Wed, 16 Feb 2011 13:30:31 -0800
7013964: openjdk LICENSE file needs rebranding
Reviewed-by: darcy, katleman, jjg
1 /*
2 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 # define __STDC_FORMAT_MACROS
27 // no precompiled headers
28 #include "classfile/classLoader.hpp"
29 #include "classfile/systemDictionary.hpp"
30 #include "classfile/vmSymbols.hpp"
31 #include "code/icBuffer.hpp"
32 #include "code/vtableStubs.hpp"
33 #include "compiler/compileBroker.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "jvm_linux.h"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "mutex_linux.inline.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/extendedPC.hpp"
46 #include "runtime/globals.hpp"
47 #include "runtime/interfaceSupport.hpp"
48 #include "runtime/java.hpp"
49 #include "runtime/javaCalls.hpp"
50 #include "runtime/mutexLocker.hpp"
51 #include "runtime/objectMonitor.hpp"
52 #include "runtime/osThread.hpp"
53 #include "runtime/perfMemory.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #include "runtime/statSampler.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "runtime/threadCritical.hpp"
58 #include "runtime/timer.hpp"
59 #include "services/attachListener.hpp"
60 #include "services/runtimeService.hpp"
61 #include "thread_linux.inline.hpp"
62 #include "utilities/decoder.hpp"
63 #include "utilities/defaultStream.hpp"
64 #include "utilities/events.hpp"
65 #include "utilities/growableArray.hpp"
66 #include "utilities/vmError.hpp"
67 #ifdef TARGET_ARCH_x86
68 # include "assembler_x86.inline.hpp"
69 # include "nativeInst_x86.hpp"
70 #endif
71 #ifdef TARGET_ARCH_sparc
72 # include "assembler_sparc.inline.hpp"
73 # include "nativeInst_sparc.hpp"
74 #endif
75 #ifdef TARGET_ARCH_zero
76 # include "assembler_zero.inline.hpp"
77 # include "nativeInst_zero.hpp"
78 #endif
79 #ifdef COMPILER1
80 #include "c1/c1_Runtime1.hpp"
81 #endif
82 #ifdef COMPILER2
83 #include "opto/runtime.hpp"
84 #endif
86 // put OS-includes here
87 # include <sys/types.h>
88 # include <sys/mman.h>
89 # include <sys/stat.h>
90 # include <sys/select.h>
91 # include <pthread.h>
92 # include <signal.h>
93 # include <errno.h>
94 # include <dlfcn.h>
95 # include <stdio.h>
96 # include <unistd.h>
97 # include <sys/resource.h>
98 # include <pthread.h>
99 # include <sys/stat.h>
100 # include <sys/time.h>
101 # include <sys/times.h>
102 # include <sys/utsname.h>
103 # include <sys/socket.h>
104 # include <sys/wait.h>
105 # include <pwd.h>
106 # include <poll.h>
107 # include <semaphore.h>
108 # include <fcntl.h>
109 # include <string.h>
110 # include <syscall.h>
111 # include <sys/sysinfo.h>
112 # include <gnu/libc-version.h>
113 # include <sys/ipc.h>
114 # include <sys/shm.h>
115 # include <link.h>
116 # include <stdint.h>
117 # include <inttypes.h>
118 # include <sys/ioctl.h>
120 #define MAX_PATH (2 * K)
122 // for timer info max values which include all bits
123 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
124 #define SEC_IN_NANOSECS 1000000000LL
126 ////////////////////////////////////////////////////////////////////////////////
127 // global variables
128 julong os::Linux::_physical_memory = 0;
130 address os::Linux::_initial_thread_stack_bottom = NULL;
131 uintptr_t os::Linux::_initial_thread_stack_size = 0;
133 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
134 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
135 Mutex* os::Linux::_createThread_lock = NULL;
136 pthread_t os::Linux::_main_thread;
137 int os::Linux::_page_size = -1;
138 bool os::Linux::_is_floating_stack = false;
139 bool os::Linux::_is_NPTL = false;
140 bool os::Linux::_supports_fast_thread_cpu_time = false;
141 const char * os::Linux::_glibc_version = NULL;
142 const char * os::Linux::_libpthread_version = NULL;
144 static jlong initial_time_count=0;
146 static int clock_tics_per_sec = 100;
148 // For diagnostics to print a message once. see run_periodic_checks
149 static sigset_t check_signal_done;
150 static bool check_signals = true;;
152 static pid_t _initial_pid = 0;
154 /* Signal number used to suspend/resume a thread */
156 /* do not use any signal number less than SIGSEGV, see 4355769 */
157 static int SR_signum = SIGUSR2;
158 sigset_t SR_sigset;
160 /* Used to protect dlsym() calls */
161 static pthread_mutex_t dl_mutex;
163 ////////////////////////////////////////////////////////////////////////////////
164 // utility functions
166 static int SR_initialize();
167 static int SR_finalize();
169 julong os::available_memory() {
170 return Linux::available_memory();
171 }
173 julong os::Linux::available_memory() {
174 // values in struct sysinfo are "unsigned long"
175 struct sysinfo si;
176 sysinfo(&si);
178 return (julong)si.freeram * si.mem_unit;
179 }
181 julong os::physical_memory() {
182 return Linux::physical_memory();
183 }
185 julong os::allocatable_physical_memory(julong size) {
186 #ifdef _LP64
187 return size;
188 #else
189 julong result = MIN2(size, (julong)3800*M);
190 if (!is_allocatable(result)) {
191 // See comments under solaris for alignment considerations
192 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
193 result = MIN2(size, reasonable_size);
194 }
195 return result;
196 #endif // _LP64
197 }
199 ////////////////////////////////////////////////////////////////////////////////
200 // environment support
202 bool os::getenv(const char* name, char* buf, int len) {
203 const char* val = ::getenv(name);
204 if (val != NULL && strlen(val) < (size_t)len) {
205 strcpy(buf, val);
206 return true;
207 }
208 if (len > 0) buf[0] = 0; // return a null string
209 return false;
210 }
213 // Return true if user is running as root.
215 bool os::have_special_privileges() {
216 static bool init = false;
217 static bool privileges = false;
218 if (!init) {
219 privileges = (getuid() != geteuid()) || (getgid() != getegid());
220 init = true;
221 }
222 return privileges;
223 }
226 #ifndef SYS_gettid
227 // i386: 224, ia64: 1105, amd64: 186, sparc 143
228 #ifdef __ia64__
229 #define SYS_gettid 1105
230 #elif __i386__
231 #define SYS_gettid 224
232 #elif __amd64__
233 #define SYS_gettid 186
234 #elif __sparc__
235 #define SYS_gettid 143
236 #else
237 #error define gettid for the arch
238 #endif
239 #endif
241 // Cpu architecture string
242 #if defined(ZERO)
243 static char cpu_arch[] = ZERO_LIBARCH;
244 #elif defined(IA64)
245 static char cpu_arch[] = "ia64";
246 #elif defined(IA32)
247 static char cpu_arch[] = "i386";
248 #elif defined(AMD64)
249 static char cpu_arch[] = "amd64";
250 #elif defined(ARM)
251 static char cpu_arch[] = "arm";
252 #elif defined(PPC)
253 static char cpu_arch[] = "ppc";
254 #elif defined(SPARC)
255 # ifdef _LP64
256 static char cpu_arch[] = "sparcv9";
257 # else
258 static char cpu_arch[] = "sparc";
259 # endif
260 #else
261 #error Add appropriate cpu_arch setting
262 #endif
265 // pid_t gettid()
266 //
267 // Returns the kernel thread id of the currently running thread. Kernel
268 // thread id is used to access /proc.
269 //
270 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
271 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
272 //
273 pid_t os::Linux::gettid() {
274 int rslt = syscall(SYS_gettid);
275 if (rslt == -1) {
276 // old kernel, no NPTL support
277 return getpid();
278 } else {
279 return (pid_t)rslt;
280 }
281 }
283 // Most versions of linux have a bug where the number of processors are
284 // determined by looking at the /proc file system. In a chroot environment,
285 // the system call returns 1. This causes the VM to act as if it is
286 // a single processor and elide locking (see is_MP() call).
287 static bool unsafe_chroot_detected = false;
288 static const char *unstable_chroot_error = "/proc file system not found.\n"
289 "Java may be unstable running multithreaded in a chroot "
290 "environment on Linux when /proc filesystem is not mounted.";
292 void os::Linux::initialize_system_info() {
293 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
294 if (processor_count() == 1) {
295 pid_t pid = os::Linux::gettid();
296 char fname[32];
297 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
298 FILE *fp = fopen(fname, "r");
299 if (fp == NULL) {
300 unsafe_chroot_detected = true;
301 } else {
302 fclose(fp);
303 }
304 }
305 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
306 assert(processor_count() > 0, "linux error");
307 }
309 void os::init_system_properties_values() {
310 // char arch[12];
311 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
313 // The next steps are taken in the product version:
314 //
315 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
316 // This library should be located at:
317 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
318 //
319 // If "/jre/lib/" appears at the right place in the path, then we
320 // assume libjvm[_g].so is installed in a JDK and we use this path.
321 //
322 // Otherwise exit with message: "Could not create the Java virtual machine."
323 //
324 // The following extra steps are taken in the debugging version:
325 //
326 // If "/jre/lib/" does NOT appear at the right place in the path
327 // instead of exit check for $JAVA_HOME environment variable.
328 //
329 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
330 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
331 // it looks like libjvm[_g].so is installed there
332 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
333 //
334 // Otherwise exit.
335 //
336 // Important note: if the location of libjvm.so changes this
337 // code needs to be changed accordingly.
339 // The next few definitions allow the code to be verbatim:
340 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
341 #define getenv(n) ::getenv(n)
343 /*
344 * See ld(1):
345 * The linker uses the following search paths to locate required
346 * shared libraries:
347 * 1: ...
348 * ...
349 * 7: The default directories, normally /lib and /usr/lib.
350 */
351 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
352 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
353 #else
354 #define DEFAULT_LIBPATH "/lib:/usr/lib"
355 #endif
357 #define EXTENSIONS_DIR "/lib/ext"
358 #define ENDORSED_DIR "/lib/endorsed"
359 #define REG_DIR "/usr/java/packages"
361 {
362 /* sysclasspath, java_home, dll_dir */
363 {
364 char *home_path;
365 char *dll_path;
366 char *pslash;
367 char buf[MAXPATHLEN];
368 os::jvm_path(buf, sizeof(buf));
370 // Found the full path to libjvm.so.
371 // Now cut the path to <java_home>/jre if we can.
372 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
373 pslash = strrchr(buf, '/');
374 if (pslash != NULL)
375 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
376 dll_path = malloc(strlen(buf) + 1);
377 if (dll_path == NULL)
378 return;
379 strcpy(dll_path, buf);
380 Arguments::set_dll_dir(dll_path);
382 if (pslash != NULL) {
383 pslash = strrchr(buf, '/');
384 if (pslash != NULL) {
385 *pslash = '\0'; /* get rid of /<arch> */
386 pslash = strrchr(buf, '/');
387 if (pslash != NULL)
388 *pslash = '\0'; /* get rid of /lib */
389 }
390 }
392 home_path = malloc(strlen(buf) + 1);
393 if (home_path == NULL)
394 return;
395 strcpy(home_path, buf);
396 Arguments::set_java_home(home_path);
398 if (!set_boot_path('/', ':'))
399 return;
400 }
402 /*
403 * Where to look for native libraries
404 *
405 * Note: Due to a legacy implementation, most of the library path
406 * is set in the launcher. This was to accomodate linking restrictions
407 * on legacy Linux implementations (which are no longer supported).
408 * Eventually, all the library path setting will be done here.
409 *
410 * However, to prevent the proliferation of improperly built native
411 * libraries, the new path component /usr/java/packages is added here.
412 * Eventually, all the library path setting will be done here.
413 */
414 {
415 char *ld_library_path;
417 /*
418 * Construct the invariant part of ld_library_path. Note that the
419 * space for the colon and the trailing null are provided by the
420 * nulls included by the sizeof operator (so actually we allocate
421 * a byte more than necessary).
422 */
423 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
424 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
425 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
427 /*
428 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
429 * should always exist (until the legacy problem cited above is
430 * addressed).
431 */
432 char *v = getenv("LD_LIBRARY_PATH");
433 if (v != NULL) {
434 char *t = ld_library_path;
435 /* That's +1 for the colon and +1 for the trailing '\0' */
436 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
437 sprintf(ld_library_path, "%s:%s", v, t);
438 }
439 Arguments::set_library_path(ld_library_path);
440 }
442 /*
443 * Extensions directories.
444 *
445 * Note that the space for the colon and the trailing null are provided
446 * by the nulls included by the sizeof operator (so actually one byte more
447 * than necessary is allocated).
448 */
449 {
450 char *buf = malloc(strlen(Arguments::get_java_home()) +
451 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
452 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
453 Arguments::get_java_home());
454 Arguments::set_ext_dirs(buf);
455 }
457 /* Endorsed standards default directory. */
458 {
459 char * buf;
460 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
461 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
462 Arguments::set_endorsed_dirs(buf);
463 }
464 }
466 #undef malloc
467 #undef getenv
468 #undef EXTENSIONS_DIR
469 #undef ENDORSED_DIR
471 // Done
472 return;
473 }
475 ////////////////////////////////////////////////////////////////////////////////
476 // breakpoint support
478 void os::breakpoint() {
479 BREAKPOINT;
480 }
482 extern "C" void breakpoint() {
483 // use debugger to set breakpoint here
484 }
486 ////////////////////////////////////////////////////////////////////////////////
487 // signal support
489 debug_only(static bool signal_sets_initialized = false);
490 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
492 bool os::Linux::is_sig_ignored(int sig) {
493 struct sigaction oact;
494 sigaction(sig, (struct sigaction*)NULL, &oact);
495 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
496 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
497 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
498 return true;
499 else
500 return false;
501 }
503 void os::Linux::signal_sets_init() {
504 // Should also have an assertion stating we are still single-threaded.
505 assert(!signal_sets_initialized, "Already initialized");
506 // Fill in signals that are necessarily unblocked for all threads in
507 // the VM. Currently, we unblock the following signals:
508 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
509 // by -Xrs (=ReduceSignalUsage));
510 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
511 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
512 // the dispositions or masks wrt these signals.
513 // Programs embedding the VM that want to use the above signals for their
514 // own purposes must, at this time, use the "-Xrs" option to prevent
515 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
516 // (See bug 4345157, and other related bugs).
517 // In reality, though, unblocking these signals is really a nop, since
518 // these signals are not blocked by default.
519 sigemptyset(&unblocked_sigs);
520 sigemptyset(&allowdebug_blocked_sigs);
521 sigaddset(&unblocked_sigs, SIGILL);
522 sigaddset(&unblocked_sigs, SIGSEGV);
523 sigaddset(&unblocked_sigs, SIGBUS);
524 sigaddset(&unblocked_sigs, SIGFPE);
525 sigaddset(&unblocked_sigs, SR_signum);
527 if (!ReduceSignalUsage) {
528 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
529 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
530 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
531 }
532 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
533 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
534 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
535 }
536 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
537 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
538 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
539 }
540 }
541 // Fill in signals that are blocked by all but the VM thread.
542 sigemptyset(&vm_sigs);
543 if (!ReduceSignalUsage)
544 sigaddset(&vm_sigs, BREAK_SIGNAL);
545 debug_only(signal_sets_initialized = true);
547 }
549 // These are signals that are unblocked while a thread is running Java.
550 // (For some reason, they get blocked by default.)
551 sigset_t* os::Linux::unblocked_signals() {
552 assert(signal_sets_initialized, "Not initialized");
553 return &unblocked_sigs;
554 }
556 // These are the signals that are blocked while a (non-VM) thread is
557 // running Java. Only the VM thread handles these signals.
558 sigset_t* os::Linux::vm_signals() {
559 assert(signal_sets_initialized, "Not initialized");
560 return &vm_sigs;
561 }
563 // These are signals that are blocked during cond_wait to allow debugger in
564 sigset_t* os::Linux::allowdebug_blocked_signals() {
565 assert(signal_sets_initialized, "Not initialized");
566 return &allowdebug_blocked_sigs;
567 }
569 void os::Linux::hotspot_sigmask(Thread* thread) {
571 //Save caller's signal mask before setting VM signal mask
572 sigset_t caller_sigmask;
573 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
575 OSThread* osthread = thread->osthread();
576 osthread->set_caller_sigmask(caller_sigmask);
578 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
580 if (!ReduceSignalUsage) {
581 if (thread->is_VM_thread()) {
582 // Only the VM thread handles BREAK_SIGNAL ...
583 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
584 } else {
585 // ... all other threads block BREAK_SIGNAL
586 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
587 }
588 }
589 }
591 //////////////////////////////////////////////////////////////////////////////
592 // detecting pthread library
594 void os::Linux::libpthread_init() {
595 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
596 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
597 // generic name for earlier versions.
598 // Define macros here so we can build HotSpot on old systems.
599 # ifndef _CS_GNU_LIBC_VERSION
600 # define _CS_GNU_LIBC_VERSION 2
601 # endif
602 # ifndef _CS_GNU_LIBPTHREAD_VERSION
603 # define _CS_GNU_LIBPTHREAD_VERSION 3
604 # endif
606 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
607 if (n > 0) {
608 char *str = (char *)malloc(n);
609 confstr(_CS_GNU_LIBC_VERSION, str, n);
610 os::Linux::set_glibc_version(str);
611 } else {
612 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
613 static char _gnu_libc_version[32];
614 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
615 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
616 os::Linux::set_glibc_version(_gnu_libc_version);
617 }
619 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
620 if (n > 0) {
621 char *str = (char *)malloc(n);
622 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
623 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
624 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
625 // is the case. LinuxThreads has a hard limit on max number of threads.
626 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
627 // On the other hand, NPTL does not have such a limit, sysconf()
628 // will return -1 and errno is not changed. Check if it is really NPTL.
629 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
630 strstr(str, "NPTL") &&
631 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
632 free(str);
633 os::Linux::set_libpthread_version("linuxthreads");
634 } else {
635 os::Linux::set_libpthread_version(str);
636 }
637 } else {
638 // glibc before 2.3.2 only has LinuxThreads.
639 os::Linux::set_libpthread_version("linuxthreads");
640 }
642 if (strstr(libpthread_version(), "NPTL")) {
643 os::Linux::set_is_NPTL();
644 } else {
645 os::Linux::set_is_LinuxThreads();
646 }
648 // LinuxThreads have two flavors: floating-stack mode, which allows variable
649 // stack size; and fixed-stack mode. NPTL is always floating-stack.
650 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
651 os::Linux::set_is_floating_stack();
652 }
653 }
655 /////////////////////////////////////////////////////////////////////////////
656 // thread stack
658 // Force Linux kernel to expand current thread stack. If "bottom" is close
659 // to the stack guard, caller should block all signals.
660 //
661 // MAP_GROWSDOWN:
662 // A special mmap() flag that is used to implement thread stacks. It tells
663 // kernel that the memory region should extend downwards when needed. This
664 // allows early versions of LinuxThreads to only mmap the first few pages
665 // when creating a new thread. Linux kernel will automatically expand thread
666 // stack as needed (on page faults).
667 //
668 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
669 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
670 // region, it's hard to tell if the fault is due to a legitimate stack
671 // access or because of reading/writing non-exist memory (e.g. buffer
672 // overrun). As a rule, if the fault happens below current stack pointer,
673 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
674 // application (see Linux kernel fault.c).
