);
4632 assert_status(status == 0, status, "mutex_lock");
4633 guarantee (_nParked == 0, "invariant") ;
4634 ++ _nParked ;
4635 while (_Event < 0) {
4636 status = pthread_cond_wait(_cond, _mutex);
4637 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4638 // Treat this the same as if the wait was interrupted
4639 if (status == ETIME) { status = EINTR; }
4640 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4641 }
4642 -- _nParked ;
4643
4644 // In theory we could move the ST of 0 into _Event past the unlock(),
4645 // but then we'd need a MEMBAR after the ST.
4646 _Event = 0 ;
4647 status = pthread_mutex_unlock(_mutex);
4648 assert_status(status == 0, status, "mutex_unlock");
4649 }
4650 guarantee (_Event >= 0, "invariant") ;
4651 }
4652
4653 int os::PlatformEvent::park(jlong millis) {
4654 guarantee (_nParked == 0, "invariant") ;
4655
4656 int v ;
4657 for (;;) {
4658 v = _Event ;
4659 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4660 }
4661 guarantee (v >= 0, "invariant") ;
4662 if (v != 0) return OS_OK ;
4663
4664 // We do this the hard way, by blocking the thread.
4665 // Consider enforcing a minimum timeout value.
4666 struct timespec abst;
4667 compute_abstime(&abst, millis);
4668
4669 int ret = OS_TIMEOUT;
4670 int status = pthread_mutex_lock(_mutex);
4671 assert_status(status == 0, status, "mutex_lock");
4672 guarantee (_nParked == 0, "invariant") ;
4673 ++_nParked ;
4674
4675 // Object.wait(timo) will return because of
4676 // (a) notification
4677 // (b) timeout
4678 // (c) thread.interrupt
4679 //
4680 // Thread.interrupt and object.notify{All} both call Event::set.
4681 // That is, we treat thread.interrupt as a special case of notification.
4682 // The underlying Solaris implementation, cond_timedwait, admits
4683 // spurious/premature wakeups, but the JLS/JVM spec prevents the
4684 // JVM from making those visible to Java code. As such, we must
4685 // filter out spurious wakeups. We assume all ETIME returns are valid.
4686 //
4687 // TODO: properly differentiate simultaneous notify+interrupt.
4688 // In that case, we should propagate the notify to another waiter.
4689
4690 while (_Event < 0) {
4691 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4692 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4693 pthread_cond_destroy (_cond);
4694 pthread_cond_init (_cond, NULL) ;
4695 }
4696 assert_status(status == 0 || status == EINTR ||
4697 status == ETIME || status == ETIMEDOUT,
4698 status, "cond_timedwait");
4699 if (!FilterSpuriousWakeups) break ; // previous semantics
4700 if (status == ETIME || status == ETIMEDOUT) break ;
4701 // We consume and ignore EINTR and spurious wakeups.
4702 }
4703 --_nParked ;
4704 if (_Event >= 0) {
4705 ret = OS_OK;
4706 }
4707 _Event = 0 ;
4708 status = pthread_mutex_unlock(_mutex);
4709 assert_status(status == 0, status, "mutex_unlock");
4710 assert (_nParked == 0, "invariant") ;
4711 return ret;
4712 }
4713
4714 void os::PlatformEvent::unpark() {
4715 int v, AnyWaiters ;
4716 for (;;) {
4717 v = _Event ;
4718 if (v > 0) {
4719 // The LD of _Event could have reordered or be satisfied
4720 // by a read-aside from this processor's write buffer.
4721 // To avoid problems execute a barrier and then
4722 // ratify the value.
4723 OrderAccess::fence() ;
4724 if (_Event == v) return ;
4725 continue ;
4726 }
4727 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4728 }
4729 if (v < 0) {
4730 // Wait for the thread associated with the event to vacate
4731 int status = pthread_mutex_lock(_mutex);
4732 assert_status(status == 0, status, "mutex_lock");
4733 AnyWaiters = _nParked ;
4734 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4735 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4736 AnyWaiters = 0 ;
4737 pthread_cond_signal (_cond);
4738 }
4739 status = pthread_mutex_unlock(_mutex);
4740 assert_status(status == 0, status, "mutex_unlock");
4741 if (AnyWaiters != 0) {
4742 status = pthread_cond_signal(_cond);
4743 assert_status(status == 0, status, "cond_signal");
4744 }
4745 }
4746
4747 // Note that we signal() _after dropping the lock for "immortal" Events.
4748 // This is safe and avoids a common class of futile wakeups. In rare
4749 // circumstances this can cause a thread to return prematurely from
4750 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4751 // simply re-test the condition and re-park itself.
4752 }
4753
4754
4755 // JSR166
4756 // -------------------------------------------------------
4757
4758 /*
4759 * The solaris and linux implementations of park/unpark are fairly
4760 * conservative for now, but can be improved. They currently use a
4761 * mutex/condvar pair, plus a a count.
