1 /*
2 * Copyright 1999-2008 Sun Microsystems, Inc. All Rights Reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
24
25 // do not include precompiled header file
26 # include "incls/_os_linux.cpp.incl"
27
28 // put OS-includes here
29 # include <sys/types.h>
30 # include <sys/mman.h>
31 # include <pthread.h>
32 # include <signal.h>
33 # include <errno.h>
34 # include <dlfcn.h>
35 # include <stdio.h>
36 # include <unistd.h>
37 # include <sys/resource.h>
38 # include <pthread.h>
39 # include <sys/stat.h>
40 # include <sys/time.h>
41 # include <sys/times.h>
42 # include <sys/utsname.h>
43 # include <sys/socket.h>
44 # include <sys/wait.h>
45 # include <pwd.h>
46 # include <poll.h>
47 # include <semaphore.h>
48 # include <fcntl.h>
49 # include <string.h>
50 # include <syscall.h>
51 # include <sys/sysinfo.h>
52 # include <gnu/libc-version.h>
53 # include <sys/ipc.h>
54 # include <sys/shm.h>
55 # include <link.h>
56
57 #define MAX_PATH (2 * K)
58
59 // for timer info max values which include all bits
60 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
61 #define SEC_IN_NANOSECS 1000000000LL
62
63 ////////////////////////////////////////////////////////////////////////////////
64 // global variables
65 julong os::Linux::_physical_memory = 0;
66
67 address os::Linux::_initial_thread_stack_bottom = NULL;
68 uintptr_t os::Linux::_initial_thread_stack_size = 0;
69
70 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
71 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
72 Mutex* os::Linux::_createThread_lock = NULL;
73 pthread_t os::Linux::_main_thread;
74 int os::Linux::_page_size = -1;
75 bool os::Linux::_is_floating_stack = false;
76 bool os::Linux::_is_NPTL = false;
77 bool os::Linux::_supports_fast_thread_cpu_time = false;
78 const char * os::Linux::_glibc_version = NULL;
79 const char * os::Linux::_libpthread_version = NULL;
80
81 static jlong initial_time_count=0;
82
83 static int clock_tics_per_sec = 100;
84
85 // For diagnostics to print a message once. see run_periodic_checks
86 static sigset_t check_signal_done;
87 static bool check_signals = true;;
88
89 static pid_t _initial_pid = 0;
90
91 /* Signal number used to suspend/resume a thread */
92
93 /* do not use any signal number less than SIGSEGV, see 4355769 */
94 static int SR_signum = SIGUSR2;
95 sigset_t SR_sigset;
96
97 /* Used to protect dlsym() calls */
98 static pthread_mutex_t dl_mutex;
99
100 ////////////////////////////////////////////////////////////////////////////////
101 // utility functions
102
103 static int SR_initialize();
104 static int SR_finalize();
105
106 julong os::available_memory() {
107 return Linux::available_memory();
108 }
109
110 julong os::Linux::available_memory() {
111 // values in struct sysinfo are "unsigned long"
112 struct sysinfo si;
113 sysinfo(&si);
114
115 return (julong)si.freeram * si.mem_unit;
116 }
117
118 julong os::physical_memory() {
119 return Linux::physical_memory();
120 }
121
122 julong os::allocatable_physical_memory(julong size) {
123 #ifdef _LP64
124 return size;
125 #else
126 julong result = MIN2(size, (julong)3800*M);
127 if (!is_allocatable(result)) {
128 // See comments under solaris for alignment considerations
129 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
130 result = MIN2(size, reasonable_size);
131 }
132 return result;
133 #endif // _LP64
134 }
135
136 ////////////////////////////////////////////////////////////////////////////////
137 // environment support
138
139 bool os::getenv(const char* name, char* buf, int len) {
140 const char* val = ::getenv(name);
141 if (val != NULL && strlen(val) < (size_t)len) {
142 strcpy(buf, val);
143 return true;
144 }
145 if (len > 0) buf[0] = 0; // return a null string
146 return false;
147 }
148
149
150 // Return true if user is running as root.
151
152 bool os::have_special_privileges() {
153 static bool init = false;
154 static bool privileges = false;
155 if (!init) {
156 privileges = (getuid() != geteuid()) || (getgid() != getegid());
157 init = true;
158 }
159 return privileges;
160 }
161
162
163 #ifndef SYS_gettid
164 // i386: 224, ia64: 1105, amd64: 186, sparc 143
165 #ifdef __ia64__
166 #define SYS_gettid 1105
167 #elif __i386__
168 #define SYS_gettid 224
169 #elif __amd64__
170 #define SYS_gettid 186
171 #elif __sparc__
172 #define SYS_gettid 143
173 #else
174 #error define gettid for the arch
175 #endif
176 #endif
177
178 // Cpu architecture string
179 #if defined(IA64)
180 static char cpu_arch[] = "ia64";
181 #elif defined(IA32)
182 static char cpu_arch[] = "i386";
183 #elif defined(AMD64)
184 static char cpu_arch[] = "amd64";
185 #elif defined(SPARC)
186 # ifdef _LP64
187 static char cpu_arch[] = "sparcv9";
188 # else
189 static char cpu_arch[] = "sparc";
190 # endif
191 #else
192 #error Add appropriate cpu_arch setting
193 #endif
194
195
196 // pid_t gettid()
197 //
198 // Returns the kernel thread id of the currently running thread. Kernel
199 // thread id is used to access /proc.
200 //
201 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
202 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
203 //
204 pid_t os::Linux::gettid() {
205 int rslt = syscall(SYS_gettid);
206 if (rslt == -1) {
207 // old kernel, no NPTL support
208 return getpid();
209 } else {
210 return (pid_t)rslt;
211 }
212 }
213
214 // Most versions of linux have a bug where the number of processors are
215 // determined by looking at the /proc file system. In a chroot environment,
216 // the system call returns 1. This causes the VM to act as if it is
217 // a single processor and elide locking (see is_MP() call).
218 static bool unsafe_chroot_detected = false;
219 static const char *unstable_chroot_error = "/proc file system not found.\n"
220 "Java may be unstable running multithreaded in a chroot "
221 "environment on Linux when /proc filesystem is not mounted.";
222
223 void os::Linux::initialize_system_info() {
224 _processor_count = sysconf(_SC_NPROCESSORS_CONF);
225 if (_processor_count == 1) {
226 pid_t pid = os::Linux::gettid();
227 char fname[32];
228 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
229 FILE *fp = fopen(fname, "r");
230 if (fp == NULL) {
231 unsafe_chroot_detected = true;
232 } else {
233 fclose(fp);
234 }
235 }
236 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
237 assert(_processor_count > 0, "linux error");
238 }
239
240 void os::init_system_properties_values() {
241 // char arch[12];
242 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
243
244 // The next steps are taken in the product version:
245 //
246 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
247 // This library should be located at:
248 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
249 //
250 // If "/jre/lib/" appears at the right place in the path, then we
251 // assume libjvm[_g].so is installed in a JDK and we use this path.
252 //
253 // Otherwise exit with message: "Could not create the Java virtual machine."
254 //
255 // The following extra steps are taken in the debugging version:
256 //
257 // If "/jre/lib/" does NOT appear at the right place in the path
258 // instead of exit check for $JAVA_HOME environment variable.
259 //
260 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
261 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
262 // it looks like libjvm[_g].so is installed there
263 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
264 //
265 // Otherwise exit.
266 //
267 // Important note: if the location of libjvm.so changes this
268 // code needs to be changed accordingly.
269
270 // The next few definitions allow the code to be verbatim:
271 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
272 #define getenv(n) ::getenv(n)
273
274 /*
275 * See ld(1):
276 * The linker uses the following search paths to locate required
277 * shared libraries:
278 * 1: ...
279 * ...
280 * 7: The default directories, normally /lib and /usr/lib.
281 */
282
283 #if defined(AMD64) || defined(_LP64) && (defined(PPC) || defined(S390))
284 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
285 #else
286 #define DEFAULT_LIBPATH "/lib:/usr/lib"
287 #endif
288
289 #define EXTENSIONS_DIR "/lib/ext"
290 #define ENDORSED_DIR "/lib/endorsed"
291 #define REG_DIR "/usr/java/packages"
292
293 {
294 /* sysclasspath, java_home, dll_dir */
295 {
296 char *home_path;
297 char *dll_path;
298 char *pslash;
299 char buf[MAXPATHLEN];
300 os::jvm_path(buf, sizeof(buf));
301
302 // Found the full path to libjvm.so.
303 // Now cut the path to <java_home>/jre if we can.
304 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
305 pslash = strrchr(buf, '/');
306 if (pslash != NULL)
307 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
308 dll_path = malloc(strlen(buf) + 1);
309 if (dll_path == NULL)
310 return;
311 strcpy(dll_path, buf);
312 Arguments::set_dll_dir(dll_path);
313
314 if (pslash != NULL) {
315 pslash = strrchr(buf, '/');
316 if (pslash != NULL) {
317 *pslash = '\0'; /* get rid of /<arch> */
318 pslash = strrchr(buf, '/');
319 if (pslash != NULL)
320 *pslash = '\0'; /* get rid of /lib */
321 }
322 }
323
324 home_path = malloc(strlen(buf) + 1);
325 if (home_path == NULL)
326 return;
327 strcpy(home_path, buf);
328 Arguments::set_java_home(home_path);
329
330 if (!set_boot_path('/', ':'))
331 return;
332 }
333
334 /*
335 * Where to look for native libraries
336 *
337 * Note: Due to a legacy implementation, most of the library path
338 * is set in the launcher. This was to accomodate linking restrictions
339 * on legacy Linux implementations (which are no longer supported).
340 * Eventually, all the library path setting will be done here.
341 *
342 * However, to prevent the proliferation of improperly built native
343 * libraries, the new path component /usr/java/packages is added here.
344 * Eventually, all the library path setting will be done here.
345 */
346 {
347 char *ld_library_path;
348
349 /*
350 * Construct the invariant part of ld_library_path. Note that the
351 * space for the colon and the trailing null are provided by the
352 * nulls included by the sizeof operator (so actually we allocate
353 * a byte more than necessary).
354 */
355 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
356 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
357 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
358
359 /*
360 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
361 * should always exist (until the legacy problem cited above is
362 * addressed).
363 */
364 char *v = getenv("LD_LIBRARY_PATH");
365 if (v != NULL) {
366 char *t = ld_library_path;
367 /* That's +1 for the colon and +1 for the trailing '\0' */
368 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
369 sprintf(ld_library_path, "%s:%s", v, t);
370 }
371 Arguments::set_library_path(ld_library_path);
372 }
373
374 /*
375 * Extensions directories.
376 *
377 * Note that the space for the colon and the trailing null are provided
378 * by the nulls included by the sizeof operator (so actually one byte more
379 * than necessary is allocated).
380 */
381 {
382 char *buf = malloc(strlen(Arguments::get_java_home()) +
383 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
384 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
385 Arguments::get_java_home());
386 Arguments::set_ext_dirs(buf);
387 }
388
389 /* Endorsed standards default directory. */
390 {
391 char * buf;
392 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
393 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
394 Arguments::set_endorsed_dirs(buf);
395 }
396 }
397
398 #undef malloc
399 #undef getenv
400 #undef EXTENSIONS_DIR
401 #undef ENDORSED_DIR
402
403 // Done
404 return;
405 }
406
407 ////////////////////////////////////////////////////////////////////////////////
408 // breakpoint support
409
410 void os::breakpoint() {
411 BREAKPOINT;
412 }
413
414 extern "C" void breakpoint() {
415 // use debugger to set breakpoint here
416 }
417
418 ////////////////////////////////////////////////////////////////////////////////
419 // signal support
420
421 debug_only(static bool signal_sets_initialized = false);
422 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
423
424 bool os::Linux::is_sig_ignored(int sig) {
425 struct sigaction oact;
426 sigaction(sig, (struct sigaction*)NULL, &oact);
427 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
428 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
429 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
430 return true;
431 else
432 return false;
433 }
434
435 void os::Linux::signal_sets_init() {
436 // Should also have an assertion stating we are still single-threaded.
437 assert(!signal_sets_initialized, "Already initialized");
438 // Fill in signals that are necessarily unblocked for all threads in
439 // the VM. Currently, we unblock the following signals:
440 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
441 // by -Xrs (=ReduceSignalUsage));
442 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
443 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
444 // the dispositions or masks wrt these signals.
445 // Programs embedding the VM that want to use the above signals for their
446 // own purposes must, at this time, use the "-Xrs" option to prevent
447 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
448 // (See bug 4345157, and other related bugs).
449 // In reality, though, unblocking these signals is really a nop, since
450 // these signals are not blocked by default.
451 sigemptyset(&unblocked_sigs);
452 sigemptyset(&allowdebug_blocked_sigs);
453 sigaddset(&unblocked_sigs, SIGILL);
454 sigaddset(&unblocked_sigs, SIGSEGV);
455 sigaddset(&unblocked_sigs, SIGBUS);
456 sigaddset(&unblocked_sigs, SIGFPE);
457 sigaddset(&unblocked_sigs, SR_signum);
458
459 if (!ReduceSignalUsage) {
460 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
461 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
462 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
463 }
464 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
465 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
466 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
467 }
468 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
469 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
470 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
471 }
472 }
473 // Fill in signals that are blocked by all but the VM thread.
474 sigemptyset(&vm_sigs);
475 if (!ReduceSignalUsage)
476 sigaddset(&vm_sigs, BREAK_SIGNAL);
477 debug_only(signal_sets_initialized = true);
478
479 }
480
481 // These are signals that are unblocked while a thread is running Java.
482 // (For some reason, they get blocked by default.)
483 sigset_t* os::Linux::unblocked_signals() {
484 assert(signal_sets_initialized, "Not initialized");
485 return &unblocked_sigs;
486 }
487
488 // These are the signals that are blocked while a (non-VM) thread is
489 // running Java. Only the VM thread handles these signals.
490 sigset_t* os::Linux::vm_signals() {
491 assert(signal_sets_initialized, "Not initialized");
492 return &vm_sigs;
493 }
494
495 // These are signals that are blocked during cond_wait to allow debugger in
496 sigset_t* os::Linux::allowdebug_blocked_signals() {
497 assert(signal_sets_initialized, "Not initialized");
498 return &allowdebug_blocked_sigs;
499 }
500
501 void os::Linux::hotspot_sigmask(Thread* thread) {
502
503 //Save caller's signal mask before setting VM signal mask
504 sigset_t caller_sigmask;
505 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
506
507 OSThread* osthread = thread->osthread();
508 osthread->set_caller_sigmask(caller_sigmask);
509
510 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
511
512 if (!ReduceSignalUsage) {
513 if (thread->is_VM_thread()) {
514 // Only the VM thread handles BREAK_SIGNAL ...
515 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
516 } else {
517 // ... all other threads block BREAK_SIGNAL
518 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
519 }
520 }
521 }
522
523 //////////////////////////////////////////////////////////////////////////////
524 // detecting pthread library
525
526 void os::Linux::libpthread_init() {
527 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
528 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
529 // generic name for earlier versions.
530 // Define macros here so we can build HotSpot on old systems.
531 # ifndef _CS_GNU_LIBC_VERSION
532 # define _CS_GNU_LIBC_VERSION 2
533 # endif
534 # ifndef _CS_GNU_LIBPTHREAD_VERSION
535 # define _CS_GNU_LIBPTHREAD_VERSION 3
536 # endif
537
538 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
539 if (n > 0) {
540 char *str = (char *)malloc(n);
541 confstr(_CS_GNU_LIBC_VERSION, str, n);
542 os::Linux::set_glibc_version(str);
543 } else {
544 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
545 static char _gnu_libc_version[32];
546 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
547 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
548 os::Linux::set_glibc_version(_gnu_libc_version);
549 }
550
551 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
552 if (n > 0) {
553 char *str = (char *)malloc(n);
554 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
555 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
556 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
557 // is the case. LinuxThreads has a hard limit on max number of threads.
558 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
559 // On the other hand, NPTL does not have such a limit, sysconf()
560 // will return -1 and errno is not changed. Check if it is really NPTL.
561 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
562 strstr(str, "NPTL") &&
563 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
564 free(str);
565 os::Linux::set_libpthread_version("linuxthreads");
566 } else {
567 os::Linux::set_libpthread_version(str);
568 }
569 } else {
570 // glibc before 2.3.2 only has LinuxThreads.
571 os::Linux::set_libpthread_version("linuxthreads");
572 }
573
574 if (strstr(libpthread_version(), "NPTL")) {
575 os::Linux::set_is_NPTL();
576 } else {
577 os::Linux::set_is_LinuxThreads();
578 }
579
580 // LinuxThreads have two flavors: floating-stack mode, which allows variable
581 // stack size; and fixed-stack mode. NPTL is always floating-stack.
582 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
583 os::Linux::set_is_floating_stack();
584 }
585 }
586
587 /////////////////////////////////////////////////////////////////////////////
588 // thread stack
589
590 // Force Linux kernel to expand current thread stack. If "bottom" is close
591 // to the stack guard, caller should block all signals.
592 //
593 // MAP_GROWSDOWN:
594 // A special mmap() flag that is used to implement thread stacks. It tells
595 // kernel that the memory region should extend downwards when needed. This
596 // allows early versions of LinuxThreads to only mmap the first few pages
597 // when creating a new thread. Linux kernel will automatically expand thread
598 // stack as needed (on page faults).
599 //
600 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
601 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
602 // region, it's hard to tell if the fault is due to a legitimate stack
603 // access or because of reading/writing non-exist memory (e.g. buffer
604 // overrun). As a rule, if the fault happens below current stack pointer,
605 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
606 // application (see Linux kernel fault.c).
607 //
608 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
609 // stack overflow detection.
610 //
611 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
612 // not use this flag. However, the stack of initial thread is not created
613 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
614 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
615 // and then attach the thread to JVM.
616 //
617 // To get around the problem and allow stack banging on Linux, we need to
618 // manually expand thread stack after receiving the SIGSEGV.
619 //
620 // There are two ways to expand thread stack to address "bottom", we used
621 // both of them in JVM before 1.5:
622 // 1. adjust stack pointer first so that it is below "bottom", and then
623 // touch "bottom"
624 // 2. mmap() the page in question
625 //
626 // Now alternate signal stack is gone, it's harder to use 2. For instance,
627 // if current sp is already near the lower end of page 101, and we need to
628 // call mmap() to map page 100, it is possible that part of the mmap() frame
629 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
630 // That will destroy the mmap() frame and cause VM to crash.
