annotate src/os/linux/vm/os_linux.cpp @ 998:b9408ac1b596

Added tag hs16-b13 for changeset 62926c7f67a3
author trims
date Thu, 11 Feb 2010 20:27:54 -0800
parents cf71f149d7ae
children
rev   line source
duke@0 1 /*
xdono@579 2 * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved.
duke@0 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@0 4 *
duke@0 5 * This code is free software; you can redistribute it and/or modify it
duke@0 6 * under the terms of the GNU General Public License version 2 only, as
duke@0 7 * published by the Free Software Foundation.
duke@0 8 *
duke@0 9 * This code is distributed in the hope that it will be useful, but WITHOUT
duke@0 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@0 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
duke@0 12 * version 2 for more details (a copy is included in the LICENSE file that
duke@0 13 * accompanied this code).
duke@0 14 *
duke@0 15 * You should have received a copy of the GNU General Public License version
duke@0 16 * 2 along with this work; if not, write to the Free Software Foundation,
duke@0 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@0 18 *
duke@0 19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
duke@0 20 * CA 95054 USA or visit www.sun.com if you need additional information or
duke@0 21 * have any questions.
duke@0 22 *
duke@0 23 */
duke@0 24
duke@0 25 // do not include precompiled header file
duke@0 26 # include "incls/_os_linux.cpp.incl"
duke@0 27
duke@0 28 // put OS-includes here
duke@0 29 # include <sys/types.h>
duke@0 30 # include <sys/mman.h>
duke@0 31 # include <pthread.h>
duke@0 32 # include <signal.h>
duke@0 33 # include <errno.h>
duke@0 34 # include <dlfcn.h>
duke@0 35 # include <stdio.h>
duke@0 36 # include <unistd.h>
duke@0 37 # include <sys/resource.h>
duke@0 38 # include <pthread.h>
duke@0 39 # include <sys/stat.h>
duke@0 40 # include <sys/time.h>
duke@0 41 # include <sys/times.h>
duke@0 42 # include <sys/utsname.h>
duke@0 43 # include <sys/socket.h>
duke@0 44 # include <sys/wait.h>
duke@0 45 # include <pwd.h>
duke@0 46 # include <poll.h>
duke@0 47 # include <semaphore.h>
duke@0 48 # include <fcntl.h>
duke@0 49 # include <string.h>
duke@0 50 # include <syscall.h>
duke@0 51 # include <sys/sysinfo.h>
duke@0 52 # include <gnu/libc-version.h>
duke@0 53 # include <sys/ipc.h>
duke@0 54 # include <sys/shm.h>
duke@0 55 # include <link.h>
duke@0 56
duke@0 57 #define MAX_PATH (2 * K)
duke@0 58
duke@0 59 // for timer info max values which include all bits
duke@0 60 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
duke@0 61 #define SEC_IN_NANOSECS 1000000000LL
duke@0 62
duke@0 63 ////////////////////////////////////////////////////////////////////////////////
duke@0 64 // global variables
duke@0 65 julong os::Linux::_physical_memory = 0;
duke@0 66
duke@0 67 address os::Linux::_initial_thread_stack_bottom = NULL;
duke@0 68 uintptr_t os::Linux::_initial_thread_stack_size = 0;
duke@0 69
duke@0 70 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
duke@0 71 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
duke@0 72 Mutex* os::Linux::_createThread_lock = NULL;
duke@0 73 pthread_t os::Linux::_main_thread;
duke@0 74 int os::Linux::_page_size = -1;
duke@0 75 bool os::Linux::_is_floating_stack = false;
duke@0 76 bool os::Linux::_is_NPTL = false;
duke@0 77 bool os::Linux::_supports_fast_thread_cpu_time = false;
xlu@196 78 const char * os::Linux::_glibc_version = NULL;
xlu@196 79 const char * os::Linux::_libpthread_version = NULL;
duke@0 80
duke@0 81 static jlong initial_time_count=0;
duke@0 82
duke@0 83 static int clock_tics_per_sec = 100;
duke@0 84
duke@0 85 // For diagnostics to print a message once. see run_periodic_checks
duke@0 86 static sigset_t check_signal_done;
duke@0 87 static bool check_signals = true;;
duke@0 88
duke@0 89 static pid_t _initial_pid = 0;
duke@0 90
duke@0 91 /* Signal number used to suspend/resume a thread */
duke@0 92
duke@0 93 /* do not use any signal number less than SIGSEGV, see 4355769 */
duke@0 94 static int SR_signum = SIGUSR2;
duke@0 95 sigset_t SR_sigset;
duke@0 96
kamg@241 97 /* Used to protect dlsym() calls */
kamg@241 98 static pthread_mutex_t dl_mutex;
kamg@241 99
duke@0 100 ////////////////////////////////////////////////////////////////////////////////
duke@0 101 // utility functions
duke@0 102
duke@0 103 static int SR_initialize();
duke@0 104 static int SR_finalize();
duke@0 105
duke@0 106 julong os::available_memory() {
duke@0 107 return Linux::available_memory();
duke@0 108 }
duke@0 109
duke@0 110 julong os::Linux::available_memory() {
duke@0 111 // values in struct sysinfo are "unsigned long"
duke@0 112 struct sysinfo si;
duke@0 113 sysinfo(&si);
duke@0 114
duke@0 115 return (julong)si.freeram * si.mem_unit;
duke@0 116 }
duke@0 117
duke@0 118 julong os::physical_memory() {
duke@0 119 return Linux::physical_memory();
duke@0 120 }
duke@0 121
phh@20 122 julong os::allocatable_physical_memory(julong size) {
phh@20 123 #ifdef _LP64
phh@20 124 return size;
phh@20 125 #else
phh@20 126 julong result = MIN2(size, (julong)3800*M);
phh@20 127 if (!is_allocatable(result)) {
phh@20 128 // See comments under solaris for alignment considerations
phh@20 129 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
phh@20 130 result = MIN2(size, reasonable_size);
phh@20 131 }
phh@20 132 return result;
phh@20 133 #endif // _LP64
phh@20 134 }
phh@20 135
duke@0 136 ////////////////////////////////////////////////////////////////////////////////
duke@0 137 // environment support
duke@0 138
duke@0 139 bool os::getenv(const char* name, char* buf, int len) {
duke@0 140 const char* val = ::getenv(name);
duke@0 141 if (val != NULL && strlen(val) < (size_t)len) {
duke@0 142 strcpy(buf, val);
duke@0 143 return true;
duke@0 144 }
duke@0 145 if (len > 0) buf[0] = 0; // return a null string
duke@0 146 return false;
duke@0 147 }
duke@0 148
duke@0 149
duke@0 150 // Return true if user is running as root.