675 //
676 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
677 // stack overflow detection.
678 //
679 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
680 // not use this flag. However, the stack of initial thread is not created
681 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
682 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
683 // and then attach the thread to JVM.
684 //
685 // To get around the problem and allow stack banging on Linux, we need to
686 // manually expand thread stack after receiving the SIGSEGV.
687 //
688 // There are two ways to expand thread stack to address "bottom", we used
689 // both of them in JVM before 1.5:
690 // 1. adjust stack pointer first so that it is below "bottom", and then
691 // touch "bottom"
692 // 2. mmap() the page in question
693 //
694 // Now alternate signal stack is gone, it's harder to use 2. For instance,
695 // if current sp is already near the lower end of page 101, and we need to
696 // call mmap() to map page 100, it is possible that part of the mmap() frame
697 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
698 // That will destroy the mmap() frame and cause VM to crash.
699 //
700 // The following code works by adjusting sp first, then accessing the "bottom"
701 // page to force a page fault. Linux kernel will then automatically expand the
702 // stack mapping.
703 //
704 // _expand_stack_to() assumes its frame size is less than page size, which
705 // should always be true if the function is not inlined.
707 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
708 #define NOINLINE
709 #else
710 #define NOINLINE __attribute__ ((noinline))
711 #endif
713 static void _expand_stack_to(address bottom) NOINLINE;
715 static void _expand_stack_to(address bottom) {
716 address sp;
717 size_t size;
718 volatile char *p;
720 // Adjust bottom to point to the largest address within the same page, it
721 // gives us a one-page buffer if alloca() allocates slightly more memory.
722 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
723 bottom += os::Linux::page_size() - 1;
725 // sp might be slightly above current stack pointer; if that's the case, we
726 // will alloca() a little more space than necessary, which is OK. Don't use
727 // os::current_stack_pointer(), as its result can be slightly below current
728 // stack pointer, causing us to not alloca enough to reach "bottom".
729 sp = (address)&sp;
731 if (sp > bottom) {
732 size = sp - bottom;
733 p = (volatile char *)alloca(size);
734 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
735 p[0] = '\0';
736 }
737 }
739 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
740 assert(t!=NULL, "just checking");
741 assert(t->osthread()->expanding_stack(), "expand should be set");
742 assert(t->stack_base() != NULL, "stack_base was not initialized");
744 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
745 sigset_t mask_all, old_sigset;
746 sigfillset(&mask_all);
747 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
748 _expand_stack_to(addr);
749 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
750 return true;
751 }
752 return false;
753 }
755 //////////////////////////////////////////////////////////////////////////////
756 // create new thread
758 static address highest_vm_reserved_address();
760 // check if it's safe to start a new thread
761 static bool _thread_safety_check(Thread* thread) {
762 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
763 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
764 // Heap is mmap'ed at lower end of memory space. Thread stacks are
765 // allocated (MAP_FIXED) from high address space. Every thread stack
766 // occupies a fixed size slot (usually 2Mbytes, but user can change
767 // it to other values if they rebuild LinuxThreads).
768 //
769 // Problem with MAP_FIXED is that mmap() can still succeed even part of
770 // the memory region has already been mmap'ed. That means if we have too
771 // many threads and/or very large heap, eventually thread stack will
772 // collide with heap.
773 //
774 // Here we try to prevent heap/stack collision by comparing current
775 // stack bottom with the highest address that has been mmap'ed by JVM
776 // plus a safety margin for memory maps created by native code.
777 //
778 // This feature can be disabled by setting ThreadSafetyMargin to 0
779 //
780 if (ThreadSafetyMargin > 0) {
781 address stack_bottom = os::current_stack_base() - os::current_stack_size();
783 // not safe if our stack extends below the safety margin
784 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
785 } else {
786 return true;
787 }
788 } else {
789 // Floating stack LinuxThreads or NPTL:
790 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
791 // there's not enough space left, pthread_create() will fail. If we come
792 // here, that means enough space has been reserved for stack.
793 return true;
794 }
795 }
797 // Thread start routine for all newly created threads
798 static void *java_start(Thread *thread) {
799 // Try to randomize the cache line index of hot stack frames.
800 // This helps when threads of the same stack traces evict each other's
801 // cache lines. The threads can be either from the same JVM instance, or
802 // from different JVM instances. The benefit is especially true for
803 // processors with hyperthreading technology.
804 static int counter = 0;
805 int pid = os::current_process_id();
806 alloca(((pid ^ counter++) & 7) * 128);
808 ThreadLocalStorage::set_thread(thread);
810 OSThread* osthread = thread->osthread();
811 Monitor* sync = osthread->startThread_lock();
813 // non floating stack LinuxThreads needs extra check, see above
814 if (!_thread_safety_check(thread)) {
815 // notify parent thread
816 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
817 osthread->set_state(ZOMBIE);
818 sync->notify_all();
819 return NULL;
820 }
822 // thread_id is kernel thread id (similar to Solaris LWP id)
823 osthread->set_thread_id(os::Linux::gettid());
825 if (UseNUMA) {
826 int lgrp_id = os::numa_get_group_id();
827 if (lgrp_id != -1) {
828 thread->set_lgrp_id(lgrp_id);
829 }
830 }
831 // initialize signal mask for this thread
832 os::Linux::hotspot_sigmask(thread);
834 // initialize floating point control register
835 os::Linux::init_thread_fpu_state();
837 // handshaking with parent thread
838 {
839 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
841 // notify parent thread
842 osthread->set_state(INITIALIZED);
843 sync->notify_all();
845 // wait until os::start_thread()
846 while (osthread->get_state() == INITIALIZED) {
847 sync->wait(Mutex::_no_safepoint_check_flag);
848 }
849 }
851 // call one more level start routine
852 thread->run();
854 return 0;
855 }
857 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
858 assert(thread->osthread() == NULL, "caller responsible");
860 // Allocate the OSThread object
861 OSThread* osthread = new OSThread(NULL, NULL);
862 if (osthread == NULL) {
863 return false;
864 }
866 // set the correct thread state
867 osthread->set_thread_type(thr_type);
869 // Initial state is ALLOCATED but not INITIALIZED
870 osthread->set_state(ALLOCATED);
872 thread->set_osthread(osthread);
874 // init thread attributes
875 pthread_attr_t attr;
876 pthread_attr_init(&attr);
877 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
879 // stack size
880 if (os::Linux::supports_variable_stack_size()) {
881 // calculate stack size if it's not specified by caller
882 if (stack_size == 0) {
883 stack_size = os::Linux::default_stack_size(thr_type);
885 switch (thr_type) {
886 case os::java_thread:
887 // Java threads use ThreadStackSize which default value can be
888 // changed with the flag -Xss
889 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
890 stack_size = JavaThread::stack_size_at_create();
891 break;
892 case os::compiler_thread:
893 if (CompilerThreadStackSize > 0) {
894 stack_size = (size_t)(CompilerThreadStackSize * K);
895 break;
896 } // else fall through:
897 // use VMThreadStackSize if CompilerThreadStackSize is not defined
898 case os::vm_thread:
899 case os::pgc_thread:
900 case os::cgc_thread:
901 case os::watcher_thread:
902 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
903 break;
904 }
905 }
907 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
908 pthread_attr_setstacksize(&attr, stack_size);
909 } else {
910 // let pthread_create() pick the default value.
911 }
913 // glibc guard page
914 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
916 ThreadState state;
918 {
919 // Serialize thread creation if we are running with fixed stack LinuxThreads
920 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
921 if (lock) {
922 os::Linux::createThread_lock()->lock_without_safepoint_check();
923 }
925 pthread_t tid;
926 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
928 pthread_attr_destroy(&attr);
930 if (ret != 0) {
931 if (PrintMiscellaneous && (Verbose || WizardMode)) {
932 perror("pthread_create()");
933 }
934 // Need to clean up stuff we've allocated so far
935 thread->set_osthread(NULL);
936 delete osthread;
937 if (lock) os::Linux::createThread_lock()->unlock();
938 return false;
939 }
941 // Store pthread info into the OSThread
942 osthread->set_pthread_id(tid);
944 // Wait until child thread is either initialized or aborted
945 {
946 Monitor* sync_with_child = osthread->startThread_lock();
947 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
948 while ((state = osthread->get_state()) == ALLOCATED) {
949 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
950 }
951 }
953 if (lock) {
954 os::Linux::createThread_lock()->unlock();
955 }
956 }
958 // Aborted due to thread limit being reached
959 if (state == ZOMBIE) {
960 thread->set_osthread(NULL);
961 delete osthread;
962 return false;
963 }
965 // The thread is returned suspended (in state INITIALIZED),
966 // and is started higher up in the call chain
967 assert(state == INITIALIZED, "race condition");
968 return true;
969 }
971 /////////////////////////////////////////////////////////////////////////////
972 // attach existing thread
974 // bootstrap the main thread
975 bool os::create_main_thread(JavaThread* thread) {
976 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
977 return create_attached_thread(thread);
978 }
980 bool os::create_attached_thread(JavaThread* thread) {
981 #ifdef ASSERT
982 thread->verify_not_published();
983 #endif
985 // Allocate the OSThread object
986 OSThread* osthread = new OSThread(NULL, NULL);
988 if (osthread == NULL) {
989 return false;
990 }
992 // Store pthread info into the OSThread
993 osthread->set_thread_id(os::Linux::gettid());
994 osthread->set_pthread_id(::pthread_self());
996 // initialize floating point control register
997 os::Linux::init_thread_fpu_state();
999 // Initial thread state is RUNNABLE
1000 osthread->set_state(RUNNABLE);
1002 thread->set_osthread(osthread);
1004 if (UseNUMA) {
1005 int lgrp_id = os::numa_get_group_id();
1006 if (lgrp_id != -1) {
1007 thread->set_lgrp_id(lgrp_id);
1008 }
1009 }
1011 if (os::Linux::is_initial_thread()) {
1012 // If current thread is initial thread, its stack is mapped on demand,
1013 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1014 // the entire stack region to avoid SEGV in stack banging.
1015 // It is also useful to get around the heap-stack-gap problem on SuSE
1016 // kernel (see 4821821 for details). We first expand stack to the top
1017 // of yellow zone, then enable stack yellow zone (order is significant,
1018 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1019 // is no gap between the last two virtual memory regions.
1021 JavaThread *jt = (JavaThread *)thread;
1022 address addr = jt->stack_yellow_zone_base();
1023 assert(addr != NULL, "initialization problem?");
1024 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1026 osthread->set_expanding_stack();
1027 os::Linux::manually_expand_stack(jt, addr);
1028 osthread->clear_expanding_stack();
1029 }
1031 // initialize signal mask for this thread
1032 // and save the caller's signal mask
1033 os::Linux::hotspot_sigmask(thread);
1035 return true;
1036 }
1038 void os::pd_start_thread(Thread* thread) {
1039 OSThread * osthread = thread->osthread();
1040 assert(osthread->get_state() != INITIALIZED, "just checking");
1041 Monitor* sync_with_child = osthread->startThread_lock();
1042 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1043 sync_with_child->notify();
1044 }
1046 // Free Linux resources related to the OSThread
1047 void os::free_thread(OSThread* osthread) {
1048 assert(osthread != NULL, "osthread not set");
1050 if (Thread::current()->osthread() == osthread) {
1051 // Restore caller's signal mask
1052 sigset_t sigmask = osthread->caller_sigmask();
1053 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1054 }
1056 delete osthread;
1057 }
1059 //////////////////////////////////////////////////////////////////////////////
1060 // thread local storage
1062 int os::allocate_thread_local_storage() {
1063 pthread_key_t key;
1064 int rslt = pthread_key_create(&key, NULL);
1065 assert(rslt == 0, "cannot allocate thread local storage");
1066 return (int)key;
1067 }
1069 // Note: This is currently not used by VM, as we don't destroy TLS key
1070 // on VM exit.
1071 void os::free_thread_local_storage(int index) {
1072 int rslt = pthread_key_delete((pthread_key_t)index);
1073 assert(rslt == 0, "invalid index");
1074 }
1076 void os::thread_local_storage_at_put(int index, void* value) {
1077 int rslt = pthread_setspecific((pthread_key_t)index, value);
1078 assert(rslt == 0, "pthread_setspecific failed");
1079 }
1081 extern "C" Thread* get_thread() {
1082 return ThreadLocalStorage::thread();
1083 }
1085 //////////////////////////////////////////////////////////////////////////////
1086 // initial thread
1088 // Check if current thread is the initial thread, similar to Solaris thr_main.
1089 bool os::Linux::is_initial_thread(void) {
1090 char dummy;
1091 // If called before init complete, thread stack bottom will be null.
1092 // Can be called if fatal error occurs before initialization.
1093 if (initial_thread_stack_bottom() == NULL) return false;
1094 assert(initial_thread_stack_bottom() != NULL &&
1095 initial_thread_stack_size() != 0,
1096 "os::init did not locate initial thread's stack region");
1097 if ((address)&dummy >= initial_thread_stack_bottom() &&
1098 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1099 return true;
1100 else return false;
1101 }
1103 // Find the virtual memory area that contains addr
1104 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1105 FILE *fp = fopen("/proc/self/maps", "r");
1106 if (fp) {
1107 address low, high;
1108 while (!feof(fp)) {
1109 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1110 if (low <= addr && addr < high) {
1111 if (vma_low) *vma_low = low;
1112 if (vma_high) *vma_high = high;
1113 fclose (fp);
1114 return true;
1115 }
1116 }
1117 for (;;) {
1118 int ch = fgetc(fp);
1119 if (ch == EOF || ch == (int)'\n') break;
1120 }
1121 }
1122 fclose(fp);
1123 }
1124 return false;
1125 }
1127 // Locate initial thread stack. This special handling of initial thread stack
1128 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1129 // bogus value for initial thread.
1130 void os::Linux::capture_initial_stack(size_t max_size) {
1131 // stack size is the easy part, get it from RLIMIT_STACK
1132 size_t stack_size;
1133 struct rlimit rlim;
1134 getrlimit(RLIMIT_STACK, &rlim);
1135 stack_size = rlim.rlim_cur;
1137 // 6308388: a bug in ld.so will relocate its own .data section to the
1138 // lower end of primordial stack; reduce ulimit -s value a little bit
1139 // so we won't install guard page on ld.so's data section.
1140 stack_size -= 2 * page_size();
1142 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1143 // 7.1, in both cases we will get 2G in return value.
1144 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1145 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1146 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1147 // in case other parts in glibc still assumes 2M max stack size.
1148 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1149 #ifndef IA64
1150 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1151 #else
1152 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1153 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1154 #endif
1156 // Try to figure out where the stack base (top) is. This is harder.
1157 //
1158 // When an application is started, glibc saves the initial stack pointer in
1159 // a global variable "__libc_stack_end", which is then used by system
1160 // libraries. __libc_stack_end should be pretty close to stack top. The
1161 // variable is available since the very early days. However, because it is
1162 // a private interface, it could disappear in the future.
1163 //
1164 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1165 // to __libc_stack_end, it is very close to stack top, but isn't the real
1166 // stack top. Note that /proc may not exist if VM is running as a chroot
1167 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1168 // /proc/<pid>/stat could change in the future (though unlikely).
1169 //
1170 // We try __libc_stack_end first. If that doesn't work, look for
1171 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1172 // as a hint, which should work well in most cases.
1174 uintptr_t stack_start;
1176 // try __libc_stack_end first
1177 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1178 if (p && *p) {
1179 stack_start = *p;
1180 } else {
1181 // see if we can get the start_stack field from /proc/self/stat
1182 FILE *fp;
1183 int pid;
1184 char state;
1185 int ppid;
1186 int pgrp;
1187 int session;
1188 int nr;
1189 int tpgrp;
1190 unsigned long flags;
1191 unsigned long minflt;
1192 unsigned long cminflt;
1193 unsigned long majflt;
1194 unsigned long cmajflt;
1195 unsigned long utime;
1196 unsigned long stime;
1197 long cutime;
1198 long cstime;
1199 long prio;
1200 long nice;
1201 long junk;
1202 long it_real;
1203 uintptr_t start;
1204 uintptr_t vsize;
1205 intptr_t rss;
1206 uintptr_t rsslim;
1207 uintptr_t scodes;
1208 uintptr_t ecode;
1209 int i;
1211 // Figure what the primordial thread stack base is. Code is inspired
1212 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1213 // followed by command name surrounded by parentheses, state, etc.
1214 char stat[2048];
1215 int statlen;
1217 fp = fopen("/proc/self/stat", "r");
1218 if (fp) {
1219 statlen = fread(stat, 1, 2047, fp);
1220 stat[statlen] = '\0';
1221 fclose(fp);
1223 // Skip pid and the command string. Note that we could be dealing with
1224 // weird command names, e.g. user could decide to rename java launcher
1225 // to "java 1.4.2 :)", then the stat file would look like
1226 // 1234 (java 1.4.2 :)) R ... ...
1227 // We don't really need to know the command string, just find the last
1228 // occurrence of ")" and then start parsing from there. See bug 4726580.
1229 char * s = strrchr(stat, ')');
1231 i = 0;
1232 if (s) {
1233 // Skip blank chars
1234 do s++; while (isspace(*s));
1236 #define _UFM UINTX_FORMAT
1237 #define _DFM INTX_FORMAT
1239 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1240 /* 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 */
1241 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1242 &state, /* 3 %c */
1243 &ppid, /* 4 %d */
1244 &pgrp, /* 5 %d */
1245 &session, /* 6 %d */
1246 &nr, /* 7 %d */
1247 &tpgrp, /* 8 %d */
1248 &flags, /* 9 %lu */
1249 &minflt, /* 10 %lu */
1250 &cminflt, /* 11 %lu */
1251 &majflt, /* 12 %lu */
1252 &cmajflt, /* 13 %lu */
1253 &utime, /* 14 %lu */
1254 &stime, /* 15 %lu */
1255 &cutime, /* 16 %ld */
1256 &cstime, /* 17 %ld */
1257 &prio, /* 18 %ld */
1258 &nice, /* 19 %ld */
1259 &junk, /* 20 %ld */
1260 &it_real, /* 21 %ld */
1261 &start, /* 22 UINTX_FORMAT */
1262 &vsize, /* 23 UINTX_FORMAT */
1263 &rss, /* 24 INTX_FORMAT */
1264 &rsslim, /* 25 UINTX_FORMAT */
1265 &scodes, /* 26 UINTX_FORMAT */
1266 &ecode, /* 27 UINTX_FORMAT */
1267 &stack_start); /* 28 UINTX_FORMAT */
1268 }
1270 #undef _UFM
1271 #undef _DFM
1273 if (i != 28 - 2) {
1274 assert(false, "Bad conversion from /proc/self/stat");
1275 // product mode - assume we are the initial thread, good luck in the
1276 // embedded case.
1277 warning("Can't detect initial thread stack location - bad conversion");
1278 stack_start = (uintptr_t) &rlim;
1279 }
1280 } else {
1281 // For some reason we can't open /proc/self/stat (for example, running on
1282 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1283 // most cases, so don't abort:
1284 warning("Can't detect initial thread stack location - no /proc/self/stat");
1285 stack_start = (uintptr_t) &rlim;
1286 }
1287 }
1289 // Now we have a pointer (stack_start) very close to the stack top, the
1290 // next thing to do is to figure out the exact location of stack top. We
1291 // can find out the virtual memory area that contains stack_start by
1292 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1293 // and its upper limit is the real stack top. (again, this would fail if
1294 // running inside chroot, because /proc may not exist.)