631 //
632 // The following code works by adjusting sp first, then accessing the "bottom"
633 // page to force a page fault. Linux kernel will then automatically expand the
634 // stack mapping.
635 //
636 // _expand_stack_to() assumes its frame size is less than page size, which
637 // should always be true if the function is not inlined.
638
639 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
640 #define NOINLINE
641 #else
642 #define NOINLINE __attribute__ ((noinline))
643 #endif
644
645 static void _expand_stack_to(address bottom) NOINLINE;
646
647 static void _expand_stack_to(address bottom) {
648 address sp;
649 size_t size;
650 volatile char *p;
651
652 // Adjust bottom to point to the largest address within the same page, it
653 // gives us a one-page buffer if alloca() allocates slightly more memory.
654 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
655 bottom += os::Linux::page_size() - 1;
656
657 // sp might be slightly above current stack pointer; if that's the case, we
658 // will alloca() a little more space than necessary, which is OK. Don't use
659 // os::current_stack_pointer(), as its result can be slightly below current
660 // stack pointer, causing us to not alloca enough to reach "bottom".
661 sp = (address)&sp;
662
663 if (sp > bottom) {
664 size = sp - bottom;
665 p = (volatile char *)alloca(size);
666 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
667 p[0] = '\0';
668 }
669 }
670
671 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
672 assert(t!=NULL, "just checking");
673 assert(t->osthread()->expanding_stack(), "expand should be set");
674 assert(t->stack_base() != NULL, "stack_base was not initialized");
675
676 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
677 sigset_t mask_all, old_sigset;
678 sigfillset(&mask_all);
679 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
680 _expand_stack_to(addr);
681 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
682 return true;
683 }
684 return false;
685 }
686
687 //////////////////////////////////////////////////////////////////////////////
688 // create new thread
689
690 static address highest_vm_reserved_address();
691
692 // check if it's safe to start a new thread
693 static bool _thread_safety_check(Thread* thread) {
694 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
695 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
696 // Heap is mmap'ed at lower end of memory space. Thread stacks are
697 // allocated (MAP_FIXED) from high address space. Every thread stack
698 // occupies a fixed size slot (usually 2Mbytes, but user can change
699 // it to other values if they rebuild LinuxThreads).
700 //
701 // Problem with MAP_FIXED is that mmap() can still succeed even part of
702 // the memory region has already been mmap'ed. That means if we have too
703 // many threads and/or very large heap, eventually thread stack will
704 // collide with heap.
705 //
706 // Here we try to prevent heap/stack collision by comparing current
707 // stack bottom with the highest address that has been mmap'ed by JVM
708 // plus a safety margin for memory maps created by native code.
709 //
710 // This feature can be disabled by setting ThreadSafetyMargin to 0
711 //
712 if (ThreadSafetyMargin > 0) {
713 address stack_bottom = os::current_stack_base() - os::current_stack_size();
714
715 // not safe if our stack extends below the safety margin
716 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
717 } else {
718 return true;
719 }
720 } else {
721 // Floating stack LinuxThreads or NPTL:
722 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
723 // there's not enough space left, pthread_create() will fail. If we come
724 // here, that means enough space has been reserved for stack.
725 return true;
726 }
727 }
728
729 // Thread start routine for all newly created threads
730 static void *java_start(Thread *thread) {
731 // Try to randomize the cache line index of hot stack frames.
732 // This helps when threads of the same stack traces evict each other's
733 // cache lines. The threads can be either from the same JVM instance, or
734 // from different JVM instances. The benefit is especially true for
735 // processors with hyperthreading technology.
736 static int counter = 0;
737 int pid = os::current_process_id();
738 alloca(((pid ^ counter++) & 7) * 128);
739
740 ThreadLocalStorage::set_thread(thread);
741
742 OSThread* osthread = thread->osthread();
743 Monitor* sync = osthread->startThread_lock();
744
745 // non floating stack LinuxThreads needs extra check, see above
746 if (!_thread_safety_check(thread)) {
747 // notify parent thread
748 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
749 osthread->set_state(ZOMBIE);
750 sync->notify_all();
751 return NULL;
752 }
753
754 // thread_id is kernel thread id (similar to Solaris LWP id)
755 osthread->set_thread_id(os::Linux::gettid());
756
757 if (UseNUMA) {
758 int lgrp_id = os::numa_get_group_id();
759 if (lgrp_id != -1) {
760 thread->set_lgrp_id(lgrp_id);
761 }
762 }
763 // initialize signal mask for this thread
764 os::Linux::hotspot_sigmask(thread);
765
766 // initialize floating point control register
767 os::Linux::init_thread_fpu_state();
768
769 // handshaking with parent thread
770 {
771 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
772
773 // notify parent thread
774 osthread->set_state(INITIALIZED);
775 sync->notify_all();
776
777 // wait until os::start_thread()
778 while (osthread->get_state() == INITIALIZED) {
779 sync->wait(Mutex::_no_safepoint_check_flag);
780 }
781 }
782
783 // call one more level start routine
784 thread->run();
785
786 return 0;
787 }
788
789 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
790 assert(thread->osthread() == NULL, "caller responsible");
791
792 // Allocate the OSThread object
793 OSThread* osthread = new OSThread(NULL, NULL);
794 if (osthread == NULL) {
795 return false;
796 }
797
798 // set the correct thread state
799 osthread->set_thread_type(thr_type);
800
801 // Initial state is ALLOCATED but not INITIALIZED
802 osthread->set_state(ALLOCATED);
803
804 thread->set_osthread(osthread);
805
806 // init thread attributes
807 pthread_attr_t attr;
808 pthread_attr_init(&attr);
809 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
810
811 // stack size
812 if (os::Linux::supports_variable_stack_size()) {
813 // calculate stack size if it's not specified by caller
814 if (stack_size == 0) {
815 stack_size = os::Linux::default_stack_size(thr_type);
816
817 switch (thr_type) {
818 case os::java_thread:
819 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
820 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
821 break;
822 case os::compiler_thread:
823 if (CompilerThreadStackSize > 0) {
824 stack_size = (size_t)(CompilerThreadStackSize * K);
825 break;
826 } // else fall through:
827 // use VMThreadStackSize if CompilerThreadStackSize is not defined
828 case os::vm_thread:
829 case os::pgc_thread:
830 case os::cgc_thread:
831 case os::watcher_thread:
832 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
833 break;
834 }
835 }
836
837 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
838 pthread_attr_setstacksize(&attr, stack_size);
839 } else {
840 // let pthread_create() pick the default value.
841 }
842
843 // glibc guard page
844 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
845
846 ThreadState state;
847
848 {
849 // Serialize thread creation if we are running with fixed stack LinuxThreads
850 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
851 if (lock) {
852 os::Linux::createThread_lock()->lock_without_safepoint_check();
853 }
854
855 pthread_t tid;
856 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
857
858 pthread_attr_destroy(&attr);
859
860 if (ret != 0) {
861 if (PrintMiscellaneous && (Verbose || WizardMode)) {
862 perror("pthread_create()");
863 }
864 // Need to clean up stuff we've allocated so far
865 thread->set_osthread(NULL);
866 delete osthread;
867 if (lock) os::Linux::createThread_lock()->unlock();
868 return false;
869 }
870
871 // Store pthread info into the OSThread
872 osthread->set_pthread_id(tid);
873
874 // Wait until child thread is either initialized or aborted
875 {
876 Monitor* sync_with_child = osthread->startThread_lock();
877 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
878 while ((state = osthread->get_state()) == ALLOCATED) {
879 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
880 }
881 }
882
883 if (lock) {
884 os::Linux::createThread_lock()->unlock();
885 }
886 }
887
888 // Aborted due to thread limit being reached
889 if (state == ZOMBIE) {
890 thread->set_osthread(NULL);
891 delete osthread;
892 return false;
893 }
894
895 // The thread is returned suspended (in state INITIALIZED),
896 // and is started higher up in the call chain
897 assert(state == INITIALIZED, "race condition");
898 return true;
899 }
900
901 /////////////////////////////////////////////////////////////////////////////
902 // attach existing thread
903
904 // bootstrap the main thread
905 bool os::create_main_thread(JavaThread* thread) {
906 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
907 return create_attached_thread(thread);
908 }
909
910 bool os::create_attached_thread(JavaThread* thread) {
911 #ifdef ASSERT
912 thread->verify_not_published();
913 #endif
914
915 // Allocate the OSThread object
916 OSThread* osthread = new OSThread(NULL, NULL);
917
918 if (osthread == NULL) {
919 return false;
920 }
921
922 // Store pthread info into the OSThread
923 osthread->set_thread_id(os::Linux::gettid());
924 osthread->set_pthread_id(::pthread_self());
925
926 // initialize floating point control register
927 os::Linux::init_thread_fpu_state();
928
929 // Initial thread state is RUNNABLE
930 osthread->set_state(RUNNABLE);
931
932 thread->set_osthread(osthread);
933
934 if (UseNUMA) {
935 int lgrp_id = os::numa_get_group_id();
936 if (lgrp_id != -1) {
937 thread->set_lgrp_id(lgrp_id);
938 }
939 }
940
941 if (os::Linux::is_initial_thread()) {
942 // If current thread is initial thread, its stack is mapped on demand,
943 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
944 // the entire stack region to avoid SEGV in stack banging.
945 // It is also useful to get around the heap-stack-gap problem on SuSE
946 // kernel (see 4821821 for details). We first expand stack to the top
947 // of yellow zone, then enable stack yellow zone (order is significant,
948 // enabling yellow zone first will crash JVM on SuSE Linux), so there
949 // is no gap between the last two virtual memory regions.
950
951 JavaThread *jt = (JavaThread *)thread;
952 address addr = jt->stack_yellow_zone_base();
953 assert(addr != NULL, "initialization problem?");
954 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
955
956 osthread->set_expanding_stack();
957 os::Linux::manually_expand_stack(jt, addr);
958 osthread->clear_expanding_stack();
959 }
960
961 // initialize signal mask for this thread
962 // and save the caller's signal mask
963 os::Linux::hotspot_sigmask(thread);
964
965 return true;
966 }
967
968 void os::pd_start_thread(Thread* thread) {
969 OSThread * osthread = thread->osthread();
970 assert(osthread->get_state() != INITIALIZED, "just checking");
971 Monitor* sync_with_child = osthread->startThread_lock();
972 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
973 sync_with_child->notify();
974 }
975
976 // Free Linux resources related to the OSThread
977 void os::free_thread(OSThread* osthread) {
978 assert(osthread != NULL, "osthread not set");
979
980 if (Thread::current()->osthread() == osthread) {
981 // Restore caller's signal mask
982 sigset_t sigmask = osthread->caller_sigmask();
983 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
984 }
985
986 delete osthread;
987 }
988
989 //////////////////////////////////////////////////////////////////////////////
990 // thread local storage
991
992 int os::allocate_thread_local_storage() {
993 pthread_key_t key;
994 int rslt = pthread_key_create(&key, NULL);
995 assert(rslt == 0, "cannot allocate thread local storage");
996 return (int)key;
997 }
998
999 // Note: This is currently not used by VM, as we don't destroy TLS key
1000 // on VM exit.
1001 void os::free_thread_local_storage(int index) {
1002 int rslt = pthread_key_delete((pthread_key_t)index);
1003 assert(rslt == 0, "invalid index");
1004 }
1005
1006 void os::thread_local_storage_at_put(int index, void* value) {
1007 int rslt = pthread_setspecific((pthread_key_t)index, value);
1008 assert(rslt == 0, "pthread_setspecific failed");
1009 }
1010
1011 extern "C" Thread* get_thread() {
1012 return ThreadLocalStorage::thread();
1013 }
1014
1015 //////////////////////////////////////////////////////////////////////////////
1016 // initial thread
1017
1018 // Check if current thread is the initial thread, similar to Solaris thr_main.
1019 bool os::Linux::is_initial_thread(void) {
1020 char dummy;
1021 // If called before init complete, thread stack bottom will be null.
1022 // Can be called if fatal error occurs before initialization.
1023 if (initial_thread_stack_bottom() == NULL) return false;
1024 assert(initial_thread_stack_bottom() != NULL &&
1025 initial_thread_stack_size() != 0,
1026 "os::init did not locate initial thread's stack region");
1027 if ((address)&dummy >= initial_thread_stack_bottom() &&
1028 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1029 return true;
1030 else return false;
1031 }
1032
1033 // Find the virtual memory area that contains addr
1034 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1035 FILE *fp = fopen("/proc/self/maps", "r");
1036 if (fp) {
1037 address low, high;
1038 while (!feof(fp)) {
1039 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1040 if (low <= addr && addr < high) {
1041 if (vma_low) *vma_low = low;
1042 if (vma_high) *vma_high = high;
1043 fclose (fp);
1044 return true;
1045 }
1046 }
1047 for (;;) {
1048 int ch = fgetc(fp);
1049 if (ch == EOF || ch == (int)'\n') break;
1050 }
1051 }
1052 fclose(fp);
1053 }
1054 return false;
1055 }
1056
1057 // Locate initial thread stack. This special handling of initial thread stack
1058 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1059 // bogus value for initial thread.
1060 void os::Linux::capture_initial_stack(size_t max_size) {
1061 // stack size is the easy part, get it from RLIMIT_STACK
1062 size_t stack_size;
1063 struct rlimit rlim;
1064 getrlimit(RLIMIT_STACK, &rlim);
1065 stack_size = rlim.rlim_cur;
1066
1067 // 6308388: a bug in ld.so will relocate its own .data section to the
1068 // lower end of primordial stack; reduce ulimit -s value a little bit
1069 // so we won't install guard page on ld.so's data section.
1070 stack_size -= 2 * page_size();
1071
1072 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1073 // 7.1, in both cases we will get 2G in return value.
1074 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1075 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1076 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1077 // in case other parts in glibc still assumes 2M max stack size.
1078 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1079 #ifndef IA64
1080 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1081 #else
1082 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1083 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1084 #endif
1085
1086 // Try to figure out where the stack base (top) is. This is harder.
1087 //
1088 // When an application is started, glibc saves the initial stack pointer in
1089 // a global variable "__libc_stack_end", which is then used by system
1090 // libraries. __libc_stack_end should be pretty close to stack top. The
1091 // variable is available since the very early days. However, because it is
1092 // a private interface, it could disappear in the future.
1093 //
1094 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1095 // to __libc_stack_end, it is very close to stack top, but isn't the real
1096 // stack top. Note that /proc may not exist if VM is running as a chroot
1097 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1098 // /proc/<pid>/stat could change in the future (though unlikely).
1099 //
1100 // We try __libc_stack_end first. If that doesn't work, look for
1101 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1102 // as a hint, which should work well in most cases.
1103
1104 uintptr_t stack_start;
1105
1106 // try __libc_stack_end first
1107 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1108 if (p && *p) {
1109 stack_start = *p;
1110 } else {
1111 // see if we can get the start_stack field from /proc/self/stat
1112 FILE *fp;
1113 int pid;
1114 char state;
1115 int ppid;
1116 int pgrp;
1117 int session;
1118 int nr;
1119 int tpgrp;
1120 unsigned long flags;
1121 unsigned long minflt;
1122 unsigned long cminflt;
1123 unsigned long majflt;
1124 unsigned long cmajflt;
1125 unsigned long utime;
1126 unsigned long stime;
1127 long cutime;
1128 long cstime;
1129 long prio;
1130 long nice;
1131 long junk;
1132 long it_real;
1133 uintptr_t start;
1134 uintptr_t vsize;
1135 uintptr_t rss;
1136 unsigned long rsslim;
1137 uintptr_t scodes;
1138 uintptr_t ecode;
1139 int i;
1140
1141 // Figure what the primordial thread stack base is. Code is inspired
1142 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1143 // followed by command name surrounded by parentheses, state, etc.
1144 char stat[2048];
1145 int statlen;
1146
1147 fp = fopen("/proc/self/stat", "r");
1148 if (fp) {
1149 statlen = fread(stat, 1, 2047, fp);
1150 stat[statlen] = '\0';
1151 fclose(fp);
1152
1153 // Skip pid and the command string. Note that we could be dealing with
1154 // weird command names, e.g. user could decide to rename java launcher
1155 // to "java 1.4.2 :)", then the stat file would look like
1156 // 1234 (java 1.4.2 :)) R ... ...
1157 // We don't really need to know the command string, just find the last
1158 // occurrence of ")" and then start parsing from there. See bug 4726580.
1159 char * s = strrchr(stat, ')');
1160
1161 i = 0;
1162 if (s) {
1163 // Skip blank chars
1164 do s++; while (isspace(*s));
1165
1166 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1167 /* 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 */
1168 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld %lu %lu %ld %lu %lu %lu %lu",
1169 &state, /* 3 %c */
1170 &ppid, /* 4 %d */
1171 &pgrp, /* 5 %d */
1172 &session, /* 6 %d */
1173 &nr, /* 7 %d */
1174 &tpgrp, /* 8 %d */
1175 &flags, /* 9 %lu */
1176 &minflt, /* 10 %lu */
1177 &cminflt, /* 11 %lu */
1178 &majflt, /* 12 %lu */
1179 &cmajflt, /* 13 %lu */
1180 &utime, /* 14 %lu */
1181 &stime, /* 15 %lu */
1182 &cutime, /* 16 %ld */
1183 &cstime, /* 17 %ld */
1184 &prio, /* 18 %ld */
1185 &nice, /* 19 %ld */
1186 &junk, /* 20 %ld */
1187 &it_real, /* 21 %ld */
1188 &start, /* 22 %lu */
1189 &vsize, /* 23 %lu */
1190 &rss, /* 24 %ld */
1191 &rsslim, /* 25 %lu */
1192 &scodes, /* 26 %lu */
1193 &ecode, /* 27 %lu */
1194 &stack_start); /* 28 %lu */
1195 }
1196
1197 if (i != 28 - 2) {
1198 assert(false, "Bad conversion from /proc/self/stat");
1199 // product mode - assume we are the initial thread, good luck in the
1200 // embedded case.