4762 * Park decrements count if > 0, else does a condvar wait. Unpark
4763 * sets count to 1 and signals condvar. Only one thread ever waits
4764 * on the condvar. Contention seen when trying to park implies that someone
4765 * is unparking you, so don't wait. And spurious returns are fine, so there
4766 * is no need to track notifications.
4767 */
4768
4769
4770 #define NANOSECS_PER_SEC 1000000000
4771 #define NANOSECS_PER_MILLISEC 1000000
4772 #define MAX_SECS 100000000
4773 /*
4774 * This code is common to linux and solaris and will be moved to a
4775 * common place in dolphin.
4776 *
4777 * The passed in time value is either a relative time in nanoseconds
4778 * or an absolute time in milliseconds. Either way it has to be unpacked
4779 * into suitable seconds and nanoseconds components and stored in the
4780 * given timespec structure.
4781 * Given time is a 64-bit value and the time_t used in the timespec is only
4782 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4783 * overflow if times way in the future are given. Further on Solaris versions
4784 * prior to 10 there is a restriction (see cond_timedwait) that the specified
4785 * number of seconds, in abstime, is less than current_time + 100,000,000.
4786 * As it will be 28 years before "now + 100000000" will overflow we can
4787 * ignore overflow and just impose a hard-limit on seconds using the value
4788 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4789 * years from "now".
4790 */
4791
4792 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4793 assert (time > 0, "convertTime");
4794
4795 struct timeval now;
4796 int status = gettimeofday(&now, NULL);
4797 assert(status == 0, "gettimeofday");
4798
4799 time_t max_secs = now.tv_sec + MAX_SECS;
4800
4801 if (isAbsolute) {
4802 jlong secs = time / 1000;
4803 if (secs > max_secs) {
4804 absTime->tv_sec = max_secs;
4805 }
4806 else {
4807 absTime->tv_sec = secs;
4808 }
4809 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4810 }
4811 else {
4812 jlong secs = time / NANOSECS_PER_SEC;
4813 if (secs >= MAX_SECS) {
4814 absTime->tv_sec = max_secs;
4815 absTime->tv_nsec = 0;
4816 }
4817 else {
4818 absTime->tv_sec = now.tv_sec + secs;
4819 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4820 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4821 absTime->tv_nsec -= NANOSECS_PER_SEC;
4822 ++absTime->tv_sec; // note: this must be <= max_secs
4823 }
4824 }
4825 }
4826 assert(absTime->tv_sec >= 0, "tv_sec < 0");
4827 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4828 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4829 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4830 }
4831
4832 void Parker::park(bool isAbsolute, jlong time) {
4833 // Optional fast-path check:
4834 // Return immediately if a permit is available.
4835 if (_counter > 0) {
4836 _counter = 0 ;
4837 OrderAccess::fence();
4838 return ;
4839 }
4840
4841 Thread* thread = Thread::current();
4842 assert(thread->is_Java_thread(), "Must be JavaThread");
4843 JavaThread *jt = (JavaThread *)thread;
4844
4845 // Optional optimization -- avoid state transitions if there's an interrupt pending.
4846 // Check interrupt before trying to wait
4847 if (Thread::is_interrupted(thread, false)) {
4848 return;
4849 }
4850
4851 // Next, demultiplex/decode time arguments
4852 timespec absTime;
4853 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
4854 return;
4855 }
4856 if (time > 0) {
4857 unpackTime(&absTime, isAbsolute, time);
4858 }
4859
4860
4861 // Enter safepoint region
4862 // Beware of deadlocks such as 6317397.
4863 // The per-thread Parker:: mutex is a classic leaf-lock.
4864 // In particular a thread must never block on the Threads_lock while
4865 // holding the Parker:: mutex. If safepoints are pending both the
4866 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4867 ThreadBlockInVM tbivm(jt);
4868
4869 // Don't wait if cannot get lock since interference arises from
4870 // unblocking. Also. check interrupt before trying wait
4871 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4872 return;
4873 }
4874
4875 int status ;
4876 if (_counter > 0) { // no wait needed
4877 _counter = 0;
4878 status = pthread_mutex_unlock(_mutex);
4879 assert (status == 0, "invariant") ;
4880 OrderAccess::fence();
4881 return;
4882 }
4883
4884 #ifdef ASSERT
4885 // Don't catch signals while blocked; let the running threads have the signals.
4886 // (This allows a debugger to break into the running thread.)