1296 uintptr_t stack_top;
1297 address low, high;
1298 if (find_vma((address)stack_start, &low, &high)) {
1299 // success, "high" is the true stack top. (ignore "low", because initial
1300 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1301 stack_top = (uintptr_t)high;
1302 } else {
1303 // failed, likely because /proc/self/maps does not exist
1304 warning("Can't detect initial thread stack location - find_vma failed");
1305 // best effort: stack_start is normally within a few pages below the real
1306 // stack top, use it as stack top, and reduce stack size so we won't put
1307 // guard page outside stack.
1308 stack_top = stack_start;
1309 stack_size -= 16 * page_size();
1310 }
1312 // stack_top could be partially down the page so align it
1313 stack_top = align_size_up(stack_top, page_size());
1315 if (max_size && stack_size > max_size) {
1316 _initial_thread_stack_size = max_size;
1317 } else {
1318 _initial_thread_stack_size = stack_size;
1319 }
1321 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1322 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1323 }
1325 ////////////////////////////////////////////////////////////////////////////////
1326 // time support
1328 // Time since start-up in seconds to a fine granularity.
1329 // Used by VMSelfDestructTimer and the MemProfiler.
1330 double os::elapsedTime() {
1332 return (double)(os::elapsed_counter()) * 0.000001;
1333 }
1335 jlong os::elapsed_counter() {
1336 timeval time;
1337 int status = gettimeofday(&time, NULL);
1338 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1339 }
1341 jlong os::elapsed_frequency() {
1342 return (1000 * 1000);
1343 }
1345 // For now, we say that linux does not support vtime. I have no idea
1346 // whether it can actually be made to (DLD, 9/13/05).
1348 bool os::supports_vtime() { return false; }
1349 bool os::enable_vtime() { return false; }
1350 bool os::vtime_enabled() { return false; }
1351 double os::elapsedVTime() {
1352 // better than nothing, but not much
1353 return elapsedTime();
1354 }
1356 jlong os::javaTimeMillis() {
1357 timeval time;
1358 int status = gettimeofday(&time, NULL);
1359 assert(status != -1, "linux error");
1360 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1361 }
1363 #ifndef CLOCK_MONOTONIC
1364 #define CLOCK_MONOTONIC (1)
1365 #endif
1367 void os::Linux::clock_init() {
1368 // we do dlopen's in this particular order due to bug in linux
1369 // dynamical loader (see 6348968) leading to crash on exit
1370 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1371 if (handle == NULL) {
1372 handle = dlopen("librt.so", RTLD_LAZY);
1373 }
1375 if (handle) {
1376 int (*clock_getres_func)(clockid_t, struct timespec*) =
1377 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1378 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1379 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1380 if (clock_getres_func && clock_gettime_func) {
1381 // See if monotonic clock is supported by the kernel. Note that some
1382 // early implementations simply return kernel jiffies (updated every
1383 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1384 // for nano time (though the monotonic property is still nice to have).
1385 // It's fixed in newer kernels, however clock_getres() still returns
1386 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1387 // resolution for now. Hopefully as people move to new kernels, this
1388 // won't be a problem.
1389 struct timespec res;
1390 struct timespec tp;
1391 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1392 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1393 // yes, monotonic clock is supported
1394 _clock_gettime = clock_gettime_func;
1395 } else {
1396 // close librt if there is no monotonic clock
1397 dlclose(handle);
1398 }
1399 }
1400 }
1401 }
1403 #ifndef SYS_clock_getres
1405 #if defined(IA32) || defined(AMD64)
1406 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1407 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1408 #else
1409 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1410 #define sys_clock_getres(x,y) -1
1411 #endif
1413 #else
1414 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1415 #endif
1417 void os::Linux::fast_thread_clock_init() {
1418 if (!UseLinuxPosixThreadCPUClocks) {
1419 return;
1420 }
1421 clockid_t clockid;
1422 struct timespec tp;
1423 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1424 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1426 // Switch to using fast clocks for thread cpu time if
1427 // the sys_clock_getres() returns 0 error code.
1428 // Note, that some kernels may support the current thread
1429 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1430 // returned by the pthread_getcpuclockid().
1431 // If the fast Posix clocks are supported then the sys_clock_getres()
1432 // must return at least tp.tv_sec == 0 which means a resolution
1433 // better than 1 sec. This is extra check for reliability.
1435 if(pthread_getcpuclockid_func &&
1436 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1437 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1439 _supports_fast_thread_cpu_time = true;
1440 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1441 }
1442 }
1444 jlong os::javaTimeNanos() {
1445 if (Linux::supports_monotonic_clock()) {
1446 struct timespec tp;
1447 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1448 assert(status == 0, "gettime error");
1449 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1450 return result;
1451 } else {
1452 timeval time;
1453 int status = gettimeofday(&time, NULL);
1454 assert(status != -1, "linux error");
1455 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1456 return 1000 * usecs;
1457 }
1458 }
1460 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1461 if (Linux::supports_monotonic_clock()) {
1462 info_ptr->max_value = ALL_64_BITS;
1464 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1465 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1466 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1467 } else {
1468 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1469 info_ptr->max_value = ALL_64_BITS;
1471 // gettimeofday is a real time clock so it skips
1472 info_ptr->may_skip_backward = true;
1473 info_ptr->may_skip_forward = true;
1474 }
1476 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1477 }
1479 // Return the real, user, and system times in seconds from an
1480 // arbitrary fixed point in the past.
1481 bool os::getTimesSecs(double* process_real_time,
1482 double* process_user_time,
1483 double* process_system_time) {
1484 struct tms ticks;
1485 clock_t real_ticks = times(&ticks);
1487 if (real_ticks == (clock_t) (-1)) {
1488 return false;
1489 } else {
1490 double ticks_per_second = (double) clock_tics_per_sec;
1491 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1492 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1493 *process_real_time = ((double) real_ticks) / ticks_per_second;
1495 return true;
1496 }
1497 }
1500 char * os::local_time_string(char *buf, size_t buflen) {
1501 struct tm t;
1502 time_t long_time;
1503 time(&long_time);
1504 localtime_r(&long_time, &t);
1505 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1506 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1507 t.tm_hour, t.tm_min, t.tm_sec);
1508 return buf;
1509 }
1511 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1512 return localtime_r(clock, res);
1513 }
1515 ////////////////////////////////////////////////////////////////////////////////
1516 // runtime exit support
1518 // Note: os::shutdown() might be called very early during initialization, or
1519 // called from signal handler. Before adding something to os::shutdown(), make
1520 // sure it is async-safe and can handle partially initialized VM.
1521 void os::shutdown() {
1523 // allow PerfMemory to attempt cleanup of any persistent resources
1524 perfMemory_exit();
1526 // needs to remove object in file system
1527 AttachListener::abort();
1529 // flush buffered output, finish log files
1530 ostream_abort();
1532 // Check for abort hook
1533 abort_hook_t abort_hook = Arguments::abort_hook();
1534 if (abort_hook != NULL) {
1535 abort_hook();
1536 }
1538 }
1540 // Note: os::abort() might be called very early during initialization, or
1541 // called from signal handler. Before adding something to os::abort(), make
1542 // sure it is async-safe and can handle partially initialized VM.
1543 void os::abort(bool dump_core) {
1544 os::shutdown();
1545 if (dump_core) {
1546 #ifndef PRODUCT
1547 fdStream out(defaultStream::output_fd());
1548 out.print_raw("Current thread is ");
1549 char buf[16];
1550 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1551 out.print_raw_cr(buf);
1552 out.print_raw_cr("Dumping core ...");
1553 #endif
1554 ::abort(); // dump core
1555 }
1557 ::exit(1);
1558 }
1560 // Die immediately, no exit hook, no abort hook, no cleanup.
1561 void os::die() {
1562 // _exit() on LinuxThreads only kills current thread
1563 ::abort();
1564 }
1566 // unused on linux for now.
1567 void os::set_error_file(const char *logfile) {}
1570 // This method is a copy of JDK's sysGetLastErrorString
1571 // from src/solaris/hpi/src/system_md.c
1573 size_t os::lasterror(char *buf, size_t len) {
1575 if (errno == 0) return 0;
1577 const char *s = ::strerror(errno);
1578 size_t n = ::strlen(s);
1579 if (n >= len) {
1580 n = len - 1;
1581 }
1582 ::strncpy(buf, s, n);
1583 buf[n] = '\0';
1584 return n;
1585 }
1587 intx os::current_thread_id() { return (intx)pthread_self(); }
1588 int os::current_process_id() {
1590 // Under the old linux thread library, linux gives each thread
1591 // its own process id. Because of this each thread will return
1592 // a different pid if this method were to return the result
1593 // of getpid(2). Linux provides no api that returns the pid
1594 // of the launcher thread for the vm. This implementation
1595 // returns a unique pid, the pid of the launcher thread
1596 // that starts the vm 'process'.
1598 // Under the NPTL, getpid() returns the same pid as the
1599 // launcher thread rather than a unique pid per thread.
1600 // Use gettid() if you want the old pre NPTL behaviour.
1602 // if you are looking for the result of a call to getpid() that
1603 // returns a unique pid for the calling thread, then look at the
1604 // OSThread::thread_id() method in osThread_linux.hpp file
1606 return (int)(_initial_pid ? _initial_pid : getpid());
1607 }
1609 // DLL functions
1611 const char* os::dll_file_extension() { return ".so"; }
1613 // This must be hard coded because it's the system's temporary
1614 // directory not the java application's temp directory, ala java.io.tmpdir.
1615 const char* os::get_temp_directory() { return "/tmp"; }
1617 static bool file_exists(const char* filename) {
1618 struct stat statbuf;
1619 if (filename == NULL || strlen(filename) == 0) {
1620 return false;
1621 }
1622 return os::stat(filename, &statbuf) == 0;
1623 }
1625 void os::dll_build_name(char* buffer, size_t buflen,
1626 const char* pname, const char* fname) {
1627 // Copied from libhpi
1628 const size_t pnamelen = pname ? strlen(pname) : 0;
1630 // Quietly truncate on buffer overflow. Should be an error.
1631 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1632 *buffer = '\0';
1633 return;
1634 }
1636 if (pnamelen == 0) {
1637 snprintf(buffer, buflen, "lib%s.so", fname);
1638 } else if (strchr(pname, *os::path_separator()) != NULL) {
1639 int n;
1640 char** pelements = split_path(pname, &n);
1641 for (int i = 0 ; i < n ; i++) {
1642 // Really shouldn't be NULL, but check can't hurt
1643 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1644 continue; // skip the empty path values
1645 }
1646 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1647 if (file_exists(buffer)) {
1648 break;
1649 }
1650 }
1651 // release the storage
1652 for (int i = 0 ; i < n ; i++) {
1653 if (pelements[i] != NULL) {
1654 FREE_C_HEAP_ARRAY(char, pelements[i]);
1655 }
1656 }
1657 if (pelements != NULL) {
1658 FREE_C_HEAP_ARRAY(char*, pelements);
1659 }
1660 } else {
1661 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1662 }
1663 }
1665 const char* os::get_current_directory(char *buf, int buflen) {
1666 return getcwd(buf, buflen);
1667 }
1669 // check if addr is inside libjvm[_g].so
1670 bool os::address_is_in_vm(address addr) {
1671 static address libjvm_base_addr;
1672 Dl_info dlinfo;
1674 if (libjvm_base_addr == NULL) {
1675 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1676 libjvm_base_addr = (address)dlinfo.dli_fbase;
1677 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1678 }
1680 if (dladdr((void *)addr, &dlinfo)) {
1681 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1682 }
1684 return false;
1685 }
1687 bool os::dll_address_to_function_name(address addr, char *buf,
1688 int buflen, int *offset) {
1689 Dl_info dlinfo;
1691 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1692 if (buf != NULL) {
1693 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1694 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1695 }
1696 }
1697 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1698 return true;
1699 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1700 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1701 dlinfo.dli_fname, buf, buflen, offset) == Decoder::no_error) {
1702 return true;
1703 }
1704 }
1706 if (buf != NULL) buf[0] = '\0';
1707 if (offset != NULL) *offset = -1;
1708 return false;
1709 }
1711 struct _address_to_library_name {
1712 address addr; // input : memory address
1713 size_t buflen; // size of fname
1714 char* fname; // output: library name
1715 address base; // library base addr
1716 };
1718 static int address_to_library_name_callback(struct dl_phdr_info *info,
1719 size_t size, void *data) {
1720 int i;
1721 bool found = false;
1722 address libbase = NULL;
1723 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1725 // iterate through all loadable segments
1726 for (i = 0; i < info->dlpi_phnum; i++) {
1727 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1728 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1729 // base address of a library is the lowest address of its loaded
1730 // segments.
1731 if (libbase == NULL || libbase > segbase) {
1732 libbase = segbase;
1733 }
1734 // see if 'addr' is within current segment
1735 if (segbase <= d->addr &&
1736 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1737 found = true;
1738 }
1739 }
1740 }
1742 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1743 // so dll_address_to_library_name() can fall through to use dladdr() which
1744 // can figure out executable name from argv[0].
1745 if (found && info->dlpi_name && info->dlpi_name[0]) {
1746 d->base = libbase;
1747 if (d->fname) {
1748 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1749 }
1750 return 1;
1751 }
1752 return 0;
1753 }
1755 bool os::dll_address_to_library_name(address addr, char* buf,
1756 int buflen, int* offset) {
1757 Dl_info dlinfo;
1758 struct _address_to_library_name data;
1760 // There is a bug in old glibc dladdr() implementation that it could resolve
1761 // to wrong library name if the .so file has a base address != NULL. Here
1762 // we iterate through the program headers of all loaded libraries to find
1763 // out which library 'addr' really belongs to. This workaround can be
1764 // removed once the minimum requirement for glibc is moved to 2.3.x.
1765 data.addr = addr;
1766 data.fname = buf;
1767 data.buflen = buflen;
1768 data.base = NULL;
1769 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1771 if (rslt) {
1772 // buf already contains library name
1773 if (offset) *offset = addr - data.base;
1774 return true;
1775 } else if (dladdr((void*)addr, &dlinfo)){
1776 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1777 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1778 return true;
1779 } else {
1780 if (buf) buf[0] = '\0';
1781 if (offset) *offset = -1;
1782 return false;
1783 }
1784 }
1786 // Loads .dll/.so and
1787 // in case of error it checks if .dll/.so was built for the
1788 // same architecture as Hotspot is running on
1790 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1791 {
1792 void * result= ::dlopen(filename, RTLD_LAZY);
1793 if (result != NULL) {
1794 // Successful loading
1795 return result;
1796 }
1798 Elf32_Ehdr elf_head;
1800 // Read system error message into ebuf
1801 // It may or may not be overwritten below
1802 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1803 ebuf[ebuflen-1]='\0';
1804 int diag_msg_max_length=ebuflen-strlen(ebuf);
1805 char* diag_msg_buf=ebuf+strlen(ebuf);
1807 if (diag_msg_max_length==0) {
1808 // No more space in ebuf for additional diagnostics message
1809 return NULL;
1810 }
1813 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1815 if (file_descriptor < 0) {
1816 // Can't open library, report dlerror() message
1817 return NULL;
1818 }
1820 bool failed_to_read_elf_head=
1821 (sizeof(elf_head)!=
1822 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1824 ::close(file_descriptor);
1825 if (failed_to_read_elf_head) {
1826 // file i/o error - report dlerror() msg
1827 return NULL;
1828 }
1830 typedef struct {
1831 Elf32_Half code; // Actual value as defined in elf.h
1832 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1833 char elf_class; // 32 or 64 bit
1834 char endianess; // MSB or LSB
1835 char* name; // String representation
1836 } arch_t;
1838 #ifndef EM_486
1839 #define EM_486 6 /* Intel 80486 */
1840 #endif
1842 static const arch_t arch_array[]={
1843 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1844 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1845 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1846 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1847 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1848 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1849 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1850 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1851 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1852 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1853 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1854 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1855 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1856 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1857 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1858 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1859 };
1861 #if (defined IA32)
1862 static Elf32_Half running_arch_code=EM_386;
1863 #elif (defined AMD64)
1864 static Elf32_Half running_arch_code=EM_X86_64;
1865 #elif (defined IA64)
1866 static Elf32_Half running_arch_code=EM_IA_64;
1867 #elif (defined __sparc) && (defined _LP64)
1868 static Elf32_Half running_arch_code=EM_SPARCV9;
1869 #elif (defined __sparc) && (!defined _LP64)
1870 static Elf32_Half running_arch_code=EM_SPARC;
1871 #elif (defined __powerpc64__)
1872 static Elf32_Half running_arch_code=EM_PPC64;
1873 #elif (defined __powerpc__)
1874 static Elf32_Half running_arch_code=EM_PPC;
1875 #elif (defined ARM)
1876 static Elf32_Half running_arch_code=EM_ARM;
1877 #elif (defined S390)
1878 static Elf32_Half running_arch_code=EM_S390;
1879 #elif (defined ALPHA)
1880 static Elf32_Half running_arch_code=EM_ALPHA;
1881 #elif (defined MIPSEL)
1882 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1883 #elif (defined PARISC)
1884 static Elf32_Half running_arch_code=EM_PARISC;
1885 #elif (defined MIPS)
1886 static Elf32_Half running_arch_code=EM_MIPS;
1887 #elif (defined M68K)
1888 static Elf32_Half running_arch_code=EM_68K;
1889 #else
1890 #error Method os::dll_load requires that one of following is defined:\
1891 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1892 #endif
1894 // Identify compatability class for VM's architecture and library's architecture
1895 // Obtain string descriptions for architectures
1897 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1898 int running_arch_index=-1;
1900 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1901 if (running_arch_code == arch_array[i].code) {
1902 running_arch_index = i;
1903 }
1904 if (lib_arch.code == arch_array[i].code) {
1905 lib_arch.compat_class = arch_array[i].compat_class;
1906 lib_arch.name = arch_array[i].name;
1907 }
1908 }
1910 assert(running_arch_index != -1,
1911 "Didn't find running architecture code (running_arch_code) in arch_array");
1912 if (running_arch_index == -1) {
1913 // Even though running architecture detection failed
1914 // we may still continue with reporting dlerror() message
1915 return NULL;
1916 }
1918 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1919 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1920 return NULL;
1921 }
1923 #ifndef S390
1924 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1925 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1926 return NULL;
1927 }
1928 #endif // !S390
1930 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1931 if ( lib_arch.name!=NULL ) {
1932 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1933 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1934 lib_arch.name, arch_array[running_arch_index].name);
1935 } else {
1936 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1937 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1938 lib_arch.code,
1939 arch_array[running_arch_index].name);
1940 }
1941 }
1943 return NULL;
1944 }
1946 /*
1947 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1948 * chances are you might want to run the generated bits against glibc-2.0
1949 * libdl.so, so always use locking for any version of glibc.
1950 */
1951 void* os::dll_lookup(void* handle, const char* name) {
1952 pthread_mutex_lock(&dl_mutex);
1953 void* res = dlsym(handle, name);
1954 pthread_mutex_unlock(&dl_mutex);
1955 return res;
1956 }
1959 static bool _print_ascii_file(const char* filename, outputStream* st) {
1960 int fd = ::open(filename, O_RDONLY);
1961 if (fd == -1) {
1962 return false;
1963 }
1965 char buf[32];
1966 int bytes;
1967 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1968 st->print_raw(buf, bytes);
1969 }
1971 ::close(fd);
1973 return true;
1974 }
1976 void os::print_dll_info(outputStream *st) {
1977 st->print_cr("Dynamic libraries:");
1979 char fname[32];
1980 pid_t pid = os::Linux::gettid();
1982 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1984 if (!_print_ascii_file(fname, st)) {
1985 st->print("Can not get library information for pid = %d\n", pid);
1986 }
1987 }
1990 void os::print_os_info(outputStream* st) {
1991 st->print("OS:");
1993 // Try to identify popular distros.