duke@0 151
duke@0 152 bool os::have_special_privileges() {
duke@0 153 static bool init = false;
duke@0 154 static bool privileges = false;
duke@0 155 if (!init) {
duke@0 156 privileges = (getuid() != geteuid()) || (getgid() != getegid());
duke@0 157 init = true;
duke@0 158 }
duke@0 159 return privileges;
duke@0 160 }
duke@0 161
duke@0 162
duke@0 163 #ifndef SYS_gettid
duke@0 164 // i386: 224, ia64: 1105, amd64: 186, sparc 143
duke@0 165 #ifdef __ia64__
duke@0 166 #define SYS_gettid 1105
duke@0 167 #elif __i386__
duke@0 168 #define SYS_gettid 224
duke@0 169 #elif __amd64__
duke@0 170 #define SYS_gettid 186
duke@0 171 #elif __sparc__
duke@0 172 #define SYS_gettid 143
duke@0 173 #else
duke@0 174 #error define gettid for the arch
duke@0 175 #endif
duke@0 176 #endif
duke@0 177
duke@0 178 // Cpu architecture string
duke@0 179 #if defined(IA64)
duke@0 180 static char cpu_arch[] = "ia64";
duke@0 181 #elif defined(IA32)
duke@0 182 static char cpu_arch[] = "i386";
duke@0 183 #elif defined(AMD64)
duke@0 184 static char cpu_arch[] = "amd64";
duke@0 185 #elif defined(SPARC)
duke@0 186 # ifdef _LP64
duke@0 187 static char cpu_arch[] = "sparcv9";
duke@0 188 # else
duke@0 189 static char cpu_arch[] = "sparc";
duke@0 190 # endif
duke@0 191 #else
duke@0 192 #error Add appropriate cpu_arch setting
duke@0 193 #endif
duke@0 194
duke@0 195
duke@0 196 // pid_t gettid()
duke@0 197 //
duke@0 198 // Returns the kernel thread id of the currently running thread. Kernel
duke@0 199 // thread id is used to access /proc.
duke@0 200 //
duke@0 201 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
duke@0 202 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
duke@0 203 //
duke@0 204 pid_t os::Linux::gettid() {
duke@0 205 int rslt = syscall(SYS_gettid);
duke@0 206 if (rslt == -1) {
duke@0 207 // old kernel, no NPTL support
duke@0 208 return getpid();
duke@0 209 } else {
duke@0 210 return (pid_t)rslt;
duke@0 211 }
duke@0 212 }
duke@0 213
duke@0 214 // Most versions of linux have a bug where the number of processors are
duke@0 215 // determined by looking at the /proc file system. In a chroot environment,
duke@0 216 // the system call returns 1. This causes the VM to act as if it is
duke@0 217 // a single processor and elide locking (see is_MP() call).
duke@0 218 static bool unsafe_chroot_detected = false;
xlu@196 219 static const char *unstable_chroot_error = "/proc file system not found.\n"
xlu@196 220 "Java may be unstable running multithreaded in a chroot "
xlu@196 221 "environment on Linux when /proc filesystem is not mounted.";
duke@0 222
duke@0 223 void os::Linux::initialize_system_info() {
duke@0 224 _processor_count = sysconf(_SC_NPROCESSORS_CONF);
duke@0 225 if (_processor_count == 1) {
duke@0 226 pid_t pid = os::Linux::gettid();
duke@0 227 char fname[32];
duke@0 228 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
duke@0 229 FILE *fp = fopen(fname, "r");
duke@0 230 if (fp == NULL) {
duke@0 231 unsafe_chroot_detected = true;
duke@0 232 } else {
duke@0 233 fclose(fp);
duke@0 234 }
duke@0 235 }
duke@0 236 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
duke@0 237 assert(_processor_count > 0, "linux error");
duke@0 238 }
duke@0 239
duke@0 240 void os::init_system_properties_values() {
duke@0 241 // char arch[12];
duke@0 242 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
duke@0 243
duke@0 244 // The next steps are taken in the product version:
duke@0 245 //
duke@0 246 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
duke@0 247 // This library should be located at:
duke@0 248 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
duke@0 249 //
duke@0 250 // If "/jre/lib/" appears at the right place in the path, then we
duke@0 251 // assume libjvm[_g].so is installed in a JDK and we use this path.
duke@0 252 //
duke@0 253 // Otherwise exit with message: "Could not create the Java virtual machine."
duke@0 254 //
duke@0 255 // The following extra steps are taken in the debugging version:
duke@0 256 //
duke@0 257 // If "/jre/lib/" does NOT appear at the right place in the path
duke@0 258 // instead of exit check for $JAVA_HOME environment variable.
duke@0 259 //
duke@0 260 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
duke@0 261 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
duke@0 262 // it looks like libjvm[_g].so is installed there
duke@0 263 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
duke@0 264 //
duke@0 265 // Otherwise exit.
duke@0 266 //
duke@0 267 // Important note: if the location of libjvm.so changes this
duke@0 268 // code needs to be changed accordingly.
duke@0 269
duke@0 270 // The next few definitions allow the code to be verbatim:
duke@0 271 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
duke@0 272 #define getenv(n) ::getenv(n)
duke@0 273
duke@0 274 /*
duke@0 275 * See ld(1):
duke@0 276 * The linker uses the following search paths to locate required
duke@0 277 * shared libraries:
duke@0 278 * 1: ...
duke@0 279 * ...
duke@0 280 * 7: The default directories, normally /lib and /usr/lib.
duke@0 281 */
kvn@508 282 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
kvn@508 283 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
kvn@508 284 #else
duke@0 285 #define DEFAULT_LIBPATH "/lib:/usr/lib"
kvn@508 286 #endif
duke@0 287
duke@0 288 #define EXTENSIONS_DIR "/lib/ext"
duke@0 289 #define ENDORSED_DIR "/lib/endorsed"
duke@0 290 #define REG_DIR "/usr/java/packages"
duke@0 291
duke@0 292 {
duke@0 293 /* sysclasspath, java_home, dll_dir */
duke@0 294 {
duke@0 295 char *home_path;
duke@0 296 char *dll_path;
duke@0 297 char *pslash;
duke@0 298 char buf[MAXPATHLEN];
duke@0 299 os::jvm_path(buf, sizeof(buf));
duke@0 300
duke@0 301 // Found the full path to libjvm.so.
duke@0 302 // Now cut the path to <java_home>/jre if we can.
duke@0 303 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
duke@0 304 pslash = strrchr(buf, '/');
duke@0 305 if (pslash != NULL)
duke@0 306 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
duke@0 307 dll_path = malloc(strlen(buf) + 1);
duke@0 308 if (dll_path == NULL)
duke@0 309 return;
duke@0 310 strcpy(dll_path, buf);
duke@0 311 Arguments::set_dll_dir(dll_path);
duke@0 312
duke@0 313 if (pslash != NULL) {
duke@0 314 pslash = strrchr(buf, '/');
duke@0 315 if (pslash != NULL) {
duke@0 316 *pslash = '\0'; /* get rid of /<arch> */
duke@0 317 pslash = strrchr(buf, '/');
duke@0 318 if (pslash != NULL)
duke@0 319 *pslash = '\0'; /* get rid of /lib */
duke@0 320 }
duke@0 321 }
duke@0 322
duke@0 323 home_path = malloc(strlen(buf) + 1);
duke@0 324 if (home_path == NULL)
duke@0 325 return;
duke@0 326 strcpy(home_path, buf);
duke@0 327 Arguments::set_java_home(home_path);
duke@0 328
duke@0 329 if (!set_boot_path('/', ':'))
duke@0 330 return;
duke@0 331 }
duke@0 332
duke@0 333 /*
duke@0 334 * Where to look for native libraries
duke@0 335 *
duke@0 336 * Note: Due to a legacy implementation, most of the library path
duke@0 337 * is set in the launcher. This was to accomodate linking restrictions
duke@0 338 * on legacy Linux implementations (which are no longer supported).
duke@0 339 * Eventually, all the library path setting will be done here.
duke@0 340 *
duke@0 341 * However, to prevent the proliferation of improperly built native
duke@0 342 * libraries, the new path component /usr/java/packages is added here.
duke@0 343 * Eventually, all the library path setting will be done here.