1201 warning("Can't detect initial thread stack location - bad conversion");
1202 stack_start = (uintptr_t) &rlim;
1203 }
1204 } else {
1205 // For some reason we can't open /proc/self/stat (for example, running on
1206 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1207 // most cases, so don't abort:
1208 warning("Can't detect initial thread stack location - no /proc/self/stat");
1209 stack_start = (uintptr_t) &rlim;
1210 }
1211 }
1212
1213 // Now we have a pointer (stack_start) very close to the stack top, the
1214 // next thing to do is to figure out the exact location of stack top. We
1215 // can find out the virtual memory area that contains stack_start by
1216 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1217 // and its upper limit is the real stack top. (again, this would fail if
1218 // running inside chroot, because /proc may not exist.)
1219
1220 uintptr_t stack_top;
1221 address low, high;
1222 if (find_vma((address)stack_start, &low, &high)) {
1223 // success, "high" is the true stack top. (ignore "low", because initial
1224 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1225 stack_top = (uintptr_t)high;
1226 } else {
1227 // failed, likely because /proc/self/maps does not exist
1228 warning("Can't detect initial thread stack location - find_vma failed");
1229 // best effort: stack_start is normally within a few pages below the real
1230 // stack top, use it as stack top, and reduce stack size so we won't put
1231 // guard page outside stack.
1232 stack_top = stack_start;
1233 stack_size -= 16 * page_size();
1234 }
1235
1236 // stack_top could be partially down the page so align it
1237 stack_top = align_size_up(stack_top, page_size());
1238
1239 if (max_size && stack_size > max_size) {
1240 _initial_thread_stack_size = max_size;
1241 } else {
1242 _initial_thread_stack_size = stack_size;
1243 }
1244
1245 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1246 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1247 }
1248
1249 ////////////////////////////////////////////////////////////////////////////////
1250 // time support
1251
1252 // Time since start-up in seconds to a fine granularity.
1253 // Used by VMSelfDestructTimer and the MemProfiler.
1254 double os::elapsedTime() {
1255
1256 return (double)(os::elapsed_counter()) * 0.000001;
1257 }
1258
1259 jlong os::elapsed_counter() {
1260 timeval time;
1261 int status = gettimeofday(&time, NULL);
1262 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1263 }
1264
1265 jlong os::elapsed_frequency() {
1266 return (1000 * 1000);
1267 }
1268
1269 // For now, we say that linux does not support vtime. I have no idea
1270 // whether it can actually be made to (DLD, 9/13/05).
1271
1272 bool os::supports_vtime() { return false; }
1273 bool os::enable_vtime() { return false; }
1274 bool os::vtime_enabled() { return false; }
1275 double os::elapsedVTime() {
1276 // better than nothing, but not much
1277 return elapsedTime();
1278 }
1279
1280 jlong os::javaTimeMillis() {
1281 timeval time;
1282 int status = gettimeofday(&time, NULL);
1283 assert(status != -1, "linux error");
1284 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1285 }
1286
1287 #ifndef CLOCK_MONOTONIC
1288 #define CLOCK_MONOTONIC (1)
1289 #endif
1290
1291 void os::Linux::clock_init() {
1292 // we do dlopen's in this particular order due to bug in linux
1293 // dynamical loader (see 6348968) leading to crash on exit
1294 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1295 if (handle == NULL) {
1296 handle = dlopen("librt.so", RTLD_LAZY);
1297 }
1298
1299 if (handle) {
1300 int (*clock_getres_func)(clockid_t, struct timespec*) =
1301 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1302 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1303 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1304 if (clock_getres_func && clock_gettime_func) {
1305 // See if monotonic clock is supported by the kernel. Note that some
1306 // early implementations simply return kernel jiffies (updated every
1307 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1308 // for nano time (though the monotonic property is still nice to have).
1309 // It's fixed in newer kernels, however clock_getres() still returns
1310 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1311 // resolution for now. Hopefully as people move to new kernels, this
1312 // won't be a problem.
1313 struct timespec res;
1314 struct timespec tp;
1315 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1316 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1317 // yes, monotonic clock is supported
1318 _clock_gettime = clock_gettime_func;
1319 } else {
1320 // close librt if there is no monotonic clock
1321 dlclose(handle);
1322 }
1323 }
1324 }
1325 }
1326
1327 #ifndef SYS_clock_getres
1328
1329 #if defined(IA32) || defined(AMD64)
1330 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1331 #else
1332 #error Value of SYS_clock_getres not known on this platform
1333 #endif
1334
1335 #endif
1336
1337 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1338
1339 void os::Linux::fast_thread_clock_init() {
1340 if (!UseLinuxPosixThreadCPUClocks) {
1341 return;
1342 }
1343 clockid_t clockid;
1344 struct timespec tp;
1345 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1346 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1347
1348 // Switch to using fast clocks for thread cpu time if
1349 // the sys_clock_getres() returns 0 error code.
1350 // Note, that some kernels may support the current thread
1351 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1352 // returned by the pthread_getcpuclockid().
1353 // If the fast Posix clocks are supported then the sys_clock_getres()
1354 // must return at least tp.tv_sec == 0 which means a resolution
1355 // better than 1 sec. This is extra check for reliability.
1356
1357 if(pthread_getcpuclockid_func &&
1358 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1359 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1360
1361 _supports_fast_thread_cpu_time = true;
1362 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1363 }
1364 }
1365
1366 jlong os::javaTimeNanos() {
1367 if (Linux::supports_monotonic_clock()) {
1368 struct timespec tp;
1369 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1370 assert(status == 0, "gettime error");
1371 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1372 return result;
1373 } else {
1374 timeval time;
1375 int status = gettimeofday(&time, NULL);
1376 assert(status != -1, "linux error");
1377 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1378 return 1000 * usecs;
1379 }
1380 }
1381
1382 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1383 if (Linux::supports_monotonic_clock()) {
1384 info_ptr->max_value = ALL_64_BITS;
1385
1386 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1387 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1388 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1389 } else {
1390 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1391 info_ptr->max_value = ALL_64_BITS;
1392
1393 // gettimeofday is a real time clock so it skips
1394 info_ptr->may_skip_backward = true;
1395 info_ptr->may_skip_forward = true;
1396 }
1397
1398 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1399 }
1400
1401 // Return the real, user, and system times in seconds from an
1402 // arbitrary fixed point in the past.
1403 bool os::getTimesSecs(double* process_real_time,
1404 double* process_user_time,
1405 double* process_system_time) {
1406 struct tms ticks;
1407 clock_t real_ticks = times(&ticks);
1408
1409 if (real_ticks == (clock_t) (-1)) {
1410 return false;
1411 } else {
1412 double ticks_per_second = (double) clock_tics_per_sec;
1413 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1414 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1415 *process_real_time = ((double) real_ticks) / ticks_per_second;
1416
1417 return true;
1418 }
1419 }
1420
1421
1422 char * os::local_time_string(char *buf, size_t buflen) {
1423 struct tm t;
1424 time_t long_time;
1425 time(&long_time);
1426 localtime_r(&long_time, &t);
1427 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1428 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1429 t.tm_hour, t.tm_min, t.tm_sec);
1430 return buf;
1431 }
1432
1433 ////////////////////////////////////////////////////////////////////////////////
1434 // runtime exit support
1435
1436 // Note: os::shutdown() might be called very early during initialization, or
1437 // called from signal handler. Before adding something to os::shutdown(), make
1438 // sure it is async-safe and can handle partially initialized VM.
1439 void os::shutdown() {
1440
1441 // allow PerfMemory to attempt cleanup of any persistent resources
1442 perfMemory_exit();
1443
1444 // needs to remove object in file system
1445 AttachListener::abort();
1446
1447 // flush buffered output, finish log files
1448 ostream_abort();
1449
1450 // Check for abort hook
1451 abort_hook_t abort_hook = Arguments::abort_hook();
1452 if (abort_hook != NULL) {
1453 abort_hook();
1454 }
1455
1456 }
1457
1458 // Note: os::abort() might be called very early during initialization, or
1459 // called from signal handler. Before adding something to os::abort(), make
1460 // sure it is async-safe and can handle partially initialized VM.
1461 void os::abort(bool dump_core) {
1462 os::shutdown();
1463 if (dump_core) {
1464 #ifndef PRODUCT
1465 fdStream out(defaultStream::output_fd());
1466 out.print_raw("Current thread is ");
1467 char buf[16];
1468 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1469 out.print_raw_cr(buf);
1470 out.print_raw_cr("Dumping core ...");
1471 #endif
1472 ::abort(); // dump core
1473 }
1474
1475 ::exit(1);
1476 }
1477
1478 // Die immediately, no exit hook, no abort hook, no cleanup.
1479 void os::die() {
1480 // _exit() on LinuxThreads only kills current thread
1481 ::abort();
1482 }
1483
1484 // unused on linux for now.
1485 void os::set_error_file(const char *logfile) {}
1486
1487 intx os::current_thread_id() { return (intx)pthread_self(); }
1488 int os::current_process_id() {
1489
1490 // Under the old linux thread library, linux gives each thread
1491 // its own process id. Because of this each thread will return
1492 // a different pid if this method were to return the result
1493 // of getpid(2). Linux provides no api that returns the pid
1494 // of the launcher thread for the vm. This implementation
1495 // returns a unique pid, the pid of the launcher thread
1496 // that starts the vm 'process'.
1497
1498 // Under the NPTL, getpid() returns the same pid as the
1499 // launcher thread rather than a unique pid per thread.
1500 // Use gettid() if you want the old pre NPTL behaviour.
1501
1502 // if you are looking for the result of a call to getpid() that
1503 // returns a unique pid for the calling thread, then look at the
1504 // OSThread::thread_id() method in osThread_linux.hpp file
1505
1506 return (int)(_initial_pid ? _initial_pid : getpid());
1507 }
1508
1509 // DLL functions
1510
1511 const char* os::dll_file_extension() { return ".so"; }
1512
1513 const char* os::get_temp_directory() { return "/tmp/"; }
1514
1515 void os::dll_build_name(
1516 char* buffer, size_t buflen, const char* pname, const char* fname) {
1517 // copied from libhpi
1518 const size_t pnamelen = pname ? strlen(pname) : 0;
1519
1520 /* Quietly truncate on buffer overflow. Should be an error. */
1521 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1522 *buffer = '\0';
1523 return;
1524 }
1525
1526 if (pnamelen == 0) {
1527 sprintf(buffer, "lib%s.so", fname);
1528 } else {
1529 sprintf(buffer, "%s/lib%s.so", pname, fname);
1530 }
1531 }
1532
1533 const char* os::get_current_directory(char *buf, int buflen) {
1534 return getcwd(buf, buflen);
1535 }
1536
1537 // check if addr is inside libjvm[_g].so
1538 bool os::address_is_in_vm(address addr) {
1539 static address libjvm_base_addr;
1540 Dl_info dlinfo;
1541
1542 if (libjvm_base_addr == NULL) {
1543 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1544 libjvm_base_addr = (address)dlinfo.dli_fbase;
1545 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1546 }
1547
1548 if (dladdr((void *)addr, &dlinfo)) {
1549 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1550 }
1551
1552 return false;
1553 }
1554
1555 bool os::dll_address_to_function_name(address addr, char *buf,
1556 int buflen, int *offset) {
1557 Dl_info dlinfo;
1558
1559 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1560 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1561 if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1562 return true;
1563 } else {
1564 if (buf) buf[0] = '\0';
1565 if (offset) *offset = -1;
1566 return false;
1567 }
1568 }
1569
1570 struct _address_to_library_name {
1571 address addr; // input : memory address
1572 size_t buflen; // size of fname
1573 char* fname; // output: library name
1574 address base; // library base addr
1575 };
1576
1577 static int address_to_library_name_callback(struct dl_phdr_info *info,
1578 size_t size, void *data) {
1579 int i;
1580 bool found = false;
1581 address libbase = NULL;
1582 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1583
1584 // iterate through all loadable segments
1585 for (i = 0; i < info->dlpi_phnum; i++) {
1586 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1587 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1588 // base address of a library is the lowest address of its loaded
1589 // segments.
1590 if (libbase == NULL || libbase > segbase) {
1591 libbase = segbase;
1592 }
1593 // see if 'addr' is within current segment
1594 if (segbase <= d->addr &&
1595 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1596 found = true;
1597 }
1598 }
1599 }
1600
1601 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1602 // so dll_address_to_library_name() can fall through to use dladdr() which
1603 // can figure out executable name from argv[0].
1604 if (found && info->dlpi_name && info->dlpi_name[0]) {
1605 d->base = libbase;
1606 if (d->fname) {
1607 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1608 }
1609 return 1;
1610 }
1611 return 0;
1612 }
1613
1614 bool os::dll_address_to_library_name(address addr, char* buf,
1615 int buflen, int* offset) {
1616 Dl_info dlinfo;
1617 struct _address_to_library_name data;
1618
1619 // There is a bug in old glibc dladdr() implementation that it could resolve
1620 // to wrong library name if the .so file has a base address != NULL. Here
1621 // we iterate through the program headers of all loaded libraries to find
1622 // out which library 'addr' really belongs to. This workaround can be
1623 // removed once the minimum requirement for glibc is moved to 2.3.x.
1624 data.addr = addr;
1625 data.fname = buf;
1626 data.buflen = buflen;
1627 data.base = NULL;
1628 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1629
1630 if (rslt) {
1631 // buf already contains library name
1632 if (offset) *offset = addr - data.base;
1633 return true;
1634 } else if (dladdr((void*)addr, &dlinfo)){
1635 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1636 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1637 return true;
1638 } else {
1639 if (buf) buf[0] = '\0';
1640 if (offset) *offset = -1;
1641 return false;
1642 }
1643 }
1644
1645 // Loads .dll/.so and
1646 // in case of error it checks if .dll/.so was built for the
1647 // same architecture as Hotspot is running on
1648
1649 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1650 {
1651 void * result= ::dlopen(filename, RTLD_LAZY);
1652 if (result != NULL) {
1653 // Successful loading
1654 return result;
1655 }
1656
1657 Elf32_Ehdr elf_head;
1658
1659 // Read system error message into ebuf
1660 // It may or may not be overwritten below
1661 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1662 ebuf[ebuflen-1]='\0';
1663 int diag_msg_max_length=ebuflen-strlen(ebuf);
1664 char* diag_msg_buf=ebuf+strlen(ebuf);
1665
1666 if (diag_msg_max_length==0) {
1667 // No more space in ebuf for additional diagnostics message
1668 return NULL;
1669 }
1670
1671
1672 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1673
1674 if (file_descriptor < 0) {
1675 // Can't open library, report dlerror() message
1676 return NULL;
1677 }
1678
1679 bool failed_to_read_elf_head=
1680 (sizeof(elf_head)!=
1681 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1682
1683 ::close(file_descriptor);
1684 if (failed_to_read_elf_head) {
1685 // file i/o error - report dlerror() msg
1686 return NULL;
1687 }
1688
1689 typedef struct {
1690 Elf32_Half code; // Actual value as defined in elf.h
1691 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1692 char elf_class; // 32 or 64 bit
1693 char endianess; // MSB or LSB
1694 char* name; // String representation
1695 } arch_t;
1696
1697 #ifndef EM_486
1698 #define EM_486 6 /* Intel 80486 */
1699 #endif
1700
1701 static const arch_t arch_array[]={
1702 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1703 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1704 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1705 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1706 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1707 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1708 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1709 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1710 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}
1711 };
1712
1713 #if (defined IA32)
1714 static Elf32_Half running_arch_code=EM_386;
1715 #elif (defined AMD64)
1716 static Elf32_Half running_arch_code=EM_X86_64;
1717 #elif (defined IA64)
1718 static Elf32_Half running_arch_code=EM_IA_64;
1719 #elif (defined __sparc) && (defined _LP64)
1720 static Elf32_Half running_arch_code=EM_SPARCV9;
1721 #elif (defined __sparc) && (!defined _LP64)
1722 static Elf32_Half running_arch_code=EM_SPARC;
1723 #elif (defined __powerpc64__)
1724 static Elf32_Half running_arch_code=EM_PPC64;
1725 #elif (defined __powerpc__)
1726 static Elf32_Half running_arch_code=EM_PPC;
1727 #else
1728 #error Method os::dll_load requires that one of following is defined:\
1729 IA32, AMD64, IA64, __sparc, __powerpc__
1730 #endif
1731
1732 // Identify compatability class for VM's architecture and library's architecture
1733 // Obtain string descriptions for architectures
1734
1735 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1736 int running_arch_index=-1;
1737
1738 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1739 if (running_arch_code == arch_array[i].code) {
1740 running_arch_index = i;
1741 }
1742 if (lib_arch.code == arch_array[i].code) {
1743 lib_arch.compat_class = arch_array[i].compat_class;
1744 lib_arch.name = arch_array[i].name;
1745 }
1746 }
1747
1748 assert(running_arch_index != -1,
1749 "Didn't find running architecture code (running_arch_code) in arch_array");
1750 if (running_arch_index == -1) {
1751 // Even though running architecture detection failed
1752 // we may still continue with reporting dlerror() message
1753 return NULL;
1754 }
1755
1756 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1757 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1758 return NULL;
1759 }
1760
1761 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1762 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1763 return NULL;
1764 }
1765
1766 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1767 if ( lib_arch.name!=NULL ) {
1768 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1769 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1770 lib_arch.name, arch_array[running_arch_index].name);
1771 } else {
1772 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1773 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1774 lib_arch.code,
1775 arch_array[running_arch_index].name);
1776 }
1777 }
1778
1779 return NULL;
1780 }
1781
1782 /*
1783 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1784 * chances are you might want to run the generated bits against glibc-2.0
1785 * libdl.so, so always use locking for any version of glibc.
1786 */
1787 void* os::dll_lookup(void* handle, const char* name) {
1788 pthread_mutex_lock(&dl_mutex);
1789 void* res = dlsym(handle, name);
1790 pthread_mutex_unlock(&dl_mutex);
1791 return res;
1792 }
1793
1794
1795 bool _print_ascii_file(const char* filename, outputStream* st) {
1796 int fd = open(filename, O_RDONLY);
1797 if (fd == -1) {
1798 return false;
1799 }
1800
1801 char buf[32];
1802 int bytes;
1803 while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
1804 st->print_raw(buf, bytes);
1805 }
1806
1807 close(fd);
1808
1809 return true;
1810 }
1811
1812 void os::print_dll_info(outputStream *st) {
1813 st->print_cr("Dynamic libraries:");
1814
1815 char fname[32];
1816 pid_t pid = os::Linux::gettid();
1817
1818 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1819
1820 if (!_print_ascii_file(fname, st)) {
1821 st->print("Can not get library information for pid = %d\n", pid);
1822 }
1823 }
1824
1825
1826 void os::print_os_info(outputStream* st) {
1827 st->print("OS:");
1828
1829 // Try to identify popular distros.