1994 // Most Linux distributions have /etc/XXX-release file, which contains
1995 // the OS version string. Some have more than one /etc/XXX-release file
1996 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
1997 // so the order is important.
1998 if (!_print_ascii_file("/etc/mandrake-release", st) &&
1999 !_print_ascii_file("/etc/sun-release", st) &&
2000 !_print_ascii_file("/etc/redhat-release", st) &&
2001 !_print_ascii_file("/etc/SuSE-release", st) &&
2002 !_print_ascii_file("/etc/turbolinux-release", st) &&
2003 !_print_ascii_file("/etc/gentoo-release", st) &&
2004 !_print_ascii_file("/etc/debian_version", st) &&
2005 !_print_ascii_file("/etc/ltib-release", st) &&
2006 !_print_ascii_file("/etc/angstrom-version", st)) {
2007 st->print("Linux");
2008 }
2009 st->cr();
2011 // kernel
2012 st->print("uname:");
2013 struct utsname name;
2014 uname(&name);
2015 st->print(name.sysname); st->print(" ");
2016 st->print(name.release); st->print(" ");
2017 st->print(name.version); st->print(" ");
2018 st->print(name.machine);
2019 st->cr();
2021 // Print warning if unsafe chroot environment detected
2022 if (unsafe_chroot_detected) {
2023 st->print("WARNING!! ");
2024 st->print_cr(unstable_chroot_error);
2025 }
2027 // libc, pthread
2028 st->print("libc:");
2029 st->print(os::Linux::glibc_version()); st->print(" ");
2030 st->print(os::Linux::libpthread_version()); st->print(" ");
2031 if (os::Linux::is_LinuxThreads()) {
2032 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2033 }
2034 st->cr();
2036 // rlimit
2037 st->print("rlimit:");
2038 struct rlimit rlim;
2040 st->print(" STACK ");
2041 getrlimit(RLIMIT_STACK, &rlim);
2042 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2043 else st->print("%uk", rlim.rlim_cur >> 10);
2045 st->print(", CORE ");
2046 getrlimit(RLIMIT_CORE, &rlim);
2047 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2048 else st->print("%uk", rlim.rlim_cur >> 10);
2050 st->print(", NPROC ");
2051 getrlimit(RLIMIT_NPROC, &rlim);
2052 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2053 else st->print("%d", rlim.rlim_cur);
2055 st->print(", NOFILE ");
2056 getrlimit(RLIMIT_NOFILE, &rlim);
2057 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2058 else st->print("%d", rlim.rlim_cur);
2060 st->print(", AS ");
2061 getrlimit(RLIMIT_AS, &rlim);
2062 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2063 else st->print("%uk", rlim.rlim_cur >> 10);
2064 st->cr();
2066 // load average
2067 st->print("load average:");
2068 double loadavg[3];
2069 os::loadavg(loadavg, 3);
2070 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2071 st->cr();
2073 // meminfo
2074 st->print("\n/proc/meminfo:\n");
2075 _print_ascii_file("/proc/meminfo", st);
2076 st->cr();
2077 }
2079 void os::print_memory_info(outputStream* st) {
2081 st->print("Memory:");
2082 st->print(" %dk page", os::vm_page_size()>>10);
2084 // values in struct sysinfo are "unsigned long"
2085 struct sysinfo si;
2086 sysinfo(&si);
2088 st->print(", physical " UINT64_FORMAT "k",
2089 os::physical_memory() >> 10);
2090 st->print("(" UINT64_FORMAT "k free)",
2091 os::available_memory() >> 10);
2092 st->print(", swap " UINT64_FORMAT "k",
2093 ((jlong)si.totalswap * si.mem_unit) >> 10);
2094 st->print("(" UINT64_FORMAT "k free)",
2095 ((jlong)si.freeswap * si.mem_unit) >> 10);
2096 st->cr();
2097 }
2099 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2100 // but they're the same for all the linux arch that we support
2101 // and they're the same for solaris but there's no common place to put this.
2102 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2103 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2104 "ILL_COPROC", "ILL_BADSTK" };
2106 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2107 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2108 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2110 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2112 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2114 void os::print_siginfo(outputStream* st, void* siginfo) {
2115 st->print("siginfo:");
2117 const int buflen = 100;
2118 char buf[buflen];
2119 siginfo_t *si = (siginfo_t*)siginfo;
2120 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2121 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2122 st->print("si_errno=%s", buf);
2123 } else {
2124 st->print("si_errno=%d", si->si_errno);
2125 }
2126 const int c = si->si_code;
2127 assert(c > 0, "unexpected si_code");
2128 switch (si->si_signo) {
2129 case SIGILL:
2130 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2131 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2132 break;
2133 case SIGFPE:
2134 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2135 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2136 break;
2137 case SIGSEGV:
2138 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2139 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2140 break;
2141 case SIGBUS:
2142 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2143 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2144 break;
2145 default:
2146 st->print(", si_code=%d", si->si_code);
2147 // no si_addr
2148 }
2150 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2151 UseSharedSpaces) {
2152 FileMapInfo* mapinfo = FileMapInfo::current_info();
2153 if (mapinfo->is_in_shared_space(si->si_addr)) {
2154 st->print("\n\nError accessing class data sharing archive." \
2155 " Mapped file inaccessible during execution, " \
2156 " possible disk/network problem.");
2157 }
2158 }
2159 st->cr();
2160 }
2163 static void print_signal_handler(outputStream* st, int sig,
2164 char* buf, size_t buflen);
2166 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2167 st->print_cr("Signal Handlers:");
2168 print_signal_handler(st, SIGSEGV, buf, buflen);
2169 print_signal_handler(st, SIGBUS , buf, buflen);
2170 print_signal_handler(st, SIGFPE , buf, buflen);
2171 print_signal_handler(st, SIGPIPE, buf, buflen);
2172 print_signal_handler(st, SIGXFSZ, buf, buflen);
2173 print_signal_handler(st, SIGILL , buf, buflen);
2174 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2175 print_signal_handler(st, SR_signum, buf, buflen);
2176 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2177 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2178 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2179 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2180 }
2182 static char saved_jvm_path[MAXPATHLEN] = {0};
2184 // Find the full path to the current module, libjvm.so or libjvm_g.so
2185 void os::jvm_path(char *buf, jint buflen) {
2186 // Error checking.
2187 if (buflen < MAXPATHLEN) {
2188 assert(false, "must use a large-enough buffer");
2189 buf[0] = '\0';
2190 return;
2191 }
2192 // Lazy resolve the path to current module.
2193 if (saved_jvm_path[0] != 0) {
2194 strcpy(buf, saved_jvm_path);
2195 return;
2196 }
2198 char dli_fname[MAXPATHLEN];
2199 bool ret = dll_address_to_library_name(
2200 CAST_FROM_FN_PTR(address, os::jvm_path),
2201 dli_fname, sizeof(dli_fname), NULL);
2202 assert(ret != 0, "cannot locate libjvm");
2203 char *rp = realpath(dli_fname, buf);
2204 if (rp == NULL)
2205 return;
2207 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2208 // Support for the gamma launcher. Typical value for buf is
2209 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2210 // the right place in the string, then assume we are installed in a JDK and
2211 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2212 // up the path so it looks like libjvm.so is installed there (append a
2213 // fake suffix hotspot/libjvm.so).
2214 const char *p = buf + strlen(buf) - 1;
2215 for (int count = 0; p > buf && count < 5; ++count) {
2216 for (--p; p > buf && *p != '/'; --p)
2217 /* empty */ ;
2218 }
2220 if (strncmp(p, "/jre/lib/", 9) != 0) {
2221 // Look for JAVA_HOME in the environment.
2222 char* java_home_var = ::getenv("JAVA_HOME");
2223 if (java_home_var != NULL && java_home_var[0] != 0) {
2224 char* jrelib_p;
2225 int len;
2227 // Check the current module name "libjvm.so" or "libjvm_g.so".
2228 p = strrchr(buf, '/');
2229 assert(strstr(p, "/libjvm") == p, "invalid library name");
2230 p = strstr(p, "_g") ? "_g" : "";
2232 rp = realpath(java_home_var, buf);
2233 if (rp == NULL)
2234 return;
2236 // determine if this is a legacy image or modules image
2237 // modules image doesn't have "jre" subdirectory
2238 len = strlen(buf);
2239 jrelib_p = buf + len;
2240 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2241 if (0 != access(buf, F_OK)) {
2242 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2243 }
2245 if (0 == access(buf, F_OK)) {
2246 // Use current module name "libjvm[_g].so" instead of
2247 // "libjvm"debug_only("_g")".so" since for fastdebug version
2248 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2249 len = strlen(buf);
2250 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2251 } else {
2252 // Go back to path of .so
2253 rp = realpath(dli_fname, buf);
2254 if (rp == NULL)
2255 return;
2256 }
2257 }
2258 }
2259 }
2261 strcpy(saved_jvm_path, buf);
2262 }
2264 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2265 // no prefix required, not even "_"
2266 }
2268 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2269 // no suffix required
2270 }
2272 ////////////////////////////////////////////////////////////////////////////////
2273 // sun.misc.Signal support
2275 static volatile jint sigint_count = 0;
2277 static void
2278 UserHandler(int sig, void *siginfo, void *context) {
2279 // 4511530 - sem_post is serialized and handled by the manager thread. When
2280 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2281 // don't want to flood the manager thread with sem_post requests.
2282 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2283 return;
2285 // Ctrl-C is pressed during error reporting, likely because the error
2286 // handler fails to abort. Let VM die immediately.
2287 if (sig == SIGINT && is_error_reported()) {
2288 os::die();
2289 }
2291 os::signal_notify(sig);
2292 }
2294 void* os::user_handler() {
2295 return CAST_FROM_FN_PTR(void*, UserHandler);
2296 }
2298 extern "C" {
2299 typedef void (*sa_handler_t)(int);
2300 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2301 }
2303 void* os::signal(int signal_number, void* handler) {
2304 struct sigaction sigAct, oldSigAct;
2306 sigfillset(&(sigAct.sa_mask));
2307 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2308 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2310 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2311 // -1 means registration failed
2312 return (void *)-1;
2313 }
2315 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2316 }
2318 void os::signal_raise(int signal_number) {
2319 ::raise(signal_number);
2320 }
2322 /*
2323 * The following code is moved from os.cpp for making this
2324 * code platform specific, which it is by its very nature.
2325 */
2327 // Will be modified when max signal is changed to be dynamic
2328 int os::sigexitnum_pd() {
2329 return NSIG;
2330 }
2332 // a counter for each possible signal value
2333 static volatile jint pending_signals[NSIG+1] = { 0 };
2335 // Linux(POSIX) specific hand shaking semaphore.
2336 static sem_t sig_sem;
2338 void os::signal_init_pd() {
2339 // Initialize signal structures
2340 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2342 // Initialize signal semaphore
2343 ::sem_init(&sig_sem, 0, 0);
2344 }
2346 void os::signal_notify(int sig) {
2347 Atomic::inc(&pending_signals[sig]);
2348 ::sem_post(&sig_sem);
2349 }
2351 static int check_pending_signals(bool wait) {
2352 Atomic::store(0, &sigint_count);
2353 for (;;) {
2354 for (int i = 0; i < NSIG + 1; i++) {
2355 jint n = pending_signals[i];
2356 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2357 return i;
2358 }
2359 }
2360 if (!wait) {
2361 return -1;
2362 }
2363 JavaThread *thread = JavaThread::current();
2364 ThreadBlockInVM tbivm(thread);
2366 bool threadIsSuspended;
2367 do {
2368 thread->set_suspend_equivalent();
2369 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2370 ::sem_wait(&sig_sem);
2372 // were we externally suspended while we were waiting?
2373 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2374 if (threadIsSuspended) {
2375 //
2376 // The semaphore has been incremented, but while we were waiting
2377 // another thread suspended us. We don't want to continue running
2378 // while suspended because that would surprise the thread that
2379 // suspended us.
2380 //
2381 ::sem_post(&sig_sem);
2383 thread->java_suspend_self();
2384 }
2385 } while (threadIsSuspended);
2386 }
2387 }
2389 int os::signal_lookup() {
2390 return check_pending_signals(false);
2391 }
2393 int os::signal_wait() {
2394 return check_pending_signals(true);
2395 }
2397 ////////////////////////////////////////////////////////////////////////////////
2398 // Virtual Memory
2400 int os::vm_page_size() {
2401 // Seems redundant as all get out
2402 assert(os::Linux::page_size() != -1, "must call os::init");
2403 return os::Linux::page_size();
2404 }
2406 // Solaris allocates memory by pages.
2407 int os::vm_allocation_granularity() {
2408 assert(os::Linux::page_size() != -1, "must call os::init");
2409 return os::Linux::page_size();
2410 }
2412 // Rationale behind this function:
2413 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2414 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2415 // samples for JITted code. Here we create private executable mapping over the code cache
2416 // and then we can use standard (well, almost, as mapping can change) way to provide
2417 // info for the reporting script by storing timestamp and location of symbol
2418 void linux_wrap_code(char* base, size_t size) {
2419 static volatile jint cnt = 0;
2421 if (!UseOprofile) {
2422 return;
2423 }
2425 char buf[PATH_MAX+1];
2426 int num = Atomic::add(1, &cnt);
2428 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2429 os::get_temp_directory(), os::current_process_id(), num);
2430 unlink(buf);
2432 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2434 if (fd != -1) {
2435 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2436 if (rv != (off_t)-1) {
2437 if (::write(fd, "", 1) == 1) {
2438 mmap(base, size,
2439 PROT_READ|PROT_WRITE|PROT_EXEC,
2440 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2441 }
2442 }
2443 ::close(fd);
2444 unlink(buf);
2445 }
2446 }
2448 // NOTE: Linux kernel does not really reserve the pages for us.
2449 // All it does is to check if there are enough free pages
2450 // left at the time of mmap(). This could be a potential
2451 // problem.
2452 bool os::commit_memory(char* addr, size_t size, bool exec) {
2453 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2454 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2455 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2456 return res != (uintptr_t) MAP_FAILED;
2457 }
2459 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2460 bool exec) {
2461 return commit_memory(addr, size, exec);
2462 }
2464 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
2466 void os::free_memory(char *addr, size_t bytes) {
2467 ::mmap(addr, bytes, PROT_READ | PROT_WRITE,
2468 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2469 }
2471 void os::numa_make_global(char *addr, size_t bytes) {
2472 Linux::numa_interleave_memory(addr, bytes);
2473 }
2475 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2476 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2477 }
2479 bool os::numa_topology_changed() { return false; }
2481 size_t os::numa_get_groups_num() {
2482 int max_node = Linux::numa_max_node();
2483 return max_node > 0 ? max_node + 1 : 1;
2484 }
2486 int os::numa_get_group_id() {
2487 int cpu_id = Linux::sched_getcpu();
2488 if (cpu_id != -1) {
2489 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2490 if (lgrp_id != -1) {
2491 return lgrp_id;
2492 }
2493 }
2494 return 0;
2495 }
2497 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2498 for (size_t i = 0; i < size; i++) {
2499 ids[i] = i;
2500 }
2501 return size;
2502 }
2504 bool os::get_page_info(char *start, page_info* info) {
2505 return false;
2506 }
2508 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2509 return end;
2510 }
2512 extern "C" void numa_warn(int number, char *where, ...) { }
2513 extern "C" void numa_error(char *where) { }
2516 // If we are running with libnuma version > 2, then we should
2517 // be trying to use symbols with versions 1.1
2518 // If we are running with earlier version, which did not have symbol versions,
2519 // we should use the base version.
2520 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2521 void *f = dlvsym(handle, name, "libnuma_1.1");
2522 if (f == NULL) {
2523 f = dlsym(handle, name);
2524 }
2525 return f;
2526 }
2528 bool os::Linux::libnuma_init() {
2529 // sched_getcpu() should be in libc.
2530 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2531 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2533 if (sched_getcpu() != -1) { // Does it work?
2534 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2535 if (handle != NULL) {
2536 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2537 libnuma_dlsym(handle, "numa_node_to_cpus")));
2538 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2539 libnuma_dlsym(handle, "numa_max_node")));
2540 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2541 libnuma_dlsym(handle, "numa_available")));
2542 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2543 libnuma_dlsym(handle, "numa_tonode_memory")));
2544 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2545 libnuma_dlsym(handle, "numa_interleave_memory")));
2548 if (numa_available() != -1) {
2549 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2550 // Create a cpu -> node mapping
2551 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2552 rebuild_cpu_to_node_map();
2553 return true;
2554 }
2555 }
2556 }
2557 return false;
2558 }
2560 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2561 // The table is later used in get_node_by_cpu().
2562 void os::Linux::rebuild_cpu_to_node_map() {
2563 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2564 // in libnuma (possible values are starting from 16,
2565 // and continuing up with every other power of 2, but less
2566 // than the maximum number of CPUs supported by kernel), and
2567 // is a subject to change (in libnuma version 2 the requirements
2568 // are more reasonable) we'll just hardcode the number they use
2569 // in the library.
2570 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2572 size_t cpu_num = os::active_processor_count();
2573 size_t cpu_map_size = NCPUS / BitsPerCLong;
2574 size_t cpu_map_valid_size =
2575 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2577 cpu_to_node()->clear();
2578 cpu_to_node()->at_grow(cpu_num - 1);
2579 size_t node_num = numa_get_groups_num();
2581 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2582 for (size_t i = 0; i < node_num; i++) {
2583 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2584 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2585 if (cpu_map[j] != 0) {
2586 for (size_t k = 0; k < BitsPerCLong; k++) {
2587 if (cpu_map[j] & (1UL << k)) {
2588 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2589 }
2590 }
2591 }
2592 }
2593 }
2594 }
2595 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2596 }
2598 int os::Linux::get_node_by_cpu(int cpu_id) {
2599 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2600 return cpu_to_node()->at(cpu_id);
2601 }
2602 return -1;
2603 }
2605 GrowableArray<int>* os::Linux::_cpu_to_node;
2606 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2607 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2608 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2609 os::Linux::numa_available_func_t os::Linux::_numa_available;
2610 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2611 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2612 unsigned long* os::Linux::_numa_all_nodes;
2614 bool os::uncommit_memory(char* addr, size_t size) {
2615 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2616 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2617 return res != (uintptr_t) MAP_FAILED;
2618 }
2620 // Linux uses a growable mapping for the stack, and if the mapping for
2621 // the stack guard pages is not removed when we detach a thread the
2622 // stack cannot grow beyond the pages where the stack guard was
2623 // mapped. If at some point later in the process the stack expands to
2624 // that point, the Linux kernel cannot expand the stack any further
2625 // because the guard pages are in the way, and a segfault occurs.
2626 //
2627 // However, it's essential not to split the stack region by unmapping
2628 // a region (leaving a hole) that's already part of the stack mapping,
2629 // so if the stack mapping has already grown beyond the guard pages at
2630 // the time we create them, we have to truncate the stack mapping.
2631 // So, we need to know the extent of the stack mapping when
2632 // create_stack_guard_pages() is called.
2634 // Find the bounds of the stack mapping. Return true for success.
2635 //
2636 // We only need this for stacks that are growable: at the time of
2637 // writing thread stacks don't use growable mappings (i.e. those
2638 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2639 // only applies to the main thread.