duke@0 344 */
duke@0 345 {
duke@0 346 char *ld_library_path;
duke@0 347
duke@0 348 /*
duke@0 349 * Construct the invariant part of ld_library_path. Note that the
duke@0 350 * space for the colon and the trailing null are provided by the
duke@0 351 * nulls included by the sizeof operator (so actually we allocate
duke@0 352 * a byte more than necessary).
duke@0 353 */
duke@0 354 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
duke@0 355 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
duke@0 356 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
duke@0 357
duke@0 358 /*
duke@0 359 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
duke@0 360 * should always exist (until the legacy problem cited above is
duke@0 361 * addressed).
duke@0 362 */
duke@0 363 char *v = getenv("LD_LIBRARY_PATH");
duke@0 364 if (v != NULL) {
duke@0 365 char *t = ld_library_path;
duke@0 366 /* That's +1 for the colon and +1 for the trailing '\0' */
duke@0 367 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
duke@0 368 sprintf(ld_library_path, "%s:%s", v, t);
duke@0 369 }
duke@0 370 Arguments::set_library_path(ld_library_path);
duke@0 371 }
duke@0 372
duke@0 373 /*
duke@0 374 * Extensions directories.
duke@0 375 *
duke@0 376 * Note that the space for the colon and the trailing null are provided
duke@0 377 * by the nulls included by the sizeof operator (so actually one byte more
duke@0 378 * than necessary is allocated).
duke@0 379 */
duke@0 380 {
duke@0 381 char *buf = malloc(strlen(Arguments::get_java_home()) +
duke@0 382 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
duke@0 383 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
duke@0 384 Arguments::get_java_home());
duke@0 385 Arguments::set_ext_dirs(buf);
duke@0 386 }
duke@0 387
duke@0 388 /* Endorsed standards default directory. */
duke@0 389 {
duke@0 390 char * buf;
duke@0 391 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
duke@0 392 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
duke@0 393 Arguments::set_endorsed_dirs(buf);
duke@0 394 }
duke@0 395 }
duke@0 396
duke@0 397 #undef malloc
duke@0 398 #undef getenv
duke@0 399 #undef EXTENSIONS_DIR
duke@0 400 #undef ENDORSED_DIR
duke@0 401
duke@0 402 // Done
duke@0 403 return;
duke@0 404 }
duke@0 405
duke@0 406 ////////////////////////////////////////////////////////////////////////////////
duke@0 407 // breakpoint support
duke@0 408
duke@0 409 void os::breakpoint() {
duke@0 410 BREAKPOINT;
duke@0 411 }
duke@0 412
duke@0 413 extern "C" void breakpoint() {
duke@0 414 // use debugger to set breakpoint here
duke@0 415 }
duke@0 416
duke@0 417 ////////////////////////////////////////////////////////////////////////////////
duke@0 418 // signal support
duke@0 419
duke@0 420 debug_only(static bool signal_sets_initialized = false);
duke@0 421 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
duke@0 422
duke@0 423 bool os::Linux::is_sig_ignored(int sig) {
duke@0 424 struct sigaction oact;
duke@0 425 sigaction(sig, (struct sigaction*)NULL, &oact);
duke@0 426 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
duke@0 427 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
duke@0 428 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
duke@0 429 return true;
duke@0 430 else
duke@0 431 return false;
duke@0 432 }
duke@0 433
duke@0 434 void os::Linux::signal_sets_init() {
duke@0 435 // Should also have an assertion stating we are still single-threaded.
duke@0 436 assert(!signal_sets_initialized, "Already initialized");
duke@0 437 // Fill in signals that are necessarily unblocked for all threads in
duke@0 438 // the VM. Currently, we unblock the following signals:
duke@0 439 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
duke@0 440 // by -Xrs (=ReduceSignalUsage));
duke@0 441 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
duke@0 442 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
duke@0 443 // the dispositions or masks wrt these signals.
duke@0 444 // Programs embedding the VM that want to use the above signals for their
duke@0 445 // own purposes must, at this time, use the "-Xrs" option to prevent
duke@0 446 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
duke@0 447 // (See bug 4345157, and other related bugs).
duke@0 448 // In reality, though, unblocking these signals is really a nop, since
duke@0 449 // these signals are not blocked by default.
duke@0 450 sigemptyset(&unblocked_sigs);
duke@0 451 sigemptyset(&allowdebug_blocked_sigs);
duke@0 452 sigaddset(&unblocked_sigs, SIGILL);
duke@0 453 sigaddset(&unblocked_sigs, SIGSEGV);
duke@0 454 sigaddset(&unblocked_sigs, SIGBUS);
duke@0 455 sigaddset(&unblocked_sigs, SIGFPE);
duke@0 456 sigaddset(&unblocked_sigs, SR_signum);
duke@0 457
duke@0 458 if (!ReduceSignalUsage) {
duke@0 459 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
duke@0 460 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
duke@0 461 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
duke@0 462 }
duke@0 463 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
duke@0 464 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
duke@0 465 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
duke@0 466 }
duke@0 467 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
duke@0 468 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
duke@0 469 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
duke@0 470 }
duke@0 471 }
duke@0 472 // Fill in signals that are blocked by all but the VM thread.
duke@0 473 sigemptyset(&vm_sigs);
duke@0 474 if (!ReduceSignalUsage)
duke@0 475 sigaddset(&vm_sigs, BREAK_SIGNAL);
duke@0 476 debug_only(signal_sets_initialized = true);
duke@0 477
duke@0 478 }
duke@0 479
duke@0 480 // These are signals that are unblocked while a thread is running Java.
duke@0 481 // (For some reason, they get blocked by default.)
duke@0 482 sigset_t* os::Linux::unblocked_signals() {
duke@0 483 assert(signal_sets_initialized, "Not initialized");
duke@0 484 return &unblocked_sigs;
duke@0 485 }
duke@0 486
duke@0 487 // These are the signals that are blocked while a (non-VM) thread is
duke@0 488 // running Java. Only the VM thread handles these signals.
duke@0 489 sigset_t* os::Linux::vm_signals() {
duke@0 490 assert(signal_sets_initialized, "Not initialized");
duke@0 491 return &vm_sigs;
duke@0 492 }
duke@0 493
duke@0 494 // These are signals that are blocked during cond_wait to allow debugger in
duke@0 495 sigset_t* os::Linux::allowdebug_blocked_signals() {
duke@0 496 assert(signal_sets_initialized, "Not initialized");
duke@0 497 return &allowdebug_blocked_sigs;
duke@0 498 }
duke@0 499
duke@0 500 void os::Linux::hotspot_sigmask(Thread* thread) {
duke@0 501
duke@0 502 //Save caller's signal mask before setting VM signal mask
duke@0 503 sigset_t caller_sigmask;
duke@0 504 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
duke@0 505
duke@0 506 OSThread* osthread = thread->osthread();
duke@0 507 osthread->set_caller_sigmask(caller_sigmask);
duke@0 508
duke@0 509 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
duke@0 510
duke@0 511 if (!ReduceSignalUsage) {
duke@0 512 if (thread->is_VM_thread()) {
duke@0 513 // Only the VM thread handles BREAK_SIGNAL ...
duke@0 514 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
duke@0 515 } else {
duke@0 516 // ... all other threads block BREAK_SIGNAL
duke@0 517 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
duke@0 518 }
duke@0 519 }
duke@0 520 }
duke@0 521
duke@0 522 //////////////////////////////////////////////////////////////////////////////
duke@0 523 // detecting pthread library
duke@0 524
duke@0 525 void os::Linux::libpthread_init() {
duke@0 526 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
duke@0 527 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
duke@0 528 // generic name for earlier versions.