1830 // Most Linux distributions have /etc/XXX-release file, which contains
1831 // the OS version string. Some have more than one /etc/XXX-release file
1832 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1833 // so the order is important.
1834 if (!_print_ascii_file("/etc/mandrake-release", st) &&
1835 !_print_ascii_file("/etc/sun-release", st) &&
1836 !_print_ascii_file("/etc/redhat-release", st) &&
1837 !_print_ascii_file("/etc/SuSE-release", st) &&
1838 !_print_ascii_file("/etc/turbolinux-release", st) &&
1839 !_print_ascii_file("/etc/gentoo-release", st) &&
1840 !_print_ascii_file("/etc/debian_version", st)) {
1841 st->print("Linux");
1842 }
1843 st->cr();
1844
1845 // kernel
1846 st->print("uname:");
1847 struct utsname name;
1848 uname(&name);
1849 st->print(name.sysname); st->print(" ");
1850 st->print(name.release); st->print(" ");
1851 st->print(name.version); st->print(" ");
1852 st->print(name.machine);
1853 st->cr();
1854
1855 // Print warning if unsafe chroot environment detected
1856 if (unsafe_chroot_detected) {
1857 st->print("WARNING!! ");
1858 st->print_cr(unstable_chroot_error);
1859 }
1860
1861 // libc, pthread
1862 st->print("libc:");
1863 st->print(os::Linux::glibc_version()); st->print(" ");
1864 st->print(os::Linux::libpthread_version()); st->print(" ");
1865 if (os::Linux::is_LinuxThreads()) {
1866 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
1867 }
1868 st->cr();
1869
1870 // rlimit
1871 st->print("rlimit:");
1872 struct rlimit rlim;
1873
1874 st->print(" STACK ");
1875 getrlimit(RLIMIT_STACK, &rlim);
1876 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1877 else st->print("%uk", rlim.rlim_cur >> 10);
1878
1879 st->print(", CORE ");
1880 getrlimit(RLIMIT_CORE, &rlim);
1881 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1882 else st->print("%uk", rlim.rlim_cur >> 10);
1883
1884 st->print(", NPROC ");
1885 getrlimit(RLIMIT_NPROC, &rlim);
1886 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1887 else st->print("%d", rlim.rlim_cur);
1888
1889 st->print(", NOFILE ");
1890 getrlimit(RLIMIT_NOFILE, &rlim);
1891 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1892 else st->print("%d", rlim.rlim_cur);
1893
1894 st->print(", AS ");
1895 getrlimit(RLIMIT_AS, &rlim);
1896 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
1897 else st->print("%uk", rlim.rlim_cur >> 10);
1898 st->cr();
1899
1900 // load average
1901 st->print("load average:");
1902 double loadavg[3];
1903 os::loadavg(loadavg, 3);
1904 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
1905 st->cr();
1906 }
1907
1908 void os::print_memory_info(outputStream* st) {
1909
1910 st->print("Memory:");
1911 st->print(" %dk page", os::vm_page_size()>>10);
1912
1913 // values in struct sysinfo are "unsigned long"
1914 struct sysinfo si;
1915 sysinfo(&si);
1916
1917 st->print(", physical " UINT64_FORMAT "k",
1918 os::physical_memory() >> 10);
1919 st->print("(" UINT64_FORMAT "k free)",
1920 os::available_memory() >> 10);
1921 st->print(", swap " UINT64_FORMAT "k",
1922 ((jlong)si.totalswap * si.mem_unit) >> 10);
1923 st->print("(" UINT64_FORMAT "k free)",
1924 ((jlong)si.freeswap * si.mem_unit) >> 10);
1925 st->cr();
1926 }
1927
1928 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
1929 // but they're the same for all the linux arch that we support
1930 // and they're the same for solaris but there's no common place to put this.
1931 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
1932 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
1933 "ILL_COPROC", "ILL_BADSTK" };
1934
1935 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
1936 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
1937 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
1938
1939 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
1940
1941 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
1942
1943 void os::print_siginfo(outputStream* st, void* siginfo) {
1944 st->print("siginfo:");
1945
1946 const int buflen = 100;
1947 char buf[buflen];
1948 siginfo_t *si = (siginfo_t*)siginfo;
1949 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
1950 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
1951 st->print("si_errno=%s", buf);
1952 } else {
1953 st->print("si_errno=%d", si->si_errno);
1954 }
1955 const int c = si->si_code;
1956 assert(c > 0, "unexpected si_code");
1957 switch (si->si_signo) {
1958 case SIGILL:
1959 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
1960 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1961 break;
1962 case SIGFPE:
1963 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
1964 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1965 break;
1966 case SIGSEGV:
1967 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
1968 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1969 break;
1970 case SIGBUS:
1971 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
1972 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
1973 break;
1974 default:
1975 st->print(", si_code=%d", si->si_code);
1976 // no si_addr
1977 }
1978
1979 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
1980 UseSharedSpaces) {
1981 FileMapInfo* mapinfo = FileMapInfo::current_info();
1982 if (mapinfo->is_in_shared_space(si->si_addr)) {
1983 st->print("\n\nError accessing class data sharing archive." \
1984 " Mapped file inaccessible during execution, " \
1985 " possible disk/network problem.");
1986 }
1987 }
1988 st->cr();
1989 }
1990
1991
1992 static void print_signal_handler(outputStream* st, int sig,
1993 char* buf, size_t buflen);
1994
1995 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1996 st->print_cr("Signal Handlers:");
1997 print_signal_handler(st, SIGSEGV, buf, buflen);
1998 print_signal_handler(st, SIGBUS , buf, buflen);
1999 print_signal_handler(st, SIGFPE , buf, buflen);
2000 print_signal_handler(st, SIGPIPE, buf, buflen);
2001 print_signal_handler(st, SIGXFSZ, buf, buflen);
2002 print_signal_handler(st, SIGILL , buf, buflen);
2003 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2004 print_signal_handler(st, SR_signum, buf, buflen);
2005 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2006 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2007 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2008 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2009 }
2010
2011 static char saved_jvm_path[MAXPATHLEN] = {0};
2012
2013 // Find the full path to the current module, libjvm.so or libjvm_g.so
2014 void os::jvm_path(char *buf, jint len) {
2015 // Error checking.
2016 if (len < MAXPATHLEN) {
2017 assert(false, "must use a large-enough buffer");
2018 buf[0] = '\0';
2019 return;
2020 }
2021 // Lazy resolve the path to current module.
2022 if (saved_jvm_path[0] != 0) {
2023 strcpy(buf, saved_jvm_path);
2024 return;
2025 }
2026
2027 char dli_fname[MAXPATHLEN];
2028 bool ret = dll_address_to_library_name(
2029 CAST_FROM_FN_PTR(address, os::jvm_path),
2030 dli_fname, sizeof(dli_fname), NULL);
2031 assert(ret != 0, "cannot locate libjvm");
2032 realpath(dli_fname, buf);
2033
2034 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2035 // Support for the gamma launcher. Typical value for buf is
2036 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2037 // the right place in the string, then assume we are installed in a JDK and
2038 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2039 // up the path so it looks like libjvm.so is installed there (append a
2040 // fake suffix hotspot/libjvm.so).
2041 const char *p = buf + strlen(buf) - 1;
2042 for (int count = 0; p > buf && count < 5; ++count) {
2043 for (--p; p > buf && *p != '/'; --p)
2044 /* empty */ ;
2045 }
2046
2047 if (strncmp(p, "/jre/lib/", 9) != 0) {
2048 // Look for JAVA_HOME in the environment.
2049 char* java_home_var = ::getenv("JAVA_HOME");
2050 if (java_home_var != NULL && java_home_var[0] != 0) {
2051 // Check the current module name "libjvm.so" or "libjvm_g.so".
2052 p = strrchr(buf, '/');
2053 assert(strstr(p, "/libjvm") == p, "invalid library name");
2054 p = strstr(p, "_g") ? "_g" : "";
2055
2056 realpath(java_home_var, buf);
2057 sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
2058 if (0 == access(buf, F_OK)) {
2059 // Use current module name "libjvm[_g].so" instead of
2060 // "libjvm"debug_only("_g")".so" since for fastdebug version
2061 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2062 // It is used when we are choosing the HPI library's name
2063 // "libhpi[_g].so" in hpi::initialize_get_interface().
2064 sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
2065 } else {
2066 // Go back to path of .so
2067 realpath(dli_fname, buf);
2068 }
2069 }
2070 }
2071 }
2072
2073 strcpy(saved_jvm_path, buf);
2074 }
2075
2076 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2077 // no prefix required, not even "_"
2078 }
2079
2080 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2081 // no suffix required
2082 }
2083
2084 ////////////////////////////////////////////////////////////////////////////////
2085 // sun.misc.Signal support
2086
2087 static volatile jint sigint_count = 0;
2088
2089 static void
2090 UserHandler(int sig, void *siginfo, void *context) {
2091 // 4511530 - sem_post is serialized and handled by the manager thread. When
2092 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2093 // don't want to flood the manager thread with sem_post requests.
2094 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2095 return;
2096
2097 // Ctrl-C is pressed during error reporting, likely because the error
2098 // handler fails to abort. Let VM die immediately.
2099 if (sig == SIGINT && is_error_reported()) {
2100 os::die();
2101 }
2102
2103 os::signal_notify(sig);
2104 }
2105
2106 void* os::user_handler() {
2107 return CAST_FROM_FN_PTR(void*, UserHandler);
2108 }
2109
2110 extern "C" {
2111 typedef void (*sa_handler_t)(int);
2112 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2113 }
2114
2115 void* os::signal(int signal_number, void* handler) {
2116 struct sigaction sigAct, oldSigAct;
2117
2118 sigfillset(&(sigAct.sa_mask));
2119 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2120 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2121
2122 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2123 // -1 means registration failed
2124 return (void *)-1;
2125 }
2126
2127 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2128 }
2129
2130 void os::signal_raise(int signal_number) {
2131 ::raise(signal_number);
2132 }
2133
2134 /*
2135 * The following code is moved from os.cpp for making this
2136 * code platform specific, which it is by its very nature.
2137 */
2138
2139 // Will be modified when max signal is changed to be dynamic
2140 int os::sigexitnum_pd() {
2141 return NSIG;
2142 }
2143
2144 // a counter for each possible signal value
2145 static volatile jint pending_signals[NSIG+1] = { 0 };
2146
2147 // Linux(POSIX) specific hand shaking semaphore.
2148 static sem_t sig_sem;
2149
2150 void os::signal_init_pd() {
2151 // Initialize signal structures
2152 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2153
2154 // Initialize signal semaphore
2155 ::sem_init(&sig_sem, 0, 0);
2156 }
2157
2158 void os::signal_notify(int sig) {
2159 Atomic::inc(&pending_signals[sig]);
2160 ::sem_post(&sig_sem);
2161 }
2162
2163 static int check_pending_signals(bool wait) {
2164 Atomic::store(0, &sigint_count);
2165 for (;;) {
2166 for (int i = 0; i < NSIG + 1; i++) {
2167 jint n = pending_signals[i];
2168 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2169 return i;
2170 }
2171 }
2172 if (!wait) {
2173 return -1;
2174 }
2175 JavaThread *thread = JavaThread::current();
2176 ThreadBlockInVM tbivm(thread);
2177
2178 bool threadIsSuspended;
2179 do {
2180 thread->set_suspend_equivalent();
2181 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2182 ::sem_wait(&sig_sem);
2183
2184 // were we externally suspended while we were waiting?
2185 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2186 if (threadIsSuspended) {
2187 //
2188 // The semaphore has been incremented, but while we were waiting
2189 // another thread suspended us. We don't want to continue running
2190 // while suspended because that would surprise the thread that
2191 // suspended us.
2192 //
2193 ::sem_post(&sig_sem);
2194
2195 thread->java_suspend_self();
2196 }
2197 } while (threadIsSuspended);
2198 }
2199 }
2200
2201 int os::signal_lookup() {
2202 return check_pending_signals(false);
2203 }
2204
2205 int os::signal_wait() {
2206 return check_pending_signals(true);
2207 }
2208
2209 ////////////////////////////////////////////////////////////////////////////////
2210 // Virtual Memory
2211
2212 int os::vm_page_size() {
2213 // Seems redundant as all get out
2214 assert(os::Linux::page_size() != -1, "must call os::init");
2215 return os::Linux::page_size();
2216 }
2217
2218 // Solaris allocates memory by pages.
2219 int os::vm_allocation_granularity() {
2220 assert(os::Linux::page_size() != -1, "must call os::init");
2221 return os::Linux::page_size();
2222 }
2223
2224 // Rationale behind this function:
2225 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2226 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2227 // samples for JITted code. Here we create private executable mapping over the code cache
2228 // and then we can use standard (well, almost, as mapping can change) way to provide
2229 // info for the reporting script by storing timestamp and location of symbol
2230 void linux_wrap_code(char* base, size_t size) {
2231 static volatile jint cnt = 0;
2232
2233 if (!UseOprofile) {
2234 return;
2235 }
2236
2237 char buf[40];
2238 int num = Atomic::add(1, &cnt);
2239
2240 sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num);
2241 unlink(buf);
2242
2243 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
2244
2245 if (fd != -1) {
2246 off_t rv = lseek(fd, size-2, SEEK_SET);
2247 if (rv != (off_t)-1) {
2248 if (write(fd, "", 1) == 1) {
2249 mmap(base, size,
2250 PROT_READ|PROT_WRITE|PROT_EXEC,
2251 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2252 }
2253 }
2254 close(fd);
2255 unlink(buf);
2256 }
2257 }
2258
2259 // NOTE: Linux kernel does not really reserve the pages for us.
2260 // All it does is to check if there are enough free pages
2261 // left at the time of mmap(). This could be a potential
2262 // problem.
2263 bool os::commit_memory(char* addr, size_t size) {
2264 uintptr_t res = (uintptr_t) ::mmap(addr, size,
2265 PROT_READ|PROT_WRITE|PROT_EXEC,
2266 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2267 return res != (uintptr_t) MAP_FAILED;
2268 }
2269
2270 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint) {
2271 return commit_memory(addr, size);
2272 }
2273
2274 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2275
2276 void os::free_memory(char *addr, size_t bytes) {
2277 uncommit_memory(addr, bytes);
2278 }
2279
2280 void os::numa_make_global(char *addr, size_t bytes) { }
2281
2282 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2283 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2284 }
2285
2286 bool os::numa_topology_changed() { return false; }
2287
2288 size_t os::numa_get_groups_num() {
2289 int max_node = Linux::numa_max_node();
2290 return max_node > 0 ? max_node + 1 : 1;
2291 }
2292
2293 int os::numa_get_group_id() {
2294 int cpu_id = Linux::sched_getcpu();
2295 if (cpu_id != -1) {
2296 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2297 if (lgrp_id != -1) {
2298 return lgrp_id;
2299 }
2300 }
2301 return 0;
2302 }
2303
2304 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2305 for (size_t i = 0; i < size; i++) {
2306 ids[i] = i;
2307 }
2308 return size;
2309 }
2310
2311 bool os::get_page_info(char *start, page_info* info) {
2312 return false;
2313 }
2314
2315 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2316 return end;
2317 }
2318
2319 extern "C" void numa_warn(int number, char *where, ...) { }
2320 extern "C" void numa_error(char *where) { }
2321
2322 void os::Linux::libnuma_init() {
2323 // sched_getcpu() should be in libc.
2324 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2325 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2326
2327 if (sched_getcpu() != -1) { // Does it work?
2328 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2329 if (handle != NULL) {
2330 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2331 dlsym(handle, "numa_node_to_cpus")));
2332 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2333 dlsym(handle, "numa_max_node")));
2334 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2335 dlsym(handle, "numa_available")));
2336 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2337 dlsym(handle, "numa_tonode_memory")));
2338 if (numa_available() != -1) {
2339 // Create a cpu -> node mapping
2340 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2341 rebuild_cpu_to_node_map();
2342 }
2343 }
2344 }
2345 }
2346
2347 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2348 // The table is later used in get_node_by_cpu().
2349 void os::Linux::rebuild_cpu_to_node_map() {
2350 int cpu_num = os::active_processor_count();
2351 cpu_to_node()->clear();
2352 cpu_to_node()->at_grow(cpu_num - 1);
2353 int node_num = numa_get_groups_num();
2354 int cpu_map_size = (cpu_num + BitsPerLong - 1) / BitsPerLong;
2355 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2356 for (int i = 0; i < node_num; i++) {
2357 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2358 for (int j = 0; j < cpu_map_size; j++) {
2359 if (cpu_map[j] != 0) {
2360 for (int k = 0; k < BitsPerLong; k++) {
2361 if (cpu_map[j] & (1UL << k)) {
2362 cpu_to_node()->at_put(j * BitsPerLong + k, i);
2363 }
2364 }
2365 }
2366 }
2367 }
2368 }
2369 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2370 }
2371
2372 int os::Linux::get_node_by_cpu(int cpu_id) {
2373 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2374 return cpu_to_node()->at(cpu_id);
2375 }
2376 return -1;
2377 }
2378
2379 GrowableArray<int>* os::Linux::_cpu_to_node;
2380 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2381 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2382 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2383 os::Linux::numa_available_func_t os::Linux::_numa_available;
2384 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2385
2386
2387 bool os::uncommit_memory(char* addr, size_t size) {
2388 return ::mmap(addr, size,
2389 PROT_READ|PROT_WRITE|PROT_EXEC,
2390 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0)
2391 != MAP_FAILED;
2392 }
2393
2394 static address _highest_vm_reserved_address = NULL;
2395
2396 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2397 // at 'requested_addr'. If there are existing memory mappings at the same
2398 // location, however, they will be overwritten. If 'fixed' is false,
2399 // 'requested_addr' is only treated as a hint, the return value may or
2400 // may not start from the requested address. Unlike Linux mmap(), this
2401 // function returns NULL to indicate failure.