2640 static bool
2641 get_stack_bounds(uintptr_t *bottom, uintptr_t *top)
2642 {
2643 FILE *f = fopen("/proc/self/maps", "r");
2644 if (f == NULL)
2645 return false;
2647 while (!feof(f)) {
2648 size_t dummy;
2649 char *str = NULL;
2650 ssize_t len = getline(&str, &dummy, f);
2651 if (len == -1) {
2652 fclose(f);
2653 return false;
2654 }
2656 if (len > 0 && str[len-1] == '\n') {
2657 str[len-1] = 0;
2658 len--;
2659 }
2661 static const char *stack_str = "[stack]";
2662 if (len > (ssize_t)strlen(stack_str)
2663 && (strcmp(str + len - strlen(stack_str), stack_str) == 0)) {
2664 if (sscanf(str, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2665 uintptr_t sp = (uintptr_t)__builtin_frame_address(0);
2666 if (sp >= *bottom && sp <= *top) {
2667 free(str);
2668 fclose(f);
2669 return true;
2670 }
2671 }
2672 }
2673 free(str);
2674 }
2675 fclose(f);
2676 return false;
2677 }
2679 // If the (growable) stack mapping already extends beyond the point
2680 // where we're going to put our guard pages, truncate the mapping at
2681 // that point by munmap()ping it. This ensures that when we later
2682 // munmap() the guard pages we don't leave a hole in the stack
2683 // mapping. This only affects the main/initial thread, but guard
2684 // against future OS changes
2685 bool os::create_stack_guard_pages(char* addr, size_t size) {
2686 uintptr_t stack_extent, stack_base;
2687 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2688 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2689 assert(os::Linux::is_initial_thread(),
2690 "growable stack in non-initial thread");
2691 if (stack_extent < (uintptr_t)addr)
2692 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2693 }
2695 return os::commit_memory(addr, size);
2696 }
2698 // If this is a growable mapping, remove the guard pages entirely by
2699 // munmap()ping them. If not, just call uncommit_memory(). This only
2700 // affects the main/initial thread, but guard against future OS changes
2701 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2702 uintptr_t stack_extent, stack_base;
2703 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2704 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2705 assert(os::Linux::is_initial_thread(),
2706 "growable stack in non-initial thread");
2708 return ::munmap(addr, size) == 0;
2709 }
2711 return os::uncommit_memory(addr, size);
2712 }
2714 static address _highest_vm_reserved_address = NULL;
2716 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2717 // at 'requested_addr'. If there are existing memory mappings at the same
2718 // location, however, they will be overwritten. If 'fixed' is false,
2719 // 'requested_addr' is only treated as a hint, the return value may or
2720 // may not start from the requested address. Unlike Linux mmap(), this
2721 // function returns NULL to indicate failure.
2722 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2723 char * addr;
2724 int flags;
2726 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2727 if (fixed) {
2728 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2729 flags |= MAP_FIXED;
2730 }
2732 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2733 // to PROT_EXEC if executable when we commit the page.
2734 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2735 flags, -1, 0);
2737 if (addr != MAP_FAILED) {
2738 // anon_mmap() should only get called during VM initialization,
2739 // don't need lock (actually we can skip locking even it can be called
2740 // from multiple threads, because _highest_vm_reserved_address is just a
2741 // hint about the upper limit of non-stack memory regions.)
2742 if ((address)addr + bytes > _highest_vm_reserved_address) {
2743 _highest_vm_reserved_address = (address)addr + bytes;
2744 }
2745 }
2747 return addr == MAP_FAILED ? NULL : addr;
2748 }
2750 // Don't update _highest_vm_reserved_address, because there might be memory
2751 // regions above addr + size. If so, releasing a memory region only creates
2752 // a hole in the address space, it doesn't help prevent heap-stack collision.
2753 //
2754 static int anon_munmap(char * addr, size_t size) {
2755 return ::munmap(addr, size) == 0;
2756 }
2758 char* os::reserve_memory(size_t bytes, char* requested_addr,
2759 size_t alignment_hint) {
2760 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2761 }
2763 bool os::release_memory(char* addr, size_t size) {
2764 return anon_munmap(addr, size);
2765 }
2767 static address highest_vm_reserved_address() {
2768 return _highest_vm_reserved_address;
2769 }
2771 static bool linux_mprotect(char* addr, size_t size, int prot) {
2772 // Linux wants the mprotect address argument to be page aligned.
2773 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2775 // According to SUSv3, mprotect() should only be used with mappings
2776 // established by mmap(), and mmap() always maps whole pages. Unaligned
2777 // 'addr' likely indicates problem in the VM (e.g. trying to change
2778 // protection of malloc'ed or statically allocated memory). Check the
2779 // caller if you hit this assert.
2780 assert(addr == bottom, "sanity check");
2782 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2783 return ::mprotect(bottom, size, prot) == 0;
2784 }
2786 // Set protections specified
2787 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2788 bool is_committed) {
2789 unsigned int p = 0;
2790 switch (prot) {
2791 case MEM_PROT_NONE: p = PROT_NONE; break;
2792 case MEM_PROT_READ: p = PROT_READ; break;
2793 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2794 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2795 default:
2796 ShouldNotReachHere();
2797 }
2798 // is_committed is unused.
2799 return linux_mprotect(addr, bytes, p);
2800 }
2802 bool os::guard_memory(char* addr, size_t size) {
2803 return linux_mprotect(addr, size, PROT_NONE);
2804 }
2806 bool os::unguard_memory(char* addr, size_t size) {
2807 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2808 }
2810 // Large page support
2812 static size_t _large_page_size = 0;
2814 bool os::large_page_init() {
2815 if (!UseLargePages) return false;
2817 if (LargePageSizeInBytes) {
2818 _large_page_size = LargePageSizeInBytes;
2819 } else {
2820 // large_page_size on Linux is used to round up heap size. x86 uses either
2821 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2822 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2823 // page as large as 256M.
2824 //
2825 // Here we try to figure out page size by parsing /proc/meminfo and looking
2826 // for a line with the following format:
2827 // Hugepagesize: 2048 kB
2828 //
2829 // If we can't determine the value (e.g. /proc is not mounted, or the text
2830 // format has been changed), we'll use the largest page size supported by
2831 // the processor.
2833 #ifndef ZERO
2834 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
2835 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
2836 #endif // ZERO
2838 FILE *fp = fopen("/proc/meminfo", "r");
2839 if (fp) {
2840 while (!feof(fp)) {
2841 int x = 0;
2842 char buf[16];
2843 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2844 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2845 _large_page_size = x * K;
2846 break;
2847 }
2848 } else {
2849 // skip to next line
2850 for (;;) {
2851 int ch = fgetc(fp);
2852 if (ch == EOF || ch == (int)'\n') break;
2853 }
2854 }
2855 }
2856 fclose(fp);
2857 }
2858 }
2860 const size_t default_page_size = (size_t)Linux::page_size();
2861 if (_large_page_size > default_page_size) {
2862 _page_sizes[0] = _large_page_size;
2863 _page_sizes[1] = default_page_size;
2864 _page_sizes[2] = 0;
2865 }
2867 // Large page support is available on 2.6 or newer kernel, some vendors
2868 // (e.g. Redhat) have backported it to their 2.4 based distributions.
2869 // We optimistically assume the support is available. If later it turns out
2870 // not true, VM will automatically switch to use regular page size.
2871 return true;
2872 }
2874 #ifndef SHM_HUGETLB
2875 #define SHM_HUGETLB 04000
2876 #endif
2878 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
2879 // "exec" is passed in but not used. Creating the shared image for
2880 // the code cache doesn't have an SHM_X executable permission to check.
2881 assert(UseLargePages, "only for large pages");
2883 key_t key = IPC_PRIVATE;
2884 char *addr;
2886 bool warn_on_failure = UseLargePages &&
2887 (!FLAG_IS_DEFAULT(UseLargePages) ||
2888 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
2889 );
2890 char msg[128];
2892 // Create a large shared memory region to attach to based on size.
2893 // Currently, size is the total size of the heap
2894 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
2895 if (shmid == -1) {
2896 // Possible reasons for shmget failure:
2897 // 1. shmmax is too small for Java heap.
2898 // > check shmmax value: cat /proc/sys/kernel/shmmax
2899 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
2900 // 2. not enough large page memory.
2901 // > check available large pages: cat /proc/meminfo
2902 // > increase amount of large pages:
2903 // echo new_value > /proc/sys/vm/nr_hugepages
2904 // Note 1: different Linux may use different name for this property,
2905 // e.g. on Redhat AS-3 it is "hugetlb_pool".
2906 // Note 2: it's possible there's enough physical memory available but
2907 // they are so fragmented after a long run that they can't
2908 // coalesce into large pages. Try to reserve large pages when
2909 // the system is still "fresh".
2910 if (warn_on_failure) {
2911 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
2912 warning(msg);
2913 }
2914 return NULL;
2915 }
2917 // attach to the region
2918 addr = (char*)shmat(shmid, req_addr, 0);
2919 int err = errno;
2921 // Remove shmid. If shmat() is successful, the actual shared memory segment
2922 // will be deleted when it's detached by shmdt() or when the process
2923 // terminates. If shmat() is not successful this will remove the shared
2924 // segment immediately.
2925 shmctl(shmid, IPC_RMID, NULL);
2927 if ((intptr_t)addr == -1) {
2928 if (warn_on_failure) {
2929 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
2930 warning(msg);
2931 }
2932 return NULL;
2933 }
2935 return addr;
2936 }
2938 bool os::release_memory_special(char* base, size_t bytes) {
2939 // detaching the SHM segment will also delete it, see reserve_memory_special()
2940 int rslt = shmdt(base);
2941 return rslt == 0;
2942 }
2944 size_t os::large_page_size() {
2945 return _large_page_size;
2946 }
2948 // Linux does not support anonymous mmap with large page memory. The only way
2949 // to reserve large page memory without file backing is through SysV shared
2950 // memory API. The entire memory region is committed and pinned upfront.
2951 // Hopefully this will change in the future...
2952 bool os::can_commit_large_page_memory() {
2953 return false;
2954 }
2956 bool os::can_execute_large_page_memory() {
2957 return false;
2958 }
2960 // Reserve memory at an arbitrary address, only if that area is
2961 // available (and not reserved for something else).
2963 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2964 const int max_tries = 10;
2965 char* base[max_tries];
2966 size_t size[max_tries];
2967 const size_t gap = 0x000000;
2969 // Assert only that the size is a multiple of the page size, since
2970 // that's all that mmap requires, and since that's all we really know
2971 // about at this low abstraction level. If we need higher alignment,
2972 // we can either pass an alignment to this method or verify alignment
2973 // in one of the methods further up the call chain. See bug 5044738.
2974 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2976 // Repeatedly allocate blocks until the block is allocated at the
2977 // right spot. Give up after max_tries. Note that reserve_memory() will
2978 // automatically update _highest_vm_reserved_address if the call is
2979 // successful. The variable tracks the highest memory address every reserved
2980 // by JVM. It is used to detect heap-stack collision if running with
2981 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
2982 // space than needed, it could confuse the collision detecting code. To
2983 // solve the problem, save current _highest_vm_reserved_address and
2984 // calculate the correct value before return.
2985 address old_highest = _highest_vm_reserved_address;
2987 // Linux mmap allows caller to pass an address as hint; give it a try first,
2988 // if kernel honors the hint then we can return immediately.
2989 char * addr = anon_mmap(requested_addr, bytes, false);
2990 if (addr == requested_addr) {
2991 return requested_addr;
2992 }
2994 if (addr != NULL) {
2995 // mmap() is successful but it fails to reserve at the requested address
2996 anon_munmap(addr, bytes);
2997 }
2999 int i;
3000 for (i = 0; i < max_tries; ++i) {
3001 base[i] = reserve_memory(bytes);
3003 if (base[i] != NULL) {
3004 // Is this the block we wanted?
3005 if (base[i] == requested_addr) {
3006 size[i] = bytes;
3007 break;
3008 }
3010 // Does this overlap the block we wanted? Give back the overlapped
3011 // parts and try again.
3013 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3014 if (top_overlap >= 0 && top_overlap < bytes) {
3015 unmap_memory(base[i], top_overlap);
3016 base[i] += top_overlap;
3017 size[i] = bytes - top_overlap;
3018 } else {
3019 size_t bottom_overlap = base[i] + bytes - requested_addr;
3020 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3021 unmap_memory(requested_addr, bottom_overlap);
3022 size[i] = bytes - bottom_overlap;
3023 } else {
3024 size[i] = bytes;
3025 }
3026 }
3027 }
3028 }
3030 // Give back the unused reserved pieces.
3032 for (int j = 0; j < i; ++j) {
3033 if (base[j] != NULL) {
3034 unmap_memory(base[j], size[j]);
3035 }
3036 }
3038 if (i < max_tries) {
3039 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3040 return requested_addr;
3041 } else {
3042 _highest_vm_reserved_address = old_highest;
3043 return NULL;
3044 }
3045 }
3047 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3048 return ::read(fd, buf, nBytes);
3049 }
3051 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3052 // Solaris uses poll(), linux uses park().
3053 // Poll() is likely a better choice, assuming that Thread.interrupt()
3054 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3055 // SIGSEGV, see 4355769.
3057 const int NANOSECS_PER_MILLISECS = 1000000;
3059 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3060 assert(thread == Thread::current(), "thread consistency check");
3062 ParkEvent * const slp = thread->_SleepEvent ;
3063 slp->reset() ;
3064 OrderAccess::fence() ;
3066 if (interruptible) {
3067 jlong prevtime = javaTimeNanos();
3069 for (;;) {
3070 if (os::is_interrupted(thread, true)) {
3071 return OS_INTRPT;
3072 }
3074 jlong newtime = javaTimeNanos();
3076 if (newtime - prevtime < 0) {
3077 // time moving backwards, should only happen if no monotonic clock
3078 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3079 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3080 } else {
3081 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3082 }
3084 if(millis <= 0) {
3085 return OS_OK;
3086 }
3088 prevtime = newtime;
3090 {
3091 assert(thread->is_Java_thread(), "sanity check");
3092 JavaThread *jt = (JavaThread *) thread;
3093 ThreadBlockInVM tbivm(jt);
3094 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3096 jt->set_suspend_equivalent();
3097 // cleared by handle_special_suspend_equivalent_condition() or
3098 // java_suspend_self() via check_and_wait_while_suspended()
3100 slp->park(millis);
3102 // were we externally suspended while we were waiting?
3103 jt->check_and_wait_while_suspended();
3104 }
3105 }
3106 } else {
3107 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3108 jlong prevtime = javaTimeNanos();
3110 for (;;) {
3111 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3112 // the 1st iteration ...
3113 jlong newtime = javaTimeNanos();
3115 if (newtime - prevtime < 0) {
3116 // time moving backwards, should only happen if no monotonic clock
3117 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3118 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3119 } else {
3120 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3121 }
3123 if(millis <= 0) break ;
3125 prevtime = newtime;
3126 slp->park(millis);
3127 }
3128 return OS_OK ;
3129 }
3130 }
3132 int os::naked_sleep() {
3133 // %% make the sleep time an integer flag. for now use 1 millisec.
3134 return os::sleep(Thread::current(), 1, false);
3135 }
3137 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3138 void os::infinite_sleep() {
3139 while (true) { // sleep forever ...
3140 ::sleep(100); // ... 100 seconds at a time
3141 }
3142 }
3144 // Used to convert frequent JVM_Yield() to nops
3145 bool os::dont_yield() {
3146 return DontYieldALot;
3147 }
3149 void os::yield() {
3150 sched_yield();
3151 }
3153 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3155 void os::yield_all(int attempts) {
3156 // Yields to all threads, including threads with lower priorities
3157 // Threads on Linux are all with same priority. The Solaris style
3158 // os::yield_all() with nanosleep(1ms) is not necessary.
3159 sched_yield();
3160 }
3162 // Called from the tight loops to possibly influence time-sharing heuristics
3163 void os::loop_breaker(int attempts) {
3164 os::yield_all(attempts);
3165 }
3167 ////////////////////////////////////////////////////////////////////////////////
3168 // thread priority support
3170 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3171 // only supports dynamic priority, static priority must be zero. For real-time
3172 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3173 // However, for large multi-threaded applications, SCHED_RR is not only slower
3174 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3175 // of 5 runs - Sep 2005).
3176 //
3177 // The following code actually changes the niceness of kernel-thread/LWP. It
3178 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3179 // not the entire user process, and user level threads are 1:1 mapped to kernel
3180 // threads. It has always been the case, but could change in the future. For
3181 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3182 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3184 int os::java_to_os_priority[MaxPriority + 1] = {
3185 19, // 0 Entry should never be used
3187 4, // 1 MinPriority
3188 3, // 2
3189 2, // 3
3191 1, // 4
3192 0, // 5 NormPriority
3193 -1, // 6
3195 -2, // 7
3196 -3, // 8
3197 -4, // 9 NearMaxPriority
3199 -5 // 10 MaxPriority
3200 };
3202 static int prio_init() {
3203 if (ThreadPriorityPolicy == 1) {
3204 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3205 // if effective uid is not root. Perhaps, a more elegant way of doing
3206 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3207 if (geteuid() != 0) {
3208 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3209 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3210 }
3211 ThreadPriorityPolicy = 0;
3212 }
3213 }
3214 return 0;
3215 }
3217 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3218 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3220 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3221 return (ret == 0) ? OS_OK : OS_ERR;
3222 }
3224 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3225 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3226 *priority_ptr = java_to_os_priority[NormPriority];
3227 return OS_OK;
3228 }
3230 errno = 0;
3231 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3232 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3233 }
3235 // Hint to the underlying OS that a task switch would not be good.
3236 // Void return because it's a hint and can fail.
3237 void os::hint_no_preempt() {}
3239 ////////////////////////////////////////////////////////////////////////////////
3240 // suspend/resume support
3242 // the low-level signal-based suspend/resume support is a remnant from the
3243 // old VM-suspension that used to be for java-suspension, safepoints etc,
3244 // within hotspot. Now there is a single use-case for this:
3245 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3246 // that runs in the watcher thread.
3247 // The remaining code is greatly simplified from the more general suspension
3248 // code that used to be used.
3249 //
3250 // The protocol is quite simple:
3251 // - suspend:
3252 // - sends a signal to the target thread
3253 // - polls the suspend state of the osthread using a yield loop
3254 // - target thread signal handler (SR_handler) sets suspend state
3255 // and blocks in sigsuspend until continued
3256 // - resume:
3257 // - sets target osthread state to continue
3258 // - sends signal to end the sigsuspend loop in the SR_handler
3259 //
3260 // Note that the SR_lock plays no role in this suspend/resume protocol.
3261 //
3263 static void resume_clear_context(OSThread *osthread) {
3264 osthread->set_ucontext(NULL);
3265 osthread->set_siginfo(NULL);
3267 // notify the suspend action is completed, we have now resumed
3268 osthread->sr.clear_suspended();
3269 }
3271 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3272 osthread->set_ucontext(context);
3273 osthread->set_siginfo(siginfo);
3274 }
3276 //
3277 // Handler function invoked when a thread's execution is suspended or
3278 // resumed. We have to be careful that only async-safe functions are
3279 // called here (Note: most pthread functions are not async safe and
3280 // should be avoided.)