duke@0 529 // Define macros here so we can build HotSpot on old systems.
duke@0 530 # ifndef _CS_GNU_LIBC_VERSION
duke@0 531 # define _CS_GNU_LIBC_VERSION 2
duke@0 532 # endif
duke@0 533 # ifndef _CS_GNU_LIBPTHREAD_VERSION
duke@0 534 # define _CS_GNU_LIBPTHREAD_VERSION 3
duke@0 535 # endif
duke@0 536
duke@0 537 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
duke@0 538 if (n > 0) {
duke@0 539 char *str = (char *)malloc(n);
duke@0 540 confstr(_CS_GNU_LIBC_VERSION, str, n);
duke@0 541 os::Linux::set_glibc_version(str);
duke@0 542 } else {
duke@0 543 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
duke@0 544 static char _gnu_libc_version[32];
duke@0 545 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
duke@0 546 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
duke@0 547 os::Linux::set_glibc_version(_gnu_libc_version);
duke@0 548 }
duke@0 549
duke@0 550 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
duke@0 551 if (n > 0) {
duke@0 552 char *str = (char *)malloc(n);
duke@0 553 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
duke@0 554 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
duke@0 555 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
xlu@196 556 // is the case. LinuxThreads has a hard limit on max number of threads.
xlu@196 557 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
xlu@196 558 // On the other hand, NPTL does not have such a limit, sysconf()
xlu@196 559 // will return -1 and errno is not changed. Check if it is really NPTL.
duke@0 560 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
xlu@196 561 strstr(str, "NPTL") &&
xlu@196 562 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
xlu@196 563 free(str);
xlu@196 564 os::Linux::set_libpthread_version("linuxthreads");
xlu@196 565 } else {
xlu@196 566 os::Linux::set_libpthread_version(str);
duke@0 567 }
duke@0 568 } else {
xlu@196 569 // glibc before 2.3.2 only has LinuxThreads.
xlu@196 570 os::Linux::set_libpthread_version("linuxthreads");
duke@0 571 }
duke@0 572
duke@0 573 if (strstr(libpthread_version(), "NPTL")) {
duke@0 574 os::Linux::set_is_NPTL();
duke@0 575 } else {
duke@0 576 os::Linux::set_is_LinuxThreads();
duke@0 577 }
duke@0 578
duke@0 579 // LinuxThreads have two flavors: floating-stack mode, which allows variable
duke@0 580 // stack size; and fixed-stack mode. NPTL is always floating-stack.
duke@0 581 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
duke@0 582 os::Linux::set_is_floating_stack();
duke@0 583 }
duke@0 584 }
duke@0 585
duke@0 586 /////////////////////////////////////////////////////////////////////////////
duke@0 587 // thread stack
duke@0 588
duke@0 589 // Force Linux kernel to expand current thread stack. If "bottom" is close
duke@0 590 // to the stack guard, caller should block all signals.
duke@0 591 //
duke@0 592 // MAP_GROWSDOWN:
duke@0 593 // A special mmap() flag that is used to implement thread stacks. It tells
duke@0 594 // kernel that the memory region should extend downwards when needed. This
duke@0 595 // allows early versions of LinuxThreads to only mmap the first few pages
duke@0 596 // when creating a new thread. Linux kernel will automatically expand thread
duke@0 597 // stack as needed (on page faults).
duke@0 598 //
duke@0 599 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
duke@0 600 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
duke@0 601 // region, it's hard to tell if the fault is due to a legitimate stack
duke@0 602 // access or because of reading/writing non-exist memory (e.g. buffer
duke@0 603 // overrun). As a rule, if the fault happens below current stack pointer,
duke@0 604 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
duke@0 605 // application (see Linux kernel fault.c).
duke@0 606 //
duke@0 607 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
duke@0 608 // stack overflow detection.
duke@0 609 //
duke@0 610 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
duke@0 611 // not use this flag. However, the stack of initial thread is not created
duke@0 612 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
duke@0 613 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
duke@0 614 // and then attach the thread to JVM.
duke@0 615 //
duke@0 616 // To get around the problem and allow stack banging on Linux, we need to
duke@0 617 // manually expand thread stack after receiving the SIGSEGV.
duke@0 618 //
duke@0 619 // There are two ways to expand thread stack to address "bottom", we used
duke@0 620 // both of them in JVM before 1.5:
duke@0 621 // 1. adjust stack pointer first so that it is below "bottom", and then
duke@0 622 // touch "bottom"
duke@0 623 // 2. mmap() the page in question
duke@0 624 //
duke@0 625 // Now alternate signal stack is gone, it's harder to use 2. For instance,
duke@0 626 // if current sp is already near the lower end of page 101, and we need to
duke@0 627 // call mmap() to map page 100, it is possible that part of the mmap() frame
duke@0 628 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
duke@0 629 // That will destroy the mmap() frame and cause VM to crash.
duke@0 630 //
duke@0 631 // The following code works by adjusting sp first, then accessing the "bottom"
duke@0 632 // page to force a page fault. Linux kernel will then automatically expand the
duke@0 633 // stack mapping.
duke@0 634 //
duke@0 635 // _expand_stack_to() assumes its frame size is less than page size, which
duke@0 636 // should always be true if the function is not inlined.
duke@0 637
duke@0 638 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
duke@0 639 #define NOINLINE
duke@0 640 #else
duke@0 641 #define NOINLINE __attribute__ ((noinline))
duke@0 642 #endif
duke@0 643
duke@0 644 static void _expand_stack_to(address bottom) NOINLINE;
duke@0 645
duke@0 646 static void _expand_stack_to(address bottom) {
duke@0 647 address sp;
duke@0 648 size_t size;
duke@0 649 volatile char *p;
duke@0 650
duke@0 651 // Adjust bottom to point to the largest address within the same page, it
duke@0 652 // gives us a one-page buffer if alloca() allocates slightly more memory.
duke@0 653 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
duke@0 654 bottom += os::Linux::page_size() - 1;
duke@0 655
duke@0 656 // sp might be slightly above current stack pointer; if that's the case, we
duke@0 657 // will alloca() a little more space than necessary, which is OK. Don't use
duke@0 658 // os::current_stack_pointer(), as its result can be slightly below current
duke@0 659 // stack pointer, causing us to not alloca enough to reach "bottom".
duke@0 660 sp = (address)&sp;
duke@0 661
duke@0 662 if (sp > bottom) {
duke@0 663 size = sp - bottom;
duke@0 664 p = (volatile char *)alloca(size);
duke@0 665 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
duke@0 666 p[0] = '\0';
duke@0 667 }
duke@0 668 }
duke@0 669
duke@0 670 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
duke@0 671 assert(t!=NULL, "just checking");
duke@0 672 assert(t->osthread()->expanding_stack(), "expand should be set");
duke@0 673 assert(t->stack_base() != NULL, "stack_base was not initialized");
duke@0 674
duke@0 675 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
duke@0 676 sigset_t mask_all, old_sigset;
duke@0 677 sigfillset(&mask_all);
duke@0 678 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
duke@0 679 _expand_stack_to(addr);
duke@0 680 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
duke@0 681 return true;
duke@0 682 }
duke@0 683 return false;
duke@0 684 }
duke@0 685
duke@0 686 //////////////////////////////////////////////////////////////////////////////
duke@0 687 // create new thread
duke@0 688
duke@0 689 static address highest_vm_reserved_address();
duke@0 690
duke@0 691 // check if it's safe to start a new thread
duke@0 692 static bool _thread_safety_check(Thread* thread) {
duke@0 693 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
duke@0 694 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
duke@0 695 // Heap is mmap'ed at lower end of memory space. Thread stacks are
duke@0 696 // allocated (MAP_FIXED) from high address space. Every thread stack
duke@0 697 // occupies a fixed size slot (usually 2Mbytes, but user can change
duke@0 698 // it to other values if they rebuild LinuxThreads).