2402 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2403 char * addr;
2404 int flags;
2405
2406 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2407 if (fixed) {
2408 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2409 flags |= MAP_FIXED;
2410 }
2411
2412 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE|PROT_EXEC,
2413 flags, -1, 0);
2414
2415 if (addr != MAP_FAILED) {
2416 // anon_mmap() should only get called during VM initialization,
2417 // don't need lock (actually we can skip locking even it can be called
2418 // from multiple threads, because _highest_vm_reserved_address is just a
2419 // hint about the upper limit of non-stack memory regions.)
2420 if ((address)addr + bytes > _highest_vm_reserved_address) {
2421 _highest_vm_reserved_address = (address)addr + bytes;
2422 }
2423 }
2424
2425 return addr == MAP_FAILED ? NULL : addr;
2426 }
2427
2428 // Don't update _highest_vm_reserved_address, because there might be memory
2429 // regions above addr + size. If so, releasing a memory region only creates
2430 // a hole in the address space, it doesn't help prevent heap-stack collision.
2431 //
2432 static int anon_munmap(char * addr, size_t size) {
2433 return ::munmap(addr, size) == 0;
2434 }
2435
2436 char* os::reserve_memory(size_t bytes, char* requested_addr,
2437 size_t alignment_hint) {
2438 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2439 }
2440
2441 bool os::release_memory(char* addr, size_t size) {
2442 return anon_munmap(addr, size);
2443 }
2444
2445 static address highest_vm_reserved_address() {
2446 return _highest_vm_reserved_address;
2447 }
2448
2449 static bool linux_mprotect(char* addr, size_t size, int prot) {
2450 // Linux wants the mprotect address argument to be page aligned.
2451 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2452
2453 // According to SUSv3, mprotect() should only be used with mappings
2454 // established by mmap(), and mmap() always maps whole pages. Unaligned
2455 // 'addr' likely indicates problem in the VM (e.g. trying to change
2456 // protection of malloc'ed or statically allocated memory). Check the
2457 // caller if you hit this assert.
2458 assert(addr == bottom, "sanity check");
2459
2460 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2461 return ::mprotect(bottom, size, prot) == 0;
2462 }
2463
2464 // Set protections specified
2465 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2466 bool is_committed) {
2467 unsigned int p = 0;
2468 switch (prot) {
2469 case MEM_PROT_NONE: p = PROT_NONE; break;
2470 case MEM_PROT_READ: p = PROT_READ; break;
2471 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2472 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2473 default:
2474 ShouldNotReachHere();
2475 }
2476 // is_committed is unused.
2477 return linux_mprotect(addr, bytes, p);
2478 }
2479
2480 bool os::guard_memory(char* addr, size_t size) {
2481 return linux_mprotect(addr, size, PROT_NONE);
2482 }
2483
2484 bool os::unguard_memory(char* addr, size_t size) {
2485 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE|PROT_EXEC);
2486 }
2487
2488 // Large page support
2489
2490 static size_t _large_page_size = 0;
2491
2492 bool os::large_page_init() {
2493 if (!UseLargePages) return false;
2494
2495 if (LargePageSizeInBytes) {
2496 _large_page_size = LargePageSizeInBytes;
2497 } else {
2498 // large_page_size on Linux is used to round up heap size. x86 uses either
2499 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2500 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2501 // page as large as 256M.
2502 //
2503 // Here we try to figure out page size by parsing /proc/meminfo and looking
2504 // for a line with the following format:
2505 // Hugepagesize: 2048 kB
2506 //
2507 // If we can't determine the value (e.g. /proc is not mounted, or the text
2508 // format has been changed), we'll use the largest page size supported by
2509 // the processor.
2510
2511 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M);
2512
2513 FILE *fp = fopen("/proc/meminfo", "r");
2514 if (fp) {
2515 while (!feof(fp)) {
2516 int x = 0;
2517 char buf[16];
2518 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2519 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2520 _large_page_size = x * K;
2521 break;
2522 }
2523 } else {
2524 // skip to next line
2525 for (;;) {
2526 int ch = fgetc(fp);
2527 if (ch == EOF || ch == (int)'\n') break;
2528 }
2529 }
2530 }
2531 fclose(fp);
2532 }
2533 }
2534
2535 const size_t default_page_size = (size_t)Linux::page_size();
2536 if (_large_page_size > default_page_size) {
2537 _page_sizes[0] = _large_page_size;
2538 _page_sizes[1] = default_page_size;
2539 _page_sizes[2] = 0;
2540 }
2541
2542 // Large page support is available on 2.6 or newer kernel, some vendors
2543 // (e.g. Redhat) have backported it to their 2.4 based distributions.
2544 // We optimistically assume the support is available. If later it turns out
2545 // not true, VM will automatically switch to use regular page size.
2546 return true;
2547 }
2548
2549 #ifndef SHM_HUGETLB
2550 #define SHM_HUGETLB 04000
2551 #endif
2552
2553 char* os::reserve_memory_special(size_t bytes) {
2554 assert(UseLargePages, "only for large pages");
2555
2556 key_t key = IPC_PRIVATE;
2557 char *addr;
2558
2559 bool warn_on_failure = UseLargePages &&
2560 (!FLAG_IS_DEFAULT(UseLargePages) ||
2561 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2562 );
2563 char msg[128];
2564
2565 // Create a large shared memory region to attach to based on size.
2566 // Currently, size is the total size of the heap
2567 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2568 if (shmid == -1) {
2569 // Possible reasons for shmget failure:
2570 // 1. shmmax is too small for Java heap.
2571 // > check shmmax value: cat /proc/sys/kernel/shmmax
2572 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2573 // 2. not enough large page memory.
2574 // > check available large pages: cat /proc/meminfo
2575 // > increase amount of large pages:
2576 // echo new_value > /proc/sys/vm/nr_hugepages
2577 // Note 1: different Linux may use different name for this property,
2578 // e.g. on Redhat AS-3 it is "hugetlb_pool".
2579 // Note 2: it's possible there's enough physical memory available but
2580 // they are so fragmented after a long run that they can't
2581 // coalesce into large pages. Try to reserve large pages when
2582 // the system is still "fresh".
2583 if (warn_on_failure) {
2584 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2585 warning(msg);
2586 }
2587 return NULL;
2588 }
2589
2590 // attach to the region
2591 addr = (char*)shmat(shmid, NULL, 0);
2592 int err = errno;
2593
2594 // Remove shmid. If shmat() is successful, the actual shared memory segment
2595 // will be deleted when it's detached by shmdt() or when the process
2596 // terminates. If shmat() is not successful this will remove the shared
2597 // segment immediately.
2598 shmctl(shmid, IPC_RMID, NULL);
2599
2600 if ((intptr_t)addr == -1) {
2601 if (warn_on_failure) {
2602 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2603 warning(msg);
2604 }
2605 return NULL;
2606 }
2607
2608 return addr;
2609 }
2610
2611 bool os::release_memory_special(char* base, size_t bytes) {
2612 // detaching the SHM segment will also delete it, see reserve_memory_special()
2613 int rslt = shmdt(base);
2614 return rslt == 0;
2615 }
2616
2617 size_t os::large_page_size() {
2618 return _large_page_size;
2619 }
2620
2621 // Linux does not support anonymous mmap with large page memory. The only way
2622 // to reserve large page memory without file backing is through SysV shared
2623 // memory API. The entire memory region is committed and pinned upfront.
2624 // Hopefully this will change in the future...
2625 bool os::can_commit_large_page_memory() {
2626 return false;
2627 }
2628
2629 bool os::can_execute_large_page_memory() {
2630 return false;
2631 }
2632
2633 // Reserve memory at an arbitrary address, only if that area is
2634 // available (and not reserved for something else).
2635
2636 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2637 const int max_tries = 10;
2638 char* base[max_tries];
2639 size_t size[max_tries];
2640 const size_t gap = 0x000000;
2641
2642 // Assert only that the size is a multiple of the page size, since
2643 // that's all that mmap requires, and since that's all we really know
2644 // about at this low abstraction level. If we need higher alignment,
2645 // we can either pass an alignment to this method or verify alignment
2646 // in one of the methods further up the call chain. See bug 5044738.
2647 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2648
2649 // Repeatedly allocate blocks until the block is allocated at the
2650 // right spot. Give up after max_tries. Note that reserve_memory() will
2651 // automatically update _highest_vm_reserved_address if the call is
2652 // successful. The variable tracks the highest memory address every reserved
2653 // by JVM. It is used to detect heap-stack collision if running with
2654 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2655 // space than needed, it could confuse the collision detecting code. To
2656 // solve the problem, save current _highest_vm_reserved_address and
2657 // calculate the correct value before return.
2658 address old_highest = _highest_vm_reserved_address;
2659
2660 // Linux mmap allows caller to pass an address as hint; give it a try first,
2661 // if kernel honors the hint then we can return immediately.
2662 char * addr = anon_mmap(requested_addr, bytes, false);
2663 if (addr == requested_addr) {
2664 return requested_addr;
2665 }
2666
2667 if (addr != NULL) {
2668 // mmap() is successful but it fails to reserve at the requested address
2669 anon_munmap(addr, bytes);
2670 }
2671
2672 int i;
2673 for (i = 0; i < max_tries; ++i) {
2674 base[i] = reserve_memory(bytes);
2675
2676 if (base[i] != NULL) {
2677 // Is this the block we wanted?
2678 if (base[i] == requested_addr) {
2679 size[i] = bytes;
2680 break;
2681 }
2682
2683 // Does this overlap the block we wanted? Give back the overlapped
2684 // parts and try again.
2685
2686 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2687 if (top_overlap >= 0 && top_overlap < bytes) {
2688 unmap_memory(base[i], top_overlap);
2689 base[i] += top_overlap;
2690 size[i] = bytes - top_overlap;
2691 } else {
2692 size_t bottom_overlap = base[i] + bytes - requested_addr;
2693 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2694 unmap_memory(requested_addr, bottom_overlap);
2695 size[i] = bytes - bottom_overlap;
2696 } else {
2697 size[i] = bytes;
2698 }
2699 }
2700 }
2701 }
2702
2703 // Give back the unused reserved pieces.
2704
2705 for (int j = 0; j < i; ++j) {
2706 if (base[j] != NULL) {
2707 unmap_memory(base[j], size[j]);
2708 }
2709 }
2710
2711 if (i < max_tries) {
2712 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
2713 return requested_addr;
2714 } else {
2715 _highest_vm_reserved_address = old_highest;
2716 return NULL;
2717 }
2718 }
2719
2720 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2721 return ::read(fd, buf, nBytes);
2722 }
2723
2724 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
2725 // Solaris uses poll(), linux uses park().
2726 // Poll() is likely a better choice, assuming that Thread.interrupt()
2727 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
2728 // SIGSEGV, see 4355769.
2729
2730 const int NANOSECS_PER_MILLISECS = 1000000;
2731
2732 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
2733 assert(thread == Thread::current(), "thread consistency check");
2734
2735 ParkEvent * const slp = thread->_SleepEvent ;
2736 slp->reset() ;
2737 OrderAccess::fence() ;
2738
2739 if (interruptible) {
2740 jlong prevtime = javaTimeNanos();
2741
2742 for (;;) {
2743 if (os::is_interrupted(thread, true)) {
2744 return OS_INTRPT;
2745 }
2746
2747 jlong newtime = javaTimeNanos();
2748
2749 if (newtime - prevtime < 0) {
2750 // time moving backwards, should only happen if no monotonic clock
2751 // not a guarantee() because JVM should not abort on kernel/glibc bugs
2752 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2753 } else {
2754 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2755 }
2756
2757 if(millis <= 0) {
2758 return OS_OK;
2759 }
2760
2761 prevtime = newtime;
2762
2763 {
2764 assert(thread->is_Java_thread(), "sanity check");
2765 JavaThread *jt = (JavaThread *) thread;
2766 ThreadBlockInVM tbivm(jt);
2767 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
2768
2769 jt->set_suspend_equivalent();
2770 // cleared by handle_special_suspend_equivalent_condition() or
2771 // java_suspend_self() via check_and_wait_while_suspended()
2772
2773 slp->park(millis);
2774
2775 // were we externally suspended while we were waiting?
2776 jt->check_and_wait_while_suspended();
2777 }
2778 }
2779 } else {
2780 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2781 jlong prevtime = javaTimeNanos();
2782
2783 for (;;) {
2784 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
2785 // the 1st iteration ...
2786 jlong newtime = javaTimeNanos();
2787
2788 if (newtime - prevtime < 0) {
2789 // time moving backwards, should only happen if no monotonic clock
2790 // not a guarantee() because JVM should not abort on kernel/glibc bugs
2791 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
2792 } else {
2793 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
2794 }
2795
2796 if(millis <= 0) break ;
2797
2798 prevtime = newtime;
2799 slp->park(millis);
2800 }
2801 return OS_OK ;
2802 }
2803 }
2804
2805 int os::naked_sleep() {
2806 // %% make the sleep time an integer flag. for now use 1 millisec.
2807 return os::sleep(Thread::current(), 1, false);
2808 }
2809
2810 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2811 void os::infinite_sleep() {
2812 while (true) { // sleep forever ...
2813 ::sleep(100); // ... 100 seconds at a time
2814 }
2815 }
2816
2817 // Used to convert frequent JVM_Yield() to nops
2818 bool os::dont_yield() {
2819 return DontYieldALot;
2820 }
2821
2822 void os::yield() {
2823 sched_yield();
2824 }
2825
2826 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
2827
2828 void os::yield_all(int attempts) {
2829 // Yields to all threads, including threads with lower priorities
2830 // Threads on Linux are all with same priority. The Solaris style
2831 // os::yield_all() with nanosleep(1ms) is not necessary.
2832 sched_yield();
2833 }
2834
2835 // Called from the tight loops to possibly influence time-sharing heuristics
2836 void os::loop_breaker(int attempts) {
2837 os::yield_all(attempts);
2838 }
2839
2840 ////////////////////////////////////////////////////////////////////////////////
2841 // thread priority support
2842
2843 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
2844 // only supports dynamic priority, static priority must be zero. For real-time
2845 // applications, Linux supports SCHED_RR which allows static priority (1-99).
2846 // However, for large multi-threaded applications, SCHED_RR is not only slower
2847 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
2848 // of 5 runs - Sep 2005).
2849 //
2850 // The following code actually changes the niceness of kernel-thread/LWP. It
2851 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
2852 // not the entire user process, and user level threads are 1:1 mapped to kernel
2853 // threads. It has always been the case, but could change in the future. For
2854 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
2855 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
2856
2857 int os::java_to_os_priority[MaxPriority + 1] = {
2858 19, // 0 Entry should never be used
2859
2860 4, // 1 MinPriority
2861 3, // 2
2862 2, // 3
2863
2864 1, // 4
2865 0, // 5 NormPriority
2866 -1, // 6
2867
2868 -2, // 7
2869 -3, // 8
2870 -4, // 9 NearMaxPriority
2871
2872 -5 // 10 MaxPriority
2873 };
2874
2875 static int prio_init() {
2876 if (ThreadPriorityPolicy == 1) {
2877 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
2878 // if effective uid is not root. Perhaps, a more elegant way of doing
2879 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
2880 if (geteuid() != 0) {
2881 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
2882 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
2883 }
2884 ThreadPriorityPolicy = 0;
2885 }
2886 }
2887 return 0;
2888 }
2889
2890 OSReturn os::set_native_priority(Thread* thread, int newpri) {
2891 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
2892
2893 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
2894 return (ret == 0) ? OS_OK : OS_ERR;
2895 }
2896
2897 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
2898 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
2899 *priority_ptr = java_to_os_priority[NormPriority];
2900 return OS_OK;
2901 }
2902
2903 errno = 0;
2904 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
2905 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
2906 }
2907
2908 // Hint to the underlying OS that a task switch would not be good.
2909 // Void return because it's a hint and can fail.
2910 void os::hint_no_preempt() {}
2911
2912 ////////////////////////////////////////////////////////////////////////////////
2913 // suspend/resume support
2914
2915 // the low-level signal-based suspend/resume support is a remnant from the
2916 // old VM-suspension that used to be for java-suspension, safepoints etc,
2917 // within hotspot. Now there is a single use-case for this:
2918 // - calling get_thread_pc() on the VMThread by the flat-profiler task
2919 // that runs in the watcher thread.
2920 // The remaining code is greatly simplified from the more general suspension
2921 // code that used to be used.
2922 //
2923 // The protocol is quite simple:
2924 // - suspend:
2925 // - sends a signal to the target thread
2926 // - polls the suspend state of the osthread using a yield loop
2927 // - target thread signal handler (SR_handler) sets suspend state
2928 // and blocks in sigsuspend until continued
2929 // - resume:
2930 // - sets target osthread state to continue
2931 // - sends signal to end the sigsuspend loop in the SR_handler
2932 //
2933 // Note that the SR_lock plays no role in this suspend/resume protocol.
2934 //
2935
2936 static void resume_clear_context(OSThread *osthread) {
2937 osthread->set_ucontext(NULL);
2938 osthread->set_siginfo(NULL);
2939
2940 // notify the suspend action is completed, we have now resumed
2941 osthread->sr.clear_suspended();
2942 }
2943
2944 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
2945 osthread->set_ucontext(context);
2946 osthread->set_siginfo(siginfo);
2947 }
2948
2949 //
2950 // Handler function invoked when a thread's execution is suspended or
2951 // resumed. We have to be careful that only async-safe functions are
2952 // called here (Note: most pthread functions are not async safe and
2953 // should be avoided.)
2954 //
2955 // Note: sigwait() is a more natural fit than sigsuspend() from an
2956 // interface point of view, but sigwait() prevents the signal hander
2957 // from being run. libpthread would get very confused by not having
2958 // its signal handlers run and prevents sigwait()'s use with the
2959 // mutex granting granting signal.
2960 //
2961 // Currently only ever called on the VMThread
2962 //
2963 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
2964 // Save and restore errno to avoid confusing native code with EINTR
2965 // after sigsuspend.