3281 //
3282 // Note: sigwait() is a more natural fit than sigsuspend() from an
3283 // interface point of view, but sigwait() prevents the signal hander
3284 // from being run. libpthread would get very confused by not having
3285 // its signal handlers run and prevents sigwait()'s use with the
3286 // mutex granting granting signal.
3287 //
3288 // Currently only ever called on the VMThread
3289 //
3290 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3291 // Save and restore errno to avoid confusing native code with EINTR
3292 // after sigsuspend.
3293 int old_errno = errno;
3295 Thread* thread = Thread::current();
3296 OSThread* osthread = thread->osthread();
3297 assert(thread->is_VM_thread(), "Must be VMThread");
3298 // read current suspend action
3299 int action = osthread->sr.suspend_action();
3300 if (action == SR_SUSPEND) {
3301 suspend_save_context(osthread, siginfo, context);
3303 // Notify the suspend action is about to be completed. do_suspend()
3304 // waits until SR_SUSPENDED is set and then returns. We will wait
3305 // here for a resume signal and that completes the suspend-other
3306 // action. do_suspend/do_resume is always called as a pair from
3307 // the same thread - so there are no races
3309 // notify the caller
3310 osthread->sr.set_suspended();
3312 sigset_t suspend_set; // signals for sigsuspend()
3314 // get current set of blocked signals and unblock resume signal
3315 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3316 sigdelset(&suspend_set, SR_signum);
3318 // wait here until we are resumed
3319 do {
3320 sigsuspend(&suspend_set);
3321 // ignore all returns until we get a resume signal
3322 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3324 resume_clear_context(osthread);
3326 } else {
3327 assert(action == SR_CONTINUE, "unexpected sr action");
3328 // nothing special to do - just leave the handler
3329 }
3331 errno = old_errno;
3332 }
3335 static int SR_initialize() {
3336 struct sigaction act;
3337 char *s;
3338 /* Get signal number to use for suspend/resume */
3339 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3340 int sig = ::strtol(s, 0, 10);
3341 if (sig > 0 || sig < _NSIG) {
3342 SR_signum = sig;
3343 }
3344 }
3346 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3347 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3349 sigemptyset(&SR_sigset);
3350 sigaddset(&SR_sigset, SR_signum);
3352 /* Set up signal handler for suspend/resume */
3353 act.sa_flags = SA_RESTART|SA_SIGINFO;
3354 act.sa_handler = (void (*)(int)) SR_handler;
3356 // SR_signum is blocked by default.
3357 // 4528190 - We also need to block pthread restart signal (32 on all
3358 // supported Linux platforms). Note that LinuxThreads need to block
3359 // this signal for all threads to work properly. So we don't have
3360 // to use hard-coded signal number when setting up the mask.
3361 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3363 if (sigaction(SR_signum, &act, 0) == -1) {
3364 return -1;
3365 }
3367 // Save signal flag
3368 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3369 return 0;
3370 }
3372 static int SR_finalize() {
3373 return 0;
3374 }
3377 // returns true on success and false on error - really an error is fatal
3378 // but this seems the normal response to library errors
3379 static bool do_suspend(OSThread* osthread) {
3380 // mark as suspended and send signal
3381 osthread->sr.set_suspend_action(SR_SUSPEND);
3382 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3383 assert_status(status == 0, status, "pthread_kill");
3385 // check status and wait until notified of suspension
3386 if (status == 0) {
3387 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3388 os::yield_all(i);
3389 }
3390 osthread->sr.set_suspend_action(SR_NONE);
3391 return true;
3392 }
3393 else {
3394 osthread->sr.set_suspend_action(SR_NONE);
3395 return false;
3396 }
3397 }
3399 static void do_resume(OSThread* osthread) {
3400 assert(osthread->sr.is_suspended(), "thread should be suspended");
3401 osthread->sr.set_suspend_action(SR_CONTINUE);
3403 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3404 assert_status(status == 0, status, "pthread_kill");
3405 // check status and wait unit notified of resumption
3406 if (status == 0) {
3407 for (int i = 0; osthread->sr.is_suspended(); i++) {
3408 os::yield_all(i);
3409 }
3410 }
3411 osthread->sr.set_suspend_action(SR_NONE);
3412 }
3414 ////////////////////////////////////////////////////////////////////////////////
3415 // interrupt support
3417 void os::interrupt(Thread* thread) {
3418 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3419 "possibility of dangling Thread pointer");
3421 OSThread* osthread = thread->osthread();
3423 if (!osthread->interrupted()) {
3424 osthread->set_interrupted(true);
3425 // More than one thread can get here with the same value of osthread,
3426 // resulting in multiple notifications. We do, however, want the store
3427 // to interrupted() to be visible to other threads before we execute unpark().
3428 OrderAccess::fence();
3429 ParkEvent * const slp = thread->_SleepEvent ;
3430 if (slp != NULL) slp->unpark() ;
3431 }
3433 // For JSR166. Unpark even if interrupt status already was set
3434 if (thread->is_Java_thread())
3435 ((JavaThread*)thread)->parker()->unpark();
3437 ParkEvent * ev = thread->_ParkEvent ;
3438 if (ev != NULL) ev->unpark() ;
3440 }
3442 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3443 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3444 "possibility of dangling Thread pointer");
3446 OSThread* osthread = thread->osthread();
3448 bool interrupted = osthread->interrupted();
3450 if (interrupted && clear_interrupted) {
3451 osthread->set_interrupted(false);
3452 // consider thread->_SleepEvent->reset() ... optional optimization
3453 }
3455 return interrupted;
3456 }
3458 ///////////////////////////////////////////////////////////////////////////////////
3459 // signal handling (except suspend/resume)
3461 // This routine may be used by user applications as a "hook" to catch signals.
3462 // The user-defined signal handler must pass unrecognized signals to this
3463 // routine, and if it returns true (non-zero), then the signal handler must
3464 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3465 // routine will never retun false (zero), but instead will execute a VM panic
3466 // routine kill the process.
3467 //
3468 // If this routine returns false, it is OK to call it again. This allows
3469 // the user-defined signal handler to perform checks either before or after
3470 // the VM performs its own checks. Naturally, the user code would be making
3471 // a serious error if it tried to handle an exception (such as a null check
3472 // or breakpoint) that the VM was generating for its own correct operation.
3473 //
3474 // This routine may recognize any of the following kinds of signals:
3475 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3476 // It should be consulted by handlers for any of those signals.
3477 //
3478 // The caller of this routine must pass in the three arguments supplied
3479 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3480 // field of the structure passed to sigaction(). This routine assumes that
3481 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3482 //
3483 // Note that the VM will print warnings if it detects conflicting signal
3484 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3485 //
3486 extern "C" int
3487 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3488 void* ucontext, int abort_if_unrecognized);
3490 void signalHandler(int sig, siginfo_t* info, void* uc) {
3491 assert(info != NULL && uc != NULL, "it must be old kernel");
3492 JVM_handle_linux_signal(sig, info, uc, true);
3493 }
3496 // This boolean allows users to forward their own non-matching signals
3497 // to JVM_handle_linux_signal, harmlessly.
3498 bool os::Linux::signal_handlers_are_installed = false;
3500 // For signal-chaining
3501 struct sigaction os::Linux::sigact[MAXSIGNUM];
3502 unsigned int os::Linux::sigs = 0;
3503 bool os::Linux::libjsig_is_loaded = false;
3504 typedef struct sigaction *(*get_signal_t)(int);
3505 get_signal_t os::Linux::get_signal_action = NULL;
3507 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3508 struct sigaction *actp = NULL;
3510 if (libjsig_is_loaded) {
3511 // Retrieve the old signal handler from libjsig
3512 actp = (*get_signal_action)(sig);
3513 }
3514 if (actp == NULL) {
3515 // Retrieve the preinstalled signal handler from jvm
3516 actp = get_preinstalled_handler(sig);
3517 }
3519 return actp;
3520 }
3522 static bool call_chained_handler(struct sigaction *actp, int sig,
3523 siginfo_t *siginfo, void *context) {
3524 // Call the old signal handler
3525 if (actp->sa_handler == SIG_DFL) {
3526 // It's more reasonable to let jvm treat it as an unexpected exception
3527 // instead of taking the default action.
3528 return false;
3529 } else if (actp->sa_handler != SIG_IGN) {
3530 if ((actp->sa_flags & SA_NODEFER) == 0) {
3531 // automaticlly block the signal
3532 sigaddset(&(actp->sa_mask), sig);
3533 }
3535 sa_handler_t hand;
3536 sa_sigaction_t sa;
3537 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3538 // retrieve the chained handler
3539 if (siginfo_flag_set) {
3540 sa = actp->sa_sigaction;
3541 } else {
3542 hand = actp->sa_handler;
3543 }
3545 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3546 actp->sa_handler = SIG_DFL;
3547 }
3549 // try to honor the signal mask
3550 sigset_t oset;
3551 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3553 // call into the chained handler
3554 if (siginfo_flag_set) {
3555 (*sa)(sig, siginfo, context);
3556 } else {
3557 (*hand)(sig);
3558 }
3560 // restore the signal mask
3561 pthread_sigmask(SIG_SETMASK, &oset, 0);
3562 }
3563 // Tell jvm's signal handler the signal is taken care of.
3564 return true;
3565 }
3567 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3568 bool chained = false;
3569 // signal-chaining
3570 if (UseSignalChaining) {
3571 struct sigaction *actp = get_chained_signal_action(sig);
3572 if (actp != NULL) {
3573 chained = call_chained_handler(actp, sig, siginfo, context);
3574 }
3575 }
3576 return chained;
3577 }
3579 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3580 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3581 return &sigact[sig];
3582 }
3583 return NULL;
3584 }
3586 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3587 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3588 sigact[sig] = oldAct;
3589 sigs |= (unsigned int)1 << sig;
3590 }
3592 // for diagnostic
3593 int os::Linux::sigflags[MAXSIGNUM];
3595 int os::Linux::get_our_sigflags(int sig) {
3596 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3597 return sigflags[sig];
3598 }
3600 void os::Linux::set_our_sigflags(int sig, int flags) {
3601 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3602 sigflags[sig] = flags;
3603 }
3605 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3606 // Check for overwrite.
3607 struct sigaction oldAct;
3608 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3610 void* oldhand = oldAct.sa_sigaction
3611 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3612 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3613 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3614 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3615 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3616 if (AllowUserSignalHandlers || !set_installed) {
3617 // Do not overwrite; user takes responsibility to forward to us.
3618 return;
3619 } else if (UseSignalChaining) {
3620 // save the old handler in jvm
3621 save_preinstalled_handler(sig, oldAct);
3622 // libjsig also interposes the sigaction() call below and saves the
3623 // old sigaction on it own.
3624 } else {
3625 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3626 "%#lx for signal %d.", (long)oldhand, sig));
3627 }
3628 }
3630 struct sigaction sigAct;
3631 sigfillset(&(sigAct.sa_mask));
3632 sigAct.sa_handler = SIG_DFL;
3633 if (!set_installed) {
3634 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3635 } else {
3636 sigAct.sa_sigaction = signalHandler;
3637 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3638 }
3639 // Save flags, which are set by ours
3640 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3641 sigflags[sig] = sigAct.sa_flags;
3643 int ret = sigaction(sig, &sigAct, &oldAct);
3644 assert(ret == 0, "check");
3646 void* oldhand2 = oldAct.sa_sigaction
3647 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3648 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3649 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3650 }
3652 // install signal handlers for signals that HotSpot needs to
3653 // handle in order to support Java-level exception handling.
3655 void os::Linux::install_signal_handlers() {
3656 if (!signal_handlers_are_installed) {
3657 signal_handlers_are_installed = true;
3659 // signal-chaining
3660 typedef void (*signal_setting_t)();
3661 signal_setting_t begin_signal_setting = NULL;
3662 signal_setting_t end_signal_setting = NULL;
3663 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3664 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3665 if (begin_signal_setting != NULL) {
3666 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3667 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3668 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3669 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3670 libjsig_is_loaded = true;
3671 assert(UseSignalChaining, "should enable signal-chaining");
3672 }
3673 if (libjsig_is_loaded) {
3674 // Tell libjsig jvm is setting signal handlers
3675 (*begin_signal_setting)();
3676 }
3678 set_signal_handler(SIGSEGV, true);
3679 set_signal_handler(SIGPIPE, true);
3680 set_signal_handler(SIGBUS, true);
3681 set_signal_handler(SIGILL, true);
3682 set_signal_handler(SIGFPE, true);
3683 set_signal_handler(SIGXFSZ, true);
3685 if (libjsig_is_loaded) {
3686 // Tell libjsig jvm finishes setting signal handlers
3687 (*end_signal_setting)();
3688 }
3690 // We don't activate signal checker if libjsig is in place, we trust ourselves
3691 // and if UserSignalHandler is installed all bets are off
3692 if (CheckJNICalls) {
3693 if (libjsig_is_loaded) {
3694 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3695 check_signals = false;
3696 }
3697 if (AllowUserSignalHandlers) {
3698 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3699 check_signals = false;
3700 }
3701 }
3702 }
3703 }
3705 // This is the fastest way to get thread cpu time on Linux.
3706 // Returns cpu time (user+sys) for any thread, not only for current.
3707 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3708 // It might work on 2.6.10+ with a special kernel/glibc patch.
3709 // For reference, please, see IEEE Std 1003.1-2004:
3710 // http://www.unix.org/single_unix_specification
3712 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3713 struct timespec tp;
3714 int rc = os::Linux::clock_gettime(clockid, &tp);
3715 assert(rc == 0, "clock_gettime is expected to return 0 code");
3717 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3718 }
3720 /////
3721 // glibc on Linux platform uses non-documented flag
3722 // to indicate, that some special sort of signal
3723 // trampoline is used.
3724 // We will never set this flag, and we should
3725 // ignore this flag in our diagnostic
3726 #ifdef SIGNIFICANT_SIGNAL_MASK
3727 #undef SIGNIFICANT_SIGNAL_MASK
3728 #endif
3729 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3731 static const char* get_signal_handler_name(address handler,
3732 char* buf, int buflen) {
3733 int offset;
3734 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3735 if (found) {
3736 // skip directory names
3737 const char *p1, *p2;
3738 p1 = buf;
3739 size_t len = strlen(os::file_separator());
3740 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3741 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3742 } else {
3743 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3744 }
3745 return buf;
3746 }
3748 static void print_signal_handler(outputStream* st, int sig,
3749 char* buf, size_t buflen) {
3750 struct sigaction sa;
3752 sigaction(sig, NULL, &sa);
3754 // See comment for SIGNIFICANT_SIGNAL_MASK define
3755 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3757 st->print("%s: ", os::exception_name(sig, buf, buflen));
3759 address handler = (sa.sa_flags & SA_SIGINFO)
3760 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3761 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3763 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3764 st->print("SIG_DFL");
3765 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3766 st->print("SIG_IGN");
3767 } else {
3768 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3769 }
3771 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3773 address rh = VMError::get_resetted_sighandler(sig);
3774 // May be, handler was resetted by VMError?
3775 if(rh != NULL) {
3776 handler = rh;
3777 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3778 }
3780 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3782 // Check: is it our handler?
3783 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3784 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3785 // It is our signal handler
3786 // check for flags, reset system-used one!
3787 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3788 st->print(
3789 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3790 os::Linux::get_our_sigflags(sig));
3791 }
3792 }
3793 st->cr();
3794 }
3797 #define DO_SIGNAL_CHECK(sig) \
3798 if (!sigismember(&check_signal_done, sig)) \
3799 os::Linux::check_signal_handler(sig)
3801 // This method is a periodic task to check for misbehaving JNI applications
3802 // under CheckJNI, we can add any periodic checks here
3804 void os::run_periodic_checks() {
3806 if (check_signals == false) return;
3808 // SEGV and BUS if overridden could potentially prevent
3809 // generation of hs*.log in the event of a crash, debugging
3810 // such a case can be very challenging, so we absolutely
3811 // check the following for a good measure:
3812 DO_SIGNAL_CHECK(SIGSEGV);
3813 DO_SIGNAL_CHECK(SIGILL);
3814 DO_SIGNAL_CHECK(SIGFPE);
3815 DO_SIGNAL_CHECK(SIGBUS);
3816 DO_SIGNAL_CHECK(SIGPIPE);
3817 DO_SIGNAL_CHECK(SIGXFSZ);
3820 // ReduceSignalUsage allows the user to override these handlers
3821 // see comments at the very top and jvm_solaris.h
3822 if (!ReduceSignalUsage) {
3823 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3824 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3825 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3826 DO_SIGNAL_CHECK(BREAK_SIGNAL);
3827 }
3829 DO_SIGNAL_CHECK(SR_signum);
3830 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3831 }
3833 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3835 static os_sigaction_t os_sigaction = NULL;
3837 void os::Linux::check_signal_handler(int sig) {
3838 char buf[O_BUFLEN];
3839 address jvmHandler = NULL;
3842 struct sigaction act;
3843 if (os_sigaction == NULL) {
3844 // only trust the default sigaction, in case it has been interposed
3845 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3846 if (os_sigaction == NULL) return;
3847 }
3849 os_sigaction(sig, (struct sigaction*)NULL, &act);
3852 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3854 address thisHandler = (act.sa_flags & SA_SIGINFO)
3855 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3856 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3859 switch(sig) {
3860 case SIGSEGV:
3861 case SIGBUS:
3862 case SIGFPE:
3863 case SIGPIPE:
3864 case SIGILL:
3865 case SIGXFSZ:
3866 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3867 break;
3869 case SHUTDOWN1_SIGNAL:
3870 case SHUTDOWN2_SIGNAL:
3871 case SHUTDOWN3_SIGNAL:
3872 case BREAK_SIGNAL:
3873 jvmHandler = (address)user_handler();
3874 break;
3876 case INTERRUPT_SIGNAL:
3877 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
3878 break;
3880 default:
3881 if (sig == SR_signum) {
3882 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
3883 } else {
3884 return;
3885 }
3886 break;
3887 }
3889 if (thisHandler != jvmHandler) {
3890 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3891 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3892 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3893 // No need to check this sig any longer
3894 sigaddset(&check_signal_done, sig);
3895 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
3896 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3897 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
3898 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
3899 // No need to check this sig any longer
3900 sigaddset(&check_signal_done, sig);
3901 }
3903 // Dump all the signal
3904 if (sigismember(&check_signal_done, sig)) {
3905 print_signal_handlers(tty, buf, O_BUFLEN);
3906 }
3907 }
3909 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
3911 extern bool signal_name(int signo, char* buf, size_t len);
3913 const char* os::exception_name(int exception_code, char* buf, size_t size) {
3914 if (0 < exception_code && exception_code <= SIGRTMAX) {
3915 // signal
3916 if (!signal_name(exception_code, buf, size)) {
3917 jio_snprintf(buf, size, "SIG%d", exception_code);
3918 }
3919 return buf;
3920 } else {
3921 return NULL;
3922 }
3923 }
3925 // this is called _before_ the most of global arguments have been parsed
3926 void os::init(void) {
3927 char dummy; /* used to get a guess on initial stack address */
3928 // first_hrtime = gethrtime();
3930 // With LinuxThreads the JavaMain thread pid (primordial thread)
3931 // is different than the pid of the java launcher thread.
3932 // So, on Linux, the launcher thread pid is passed to the VM
3933 // via the sun.java.launcher.pid property.
3934 // Use this property instead of getpid() if it was correctly passed.
3935 // See bug 6351349.