duke@0 699 //
duke@0 700 // Problem with MAP_FIXED is that mmap() can still succeed even part of
duke@0 701 // the memory region has already been mmap'ed. That means if we have too
duke@0 702 // many threads and/or very large heap, eventually thread stack will
duke@0 703 // collide with heap.
duke@0 704 //
duke@0 705 // Here we try to prevent heap/stack collision by comparing current
duke@0 706 // stack bottom with the highest address that has been mmap'ed by JVM
duke@0 707 // plus a safety margin for memory maps created by native code.
duke@0 708 //
duke@0 709 // This feature can be disabled by setting ThreadSafetyMargin to 0
duke@0 710 //
duke@0 711 if (ThreadSafetyMargin > 0) {
duke@0 712 address stack_bottom = os::current_stack_base() - os::current_stack_size();
duke@0 713
duke@0 714 // not safe if our stack extends below the safety margin
duke@0 715 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
duke@0 716 } else {
duke@0 717 return true;
duke@0 718 }
duke@0 719 } else {
duke@0 720 // Floating stack LinuxThreads or NPTL:
duke@0 721 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
duke@0 722 // there's not enough space left, pthread_create() will fail. If we come
duke@0 723 // here, that means enough space has been reserved for stack.
duke@0 724 return true;
duke@0 725 }
duke@0 726 }
duke@0 727
duke@0 728 // Thread start routine for all newly created threads
duke@0 729 static void *java_start(Thread *thread) {
duke@0 730 // Try to randomize the cache line index of hot stack frames.
duke@0 731 // This helps when threads of the same stack traces evict each other's
duke@0 732 // cache lines. The threads can be either from the same JVM instance, or
duke@0 733 // from different JVM instances. The benefit is especially true for
duke@0 734 // processors with hyperthreading technology.
duke@0 735 static int counter = 0;
duke@0 736 int pid = os::current_process_id();
duke@0 737 alloca(((pid ^ counter++) & 7) * 128);
duke@0 738
duke@0 739 ThreadLocalStorage::set_thread(thread);
duke@0 740
duke@0 741 OSThread* osthread = thread->osthread();
duke@0 742 Monitor* sync = osthread->startThread_lock();
duke@0 743
duke@0 744 // non floating stack LinuxThreads needs extra check, see above
duke@0 745 if (!_thread_safety_check(thread)) {
duke@0 746 // notify parent thread
duke@0 747 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
duke@0 748 osthread->set_state(ZOMBIE);
duke@0 749 sync->notify_all();
duke@0 750 return NULL;
duke@0 751 }
duke@0 752
duke@0 753 // thread_id is kernel thread id (similar to Solaris LWP id)
duke@0 754 osthread->set_thread_id(os::Linux::gettid());
duke@0 755
duke@0 756 if (UseNUMA) {
duke@0 757 int lgrp_id = os::numa_get_group_id();
duke@0 758 if (lgrp_id != -1) {
duke@0 759 thread->set_lgrp_id(lgrp_id);
duke@0 760 }
duke@0 761 }
duke@0 762 // initialize signal mask for this thread
duke@0 763 os::Linux::hotspot_sigmask(thread);
duke@0 764
duke@0 765 // initialize floating point control register
duke@0 766 os::Linux::init_thread_fpu_state();
duke@0 767
duke@0 768 // handshaking with parent thread
duke@0 769 {
duke@0 770 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
duke@0 771
duke@0 772 // notify parent thread
duke@0 773 osthread->set_state(INITIALIZED);
duke@0 774 sync->notify_all();
duke@0 775
duke@0 776 // wait until os::start_thread()
duke@0 777 while (osthread->get_state() == INITIALIZED) {
duke@0 778 sync->wait(Mutex::_no_safepoint_check_flag);
duke@0 779 }
duke@0 780 }
duke@0 781
duke@0 782 // call one more level start routine
duke@0 783 thread->run();
duke@0 784
duke@0 785 return 0;
duke@0 786 }
duke@0 787
duke@0 788 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
duke@0 789 assert(thread->osthread() == NULL, "caller responsible");
duke@0 790
duke@0 791 // Allocate the OSThread object
duke@0 792 OSThread* osthread = new OSThread(NULL, NULL);
duke@0 793 if (osthread == NULL) {
duke@0 794 return false;
duke@0 795 }
duke@0 796
duke@0 797 // set the correct thread state
duke@0 798 osthread->set_thread_type(thr_type);
duke@0 799
duke@0 800 // Initial state is ALLOCATED but not INITIALIZED
duke@0 801 osthread->set_state(ALLOCATED);
duke@0 802
duke@0 803 thread->set_osthread(osthread);
duke@0 804
duke@0 805 // init thread attributes
duke@0 806 pthread_attr_t attr;
duke@0 807 pthread_attr_init(&attr);
duke@0 808 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
duke@0 809
duke@0 810 // stack size
duke@0 811 if (os::Linux::supports_variable_stack_size()) {
duke@0 812 // calculate stack size if it's not specified by caller
duke@0 813 if (stack_size == 0) {
duke@0 814 stack_size = os::Linux::default_stack_size(thr_type);
duke@0 815
duke@0 816 switch (thr_type) {
duke@0 817 case os::java_thread:
duke@0 818 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
duke@0 819 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
duke@0 820 break;
duke@0 821 case os::compiler_thread:
duke@0 822 if (CompilerThreadStackSize > 0) {
duke@0 823 stack_size = (size_t)(CompilerThreadStackSize * K);
duke@0 824 break;
duke@0 825 } // else fall through:
duke@0 826 // use VMThreadStackSize if CompilerThreadStackSize is not defined
duke@0 827 case os::vm_thread:
duke@0 828 case os::pgc_thread:
duke@0 829 case os::cgc_thread:
duke@0 830 case os::watcher_thread:
duke@0 831 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
duke@0 832 break;
duke@0 833 }
duke@0 834 }
duke@0 835
duke@0 836 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
duke@0 837 pthread_attr_setstacksize(&attr, stack_size);
duke@0 838 } else {
duke@0 839 // let pthread_create() pick the default value.