4887 sigset_t oldsigs;
4888 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4889 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4890 #endif
4891
4892 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4893 jt->set_suspend_equivalent();
4894 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4895
4896 if (time == 0) {
4897 status = pthread_cond_wait (_cond, _mutex) ;
4898 } else {
4899 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4900 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4901 pthread_cond_destroy (_cond) ;
4902 pthread_cond_init (_cond, NULL);
4903 }
4904 }
4905 assert_status(status == 0 || status == EINTR ||
4906 status == ETIME || status == ETIMEDOUT,
4907 status, "cond_timedwait");
4908
4909 #ifdef ASSERT
4910 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4911 #endif
4912
4913 _counter = 0 ;
4914 status = pthread_mutex_unlock(_mutex) ;
4915 assert_status(status == 0, status, "invariant") ;
4916 // If externally suspended while waiting, re-suspend
4917 if (jt->handle_special_suspend_equivalent_condition()) {
4918 jt->java_suspend_self();
4919 }
4920
4921 OrderAccess::fence();
4922 }
4923
4924 void Parker::unpark() {
4925 int s, status ;
4926 status = pthread_mutex_lock(_mutex);
4927 assert (status == 0, "invariant") ;
4928 s = _counter;
4929 _counter = 1;
4930 if (s < 1) {
4931 if (WorkAroundNPTLTimedWaitHang) {
4932 status = pthread_cond_signal (_cond) ;
4933 assert (status == 0, "invariant") ;
4934 status = pthread_mutex_unlock(_mutex);
4935 assert (status == 0, "invariant") ;
4936 } else {
4937 status = pthread_mutex_unlock(_mutex);
4938 assert (status == 0, "invariant") ;
4939 status = pthread_cond_signal (_cond) ;
4940 assert (status == 0, "invariant") ;
4941 }
4942 } else {
4943 pthread_mutex_unlock(_mutex);
4944 assert (status == 0, "invariant") ;
4945 }
4946 }
4947
4948
4949 extern char** environ;
4950
4951 #ifndef __NR_fork
4952 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4953 #endif
4954
4955 #ifndef __NR_execve
4956 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4957 #endif
4958
4959 // Run the specified command in a separate process. Return its exit value,
4960 // or -1 on failure (e.g. can't fork a new process).
4961 // Unlike system(), this function can be called from signal handler. It
4962 // doesn't block SIGINT et al.
4963 int os::fork_and_exec(char* cmd) {
4964 const char * argv[4] = {"sh", "-c", cmd, NULL};
4965
4966 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4967 // pthread_atfork handlers and reset pthread library. All we need is a
4968 // separate process to execve. Make a direct syscall to fork process.
4969 // On IA64 there's no fork syscall, we have to use fork() and hope for
4970 // the best...
4971 pid_t pid = NOT_IA64(syscall(__NR_fork);)
4972 IA64_ONLY(fork();)
4973
4974 if (pid < 0) {
4975 // fork failed
4976 return -1;
4977
4978 } else if (pid == 0) {
4979 // child process
4980
4981 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4982 // first to kill every thread on the thread list. Because this list is
4983 // not reset by fork() (see notes above), execve() will instead kill
4984 // every thread in the parent process. We know this is the only thread
4985 // in the new process, so make a system call directly.
4986 // IA64 should use normal execve() from glibc to match the glibc fork()
4987 // above.
4988 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4989 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
4990
4991 // execve failed
4992 _exit(-1);
4993
4994 } else {
4995 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4996 // care about the actual exit code, for now.
4997
4998 int status;
4999
5000 // Wait for the child process to exit. This returns immediately if
5001 // the child has already exited. */
5002 while (waitpid(pid, &status, 0) < 0) {
5003 switch (errno) {
5004 case ECHILD: return 0;
5005 case EINTR: break;
5006 default: return -1;
5007 }
5008 }
5009
5010 if (WIFEXITED(status)) {
5011 // The child exited normally; get its exit code.
5012 return WEXITSTATUS(status);
5013 } else if (WIFSIGNALED(status)) {
5014 // The child exited because of a signal
5015 // The best value to return is 0x80 + signal number,
5016 // because that is what all Unix shells do, and because
5017 // it allows callers to distinguish between process exit and
5018 // process death by signal.
5019 return 0x80 + WTERMSIG(status);
5020 } else {
5021 // Unknown exit code; pass it through
5022 return status;
5023 }
5024 }
5025 }
5026
5027 // is_headless_jre()
5028 //
5029 // Test for the existence of libmawt in motif21 or xawt directories
5030 // in order to report if we are running in a headless jre
5031 //
5032 bool os::is_headless_jre() {
5033 struct stat statbuf;
5034 char buf[MAXPATHLEN];
5035 char libmawtpath[MAXPATHLEN];
5036 const char *xawtstr = "/xawt/libmawt.so";
5037 const char *motifstr = "/motif21/libmawt.so";
5038 char *p;
5039
5040 // Get path to libjvm.so
5041 os::jvm_path(buf, sizeof(buf));
5042
5043 // Get rid of libjvm.so
5044 p = strrchr(buf, '/');
5045 if (p == NULL) return false;
5046 else *p = '\0';
5047
5048 // Get rid of client or server
5049 p = strrchr(buf, '/');
5050 if (p == NULL) return false;
5051 else *p = '\0';
5052
5053 // check xawt/libmawt.so
5054 strcpy(libmawtpath, buf);
5055 strcat(libmawtpath, xawtstr);
5056 if (::stat(libmawtpath, &statbuf) == 0) return false;
5057
5058 // check motif21/libmawt.so
5059 strcpy(libmawtpath, buf);
5060 strcat(libmawtpath, motifstr);
5061 if (::stat(libmawtpath, &statbuf) == 0) return false;
5062
5063 return true;
5064 }
5065