3936 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
3938 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
3940 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
3942 init_random(1234567);
3944 ThreadCritical::initialize();
3946 Linux::set_page_size(sysconf(_SC_PAGESIZE));
3947 if (Linux::page_size() == -1) {
3948 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
3949 strerror(errno)));
3950 }
3951 init_page_sizes((size_t) Linux::page_size());
3953 Linux::initialize_system_info();
3955 // main_thread points to the aboriginal thread
3956 Linux::_main_thread = pthread_self();
3958 Linux::clock_init();
3959 initial_time_count = os::elapsed_counter();
3960 pthread_mutex_init(&dl_mutex, NULL);
3961 }
3963 // To install functions for atexit system call
3964 extern "C" {
3965 static void perfMemory_exit_helper() {
3966 perfMemory_exit();
3967 }
3968 }
3970 // this is called _after_ the global arguments have been parsed
3971 jint os::init_2(void)
3972 {
3973 Linux::fast_thread_clock_init();
3975 // Allocate a single page and mark it as readable for safepoint polling
3976 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3977 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
3979 os::set_polling_page( polling_page );
3981 #ifndef PRODUCT
3982 if(Verbose && PrintMiscellaneous)
3983 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
3984 #endif
3986 if (!UseMembar) {
3987 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
3988 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
3989 os::set_memory_serialize_page( mem_serialize_page );
3991 #ifndef PRODUCT
3992 if(Verbose && PrintMiscellaneous)
3993 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
3994 #endif
3995 }
3997 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
3999 // initialize suspend/resume support - must do this before signal_sets_init()
4000 if (SR_initialize() != 0) {
4001 perror("SR_initialize failed");
4002 return JNI_ERR;
4003 }
4005 Linux::signal_sets_init();
4006 Linux::install_signal_handlers();
4008 // Check minimum allowable stack size for thread creation and to initialize
4009 // the java system classes, including StackOverflowError - depends on page
4010 // size. Add a page for compiler2 recursion in main thread.
4011 // Add in 2*BytesPerWord times page size to account for VM stack during
4012 // class initialization depending on 32 or 64 bit VM.
4013 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4014 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4015 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4017 size_t threadStackSizeInBytes = ThreadStackSize * K;
4018 if (threadStackSizeInBytes != 0 &&
4019 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4020 tty->print_cr("\nThe stack size specified is too small, "
4021 "Specify at least %dk",
4022 os::Linux::min_stack_allowed/ K);
4023 return JNI_ERR;
4024 }
4026 // Make the stack size a multiple of the page size so that
4027 // the yellow/red zones can be guarded.
4028 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4029 vm_page_size()));
4031 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4033 Linux::libpthread_init();
4034 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4035 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4036 Linux::glibc_version(), Linux::libpthread_version(),
4037 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4038 }
4040 if (UseNUMA) {
4041 if (!Linux::libnuma_init()) {
4042 UseNUMA = false;
4043 } else {
4044 if ((Linux::numa_max_node() < 1)) {
4045 // There's only one node(they start from 0), disable NUMA.
4046 UseNUMA = false;
4047 }
4048 }
4049 if (!UseNUMA && ForceNUMA) {
4050 UseNUMA = true;
4051 }
4052 }
4054 if (MaxFDLimit) {
4055 // set the number of file descriptors to max. print out error
4056 // if getrlimit/setrlimit fails but continue regardless.
4057 struct rlimit nbr_files;
4058 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4059 if (status != 0) {
4060 if (PrintMiscellaneous && (Verbose || WizardMode))
4061 perror("os::init_2 getrlimit failed");
4062 } else {
4063 nbr_files.rlim_cur = nbr_files.rlim_max;
4064 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4065 if (status != 0) {
4066 if (PrintMiscellaneous && (Verbose || WizardMode))
4067 perror("os::init_2 setrlimit failed");
4068 }
4069 }
4070 }
4072 // Initialize lock used to serialize thread creation (see os::create_thread)
4073 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4075 // at-exit methods are called in the reverse order of their registration.
4076 // atexit functions are called on return from main or as a result of a
4077 // call to exit(3C). There can be only 32 of these functions registered
4078 // and atexit() does not set errno.
4080 if (PerfAllowAtExitRegistration) {
4081 // only register atexit functions if PerfAllowAtExitRegistration is set.
4082 // atexit functions can be delayed until process exit time, which
4083 // can be problematic for embedded VM situations. Embedded VMs should
4084 // call DestroyJavaVM() to assure that VM resources are released.
4086 // note: perfMemory_exit_helper atexit function may be removed in
4087 // the future if the appropriate cleanup code can be added to the
4088 // VM_Exit VMOperation's doit method.
4089 if (atexit(perfMemory_exit_helper) != 0) {
4090 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4091 }
4092 }
4094 // initialize thread priority policy
4095 prio_init();
4097 return JNI_OK;
4098 }
4100 // this is called at the end of vm_initialization
4101 void os::init_3(void) { }
4103 // Mark the polling page as unreadable
4104 void os::make_polling_page_unreadable(void) {
4105 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4106 fatal("Could not disable polling page");
4107 };
4109 // Mark the polling page as readable
4110 void os::make_polling_page_readable(void) {
4111 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4112 fatal("Could not enable polling page");
4113 }
4114 };
4116 int os::active_processor_count() {
4117 // Linux doesn't yet have a (official) notion of processor sets,
4118 // so just return the number of online processors.
4119 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4120 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4121 return online_cpus;
4122 }
4124 bool os::distribute_processes(uint length, uint* distribution) {
4125 // Not yet implemented.
4126 return false;
4127 }
4129 bool os::bind_to_processor(uint processor_id) {
4130 // Not yet implemented.
4131 return false;
4132 }
4134 ///
4136 // Suspends the target using the signal mechanism and then grabs the PC before
4137 // resuming the target. Used by the flat-profiler only
4138 ExtendedPC os::get_thread_pc(Thread* thread) {
4139 // Make sure that it is called by the watcher for the VMThread
4140 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4141 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4143 ExtendedPC epc;
4145 OSThread* osthread = thread->osthread();
4146 if (do_suspend(osthread)) {
4147 if (osthread->ucontext() != NULL) {
4148 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4149 } else {
4150 // NULL context is unexpected, double-check this is the VMThread
4151 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4152 }
4153 do_resume(osthread);
4154 }
4155 // failure means pthread_kill failed for some reason - arguably this is
4156 // a fatal problem, but such problems are ignored elsewhere
4158 return epc;
4159 }
4161 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4162 {
4163 if (is_NPTL()) {
4164 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4165 } else {
4166 #ifndef IA64
4167 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4168 // word back to default 64bit precision if condvar is signaled. Java
4169 // wants 53bit precision. Save and restore current value.
4170 int fpu = get_fpu_control_word();
4171 #endif // IA64
4172 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4173 #ifndef IA64
4174 set_fpu_control_word(fpu);
4175 #endif // IA64
4176 return status;
4177 }
4178 }
4180 ////////////////////////////////////////////////////////////////////////////////
4181 // debug support
4183 static address same_page(address x, address y) {
4184 int page_bits = -os::vm_page_size();
4185 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4186 return x;
4187 else if (x > y)
4188 return (address)(intptr_t(y) | ~page_bits) + 1;
4189 else
4190 return (address)(intptr_t(y) & page_bits);
4191 }
4193 bool os::find(address addr, outputStream* st) {
4194 Dl_info dlinfo;
4195 memset(&dlinfo, 0, sizeof(dlinfo));
4196 if (dladdr(addr, &dlinfo)) {
4197 st->print(PTR_FORMAT ": ", addr);
4198 if (dlinfo.dli_sname != NULL) {
4199 st->print("%s+%#x", dlinfo.dli_sname,
4200 addr - (intptr_t)dlinfo.dli_saddr);
4201 } else if (dlinfo.dli_fname) {
4202 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4203 } else {
4204 st->print("<absolute address>");
4205 }
4206 if (dlinfo.dli_fname) {
4207 st->print(" in %s", dlinfo.dli_fname);
4208 }
4209 if (dlinfo.dli_fbase) {
4210 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4211 }
4212 st->cr();
4214 if (Verbose) {
4215 // decode some bytes around the PC
4216 address begin = same_page(addr-40, addr);
4217 address end = same_page(addr+40, addr);
4218 address lowest = (address) dlinfo.dli_sname;
4219 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4220 if (begin < lowest) begin = lowest;
4221 Dl_info dlinfo2;
4222 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4223 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4224 end = (address) dlinfo2.dli_saddr;
4225 Disassembler::decode(begin, end, st);
4226 }
4227 return true;
4228 }
4229 return false;
4230 }
4232 ////////////////////////////////////////////////////////////////////////////////
4233 // misc
4235 // This does not do anything on Linux. This is basically a hook for being
4236 // able to use structured exception handling (thread-local exception filters)
4237 // on, e.g., Win32.
4238 void
4239 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4240 JavaCallArguments* args, Thread* thread) {
4241 f(value, method, args, thread);
4242 }
4244 void os::print_statistics() {
4245 }
4247 int os::message_box(const char* title, const char* message) {
4248 int i;
4249 fdStream err(defaultStream::error_fd());
4250 for (i = 0; i < 78; i++) err.print_raw("=");
4251 err.cr();
4252 err.print_raw_cr(title);
4253 for (i = 0; i < 78; i++) err.print_raw("-");
4254 err.cr();
4255 err.print_raw_cr(message);
4256 for (i = 0; i < 78; i++) err.print_raw("=");
4257 err.cr();
4259 char buf[16];
4260 // Prevent process from exiting upon "read error" without consuming all CPU
4261 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4263 return buf[0] == 'y' || buf[0] == 'Y';
4264 }
4266 int os::stat(const char *path, struct stat *sbuf) {
4267 char pathbuf[MAX_PATH];
4268 if (strlen(path) > MAX_PATH - 1) {
4269 errno = ENAMETOOLONG;
4270 return -1;
4271 }
4272 os::native_path(strcpy(pathbuf, path));
4273 return ::stat(pathbuf, sbuf);
4274 }
4276 bool os::check_heap(bool force) {
4277 return true;
4278 }
4280 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4281 return ::vsnprintf(buf, count, format, args);
4282 }
4284 // Is a (classpath) directory empty?
4285 bool os::dir_is_empty(const char* path) {
4286 DIR *dir = NULL;
4287 struct dirent *ptr;
4289 dir = opendir(path);
4290 if (dir == NULL) return true;
4292 /* Scan the directory */
4293 bool result = true;
4294 char buf[sizeof(struct dirent) + MAX_PATH];
4295 while (result && (ptr = ::readdir(dir)) != NULL) {
4296 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4297 result = false;
4298 }
4299 }
4300 closedir(dir);
4301 return result;
4302 }
4304 // This code originates from JDK's sysOpen and open64_w
4305 // from src/solaris/hpi/src/system_md.c
4307 #ifndef O_DELETE
4308 #define O_DELETE 0x10000
4309 #endif
4311 // Open a file. Unlink the file immediately after open returns
4312 // if the specified oflag has the O_DELETE flag set.
4313 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4315 int os::open(const char *path, int oflag, int mode) {
4317 if (strlen(path) > MAX_PATH - 1) {
4318 errno = ENAMETOOLONG;
4319 return -1;
4320 }
4321 int fd;
4322 int o_delete = (oflag & O_DELETE);
4323 oflag = oflag & ~O_DELETE;
4325 fd = ::open64(path, oflag, mode);
4326 if (fd == -1) return -1;
4328 //If the open succeeded, the file might still be a directory
4329 {
4330 struct stat64 buf64;
4331 int ret = ::fstat64(fd, &buf64);
4332 int st_mode = buf64.st_mode;
4334 if (ret != -1) {
4335 if ((st_mode & S_IFMT) == S_IFDIR) {
4336 errno = EISDIR;
4337 ::close(fd);
4338 return -1;
4339 }
4340 } else {
4341 ::close(fd);
4342 return -1;
4343 }
4344 }
4346 /*
4347 * All file descriptors that are opened in the JVM and not
4348 * specifically destined for a subprocess should have the
4349 * close-on-exec flag set. If we don't set it, then careless 3rd
4350 * party native code might fork and exec without closing all
4351 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4352 * UNIXProcess.c), and this in turn might:
4353 *
4354 * - cause end-of-file to fail to be detected on some file
4355 * descriptors, resulting in mysterious hangs, or
4356 *
4357 * - might cause an fopen in the subprocess to fail on a system
4358 * suffering from bug 1085341.
4359 *
4360 * (Yes, the default setting of the close-on-exec flag is a Unix
4361 * design flaw)
4362 *
4363 * See:
4364 * 1085341: 32-bit stdio routines should support file descriptors >255
4365 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4366 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4367 */
4368 #ifdef FD_CLOEXEC
4369 {
4370 int flags = ::fcntl(fd, F_GETFD);
4371 if (flags != -1)
4372 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4373 }
4374 #endif
4376 if (o_delete != 0) {
4377 ::unlink(path);
4378 }
4379 return fd;
4380 }
4383 // create binary file, rewriting existing file if required
4384 int os::create_binary_file(const char* path, bool rewrite_existing) {
4385 int oflags = O_WRONLY | O_CREAT;
4386 if (!rewrite_existing) {
4387 oflags |= O_EXCL;
4388 }
4389 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4390 }
4392 // return current position of file pointer
4393 jlong os::current_file_offset(int fd) {
4394 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4395 }
4397 // move file pointer to the specified offset
4398 jlong os::seek_to_file_offset(int fd, jlong offset) {
4399 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4400 }
4402 // This code originates from JDK's sysAvailable
4403 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4405 int os::available(int fd, jlong *bytes) {
4406 jlong cur, end;
4407 int mode;
4408 struct stat64 buf64;
4410 if (::fstat64(fd, &buf64) >= 0) {
4411 mode = buf64.st_mode;
4412 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4413 /*
4414 * XXX: is the following call interruptible? If so, this might
4415 * need to go through the INTERRUPT_IO() wrapper as for other
4416 * blocking, interruptible calls in this file.
4417 */
4418 int n;
4419 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4420 *bytes = n;
4421 return 1;
4422 }
4423 }
4424 }
4425 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4426 return 0;
4427 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4428 return 0;
4429 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4430 return 0;
4431 }
4432 *bytes = end - cur;
4433 return 1;
4434 }
4436 int os::socket_available(int fd, jint *pbytes) {
4437 // Linux doc says EINTR not returned, unlike Solaris
4438 int ret = ::ioctl(fd, FIONREAD, pbytes);
4440 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4441 // is expected to return 0 on failure and 1 on success to the jdk.
4442 return (ret < 0) ? 0 : 1;
4443 }
4445 // Map a block of memory.
4446 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4447 char *addr, size_t bytes, bool read_only,
4448 bool allow_exec) {
4449 int prot;
4450 int flags;
4452 if (read_only) {
4453 prot = PROT_READ;
4454 flags = MAP_SHARED;
4455 } else {
4456 prot = PROT_READ | PROT_WRITE;
4457 flags = MAP_PRIVATE;
4458 }
4460 if (allow_exec) {
4461 prot |= PROT_EXEC;
4462 }
4464 if (addr != NULL) {
4465 flags |= MAP_FIXED;
4466 }
4468 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4469 fd, file_offset);
4470 if (mapped_address == MAP_FAILED) {
4471 return NULL;
4472 }
4473 return mapped_address;
4474 }
4477 // Remap a block of memory.
4478 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4479 char *addr, size_t bytes, bool read_only,
4480 bool allow_exec) {
4481 // same as map_memory() on this OS
4482 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4483 allow_exec);
4484 }
4487 // Unmap a block of memory.
4488 bool os::unmap_memory(char* addr, size_t bytes) {
4489 return munmap(addr, bytes) == 0;
4490 }
4492 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4494 static clockid_t thread_cpu_clockid(Thread* thread) {
4495 pthread_t tid = thread->osthread()->pthread_id();
4496 clockid_t clockid;
4498 // Get thread clockid
4499 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4500 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4501 return clockid;
4502 }
4504 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4505 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4506 // of a thread.
4507 //
4508 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4509 // the fast estimate available on the platform.
4511 jlong os::current_thread_cpu_time() {
4512 if (os::Linux::supports_fast_thread_cpu_time()) {
4513 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4514 } else {
4515 // return user + sys since the cost is the same
4516 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4517 }
4518 }
4520 jlong os::thread_cpu_time(Thread* thread) {
4521 // consistent with what current_thread_cpu_time() returns
4522 if (os::Linux::supports_fast_thread_cpu_time()) {
4523 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4524 } else {
4525 return slow_thread_cpu_time(thread, true /* user + sys */);
4526 }
4527 }
4529 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4530 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4531 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4532 } else {
4533 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4534 }
4535 }
4537 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4538 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4539 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4540 } else {
4541 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4542 }
4543 }
4545 //
4546 // -1 on error.
4547 //
4549 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4550 static bool proc_pid_cpu_avail = true;
4551 static bool proc_task_unchecked = true;
4552 static const char *proc_stat_path = "/proc/%d/stat";
4553 pid_t tid = thread->osthread()->thread_id();
4554 int i;
4555 char *s;
4556 char stat[2048];
4557 int statlen;
4558 char proc_name[64];
4559 int count;
4560 long sys_time, user_time;
4561 char string[64];
4562 char cdummy;
4563 int idummy;
4564 long ldummy;
4565 FILE *fp;
4567 // We first try accessing /proc/<pid>/cpu since this is faster to
4568 // process. If this file is not present (linux kernels 2.5 and above)
4569 // then we open /proc/<pid>/stat.
4570 if ( proc_pid_cpu_avail ) {
4571 sprintf(proc_name, "/proc/%d/cpu", tid);
4572 fp = fopen(proc_name, "r");
4573 if ( fp != NULL ) {
4574 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4575 fclose(fp);
4576 if ( count != 3 ) return -1;
4578 if (user_sys_cpu_time) {
4579 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4580 } else {
4581 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4582 }
4583 }
4584 else proc_pid_cpu_avail = false;
4585 }
4587 // The /proc/<tid>/stat aggregates per-process usage on
4588 // new Linux kernels 2.6+ where NPTL is supported.
4589 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4590 // See bug 6328462.
4591 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4592 // and possibly in some other cases, so we check its availability.
4593 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4594 // This is executed only once
4595 proc_task_unchecked = false;
4596 fp = fopen("/proc/self/task", "r");
4597 if (fp != NULL) {
4598 proc_stat_path = "/proc/self/task/%d/stat";
4599 fclose(fp);
4600 }
4601 }
4603 sprintf(proc_name, proc_stat_path, tid);
4604 fp = fopen(proc_name, "r");
4605 if ( fp == NULL ) return -1;
4606 statlen = fread(stat, 1, 2047, fp);
4607 stat[statlen] = '\0';
4608 fclose(fp);
4610 // Skip pid and the command string. Note that we could be dealing with
4611 // weird command names, e.g. user could decide to rename java launcher
4612 // to "java 1.4.2 :)", then the stat file would look like
4613 // 1234 (java 1.4.2 :)) R ... ...
4614 // We don't really need to know the command string, just find the last
4615 // occurrence of ")" and then start parsing from there. See bug 4726580.
4616 s = strrchr(stat, ')');
4617 i = 0;
4618 if (s == NULL ) return -1;
4620 // Skip blank chars
4621 do s++; while (isspace(*s));
4623 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4624 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4625 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4626 &user_time, &sys_time);
4627 if ( count != 13 ) return -1;
4628 if (user_sys_cpu_time) {
4629 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4630 } else {
4631 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4632 }
4633 }
4635 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4636 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4637 info_ptr->may_skip_backward = false; // elapsed time not wall time
4638 info_ptr->may_skip_forward = false; // elapsed time not wall time
4639 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4640 }
4642 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4643 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4644 info_ptr->may_skip_backward = false; // elapsed time not wall time
4645 info_ptr->may_skip_forward = false; // elapsed time not wall time
4646 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4647 }
4649 bool os::is_thread_cpu_time_supported() {
4650 return true;
4651 }
4653 // System loadavg support. Returns -1 if load average cannot be obtained.