duke@0 840 }
duke@0 841
duke@0 842 // glibc guard page
duke@0 843 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
duke@0 844
duke@0 845 ThreadState state;
duke@0 846
duke@0 847 {
duke@0 848 // Serialize thread creation if we are running with fixed stack LinuxThreads
duke@0 849 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
duke@0 850 if (lock) {
duke@0 851 os::Linux::createThread_lock()->lock_without_safepoint_check();
duke@0 852 }
duke@0 853
duke@0 854 pthread_t tid;
duke@0 855 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
duke@0 856
duke@0 857 pthread_attr_destroy(&attr);
duke@0 858
duke@0 859 if (ret != 0) {
duke@0 860 if (PrintMiscellaneous && (Verbose || WizardMode)) {
duke@0 861 perror("pthread_create()");
duke@0 862 }
duke@0 863 // Need to clean up stuff we've allocated so far
duke@0 864 thread->set_osthread(NULL);
duke@0 865 delete osthread;
duke@0 866 if (lock) os::Linux::createThread_lock()->unlock();
duke@0 867 return false;
duke@0 868 }
duke@0 869
duke@0 870 // Store pthread info into the OSThread
duke@0 871 osthread->set_pthread_id(tid);
duke@0 872
duke@0 873 // Wait until child thread is either initialized or aborted
duke@0 874 {
duke@0 875 Monitor* sync_with_child = osthread->startThread_lock();
duke@0 876 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
duke@0 877 while ((state = osthread->get_state()) == ALLOCATED) {
duke@0 878 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
duke@0 879 }
duke@0 880 }
duke@0 881
duke@0 882 if (lock) {
duke@0 883 os::Linux::createThread_lock()->unlock();
duke@0 884 }
duke@0 885 }
duke@0 886
duke@0 887 // Aborted due to thread limit being reached
duke@0 888 if (state == ZOMBIE) {
duke@0 889 thread->set_osthread(NULL);
duke@0 890 delete osthread;
duke@0 891 return false;
duke@0 892 }
duke@0 893
duke@0 894 // The thread is returned suspended (in state INITIALIZED),
duke@0 895 // and is started higher up in the call chain
duke@0 896 assert(state == INITIALIZED, "race condition");
duke@0 897 return true;
duke@0 898 }
duke@0 899
duke@0 900 /////////////////////////////////////////////////////////////////////////////
duke@0 901 // attach existing thread
duke@0 902
duke@0 903 // bootstrap the main thread
duke@0 904 bool os::create_main_thread(JavaThread* thread) {
duke@0 905 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
duke@0 906 return create_attached_thread(thread);
duke@0 907 }
duke@0 908
duke@0 909 bool os::create_attached_thread(JavaThread* thread) {
duke@0 910 #ifdef ASSERT
duke@0 911 thread->verify_not_published();
duke@0 912 #endif
duke@0 913
duke@0 914 // Allocate the OSThread object
duke@0 915 OSThread* osthread = new OSThread(NULL, NULL);
duke@0 916
duke@0 917 if (osthread == NULL) {
duke@0 918 return false;
duke@0 919 }
duke@0 920
duke@0 921 // Store pthread info into the OSThread
duke@0 922 osthread->set_thread_id(os::Linux::gettid());
duke@0 923 osthread->set_pthread_id(::pthread_self());
duke@0 924
duke@0 925 // initialize floating point control register
duke@0 926 os::Linux::init_thread_fpu_state();
duke@0 927
duke@0 928 // Initial thread state is RUNNABLE
duke@0 929 osthread->set_state(RUNNABLE);
duke@0 930
duke@0 931 thread->set_osthread(osthread);
duke@0 932
duke@0 933 if (UseNUMA) {
duke@0 934 int lgrp_id = os::numa_get_group_id();
duke@0 935 if (lgrp_id != -1) {
duke@0 936 thread->set_lgrp_id(lgrp_id);
duke@0 937 }
duke@0 938 }
duke@0 939
duke@0 940 if (os::Linux::is_initial_thread()) {
duke@0 941 // If current thread is initial thread, its stack is mapped on demand,
duke@0 942 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
duke@0 943 // the entire stack region to avoid SEGV in stack banging.
duke@0 944 // It is also useful to get around the heap-stack-gap problem on SuSE
duke@0 945 // kernel (see 4821821 for details). We first expand stack to the top
duke@0 946 // of yellow zone, then enable stack yellow zone (order is significant,
duke@0 947 // enabling yellow zone first will crash JVM on SuSE Linux), so there
duke@0 948 // is no gap between the last two virtual memory regions.
duke@0 949
duke@0 950 JavaThread *jt = (JavaThread *)thread;
duke@0 951 address addr = jt->stack_yellow_zone_base();
duke@0 952 assert(addr != NULL, "initialization problem?");
duke@0 953 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
duke@0 954
duke@0 955 osthread->set_expanding_stack();
duke@0 956 os::Linux::manually_expand_stack(jt, addr);
duke@0 957 osthread->clear_expanding_stack();
duke@0 958 }
duke@0 959
duke@0 960 // initialize signal mask for this thread
duke@0 961 // and save the caller's signal mask
duke@0 962 os::Linux::hotspot_sigmask(thread);
duke@0 963
duke@0 964 return true;
duke@0 965 }
duke@0 966
duke@0 967 void os::pd_start_thread(Thread* thread) {
duke@0 968 OSThread * osthread = thread->osthread();
duke@0 969 assert(osthread->get_state() != INITIALIZED, "just checking");
duke@0 970 Monitor* sync_with_child = osthread->startThread_lock();
duke@0 971 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
duke@0 972 sync_with_child->notify();
duke@0 973 }
duke@0 974
duke@0 975 // Free Linux resources related to the OSThread
duke@0 976 void os::free_thread(OSThread* osthread) {
duke@0 977 assert(osthread != NULL, "osthread not set");
duke@0 978
duke@0 979 if (Thread::current()->osthread() == osthread) {
duke@0 980 // Restore caller's signal mask
duke@0 981 sigset_t sigmask = osthread->caller_sigmask();
duke@0 982 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
duke@0 983 }
duke@0 984
duke@0 985 delete osthread;
duke@0 986 }
duke@0 987
duke@0 988 //////////////////////////////////////////////////////////////////////////////
duke@0 989 // thread local storage
duke@0 990
duke@0 991 int os::allocate_thread_local_storage() {
duke@0 992 pthread_key_t key;
duke@0 993 int rslt = pthread_key_create(&key, NULL);
duke@0 994 assert(rslt == 0, "cannot allocate thread local storage");
duke@0 995 return (int)key;
duke@0 996 }
duke@0 997
duke@0 998 // Note: This is currently not used by VM, as we don't destroy TLS key
duke@0 999 // on VM exit.
duke@0 1000 void os::free_thread_local_storage(int index) {
duke@0 1001 int rslt = pthread_key_delete((pthread_key_t)index);
duke@0 1002 assert(rslt == 0, "invalid index");
duke@0 1003 }
duke@0 1004
duke@0 1005 void os::thread_local_storage_at_put(int index, void* value) {
duke@0 1006 int rslt = pthread_setspecific((pthread_key_t)index, value);
duke@0 1007 assert(rslt == 0, "pthread_setspecific failed");
duke@0 1008 }
duke@0 1009
duke@0 1010 extern "C" Thread* get_thread() {
duke@0 1011 return ThreadLocalStorage::thread();
duke@0 1012 }
duke@0 1013
duke@0 1014 //////////////////////////////////////////////////////////////////////////////
duke@0 1015 // initial thread
duke@0 1016
duke@0 1017 // Check if current thread is the initial thread, similar to Solaris thr_main.
duke@0 1018 bool os::Linux::is_initial_thread(void) {
duke@0 1019 char dummy;
duke@0 1020 // If called before init complete, thread stack bottom will be null.
duke@0 1021 // Can be called if fatal error occurs before initialization.