2966 int old_errno = errno;
2967
2968 Thread* thread = Thread::current();
2969 OSThread* osthread = thread->osthread();
2970 assert(thread->is_VM_thread(), "Must be VMThread");
2971 // read current suspend action
2972 int action = osthread->sr.suspend_action();
2973 if (action == SR_SUSPEND) {
2974 suspend_save_context(osthread, siginfo, context);
2975
2976 // Notify the suspend action is about to be completed. do_suspend()
2977 // waits until SR_SUSPENDED is set and then returns. We will wait
2978 // here for a resume signal and that completes the suspend-other
2979 // action. do_suspend/do_resume is always called as a pair from
2980 // the same thread - so there are no races
2981
2982 // notify the caller
2983 osthread->sr.set_suspended();
2984
2985 sigset_t suspend_set; // signals for sigsuspend()
2986
2987 // get current set of blocked signals and unblock resume signal
2988 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
2989 sigdelset(&suspend_set, SR_signum);
2990
2991 // wait here until we are resumed
2992 do {
2993 sigsuspend(&suspend_set);
2994 // ignore all returns until we get a resume signal
2995 } while (osthread->sr.suspend_action() != SR_CONTINUE);
2996
2997 resume_clear_context(osthread);
2998
2999 } else {
3000 assert(action == SR_CONTINUE, "unexpected sr action");
3001 // nothing special to do - just leave the handler
3002 }
3003
3004 errno = old_errno;
3005 }
3006
3007
3008 static int SR_initialize() {
3009 struct sigaction act;
3010 char *s;
3011 /* Get signal number to use for suspend/resume */
3012 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3013 int sig = ::strtol(s, 0, 10);
3014 if (sig > 0 || sig < _NSIG) {
3015 SR_signum = sig;
3016 }
3017 }
3018
3019 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3020 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3021
3022 sigemptyset(&SR_sigset);
3023 sigaddset(&SR_sigset, SR_signum);
3024
3025 /* Set up signal handler for suspend/resume */
3026 act.sa_flags = SA_RESTART|SA_SIGINFO;
3027 act.sa_handler = (void (*)(int)) SR_handler;
3028
3029 // SR_signum is blocked by default.
3030 // 4528190 - We also need to block pthread restart signal (32 on all
3031 // supported Linux platforms). Note that LinuxThreads need to block
3032 // this signal for all threads to work properly. So we don't have
3033 // to use hard-coded signal number when setting up the mask.
3034 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3035
3036 if (sigaction(SR_signum, &act, 0) == -1) {
3037 return -1;
3038 }
3039
3040 // Save signal flag
3041 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3042 return 0;
3043 }
3044
3045 static int SR_finalize() {
3046 return 0;
3047 }
3048
3049
3050 // returns true on success and false on error - really an error is fatal
3051 // but this seems the normal response to library errors
3052 static bool do_suspend(OSThread* osthread) {
3053 // mark as suspended and send signal
3054 osthread->sr.set_suspend_action(SR_SUSPEND);
3055 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3056 assert_status(status == 0, status, "pthread_kill");
3057
3058 // check status and wait until notified of suspension
3059 if (status == 0) {
3060 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3061 os::yield_all(i);
3062 }
3063 osthread->sr.set_suspend_action(SR_NONE);
3064 return true;
3065 }
3066 else {
3067 osthread->sr.set_suspend_action(SR_NONE);
3068 return false;
3069 }
3070 }
3071
3072 static void do_resume(OSThread* osthread) {
3073 assert(osthread->sr.is_suspended(), "thread should be suspended");
3074 osthread->sr.set_suspend_action(SR_CONTINUE);
3075
3076 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3077 assert_status(status == 0, status, "pthread_kill");
3078 // check status and wait unit notified of resumption
3079 if (status == 0) {
3080 for (int i = 0; osthread->sr.is_suspended(); i++) {
3081 os::yield_all(i);
3082 }
3083 }
3084 osthread->sr.set_suspend_action(SR_NONE);
3085 }
3086
3087 ////////////////////////////////////////////////////////////////////////////////
3088 // interrupt support
3089
3090 void os::interrupt(Thread* thread) {
3091 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3092 "possibility of dangling Thread pointer");
3093
3094 OSThread* osthread = thread->osthread();
3095
3096 if (!osthread->interrupted()) {
3097 osthread->set_interrupted(true);
3098 // More than one thread can get here with the same value of osthread,
3099 // resulting in multiple notifications. We do, however, want the store
3100 // to interrupted() to be visible to other threads before we execute unpark().
3101 OrderAccess::fence();
3102 ParkEvent * const slp = thread->_SleepEvent ;
3103 if (slp != NULL) slp->unpark() ;
3104 }
3105
3106 // For JSR166. Unpark even if interrupt status already was set
3107 if (thread->is_Java_thread())
3108 ((JavaThread*)thread)->parker()->unpark();
3109
3110 ParkEvent * ev = thread->_ParkEvent ;
3111 if (ev != NULL) ev->unpark() ;
3112
3113 }
3114
3115 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3116 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3117 "possibility of dangling Thread pointer");
3118
3119 OSThread* osthread = thread->osthread();
3120
3121 bool interrupted = osthread->interrupted();
3122
3123 if (interrupted && clear_interrupted) {
3124 osthread->set_interrupted(false);
3125 // consider thread->_SleepEvent->reset() ... optional optimization
3126 }
3127
3128 return interrupted;
3129 }
3130
3131 ///////////////////////////////////////////////////////////////////////////////////
3132 // signal handling (except suspend/resume)
3133
3134 // This routine may be used by user applications as a "hook" to catch signals.
3135 // The user-defined signal handler must pass unrecognized signals to this
3136 // routine, and if it returns true (non-zero), then the signal handler must
3137 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3138 // routine will never retun false (zero), but instead will execute a VM panic
3139 // routine kill the process.
3140 //
3141 // If this routine returns false, it is OK to call it again. This allows
3142 // the user-defined signal handler to perform checks either before or after
3143 // the VM performs its own checks. Naturally, the user code would be making
3144 // a serious error if it tried to handle an exception (such as a null check
3145 // or breakpoint) that the VM was generating for its own correct operation.
3146 //
3147 // This routine may recognize any of the following kinds of signals:
3148 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3149 // It should be consulted by handlers for any of those signals.
3150 //
3151 // The caller of this routine must pass in the three arguments supplied
3152 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3153 // field of the structure passed to sigaction(). This routine assumes that
3154 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3155 //
3156 // Note that the VM will print warnings if it detects conflicting signal
3157 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3158 //
3159 extern "C" int
3160 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3161 void* ucontext, int abort_if_unrecognized);
3162
3163 void signalHandler(int sig, siginfo_t* info, void* uc) {
3164 assert(info != NULL && uc != NULL, "it must be old kernel");
3165 JVM_handle_linux_signal(sig, info, uc, true);
3166 }
3167
3168
3169 // This boolean allows users to forward their own non-matching signals
3170 // to JVM_handle_linux_signal, harmlessly.
3171 bool os::Linux::signal_handlers_are_installed = false;
3172
3173 // For signal-chaining
3174 struct sigaction os::Linux::sigact[MAXSIGNUM];
3175 unsigned int os::Linux::sigs = 0;
3176 bool os::Linux::libjsig_is_loaded = false;
3177 typedef struct sigaction *(*get_signal_t)(int);
3178 get_signal_t os::Linux::get_signal_action = NULL;
3179
3180 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3181 struct sigaction *actp = NULL;
3182
3183 if (libjsig_is_loaded) {
3184 // Retrieve the old signal handler from libjsig
3185 actp = (*get_signal_action)(sig);
3186 }
3187 if (actp == NULL) {
3188 // Retrieve the preinstalled signal handler from jvm
3189 actp = get_preinstalled_handler(sig);
3190 }
3191
3192 return actp;
3193 }
3194
3195 static bool call_chained_handler(struct sigaction *actp, int sig,
3196 siginfo_t *siginfo, void *context) {
3197 // Call the old signal handler
3198 if (actp->sa_handler == SIG_DFL) {
3199 // It's more reasonable to let jvm treat it as an unexpected exception
3200 // instead of taking the default action.
3201 return false;
3202 } else if (actp->sa_handler != SIG_IGN) {
3203 if ((actp->sa_flags & SA_NODEFER) == 0) {
3204 // automaticlly block the signal
3205 sigaddset(&(actp->sa_mask), sig);
3206 }
3207
3208 sa_handler_t hand;
3209 sa_sigaction_t sa;
3210 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3211 // retrieve the chained handler
3212 if (siginfo_flag_set) {
3213 sa = actp->sa_sigaction;
3214 } else {
3215 hand = actp->sa_handler;
3216 }
3217
3218 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3219 actp->sa_handler = SIG_DFL;
3220 }
3221
3222 // try to honor the signal mask
3223 sigset_t oset;
3224 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3225
3226 // call into the chained handler
3227 if (siginfo_flag_set) {
3228 (*sa)(sig, siginfo, context);
3229 } else {
3230 (*hand)(sig);
3231 }
3232
3233 // restore the signal mask
3234 pthread_sigmask(SIG_SETMASK, &oset, 0);
3235 }
3236 // Tell jvm's signal handler the signal is taken care of.
3237 return true;
3238 }
3239
3240 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3241 bool chained = false;
3242 // signal-chaining
3243 if (UseSignalChaining) {
3244 struct sigaction *actp = get_chained_signal_action(sig);
3245 if (actp != NULL) {
3246 chained = call_chained_handler(actp, sig, siginfo, context);
3247 }
3248 }
3249 return chained;
3250 }
3251
3252 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3253 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3254 return &sigact[sig];
3255 }
3256 return NULL;
3257 }
3258
3259 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3260 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3261 sigact[sig] = oldAct;
3262 sigs |= (unsigned int)1 << sig;
3263 }
3264
3265 // for diagnostic
3266 int os::Linux::sigflags[MAXSIGNUM];
3267
3268 int os::Linux::get_our_sigflags(int sig) {
3269 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3270 return sigflags[sig];
3271 }
3272
3273 void os::Linux::set_our_sigflags(int sig, int flags) {
3274 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3275 sigflags[sig] = flags;
3276 }
3277
3278 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3279 // Check for overwrite.
3280 struct sigaction oldAct;
3281 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3282
3283 void* oldhand = oldAct.sa_sigaction
3284 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3285 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3286 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3287 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3288 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3289 if (AllowUserSignalHandlers || !set_installed) {
3290 // Do not overwrite; user takes responsibility to forward to us.
3291 return;
3292 } else if (UseSignalChaining) {
3293 // save the old handler in jvm
3294 save_preinstalled_handler(sig, oldAct);
3295 // libjsig also interposes the sigaction() call below and saves the
3296 // old sigaction on it own.
3297 } else {
3298 fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig);
3299 }
3300 }
3301
3302 struct sigaction sigAct;
3303 sigfillset(&(sigAct.sa_mask));
3304 sigAct.sa_handler = SIG_DFL;
3305 if (!set_installed) {
3306 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3307 } else {
3308 sigAct.sa_sigaction = signalHandler;
3309 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3310 }
3311 // Save flags, which are set by ours
3312 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3313 sigflags[sig] = sigAct.sa_flags;
3314
3315 int ret = sigaction(sig, &sigAct, &oldAct);
3316 assert(ret == 0, "check");
3317
3318 void* oldhand2 = oldAct.sa_sigaction
3319 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3320 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3321 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3322 }
3323
3324 // install signal handlers for signals that HotSpot needs to
3325 // handle in order to support Java-level exception handling.
3326
3327 void os::Linux::install_signal_handlers() {
3328 if (!signal_handlers_are_installed) {
3329 signal_handlers_are_installed = true;
3330
3331 // signal-chaining
3332 typedef void (*signal_setting_t)();
3333 signal_setting_t begin_signal_setting = NULL;
3334 signal_setting_t end_signal_setting = NULL;
3335 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3336 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3337 if (begin_signal_setting != NULL) {
3338 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3339 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3340 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3341 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3342 libjsig_is_loaded = true;
3343 assert(UseSignalChaining, "should enable signal-chaining");
3344 }
3345 if (libjsig_is_loaded) {
3346 // Tell libjsig jvm is setting signal handlers
3347 (*begin_signal_setting)();
3348 }
3349
3350 set_signal_handler(SIGSEGV, true);
3351 set_signal_handler(SIGPIPE, true);
3352 set_signal_handler(SIGBUS, true);
3353 set_signal_handler(SIGILL, true);
3354 set_signal_handler(SIGFPE, true);
3355 set_signal_handler(SIGXFSZ, true);
3356
3357 if (libjsig_is_loaded) {
3358 // Tell libjsig jvm finishes setting signal handlers
3359 (*end_signal_setting)();
3360 }
3361
3362 // We don't activate signal checker if libjsig is in place, we trust ourselves
3363 // and if UserSignalHandler is installed all bets are off
3364 if (CheckJNICalls) {
3365 if (libjsig_is_loaded) {
3366 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3367 check_signals = false;
3368 }
3369 if (AllowUserSignalHandlers) {
3370 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3371 check_signals = false;
3372 }
3373 }
3374 }
3375 }
3376
3377 // This is the fastest way to get thread cpu time on Linux.
3378 // Returns cpu time (user+sys) for any thread, not only for current.
3379 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3380 // It might work on 2.6.10+ with a special kernel/glibc patch.
3381 // For reference, please, see IEEE Std 1003.1-2004:
3382 // http://www.unix.org/single_unix_specification
3383
3384 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3385 struct timespec tp;
3386 int rc = os::Linux::clock_gettime(clockid, &tp);
3387 assert(rc == 0, "clock_gettime is expected to return 0 code");
3388
3389 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3390 }
3391
3392 /////
3393 // glibc on Linux platform uses non-documented flag
3394 // to indicate, that some special sort of signal
3395 // trampoline is used.
3396 // We will never set this flag, and we should
3397 // ignore this flag in our diagnostic
3398 #ifdef SIGNIFICANT_SIGNAL_MASK
3399 #undef SIGNIFICANT_SIGNAL_MASK
3400 #endif
3401 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3402
3403 static const char* get_signal_handler_name(address handler,
3404 char* buf, int buflen) {
3405 int offset;
3406 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3407 if (found) {
3408 // skip directory names
3409 const char *p1, *p2;
3410 p1 = buf;
3411 size_t len = strlen(os::file_separator());
3412 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3413 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3414 } else {
3415 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3416 }
3417 return buf;
3418 }
3419
3420 static void print_signal_handler(outputStream* st, int sig,
3421 char* buf, size_t buflen) {
3422 struct sigaction sa;
3423
3424 sigaction(sig, NULL, &sa);
3425
3426 // See comment for SIGNIFICANT_SIGNAL_MASK define
3427 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3428
3429 st->print("%s: ", os::exception_name(sig, buf, buflen));
3430
3431 address handler = (sa.sa_flags & SA_SIGINFO)
3432 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3433 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3434
3435 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3436 st->print("SIG_DFL");
3437 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3438 st->print("SIG_IGN");
3439 } else {
3440 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3441 }
3442
3443 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3444
3445 address rh = VMError::get_resetted_sighandler(sig);
3446 // May be, handler was resetted by VMError?
3447 if(rh != NULL) {
3448 handler = rh;
3449 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3450 }
3451
3452 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3453
3454 // Check: is it our handler?
3455 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3456 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3457 // It is our signal handler
3458 // check for flags, reset system-used one!
3459 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3460 st->print(
3461 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3462 os::Linux::get_our_sigflags(sig));
3463 }
3464 }
3465 st->cr();
3466 }
3467
3468
3469 #define DO_SIGNAL_CHECK(sig) \
3470 if (!sigismember(&check_signal_done, sig)) \
3471 os::Linux::check_signal_handler(sig)
3472
3473 // This method is a periodic task to check for misbehaving JNI applications
3474 // under CheckJNI, we can add any periodic checks here
3475
3476 void os::run_periodic_checks() {
3477
3478 if (check_signals == false) return;
3479
3480 // SEGV and BUS if overridden could potentially prevent
3481 // generation of hs*.log in the event of a crash, debugging
3482 // such a case can be very challenging, so we absolutely
3483 // check the following for a good measure:
3484 DO_SIGNAL_CHECK(SIGSEGV);
3485 DO_SIGNAL_CHECK(SIGILL);
3486 DO_SIGNAL_CHECK(SIGFPE);
3487 DO_SIGNAL_CHECK(SIGBUS);
3488 DO_SIGNAL_CHECK(SIGPIPE);
3489 DO_SIGNAL_CHECK(SIGXFSZ);
3490
3491
3492 // ReduceSignalUsage allows the user to override these handlers
3493 // see comments at the very top and jvm_solaris.h
3494 if (!ReduceSignalUsage) {
3495 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3496 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3497 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3498 DO_SIGNAL_CHECK(BREAK_SIGNAL);
3499 }
3500
3501 DO_SIGNAL_CHECK(SR_signum);
3502 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3503 }
3504
3505 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3506
3507 static os_sigaction_t os_sigaction = NULL;
3508
3509 void os::Linux::check_signal_handler(int sig) {
3510 char buf[O_BUFLEN];
3511 address jvmHandler = NULL;
3512
3513
3514 struct sigaction act;
3515 if (os_sigaction == NULL) {
3516 // only trust the default sigaction, in case it has been interposed
3517 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3518 if (os_sigaction == NULL) return;
3519 }
3520
3521 os_sigaction(sig, (struct sigaction*)NULL, &act);
3522
3523
3524 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3525
3526 address thisHandler = (act.sa_flags & SA_SIGINFO)
3527 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3528 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3529
3530
3531 switch(sig) {
3532 case SIGSEGV:
3533 case SIGBUS:
3534 case SIGFPE:
3535 case SIGPIPE:
3536 case SIGILL:
3537 case SIGXFSZ:
3538 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3539 break;
3540
3541 case SHUTDOWN1_SIGNAL:
3542 case SHUTDOWN2_SIGNAL:
3543 case SHUTDOWN3_SIGNAL:
3544 case BREAK_SIGNAL:
3545 jvmHandler = (address)user_handler();
3546 break;
3547
3548 case INTERRUPT_SIGNAL:
3549 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3550 break;
3551
3552 default:
3553 if (sig == SR_signum) {
3554 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3555 } else {
3556 return;
3557 }
3558 break;
3559 }
3560
3561 if (thisHandler != jvmHandler) {
3562 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3563 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3564 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3565 // No need to check this sig any longer
3566 sigaddset(&check_signal_done, sig);
3567 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3568 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3569 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3570 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
3571 // No need to check this sig any longer
3572 sigaddset(&check_signal_done, sig);
3573 }
3574
3575 // Dump all the signal
3576 if (sigismember(&check_signal_done, sig)) {
3577 print_signal_handlers(tty, buf, O_BUFLEN);
3578 }
3579 }
3580
3581 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3582
3583 extern bool signal_name(int signo, char* buf, size_t len);
3584
3585 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3586 if (0 < exception_code && exception_code <= SIGRTMAX) {
3587 // signal
3588 if (!signal_name(exception_code, buf, size)) {
3589 jio_snprintf(buf, size, "SIG%d", exception_code);
3590 }
3591 return buf;
3592 } else {
3593 return NULL;
3594 }
3595 }
3596
3597 // this is called _before_ the most of global arguments have been parsed
3598 void os::init(void) {
3599 char dummy; /* used to get a guess on initial stack address */
3600 // first_hrtime = gethrtime();
3601
3602 // With LinuxThreads the JavaMain thread pid (primordial thread)
3603 // is different than the pid of the java launcher thread.