4654 // Linux doesn't yet have a (official) notion of processor sets,
4655 // so just return the system wide load average.
4656 int os::loadavg(double loadavg[], int nelem) {
4657 return ::getloadavg(loadavg, nelem);
4658 }
4660 void os::pause() {
4661 char filename[MAX_PATH];
4662 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4663 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4664 } else {
4665 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4666 }
4668 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4669 if (fd != -1) {
4670 struct stat buf;
4671 ::close(fd);
4672 while (::stat(filename, &buf) == 0) {
4673 (void)::poll(NULL, 0, 100);
4674 }
4675 } else {
4676 jio_fprintf(stderr,
4677 "Could not open pause file '%s', continuing immediately.\n", filename);
4678 }
4679 }
4681 extern "C" {
4683 /**
4684 * NOTE: the following code is to keep the green threads code
4685 * in the libjava.so happy. Once the green threads is removed,
4686 * these code will no longer be needed.
4687 */
4688 int
4689 jdk_waitpid(pid_t pid, int* status, int options) {
4690 return waitpid(pid, status, options);
4691 }
4693 int
4694 fork1() {
4695 return fork();
4696 }
4698 int
4699 jdk_sem_init(sem_t *sem, int pshared, unsigned int value) {
4700 return sem_init(sem, pshared, value);
4701 }
4703 int
4704 jdk_sem_post(sem_t *sem) {
4705 return sem_post(sem);
4706 }
4708 int
4709 jdk_sem_wait(sem_t *sem) {
4710 return sem_wait(sem);
4711 }
4713 int
4714 jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) {
4715 return pthread_sigmask(how , newmask, oldmask);
4716 }
4718 }
4720 // Refer to the comments in os_solaris.cpp park-unpark.
4721 //
4722 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4723 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4724 // For specifics regarding the bug see GLIBC BUGID 261237 :
4725 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4726 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4727 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4728 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4729 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4730 // and monitorenter when we're using 1-0 locking. All those operations may result in
4731 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4732 // of libpthread avoids the problem, but isn't practical.
4733 //
4734 // Possible remedies:
4735 //
4736 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4737 // This is palliative and probabilistic, however. If the thread is preempted
4738 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4739 // than the minimum period may have passed, and the abstime may be stale (in the
4740 // past) resultin in a hang. Using this technique reduces the odds of a hang
4741 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4742 //
4743 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4744 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4745 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4746 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4747 // thread.
4748 //
4749 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4750 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4751 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4752 // This also works well. In fact it avoids kernel-level scalability impediments
4753 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4754 // timers in a graceful fashion.
4755 //
4756 // 4. When the abstime value is in the past it appears that control returns
4757 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4758 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4759 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4760 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4761 // It may be possible to avoid reinitialization by checking the return
4762 // value from pthread_cond_timedwait(). In addition to reinitializing the
4763 // condvar we must establish the invariant that cond_signal() is only called
4764 // within critical sections protected by the adjunct mutex. This prevents
4765 // cond_signal() from "seeing" a condvar that's in the midst of being
4766 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4767 // desirable signal-after-unlock optimization that avoids futile context switching.
4768 //
4769 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4770 // structure when a condvar is used or initialized. cond_destroy() would
4771 // release the helper structure. Our reinitialize-after-timedwait fix
4772 // put excessive stress on malloc/free and locks protecting the c-heap.
4773 //
4774 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4775 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4776 // and only enabling the work-around for vulnerable environments.
4778 // utility to compute the abstime argument to timedwait:
4779 // millis is the relative timeout time
4780 // abstime will be the absolute timeout time
4781 // TODO: replace compute_abstime() with unpackTime()
4783 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4784 if (millis < 0) millis = 0;
4785 struct timeval now;
4786 int status = gettimeofday(&now, NULL);
4787 assert(status == 0, "gettimeofday");
4788 jlong seconds = millis / 1000;
4789 millis %= 1000;
4790 if (seconds > 50000000) { // see man cond_timedwait(3T)
4791 seconds = 50000000;
4792 }
4793 abstime->tv_sec = now.tv_sec + seconds;
4794 long usec = now.tv_usec + millis * 1000;
4795 if (usec >= 1000000) {
4796 abstime->tv_sec += 1;
4797 usec -= 1000000;
4798 }
4799 abstime->tv_nsec = usec * 1000;
4800 return abstime;
4801 }
4804 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4805 // Conceptually TryPark() should be equivalent to park(0).
4807 int os::PlatformEvent::TryPark() {
4808 for (;;) {
4809 const int v = _Event ;
4810 guarantee ((v == 0) || (v == 1), "invariant") ;
4811 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4812 }
4813 }
4815 void os::PlatformEvent::park() { // AKA "down()"
4816 // Invariant: Only the thread associated with the Event/PlatformEvent
4817 // may call park().
4818 // TODO: assert that _Assoc != NULL or _Assoc == Self
4819 int v ;
4820 for (;;) {
4821 v = _Event ;
4822 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4823 }
4824 guarantee (v >= 0, "invariant") ;
4825 if (v == 0) {
4826 // Do this the hard way by blocking ...
4827 int status = pthread_mutex_lock(_mutex);
4828 assert_status(status == 0, status, "mutex_lock");
4829 guarantee (_nParked == 0, "invariant") ;
4830 ++ _nParked ;
4831 while (_Event < 0) {
4832 status = pthread_cond_wait(_cond, _mutex);
4833 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4834 // Treat this the same as if the wait was interrupted
4835 if (status == ETIME) { status = EINTR; }
4836 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4837 }
4838 -- _nParked ;
4840 // In theory we could move the ST of 0 into _Event past the unlock(),
4841 // but then we'd need a MEMBAR after the ST.
4842 _Event = 0 ;
4843 status = pthread_mutex_unlock(_mutex);
4844 assert_status(status == 0, status, "mutex_unlock");
4845 }
4846 guarantee (_Event >= 0, "invariant") ;
4847 }
4849 int os::PlatformEvent::park(jlong millis) {
4850 guarantee (_nParked == 0, "invariant") ;
4852 int v ;
4853 for (;;) {
4854 v = _Event ;
4855 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4856 }
4857 guarantee (v >= 0, "invariant") ;
4858 if (v != 0) return OS_OK ;
4860 // We do this the hard way, by blocking the thread.
4861 // Consider enforcing a minimum timeout value.
4862 struct timespec abst;
4863 compute_abstime(&abst, millis);
4865 int ret = OS_TIMEOUT;
4866 int status = pthread_mutex_lock(_mutex);
4867 assert_status(status == 0, status, "mutex_lock");
4868 guarantee (_nParked == 0, "invariant") ;
4869 ++_nParked ;
4871 // Object.wait(timo) will return because of
4872 // (a) notification
4873 // (b) timeout
4874 // (c) thread.interrupt
4875 //
4876 // Thread.interrupt and object.notify{All} both call Event::set.
4877 // That is, we treat thread.interrupt as a special case of notification.
4878 // The underlying Solaris implementation, cond_timedwait, admits
4879 // spurious/premature wakeups, but the JLS/JVM spec prevents the
4880 // JVM from making those visible to Java code. As such, we must
4881 // filter out spurious wakeups. We assume all ETIME returns are valid.
4882 //
4883 // TODO: properly differentiate simultaneous notify+interrupt.
4884 // In that case, we should propagate the notify to another waiter.
4886 while (_Event < 0) {
4887 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4888 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4889 pthread_cond_destroy (_cond);
4890 pthread_cond_init (_cond, NULL) ;
4891 }
4892 assert_status(status == 0 || status == EINTR ||
4893 status == ETIME || status == ETIMEDOUT,
4894 status, "cond_timedwait");
4895 if (!FilterSpuriousWakeups) break ; // previous semantics
4896 if (status == ETIME || status == ETIMEDOUT) break ;
4897 // We consume and ignore EINTR and spurious wakeups.
4898 }
4899 --_nParked ;
4900 if (_Event >= 0) {
4901 ret = OS_OK;
4902 }
4903 _Event = 0 ;
4904 status = pthread_mutex_unlock(_mutex);
4905 assert_status(status == 0, status, "mutex_unlock");
4906 assert (_nParked == 0, "invariant") ;
4907 return ret;
4908 }
4910 void os::PlatformEvent::unpark() {
4911 int v, AnyWaiters ;
4912 for (;;) {
4913 v = _Event ;
4914 if (v > 0) {
4915 // The LD of _Event could have reordered or be satisfied
4916 // by a read-aside from this processor's write buffer.
4917 // To avoid problems execute a barrier and then
4918 // ratify the value.
4919 OrderAccess::fence() ;
4920 if (_Event == v) return ;
4921 continue ;
4922 }
4923 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
4924 }
4925 if (v < 0) {
4926 // Wait for the thread associated with the event to vacate
4927 int status = pthread_mutex_lock(_mutex);
4928 assert_status(status == 0, status, "mutex_lock");
4929 AnyWaiters = _nParked ;
4930 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
4931 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
4932 AnyWaiters = 0 ;
4933 pthread_cond_signal (_cond);
4934 }
4935 status = pthread_mutex_unlock(_mutex);
4936 assert_status(status == 0, status, "mutex_unlock");
4937 if (AnyWaiters != 0) {
4938 status = pthread_cond_signal(_cond);
4939 assert_status(status == 0, status, "cond_signal");
4940 }
4941 }
4943 // Note that we signal() _after dropping the lock for "immortal" Events.
4944 // This is safe and avoids a common class of futile wakeups. In rare
4945 // circumstances this can cause a thread to return prematurely from
4946 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
4947 // simply re-test the condition and re-park itself.
4948 }
4951 // JSR166
4952 // -------------------------------------------------------
4954 /*
4955 * The solaris and linux implementations of park/unpark are fairly
4956 * conservative for now, but can be improved. They currently use a
4957 * mutex/condvar pair, plus a a count.
4958 * Park decrements count if > 0, else does a condvar wait. Unpark
4959 * sets count to 1 and signals condvar. Only one thread ever waits
4960 * on the condvar. Contention seen when trying to park implies that someone
4961 * is unparking you, so don't wait. And spurious returns are fine, so there
4962 * is no need to track notifications.
4963 */
4966 #define NANOSECS_PER_SEC 1000000000
4967 #define NANOSECS_PER_MILLISEC 1000000
4968 #define MAX_SECS 100000000
4969 /*
4970 * This code is common to linux and solaris and will be moved to a
4971 * common place in dolphin.
4972 *
4973 * The passed in time value is either a relative time in nanoseconds
4974 * or an absolute time in milliseconds. Either way it has to be unpacked
4975 * into suitable seconds and nanoseconds components and stored in the
4976 * given timespec structure.
4977 * Given time is a 64-bit value and the time_t used in the timespec is only
4978 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
4979 * overflow if times way in the future are given. Further on Solaris versions
4980 * prior to 10 there is a restriction (see cond_timedwait) that the specified
4981 * number of seconds, in abstime, is less than current_time + 100,000,000.
4982 * As it will be 28 years before "now + 100000000" will overflow we can
4983 * ignore overflow and just impose a hard-limit on seconds using the value
4984 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
4985 * years from "now".
4986 */
4988 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
4989 assert (time > 0, "convertTime");
4991 struct timeval now;
4992 int status = gettimeofday(&now, NULL);
4993 assert(status == 0, "gettimeofday");
4995 time_t max_secs = now.tv_sec + MAX_SECS;
4997 if (isAbsolute) {
4998 jlong secs = time / 1000;
4999 if (secs > max_secs) {
5000 absTime->tv_sec = max_secs;
5001 }
5002 else {
5003 absTime->tv_sec = secs;
5004 }
5005 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5006 }
5007 else {
5008 jlong secs = time / NANOSECS_PER_SEC;
5009 if (secs >= MAX_SECS) {
5010 absTime->tv_sec = max_secs;
5011 absTime->tv_nsec = 0;
5012 }
5013 else {
5014 absTime->tv_sec = now.tv_sec + secs;
5015 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5016 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5017 absTime->tv_nsec -= NANOSECS_PER_SEC;
5018 ++absTime->tv_sec; // note: this must be <= max_secs
5019 }
5020 }
5021 }
5022 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5023 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5024 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5025 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5026 }
5028 void Parker::park(bool isAbsolute, jlong time) {
5029 // Optional fast-path check:
5030 // Return immediately if a permit is available.
5031 if (_counter > 0) {
5032 _counter = 0 ;
5033 OrderAccess::fence();
5034 return ;
5035 }
5037 Thread* thread = Thread::current();
5038 assert(thread->is_Java_thread(), "Must be JavaThread");
5039 JavaThread *jt = (JavaThread *)thread;
5041 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5042 // Check interrupt before trying to wait
5043 if (Thread::is_interrupted(thread, false)) {
5044 return;
5045 }
5047 // Next, demultiplex/decode time arguments
5048 timespec absTime;
5049 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5050 return;
5051 }
5052 if (time > 0) {
5053 unpackTime(&absTime, isAbsolute, time);
5054 }
5057 // Enter safepoint region
5058 // Beware of deadlocks such as 6317397.
5059 // The per-thread Parker:: mutex is a classic leaf-lock.
5060 // In particular a thread must never block on the Threads_lock while
5061 // holding the Parker:: mutex. If safepoints are pending both the
5062 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5063 ThreadBlockInVM tbivm(jt);
5065 // Don't wait if cannot get lock since interference arises from
5066 // unblocking. Also. check interrupt before trying wait
5067 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5068 return;
5069 }
5071 int status ;
5072 if (_counter > 0) { // no wait needed
5073 _counter = 0;
5074 status = pthread_mutex_unlock(_mutex);
5075 assert (status == 0, "invariant") ;
5076 OrderAccess::fence();
5077 return;
5078 }
5080 #ifdef ASSERT
5081 // Don't catch signals while blocked; let the running threads have the signals.
5082 // (This allows a debugger to break into the running thread.)
5083 sigset_t oldsigs;
5084 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5085 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5086 #endif
5088 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5089 jt->set_suspend_equivalent();
5090 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5092 if (time == 0) {
5093 status = pthread_cond_wait (_cond, _mutex) ;
5094 } else {
5095 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5096 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5097 pthread_cond_destroy (_cond) ;
5098 pthread_cond_init (_cond, NULL);
5099 }
5100 }
5101 assert_status(status == 0 || status == EINTR ||
5102 status == ETIME || status == ETIMEDOUT,
5103 status, "cond_timedwait");
5105 #ifdef ASSERT
5106 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5107 #endif
5109 _counter = 0 ;
5110 status = pthread_mutex_unlock(_mutex) ;
5111 assert_status(status == 0, status, "invariant") ;
5112 // If externally suspended while waiting, re-suspend
5113 if (jt->handle_special_suspend_equivalent_condition()) {
5114 jt->java_suspend_self();
5115 }
5117 OrderAccess::fence();
5118 }
5120 void Parker::unpark() {
5121 int s, status ;
5122 status = pthread_mutex_lock(_mutex);
5123 assert (status == 0, "invariant") ;
5124 s = _counter;
5125 _counter = 1;
5126 if (s < 1) {
5127 if (WorkAroundNPTLTimedWaitHang) {
5128 status = pthread_cond_signal (_cond) ;
5129 assert (status == 0, "invariant") ;
5130 status = pthread_mutex_unlock(_mutex);
5131 assert (status == 0, "invariant") ;
5132 } else {
5133 status = pthread_mutex_unlock(_mutex);
5134 assert (status == 0, "invariant") ;
5135 status = pthread_cond_signal (_cond) ;
5136 assert (status == 0, "invariant") ;
5137 }
5138 } else {
5139 pthread_mutex_unlock(_mutex);
5140 assert (status == 0, "invariant") ;
5141 }
5142 }
5145 extern char** environ;
5147 #ifndef __NR_fork
5148 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5149 #endif
5151 #ifndef __NR_execve
5152 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5153 #endif
5155 // Run the specified command in a separate process. Return its exit value,
5156 // or -1 on failure (e.g. can't fork a new process).
5157 // Unlike system(), this function can be called from signal handler. It
5158 // doesn't block SIGINT et al.
5159 int os::fork_and_exec(char* cmd) {
5160 const char * argv[4] = {"sh", "-c", cmd, NULL};
5162 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5163 // pthread_atfork handlers and reset pthread library. All we need is a
5164 // separate process to execve. Make a direct syscall to fork process.
5165 // On IA64 there's no fork syscall, we have to use fork() and hope for
5166 // the best...
5167 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5168 IA64_ONLY(fork();)
5170 if (pid < 0) {
5171 // fork failed
5172 return -1;
5174 } else if (pid == 0) {
5175 // child process
5177 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5178 // first to kill every thread on the thread list. Because this list is
5179 // not reset by fork() (see notes above), execve() will instead kill
5180 // every thread in the parent process. We know this is the only thread
5181 // in the new process, so make a system call directly.
5182 // IA64 should use normal execve() from glibc to match the glibc fork()
5183 // above.
5184 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5185 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5187 // execve failed
5188 _exit(-1);
5190 } else {
5191 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5192 // care about the actual exit code, for now.
5194 int status;
5196 // Wait for the child process to exit. This returns immediately if
5197 // the child has already exited. */
5198 while (waitpid(pid, &status, 0) < 0) {
5199 switch (errno) {
5200 case ECHILD: return 0;
5201 case EINTR: break;
5202 default: return -1;
5203 }
5204 }
5206 if (WIFEXITED(status)) {
5207 // The child exited normally; get its exit code.
5208 return WEXITSTATUS(status);
5209 } else if (WIFSIGNALED(status)) {
5210 // The child exited because of a signal
5211 // The best value to return is 0x80 + signal number,
5212 // because that is what all Unix shells do, and because
5213 // it allows callers to distinguish between process exit and
5214 // process death by signal.
5215 return 0x80 + WTERMSIG(status);
5216 } else {
5217 // Unknown exit code; pass it through
5218 return status;
5219 }
5220 }
5221 }
5223 // is_headless_jre()
5224 //
5225 // Test for the existence of libmawt in motif21 or xawt directories
5226 // in order to report if we are running in a headless jre
5227 //
5228 bool os::is_headless_jre() {
5229 struct stat statbuf;
5230 char buf[MAXPATHLEN];
5231 char libmawtpath[MAXPATHLEN];
5232 const char *xawtstr = "/xawt/libmawt.so";
5233 const char *motifstr = "/motif21/libmawt.so";
5234 char *p;
5236 // Get path to libjvm.so
5237 os::jvm_path(buf, sizeof(buf));
5239 // Get rid of libjvm.so
5240 p = strrchr(buf, '/');
5241 if (p == NULL) return false;
5242 else *p = '\0';
5244 // Get rid of client or server
5245 p = strrchr(buf, '/');
5246 if (p == NULL) return false;
5247 else *p = '\0';
5249 // check xawt/libmawt.so
5250 strcpy(libmawtpath, buf);
5251 strcat(libmawtpath, xawtstr);
5252 if (::stat(libmawtpath, &statbuf) == 0) return false;
5254 // check motif21/libmawt.so
5255 strcpy(libmawtpath, buf);
5256 strcat(libmawtpath, motifstr);
5257 if (::stat(libmawtpath, &statbuf) == 0) return false;
5259 return true;
5260 }