duke@0 1022 if (initial_thread_stack_bottom() == NULL) return false;
duke@0 1023 assert(initial_thread_stack_bottom() != NULL &&
duke@0 1024 initial_thread_stack_size() != 0,
duke@0 1025 "os::init did not locate initial thread's stack region");
duke@0 1026 if ((address)&dummy >= initial_thread_stack_bottom() &&
duke@0 1027 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
duke@0 1028 return true;
duke@0 1029 else return false;
duke@0 1030 }
duke@0 1031
duke@0 1032 // Find the virtual memory area that contains addr
duke@0 1033 static bool find_vma(address addr, address* vma_low, address* vma_high) {
duke@0 1034 FILE *fp = fopen("/proc/self/maps", "r");
duke@0 1035 if (fp) {
duke@0 1036 address low, high;
duke@0 1037 while (!feof(fp)) {
duke@0 1038 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
duke@0 1039 if (low <= addr && addr < high) {
duke@0 1040 if (vma_low) *vma_low = low;
duke@0 1041 if (vma_high) *vma_high = high;
duke@0 1042 fclose (fp);
duke@0 1043 return true;
duke@0 1044 }
duke@0 1045 }
duke@0 1046 for (;;) {
duke@0 1047 int ch = fgetc(fp);
duke@0 1048 if (ch == EOF || ch == (int)'\n') break;
duke@0 1049 }
duke@0 1050 }
duke@0 1051 fclose(fp);
duke@0 1052 }
duke@0 1053 return false;
duke@0 1054 }
duke@0 1055
duke@0 1056 // Locate initial thread stack. This special handling of initial thread stack
duke@0 1057 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
duke@0 1058 // bogus value for initial thread.
duke@0 1059 void os::Linux::capture_initial_stack(size_t max_size) {
duke@0 1060 // stack size is the easy part, get it from RLIMIT_STACK
duke@0 1061 size_t stack_size;
duke@0 1062 struct rlimit rlim;
duke@0 1063 getrlimit(RLIMIT_STACK, &rlim);
duke@0 1064 stack_size = rlim.rlim_cur;
duke@0 1065
duke@0 1066 // 6308388: a bug in ld.so will relocate its own .data section to the
duke@0 1067 // lower end of primordial stack; reduce ulimit -s value a little bit
duke@0 1068 // so we won't install guard page on ld.so's data section.
duke@0 1069 stack_size -= 2 * page_size();
duke@0 1070
duke@0 1071 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
duke@0 1072 // 7.1, in both cases we will get 2G in return value.
duke@0 1073 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
duke@0 1074 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
duke@0 1075 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
duke@0 1076 // in case other parts in glibc still assumes 2M max stack size.
duke@0 1077 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
duke@0 1078 #ifndef IA64
duke@0 1079 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
duke@0 1080 #else
duke@0 1081 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
duke@0 1082 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
duke@0 1083 #endif
duke@0 1084
duke@0 1085 // Try to figure out where the stack base (top) is. This is harder.
duke@0 1086 //
duke@0 1087 // When an application is started, glibc saves the initial stack pointer in
duke@0 1088 // a global variable "__libc_stack_end", which is then used by system
duke@0 1089 // libraries. __libc_stack_end should be pretty close to stack top. The
duke@0 1090 // variable is available since the very early days. However, because it is
duke@0 1091 // a private interface, it could disappear in the future.
duke@0 1092 //
duke@0 1093 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
duke@0 1094 // to __libc_stack_end, it is very close to stack top, but isn't the real
duke@0 1095 // stack top. Note that /proc may not exist if VM is running as a chroot
duke@0 1096 // program, so reading /proc/<pid>/stat could fail. Also the contents of
duke@0 1097 // /proc/<pid>/stat could change in the future (though unlikely).
duke@0 1098 //
duke@0 1099 // We try __libc_stack_end first. If that doesn't work, look for
duke@0 1100 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
duke@0 1101 // as a hint, which should work well in most cases.
duke@0 1102
duke@0 1103 uintptr_t stack_start;
duke@0 1104
duke@0 1105 // try __libc_stack_end first
duke@0 1106 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
duke@0 1107 if (p && *p) {
duke@0 1108 stack_start = *p;
duke@0 1109 } else {
duke@0 1110 // see if we can get the start_stack field from /proc/self/stat
duke@0 1111 FILE *fp;
duke@0 1112 int pid;
duke@0 1113 char state;
duke@0 1114 int ppid;
duke@0 1115 int pgrp;
duke@0 1116 int session;
duke@0 1117 int nr;
duke@0 1118 int tpgrp;
duke@0 1119 unsigned long flags;
duke@0 1120 unsigned long minflt;
duke@0 1121 unsigned long cminflt;
duke@0 1122 unsigned long majflt;
duke@0 1123 unsigned long cmajflt;
duke@0 1124 unsigned long utime;
duke@0 1125 unsigned long stime;
duke@0 1126 long cutime;
duke@0 1127 long cstime;
duke@0 1128 long prio;
duke@0 1129 long nice;
duke@0 1130 long junk;
duke@0 1131 long it_real;
duke@0 1132 uintptr_t start;
duke@0 1133 uintptr_t vsize;
duke@0 1134 uintptr_t rss;
duke@0 1135 unsigned long rsslim;
duke@0 1136 uintptr_t scodes;
duke@0 1137 uintptr_t ecode;
duke@0 1138 int i;
duke@0 1139
duke@0 1140 // Figure what the primordial thread stack base is. Code is inspired
duke@0 1141 // by email from Hans Boehm. /proc/self/stat begins with current pid,
duke@0 1142 // followed by command name surrounded by parentheses, state, etc.
duke@0 1143 char stat[2048];
duke@0 1144 int statlen;
duke@0 1145
duke@0 1146 fp = fopen("/proc/self/stat", "r");
duke@0 1147 if (fp) {
duke@0 1148 statlen = fread(stat, 1, 2047, fp);
duke@0 1149 stat[statlen] = '\0';
duke@0 1150 fclose(fp);
duke@0 1151
duke@0 1152 // Skip pid and the command string. Note that we could be dealing with
duke@0 1153 // weird command names, e.g. user could decide to rename java launcher
duke@0 1154 // to "java 1.4.2 :)", then the stat file would look like
duke@0 1155 // 1234 (java 1.4.2 :)) R ... ...
duke@0 1156 // We don't really need to know the command string, just find the last
duke@0 1157 // occurrence of ")" and then start parsing from there. See bug 4726580.
duke@0 1158 char * s = strrchr(stat, ')');
duke@0 1159
duke@0 1160 i = 0;
duke@0 1161 if (s) {
duke@0 1162 // Skip blank chars
duke@0 1163 do s++; while (isspace(*s));
duke@0 1164
duke@0 1165 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
duke@0 1166 /* 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 */
xlu@512 1167 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld "
xlu@512 1168 UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT
xlu@512 1169 " %lu "
xlu@512 1170 UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT,
duke@0 1171 &state, /* 3 %c */
duke@0 1172 &ppid, /* 4 %d */
duke@0 1173 &pgrp, /* 5 %d */
duke@0 1174 &session, /* 6 %d */
duke@0 1175 &nr, /* 7 %d */
duke@0 1176 &tpgrp, /* 8 %d */
duke@0 1177 &flags, /* 9 %lu */
duke@0 1178 &minflt, /* 10 %lu */
duke@0 1179 &cminflt, /* 11 %lu */
duke@0 1180 &majflt, /* 12 %lu */
duke@0 1181 &cmajflt, /* 13 %lu */
duke@0 1182 &utime, /* 14 %lu */
duke@0 1183 &stime, /* 15 %lu */
duke@0 1184 &cutime, /* 16 %ld */
duke@0 1185 &cstime, /* 17 %ld */
duke@0 1186 &prio, /* 18 %ld */
duke@0 1187 &nice, /* 19 %ld */
duke@0 1188 &junk, /* 20 %ld */
duke@0 1189 &it_real, /* 21 %ld */
xlu@512 1190 &start, /* 22 UINTX_FORMAT */
xlu@512 1191 &vsize, /* 23 UINTX_FORMAT */
xlu@512 1192 &rss, /* 24 UINTX_FORMAT */
duke@0 1193 &rsslim, /* 25 %lu */
xlu@512 1194 &scodes, /* 26 UINTX_FORMAT */
xlu@512 1195 &ecode, /* 27 UINTX_FORMAT */
xlu@512 1196 &stack_start); /* 28 UINTX_FORMAT */
duke@0 1197 }
duke@0 1198
duke@0 1199 if (i != 28 - 2) {
duke@0 1200 assert(false, "Bad conversion from /proc/self/stat");
duke@0 1201 // product mode - assume we are the initial thread, good luck in the
duke@0 1202 // embedded case.