3604 // So, on Linux, the launcher thread pid is passed to the VM
3605 // via the sun.java.launcher.pid property.
3606 // Use this property instead of getpid() if it was correctly passed.
3607 // See bug 6351349.
3608 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3609
3610 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3611
3612 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3613
3614 init_random(1234567);
3615
3616 ThreadCritical::initialize();
3617
3618 Linux::set_page_size(sysconf(_SC_PAGESIZE));
3619 if (Linux::page_size() == -1) {
3620 fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno));
3621 }
3622 init_page_sizes((size_t) Linux::page_size());
3623
3624 Linux::initialize_system_info();
3625
3626 // main_thread points to the aboriginal thread
3627 Linux::_main_thread = pthread_self();
3628
3629 Linux::clock_init();
3630 initial_time_count = os::elapsed_counter();
3631 pthread_mutex_init(&dl_mutex, NULL);
3632 }
3633
3634 // To install functions for atexit system call
3635 extern "C" {
3636 static void perfMemory_exit_helper() {
3637 perfMemory_exit();
3638 }
3639 }
3640
3641 // this is called _after_ the global arguments have been parsed
3642 jint os::init_2(void)
3643 {
3644 Linux::fast_thread_clock_init();
3645
3646 // Allocate a single page and mark it as readable for safepoint polling
3647 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3648 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3649
3650 os::set_polling_page( polling_page );
3651
3652 #ifndef PRODUCT
3653 if(Verbose && PrintMiscellaneous)
3654 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3655 #endif
3656
3657 if (!UseMembar) {
3658 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3659 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3660 os::set_memory_serialize_page( mem_serialize_page );
3661
3662 #ifndef PRODUCT
3663 if(Verbose && PrintMiscellaneous)
3664 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3665 #endif
3666 }
3667
3668 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3669
3670 // initialize suspend/resume support - must do this before signal_sets_init()
3671 if (SR_initialize() != 0) {
3672 perror("SR_initialize failed");
3673 return JNI_ERR;
3674 }
3675
3676 Linux::signal_sets_init();
3677 Linux::install_signal_handlers();
3678
3679 size_t threadStackSizeInBytes = ThreadStackSize * K;
3680 if (threadStackSizeInBytes != 0 &&
3681 threadStackSizeInBytes < Linux::min_stack_allowed) {
3682 tty->print_cr("\nThe stack size specified is too small, "
3683 "Specify at least %dk",
3684 Linux::min_stack_allowed / K);
3685 return JNI_ERR;
3686 }
3687
3688 // Make the stack size a multiple of the page size so that
3689 // the yellow/red zones can be guarded.
3690 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
3691 vm_page_size()));
3692
3693 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
3694
3695 Linux::libpthread_init();
3696 if (PrintMiscellaneous && (Verbose || WizardMode)) {
3697 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
3698 Linux::glibc_version(), Linux::libpthread_version(),
3699 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
3700 }
3701
3702 if (UseNUMA) {
3703 Linux::libnuma_init();
3704 }
3705
3706 if (MaxFDLimit) {
3707 // set the number of file descriptors to max. print out error
3708 // if getrlimit/setrlimit fails but continue regardless.
3709 struct rlimit nbr_files;
3710 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
3711 if (status != 0) {
3712 if (PrintMiscellaneous && (Verbose || WizardMode))
3713 perror("os::init_2 getrlimit failed");
3714 } else {
3715 nbr_files.rlim_cur = nbr_files.rlim_max;
3716 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
3717 if (status != 0) {
3718 if (PrintMiscellaneous && (Verbose || WizardMode))
3719 perror("os::init_2 setrlimit failed");
3720 }
3721 }
3722 }
3723
3724 // Initialize lock used to serialize thread creation (see os::create_thread)
3725 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
3726
3727 // Initialize HPI.
3728 jint hpi_result = hpi::initialize();
3729 if (hpi_result != JNI_OK) {
3730 tty->print_cr("There was an error trying to initialize the HPI library.");
3731 return hpi_result;
3732 }
3733
3734 // at-exit methods are called in the reverse order of their registration.
3735 // atexit functions are called on return from main or as a result of a
3736 // call to exit(3C). There can be only 32 of these functions registered
3737 // and atexit() does not set errno.
3738
3739 if (PerfAllowAtExitRegistration) {
3740 // only register atexit functions if PerfAllowAtExitRegistration is set.
3741 // atexit functions can be delayed until process exit time, which
3742 // can be problematic for embedded VM situations. Embedded VMs should
3743 // call DestroyJavaVM() to assure that VM resources are released.
3744
3745 // note: perfMemory_exit_helper atexit function may be removed in
3746 // the future if the appropriate cleanup code can be added to the
3747 // VM_Exit VMOperation's doit method.
3748 if (atexit(perfMemory_exit_helper) != 0) {
3749 warning("os::init2 atexit(perfMemory_exit_helper) failed");
3750 }
3751 }
3752
3753 // initialize thread priority policy
3754 prio_init();
3755
3756 return JNI_OK;
3757 }
3758
3759 // Mark the polling page as unreadable
3760 void os::make_polling_page_unreadable(void) {
3761 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
3762 fatal("Could not disable polling page");
3763 };
3764
3765 // Mark the polling page as readable
3766 void os::make_polling_page_readable(void) {
3767 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
3768 fatal("Could not enable polling page");
3769 }
3770 };
3771
3772 int os::active_processor_count() {
3773 // Linux doesn't yet have a (official) notion of processor sets,
3774 // so just return the number of online processors.
3775 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
3776 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
3777 return online_cpus;
3778 }
3779
3780 bool os::distribute_processes(uint length, uint* distribution) {
3781 // Not yet implemented.
3782 return false;
3783 }
3784
3785 bool os::bind_to_processor(uint processor_id) {
3786 // Not yet implemented.
3787 return false;
3788 }
3789
3790 ///
3791
3792 // Suspends the target using the signal mechanism and then grabs the PC before
3793 // resuming the target. Used by the flat-profiler only
3794 ExtendedPC os::get_thread_pc(Thread* thread) {
3795 // Make sure that it is called by the watcher for the VMThread
3796 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
3797 assert(thread->is_VM_thread(), "Can only be called for VMThread");
3798
3799 ExtendedPC epc;
3800
3801 OSThread* osthread = thread->osthread();
3802 if (do_suspend(osthread)) {
3803 if (osthread->ucontext() != NULL) {
3804 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
3805 } else {
3806 // NULL context is unexpected, double-check this is the VMThread
3807 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3808 }
3809 do_resume(osthread);
3810 }
3811 // failure means pthread_kill failed for some reason - arguably this is
3812 // a fatal problem, but such problems are ignored elsewhere
3813
3814 return epc;
3815 }
3816
3817 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
3818 {
3819 if (is_NPTL()) {
3820 return pthread_cond_timedwait(_cond, _mutex, _abstime);
3821 } else {
3822 #ifndef IA64
3823 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
3824 // word back to default 64bit precision if condvar is signaled. Java
3825 // wants 53bit precision. Save and restore current value.
3826 int fpu = get_fpu_control_word();
3827 #endif // IA64
3828 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
3829 #ifndef IA64
3830 set_fpu_control_word(fpu);
3831 #endif // IA64
3832 return status;
3833 }
3834 }
3835
3836 ////////////////////////////////////////////////////////////////////////////////
3837 // debug support
3838
3839 #ifndef PRODUCT
3840 static address same_page(address x, address y) {
3841 int page_bits = -os::vm_page_size();
3842 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
3843 return x;
3844 else if (x > y)
3845 return (address)(intptr_t(y) | ~page_bits) + 1;
3846 else
3847 return (address)(intptr_t(y) & page_bits);
3848 }
3849
3850 bool os::find(address addr) {
3851 Dl_info dlinfo;
3852 memset(&dlinfo, 0, sizeof(dlinfo));
3853 if (dladdr(addr, &dlinfo)) {
3854 tty->print(PTR_FORMAT ": ", addr);
3855 if (dlinfo.dli_sname != NULL) {
3856 tty->print("%s+%#x", dlinfo.dli_sname,
3857 addr - (intptr_t)dlinfo.dli_saddr);
3858 } else if (dlinfo.dli_fname) {
3859 tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
3860 } else {
3861 tty->print("<absolute address>");
3862 }
3863 if (dlinfo.dli_fname) {
3864 tty->print(" in %s", dlinfo.dli_fname);
3865 }
3866 if (dlinfo.dli_fbase) {
3867 tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
3868 }
3869 tty->cr();
3870
3871 if (Verbose) {
3872 // decode some bytes around the PC
3873 address begin = same_page(addr-40, addr);
3874 address end = same_page(addr+40, addr);
3875 address lowest = (address) dlinfo.dli_sname;
3876 if (!lowest) lowest = (address) dlinfo.dli_fbase;
3877 if (begin < lowest) begin = lowest;
3878 Dl_info dlinfo2;
3879 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
3880 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
3881 end = (address) dlinfo2.dli_saddr;
3882 Disassembler::decode(begin, end);
3883 }
3884 return true;
3885 }
3886 return false;
3887 }
3888
3889 #endif
3890
3891 ////////////////////////////////////////////////////////////////////////////////
3892 // misc
3893
3894 // This does not do anything on Linux. This is basically a hook for being
3895 // able to use structured exception handling (thread-local exception filters)
3896 // on, e.g., Win32.
3897 void
3898 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
3899 JavaCallArguments* args, Thread* thread) {
3900 f(value, method, args, thread);
3901 }
3902
3903 void os::print_statistics() {
3904 }
3905
3906 int os::message_box(const char* title, const char* message) {
3907 int i;
3908 fdStream err(defaultStream::error_fd());
3909 for (i = 0; i < 78; i++) err.print_raw("=");
3910 err.cr();
3911 err.print_raw_cr(title);
3912 for (i = 0; i < 78; i++) err.print_raw("-");
3913 err.cr();
3914 err.print_raw_cr(message);
3915 for (i = 0; i < 78; i++) err.print_raw("=");
3916 err.cr();
3917
3918 char buf[16];
3919 // Prevent process from exiting upon "read error" without consuming all CPU
3920 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3921
3922 return buf[0] == 'y' || buf[0] == 'Y';
3923 }
3924
3925 int os::stat(const char *path, struct stat *sbuf) {
3926 char pathbuf[MAX_PATH];
3927 if (strlen(path) > MAX_PATH - 1) {
3928 errno = ENAMETOOLONG;
3929 return -1;
3930 }
3931 hpi::native_path(strcpy(pathbuf, path));
3932 return ::stat(pathbuf, sbuf);
3933 }
3934
3935 bool os::check_heap(bool force) {
3936 return true;
3937 }
3938
3939 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
3940 return ::vsnprintf(buf, count, format, args);
3941 }
3942
3943 // Is a (classpath) directory empty?
3944 bool os::dir_is_empty(const char* path) {
3945 DIR *dir = NULL;
3946 struct dirent *ptr;
3947
3948 dir = opendir(path);
3949 if (dir == NULL) return true;
3950
3951 /* Scan the directory */
3952 bool result = true;
3953 char buf[sizeof(struct dirent) + MAX_PATH];
3954 while (result && (ptr = ::readdir(dir)) != NULL) {
3955 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
3956 result = false;
3957 }
3958 }
3959 closedir(dir);
3960 return result;
3961 }
3962
3963 // create binary file, rewriting existing file if required
3964 int os::create_binary_file(const char* path, bool rewrite_existing) {
3965 int oflags = O_WRONLY | O_CREAT;
3966 if (!rewrite_existing) {
3967 oflags |= O_EXCL;
3968 }
3969 return ::open64(path, oflags, S_IREAD | S_IWRITE);
3970 }
3971
3972 // return current position of file pointer
3973 jlong os::current_file_offset(int fd) {
3974 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
3975 }
3976
3977 // move file pointer to the specified offset
3978 jlong os::seek_to_file_offset(int fd, jlong offset) {
3979 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
3980 }
3981
3982 // Map a block of memory.
3983 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
3984 char *addr, size_t bytes, bool read_only,
3985 bool allow_exec) {
3986 int prot;
3987 int flags;
3988
3989 if (read_only) {
3990 prot = PROT_READ;
3991 flags = MAP_SHARED;
3992 } else {
3993 prot = PROT_READ | PROT_WRITE;
3994 flags = MAP_PRIVATE;
3995 }
3996
3997 if (allow_exec) {
3998 prot |= PROT_EXEC;
3999 }
4000
4001 if (addr != NULL) {
4002 flags |= MAP_FIXED;
4003 }
4004
4005 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4006 fd, file_offset);
4007 if (mapped_address == MAP_FAILED) {
4008 return NULL;
4009 }
4010 return mapped_address;
4011 }
4012
4013
4014 // Remap a block of memory.
4015 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4016 char *addr, size_t bytes, bool read_only,
4017 bool allow_exec) {
4018 // same as map_memory() on this OS
4019 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4020 allow_exec);
4021 }
4022
4023
4024 // Unmap a block of memory.
4025 bool os::unmap_memory(char* addr, size_t bytes) {
4026 return munmap(addr, bytes) == 0;
4027 }
4028
4029 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4030
4031 static clockid_t thread_cpu_clockid(Thread* thread) {
4032 pthread_t tid = thread->osthread()->pthread_id();
4033 clockid_t clockid;
4034
4035 // Get thread clockid
4036 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4037 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4038 return clockid;
4039 }
4040
4041 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4042 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4043 // of a thread.
4044 //
4045 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4046 // the fast estimate available on the platform.
4047
4048 jlong os::current_thread_cpu_time() {
4049 if (os::Linux::supports_fast_thread_cpu_time()) {
4050 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4051 } else {
4052 // return user + sys since the cost is the same
4053 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4054 }
4055 }
4056
4057 jlong os::thread_cpu_time(Thread* thread) {
4058 // consistent with what current_thread_cpu_time() returns
4059 if (os::Linux::supports_fast_thread_cpu_time()) {
4060 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4061 } else {
4062 return slow_thread_cpu_time(thread, true /* user + sys */);
4063 }
4064 }
4065
4066 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4067 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4068 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4069 } else {
4070 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4071 }
4072 }
4073
4074 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4075 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4076 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4077 } else {
4078 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4079 }
4080 }
4081
4082 //
4083 // -1 on error.
4084 //
4085
4086 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4087 static bool proc_pid_cpu_avail = true;
4088 static bool proc_task_unchecked = true;
4089 static const char *proc_stat_path = "/proc/%d/stat";
4090 pid_t tid = thread->osthread()->thread_id();
4091 int i;
4092 char *s;
4093 char stat[2048];
4094 int statlen;
4095 char proc_name[64];
4096 int count;
4097 long sys_time, user_time;
4098 char string[64];
4099 int idummy;
4100 long ldummy;
4101 FILE *fp;
4102
4103 // We first try accessing /proc/<pid>/cpu since this is faster to
4104 // process. If this file is not present (linux kernels 2.5 and above)
4105 // then we open /proc/<pid>/stat.
4106 if ( proc_pid_cpu_avail ) {
4107 sprintf(proc_name, "/proc/%d/cpu", tid);
4108 fp = fopen(proc_name, "r");
4109 if ( fp != NULL ) {
4110 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4111 fclose(fp);
4112 if ( count != 3 ) return -1;
4113
4114 if (user_sys_cpu_time) {
4115 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4116 } else {
4117 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4118 }
4119 }
4120 else proc_pid_cpu_avail = false;
4121 }
4122
4123 // The /proc/<tid>/stat aggregates per-process usage on
4124 // new Linux kernels 2.6+ where NPTL is supported.
4125 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4126 // See bug 6328462.
4127 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4128 // and possibly in some other cases, so we check its availability.
4129 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4130 // This is executed only once
4131 proc_task_unchecked = false;
4132 fp = fopen("/proc/self/task", "r");
4133 if (fp != NULL) {
4134 proc_stat_path = "/proc/self/task/%d/stat";
4135 fclose(fp);
4136 }
4137 }
4138
4139 sprintf(proc_name, proc_stat_path, tid);
4140 fp = fopen(proc_name, "r");
4141 if ( fp == NULL ) return -1;
4142 statlen = fread(stat, 1, 2047, fp);
4143 stat[statlen] = '\0';
4144 fclose(fp);
4145
4146 // Skip pid and the command string. Note that we could be dealing with
4147 // weird command names, e.g. user could decide to rename java launcher
4148 // to "java 1.4.2 :)", then the stat file would look like
4149 // 1234 (java 1.4.2 :)) R ... ...
4150 // We don't really need to know the command string, just find the last
4151 // occurrence of ")" and then start parsing from there. See bug 4726580.
4152 s = strrchr(stat, ')');
4153 i = 0;
4154 if (s == NULL ) return -1;
4155
4156 // Skip blank chars
4157 do s++; while (isspace(*s));
4158
4159 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4160 &idummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4161 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4162 &user_time, &sys_time);
4163 if ( count != 13 ) return -1;
4164 if (user_sys_cpu_time) {
4165 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4166 } else {
4167 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4168 }
4169 }
4170
4171 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4172 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4173 info_ptr->may_skip_backward = false; // elapsed time not wall time
4174 info_ptr->may_skip_forward = false; // elapsed time not wall time
4175 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4176 }
4177
4178 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4179 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4180 info_ptr->may_skip_backward = false; // elapsed time not wall time
4181 info_ptr->may_skip_forward = false; // elapsed time not wall time
4182 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4183 }
4184
4185 bool os::is_thread_cpu_time_supported() {
4186 return true;
4187 }
4188
4189 // System loadavg support. Returns -1 if load average cannot be obtained.
4190 // Linux doesn't yet have a (official) notion of processor sets,
4191 // so just return the system wide load average.
4192 int os::loadavg(double loadavg[], int nelem) {
4193 return ::getloadavg(loadavg, nelem);
4194 }
4195
4196 void os::pause() {
4197 char filename[MAX_PATH];
4198 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4199 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4200 } else {
4201 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4202 }
4203
4204 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4205 if (fd != -1) {
4206 struct stat buf;
4207 close(fd);
4208 while (::stat(filename, &buf) == 0) {
4209 (void)::poll(NULL, 0, 100);
4210 }
4211 } else {
4212 jio_fprintf(stderr,
4213 "Could not open pause file '%s', continuing immediately.\n", filename);
4214 }
4215 }
4216
4217 extern "C" {
4218
4219 /**
4220 * NOTE: the following code is to keep the green threads code
4221 * in the libjava.so happy. Once the green threads is removed,
4222 * these code will no longer be needed.
4223 */
4224 int
4225 jdk_waitpid(pid_t pid, int* status, int options) {
4226 return waitpid(pid, status, options);
4227 }
4228
4229 int
4230 fork1() {
4231 return fork();
4232 }
4233
4234 int
4235 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4236 return sem_init(sem, pshared, value);
4237 }
4238
4239 int
4240 jdk_sem_post(sem_t *sem) {
4241 return sem_post(sem);
4242 }
4243
4244 int
4245 jdk_sem_wait(sem_t *sem) {
4246 return sem_wait(sem);
4247 }
4248
4249 int
4250 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4251 return pthread_sigmask(how , newmask, oldmask);
4252 }
4253
4254 }
4255
4256 // Refer to the comments in os_solaris.cpp park-unpark.
4257 //
4258 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4259 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4260 // For specifics regarding the bug see GLIBC BUGID 261237 :
4261 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4262 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4263 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4264 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4265 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4266 // and monitorenter when we're using 1-0 locking. All those operations may result in
4267 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4268 // of libpthread avoids the problem, but isn't practical.
4269 //
4270 // Possible remedies:
4271 //
4272 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4273 // This is palliative and probabilistic, however. If the thread is preempted
4274 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4275 // than the minimum period may have passed, and the abstime may be stale (in the
4276 // past) resultin in a hang. Using this technique reduces the odds of a hang
4277 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4278 //
4279 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4280 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4281 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4282 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4283 // thread.
4284 //
4285 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4286 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4287 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4288 // This also works well. In fact it avoids kernel-level scalability impediments
4289 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4290 // timers in a graceful fashion.
4291 //
4292 // 4. When the abstime value is in the past it appears that control returns
4293 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4294 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4295 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4296 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4297 // It may be possible to avoid reinitialization by checking the return
4298 // value from pthread_cond_timedwait(). In addition to reinitializing the
4299 // condvar we must establish the invariant that cond_signal() is only called
4300 // within critical sections protected by the adjunct mutex. This prevents
4301 // cond_signal() from "seeing" a condvar that's in the midst of being
4302 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4303 // desirable signal-after-unlock optimization that avoids futile context switching.
4304 //
4305 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4306 // structure when a condvar is used or initialized. cond_destroy() would
4307 // release the helper structure. Our reinitialize-after-timedwait fix
4308 // put excessive stress on malloc/free and locks protecting the c-heap.
4309 //
4310 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4311 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4312 // and only enabling the work-around for vulnerable environments.
4313
4314 // utility to compute the abstime argument to timedwait:
4315 // millis is the relative timeout time
4316 // abstime will be the absolute timeout time
4317 // TODO: replace compute_abstime() with unpackTime()
4318
4319 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4320 if (millis < 0) millis = 0;
4321 struct timeval now;
4322 int status = gettimeofday(&now, NULL);
4323 assert(status == 0, "gettimeofday");
4324 jlong seconds = millis / 1000;
4325 millis %= 1000;
4326 if (seconds > 50000000) { // see man cond_timedwait(3T)
4327 seconds = 50000000;
4328 }
4329 abstime->tv_sec = now.tv_sec + seconds;
4330 long usec = now.tv_usec + millis * 1000;
4331 if (usec >= 1000000) {
4332 abstime->tv_sec += 1;
4333 usec -= 1000000;
4334 }
4335 abstime->tv_nsec = usec * 1000;
4336 return abstime;
4337 }
4338
4339
4340 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4341 // Conceptually TryPark() should be equivalent to park(0).
4342
4343 int os::PlatformEvent::TryPark() {
4344 for (;;) {
4345 const int v = _Event ;
4346 guarantee ((v == 0) || (v == 1), "invariant") ;
4347 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4348 }
4349 }
4350
4351 void os::PlatformEvent::park() { // AKA "down()"
4352 // Invariant: Only the thread associated with the Event/PlatformEvent
4353 // may call park().
4354 // TODO: assert that _Assoc != NULL or _Assoc == Self
4355 int v ;
4356 for (;;) {
4357 v = _Event ;
4358 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4359 }
4360 guarantee (v >= 0, "invariant") ;
4361 if (v == 0) {
4362 // Do this the hard way by blocking ...
4363 int status = pthread_mutex_lock(_mutex);
4364 assert_status(status == 0, status, "mutex_lock");
4365 guarantee (_nParked == 0, "invariant") ;
4366 ++ _nParked ;
4367 while (_Event < 0) {
4368 status = pthread_cond_wait(_cond, _mutex);
4369 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4370 // Treat this the same as if the wait was interrupted
4371 if (status == ETIME) { status = EINTR; }
4372 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4373 }
4374 -- _nParked ;
4375
4376 // In theory we could move the ST of 0 into _Event past the unlock(),
4377 // but then we'd need a MEMBAR after the ST.
4378 _Event = 0 ;
4379 status = pthread_mutex_unlock(_mutex);
4380 assert_status(status == 0, status, "mutex_unlock");
4381 }
4382 guarantee (_Event >= 0, "invariant") ;
4383 }
4384
4385 int os::PlatformEvent::park(jlong millis) {
4386 guarantee (_nParked == 0, "invariant") ;
4387
4388 int v ;
4389 for (;;) {
4390 v = _Event ;
4391 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4392 }
4393 guarantee (v >= 0, "invariant") ;
4394 if (v != 0) return OS_OK ;
4395
4396 // We do this the hard way, by blocking the thread.
4397 // Consider enforcing a minimum timeout value.
4398 struct timespec abst;
4399 compute_abstime(&abst, millis);
4400
4401 int ret = OS_TIMEOUT;
4402 int status = pthread_mutex_lock(_mutex);
4403 assert_status(status == 0, status, "mutex_lock");
4404 guarantee (_nParked == 0, "invariant") ;
4405 ++_nParked ;
4406
4407 // Object.wait(timo) will return because of
4408 // (a) notification
4409 // (b) timeout
4410 // (c) thread.interrupt
4411 //
4412 // Thread.interrupt and object.notify{All} both call Event::set.
4413 // That is, we treat thread.interrupt as a special case of notification.
4414 // The underlying Solaris implementation, cond_timedwait, admits
4415 // spurious/premature wakeups, but the JLS/JVM spec prevents the
4416 // JVM from making those visible to Java code. As such, we must
4417 // filter out spurious wakeups. We assume all ETIME returns are valid.
4418 //
4419 // TODO: properly differentiate simultaneous notify+interrupt.
4420 // In that case, we should propagate the notify to another waiter.
4421
4422 while (_Event < 0) {
4423 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4424 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4425 pthread_cond_destroy (_cond);
4426 pthread_cond_init (_cond, NULL) ;
4427 }
4428 assert_status(status == 0 || status == EINTR ||
4429 status == ETIME || status == ETIMEDOUT,
4430 status, "cond_timedwait");
4431 if (!FilterSpuriousWakeups) break ; // previous semantics
4432 if (status == ETIME || status == ETIMEDOUT) break ;
4433 // We consume and ignore EINTR and spurious wakeups.
4434 }
4435 --_nParked ;
4436 if (_Event >= 0) {
4437 ret = OS_OK;
4438 }
4439 _Event = 0 ;
4440 status = pthread_mutex_unlock(_mutex);
4441 assert_status(status == 0, status, "mutex_unlock");
4442 assert (_nParked == 0, "invariant") ;
4443 return ret;
4444 }
4445
4446 void os::PlatformEvent::unpark() {
4447 int v, AnyWaiters ;
4448 for (;;) {
4449 v = _Event ;
4450 if (v > 0) {
4451 // The LD of _Event could have reordered or be satisfied
4452 // by a read-aside from this processor's write buffer.
4453 // To avoid problems execute a barrier and then
4454 // ratify the value.
4455 OrderAccess::fence() ;
4456 if (_Event == v) return ;
4457 continue ;
4458 }
4459 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4460 }
4461 if (v < 0) {
4462 // Wait for the thread associated with the event to vacate
4463 int status = pthread_mutex_lock(_mutex);
4464 assert_status(status == 0, status, "mutex_lock");
4465 AnyWaiters = _nParked ;
4466 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4467 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4468 AnyWaiters = 0 ;
4469 pthread_cond_signal (_cond);
4470 }
4471 status = pthread_mutex_unlock(_mutex);
4472 assert_status(status == 0, status, "mutex_unlock");
4473 if (AnyWaiters != 0) {
4474 status = pthread_cond_signal(_cond);
4475 assert_status(status == 0, status, "cond_signal");
4476 }
4477 }
4478
4479 // Note that we signal() _after dropping the lock for "immortal" Events.
4480 // This is safe and avoids a common class of futile wakeups. In rare
4481 // circumstances this can cause a thread to return prematurely from
4482 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4483 // simply re-test the condition and re-park itself.
4484 }
4485
4486
4487 // JSR166
4488 // -------------------------------------------------------
4489
4490 /*
4491 * The solaris and linux implementations of park/unpark are fairly
4492 * conservative for now, but can be improved. They currently use a
4493 * mutex/condvar pair, plus a a count.
4494 * Park decrements count if > 0, else does a condvar wait. Unpark
4495 * sets count to 1 and signals condvar. Only one thread ever waits
4496 * on the condvar. Contention seen when trying to park implies that someone
4497 * is unparking you, so don't wait. And spurious returns are fine, so there
4498 * is no need to track notifications.
4499 */
4500
4501
4502 #define NANOSECS_PER_SEC 1000000000
4503 #define NANOSECS_PER_MILLISEC 1000000
4504 #define MAX_SECS 100000000
4505 /*
4506 * This code is common to linux and solaris and will be moved to a
4507 * common place in dolphin.
4508 *
4509 * The passed in time value is either a relative time in nanoseconds
4510 * or an absolute time in milliseconds. Either way it has to be unpacked
4511 * into suitable seconds and nanoseconds components and stored in the
4512 * given timespec structure.
4513 * Given time is a 64-bit value and the time_t used in the timespec is only
4514 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4515 * overflow if times way in the future are given. Further on Solaris versions
4516 * prior to 10 there is a restriction (see cond_timedwait) that the specified
4517 * number of seconds, in abstime, is less than current_time + 100,000,000.
4518 * As it will be 28 years before "now + 100000000" will overflow we can
4519 * ignore overflow and just impose a hard-limit on seconds using the value
4520 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4521 * years from "now".
4522 */
4523
4524 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4525 assert (time > 0, "convertTime");
4526
4527 struct timeval now;
4528 int status = gettimeofday(&now, NULL);
4529 assert(status == 0, "gettimeofday");
4530
4531 time_t max_secs = now.tv_sec + MAX_SECS;
4532
4533 if (isAbsolute) {
4534 jlong secs = time / 1000;
4535 if (secs > max_secs) {
4536 absTime->tv_sec = max_secs;
4537 }
4538 else {
4539 absTime->tv_sec = secs;
4540 }
4541 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
4542 }
4543 else {
4544 jlong secs = time / NANOSECS_PER_SEC;
4545 if (secs >= MAX_SECS) {
4546 absTime->tv_sec = max_secs;
4547 absTime->tv_nsec = 0;
4548 }
4549 else {
4550 absTime->tv_sec = now.tv_sec + secs;
4551 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
4552 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
4553 absTime->tv_nsec -= NANOSECS_PER_SEC;
4554 ++absTime->tv_sec; // note: this must be <= max_secs
4555 }
4556 }
4557 }
4558 assert(absTime->tv_sec >= 0, "tv_sec < 0");
4559 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
4560 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
4561 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
4562 }
4563
4564 void Parker::park(bool isAbsolute, jlong time) {
4565 // Optional fast-path check:
4566 // Return immediately if a permit is available.
4567 if (_counter > 0) {
4568 _counter = 0 ;
4569 return ;
4570 }
4571
4572 Thread* thread = Thread::current();
4573 assert(thread->is_Java_thread(), "Must be JavaThread");
4574 JavaThread *jt = (JavaThread *)thread;
4575
4576 // Optional optimization -- avoid state transitions if there's an interrupt pending.
4577 // Check interrupt before trying to wait
4578 if (Thread::is_interrupted(thread, false)) {
4579 return;
4580 }
4581
4582 // Next, demultiplex/decode time arguments
4583 timespec absTime;
4584 if (time < 0) { // don't wait at all
4585 return;
4586 }
4587 if (time > 0) {
4588 unpackTime(&absTime, isAbsolute, time);
4589 }
4590
4591
4592 // Enter safepoint region
4593 // Beware of deadlocks such as 6317397.
4594 // The per-thread Parker:: mutex is a classic leaf-lock.
4595 // In particular a thread must never block on the Threads_lock while
4596 // holding the Parker:: mutex. If safepoints are pending both the
4597 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
4598 ThreadBlockInVM tbivm(jt);
4599
4600 // Don't wait if cannot get lock since interference arises from
4601 // unblocking. Also. check interrupt before trying wait
4602 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
4603 return;
4604 }
4605
4606 int status ;
4607 if (_counter > 0) { // no wait needed
4608 _counter = 0;
4609 status = pthread_mutex_unlock(_mutex);
4610 assert (status == 0, "invariant") ;
4611 return;
4612 }
4613
4614 #ifdef ASSERT
4615 // Don't catch signals while blocked; let the running threads have the signals.
4616 // (This allows a debugger to break into the running thread.)
4617 sigset_t oldsigs;
4618 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
4619 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
4620 #endif
4621
4622 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4623 jt->set_suspend_equivalent();
4624 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
4625
4626 if (time == 0) {
4627 status = pthread_cond_wait (_cond, _mutex) ;
4628 } else {
4629 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
4630 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4631 pthread_cond_destroy (_cond) ;
4632 pthread_cond_init (_cond, NULL);
4633 }
4634 }
4635 assert_status(status == 0 || status == EINTR ||
4636 status == ETIME || status == ETIMEDOUT,
4637 status, "cond_timedwait");
4638
4639 #ifdef ASSERT
4640 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
4641 #endif
4642
4643 _counter = 0 ;
4644 status = pthread_mutex_unlock(_mutex) ;
4645 assert_status(status == 0, status, "invariant") ;
4646 // If externally suspended while waiting, re-suspend
4647 if (jt->handle_special_suspend_equivalent_condition()) {
4648 jt->java_suspend_self();
4649 }
4650
4651 }
4652
4653 void Parker::unpark() {
4654 int s, status ;
4655 status = pthread_mutex_lock(_mutex);
4656 assert (status == 0, "invariant") ;
4657 s = _counter;
4658 _counter = 1;
4659 if (s < 1) {
4660 if (WorkAroundNPTLTimedWaitHang) {
4661 status = pthread_cond_signal (_cond) ;
4662 assert (status == 0, "invariant") ;
4663 status = pthread_mutex_unlock(_mutex);
4664 assert (status == 0, "invariant") ;
4665 } else {
4666 status = pthread_mutex_unlock(_mutex);
4667 assert (status == 0, "invariant") ;
4668 status = pthread_cond_signal (_cond) ;
4669 assert (status == 0, "invariant") ;
4670 }
4671 } else {
4672 pthread_mutex_unlock(_mutex);
4673 assert (status == 0, "invariant") ;
4674 }
4675 }
4676
4677
4678 extern char** environ;
4679
4680 #ifndef __NR_fork
4681 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
4682 #endif
4683
4684 #ifndef __NR_execve
4685 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
4686 #endif
4687
4688 // Run the specified command in a separate process. Return its exit value,
4689 // or -1 on failure (e.g. can't fork a new process).
4690 // Unlike system(), this function can be called from signal handler. It
4691 // doesn't block SIGINT et al.
4692 int os::fork_and_exec(char* cmd) {
4693 const char * argv[4] = {"sh", "-c", cmd, NULL};
4694
4695 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
4696 // pthread_atfork handlers and reset pthread library. All we need is a
4697 // separate process to execve. Make a direct syscall to fork process.
4698 // On IA64 there's no fork syscall, we have to use fork() and hope for
4699 // the best...
4700 pid_t pid = NOT_IA64(syscall(__NR_fork);)
4701 IA64_ONLY(fork();)
4702
4703 if (pid < 0) {
4704 // fork failed
4705 return -1;
4706
4707 } else if (pid == 0) {
4708 // child process
4709
4710 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
4711 // first to kill every thread on the thread list. Because this list is
4712 // not reset by fork() (see notes above), execve() will instead kill
4713 // every thread in the parent process. We know this is the only thread
4714 // in the new process, so make a system call directly.
4715 // IA64 should use normal execve() from glibc to match the glibc fork()
4716 // above.
4717 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
4718 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
4719
4720 // execve failed
4721 _exit(-1);
4722
4723 } else {
4724 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
4725 // care about the actual exit code, for now.
4726
4727 int status;
4728
4729 // Wait for the child process to exit. This returns immediately if
4730 // the child has already exited. */
4731 while (waitpid(pid, &status, 0) < 0) {
4732 switch (errno) {
4733 case ECHILD: return 0;
4734 case EINTR: break;
4735 default: return -1;
4736 }
4737 }
4738
4739 if (WIFEXITED(status)) {
4740 // The child exited normally; get its exit code.
4741 return WEXITSTATUS(status);
4742 } else if (WIFSIGNALED(status)) {
4743 // The child exited because of a signal
4744 // The best value to return is 0x80 + signal number,
4745 // because that is what all Unix shells do, and because
4746 // it allows callers to distinguish between process exit and
4747 // process death by signal.
4748 return 0x80 + WTERMSIG(status);
4749 } else {
4750 // Unknown exit code; pass it through
4751 return status;
4752 }
4753 }
4754 }