duke@0 1203 warning("Can't detect initial thread stack location - bad conversion");
duke@0 1204 stack_start = (uintptr_t) &rlim;
duke@0 1205 }
duke@0 1206 } else {
duke@0 1207 // For some reason we can't open /proc/self/stat (for example, running on
duke@0 1208 // FreeBSD with a Linux emulator, or inside chroot), this should work for
duke@0 1209 // most cases, so don't abort:
duke@0 1210 warning("Can't detect initial thread stack location - no /proc/self/stat");
duke@0 1211 stack_start = (uintptr_t) &rlim;
duke@0 1212 }
duke@0 1213 }
duke@0 1214
duke@0 1215 // Now we have a pointer (stack_start) very close to the stack top, the
duke@0 1216 // next thing to do is to figure out the exact location of stack top. We
duke@0 1217 // can find out the virtual memory area that contains stack_start by
duke@0 1218 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
duke@0 1219 // and its upper limit is the real stack top. (again, this would fail if
duke@0 1220 // running inside chroot, because /proc may not exist.)
duke@0 1221
duke@0 1222 uintptr_t stack_top;
duke@0 1223 address low, high;
duke@0 1224 if (find_vma((address)stack_start, &low, &high)) {
duke@0 1225 // success, "high" is the true stack top. (ignore "low", because initial
duke@0 1226 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
duke@0 1227 stack_top = (uintptr_t)high;
duke@0 1228 } else {
duke@0 1229 // failed, likely because /proc/self/maps does not exist
duke@0 1230 warning("Can't detect initial thread stack location - find_vma failed");
duke@0 1231 // best effort: stack_start is normally within a few pages below the real
duke@0 1232 // stack top, use it as stack top, and reduce stack size so we won't put
duke@0 1233 // guard page outside stack.
duke@0 1234 stack_top = stack_start;
duke@0 1235 stack_size -= 16 * page_size();
duke@0 1236 }
duke@0 1237
duke@0 1238 // stack_top could be partially down the page so align it
duke@0 1239 stack_top = align_size_up(stack_top, page_size());
duke@0 1240
duke@0 1241 if (max_size && stack_size > max_size) {
duke@0 1242 _initial_thread_stack_size = max_size;
duke@0 1243 } else {
duke@0 1244 _initial_thread_stack_size = stack_size;
duke@0 1245 }
duke@0 1246
duke@0 1247 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
duke@0 1248 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
duke@0 1249 }
duke@0 1250
duke@0 1251 ////////////////////////////////////////////////////////////////////////////////
duke@0 1252 // time support
duke@0 1253
duke@0 1254 // Time since start-up in seconds to a fine granularity.
duke@0 1255 // Used by VMSelfDestructTimer and the MemProfiler.
duke@0 1256 double os::elapsedTime() {
duke@0 1257
duke@0 1258 return (double)(os::elapsed_counter()) * 0.000001;
duke@0 1259 }
duke@0 1260
duke@0 1261 jlong os::elapsed_counter() {
duke@0 1262 timeval time;
duke@0 1263 int status = gettimeofday(&time, NULL);
duke@0 1264 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
duke@0 1265 }
duke@0 1266
duke@0 1267 jlong os::elapsed_frequency() {
duke@0 1268 return (1000 * 1000);
duke@0 1269 }
duke@0 1270
ysr@339 1271 // For now, we say that linux does not support vtime. I have no idea
ysr@339 1272 // whether it can actually be made to (DLD, 9/13/05).
ysr@339 1273
ysr@339 1274 bool os::supports_vtime() { return false; }
ysr@339 1275 bool os::enable_vtime() { return false; }
ysr@339 1276 bool os::vtime_enabled() { return false; }
ysr@339 1277 double os::elapsedVTime() {
ysr@339 1278 // better than nothing, but not much
ysr@339 1279 return elapsedTime();
ysr@339 1280 }
ysr@339 1281
sbohne@61 1282 jlong os::javaTimeMillis() {
duke@0 1283 timeval time;
duke@0 1284 int status = gettimeofday(&time, NULL);
duke@0 1285 assert(status != -1, "linux error");
duke@0 1286 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
duke@0 1287 }
duke@0 1288
duke@0 1289 #ifndef CLOCK_MONOTONIC
duke@0 1290 #define CLOCK_MONOTONIC (1)
duke@0 1291 #endif
duke@0 1292
duke@0 1293 void os::Linux::clock_init() {
duke@0 1294 // we do dlopen's in this particular order due to bug in linux
duke@0 1295 // dynamical loader (see 6348968) leading to crash on exit
duke@0 1296 void* handle = dlopen("librt.so.1", RTLD_LAZY);
duke@0 1297 if (handle == NULL) {
duke@0 1298 handle = dlopen("librt.so", RTLD_LAZY);
duke@0 1299 }
duke@0 1300
duke@0 1301 if (handle) {
duke@0 1302 int (*clock_getres_func)(clockid_t, struct timespec*) =
duke@0 1303 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
duke@0 1304 int (*clock_gettime_func)(clockid_t, struct timespec*) =
duke@0 1305 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
duke@0 1306 if (clock_getres_func && clock_gettime_func) {
duke@0 1307 // See if monotonic clock is supported by the kernel. Note that some
duke@0 1308 // early implementations simply return kernel jiffies (updated every
duke@0 1309 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
duke@0 1310 // for nano time (though the monotonic property is still nice to have).
duke@0 1311 // It's fixed in newer kernels, however clock_getres() still returns
duke@0 1312 // 1/HZ. We check if clock_getres() works, but will ignore its reported
duke@0 1313 // resolution for now. Hopefully as people move to new kernels, this
duke@0 1314 // won't be a problem.
duke@0 1315 struct timespec res;
duke@0 1316 struct timespec tp;
duke@0 1317 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
duke@0 1318 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
duke@0 1319 // yes, monotonic clock is supported
duke@0 1320 _clock_gettime = clock_gettime_func;
duke@0 1321 } else {
duke@0 1322 // close librt if there is no monotonic clock
duke@0 1323 dlclose(handle);
duke@0 1324 }
duke@0 1325 }
duke@0 1326 }
duke@0 1327 }
duke@0 1328
duke@0 1329 #ifndef SYS_clock_getres
duke@0 1330
duke@0 1331 #if defined(IA32) || defined(AMD64)
duke@0 1332 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
duke@0 1333 #else
duke@0 1334 #error Value of SYS_clock_getres not known on this platform
duke@0 1335 #endif
duke@0 1336
duke@0 1337 #endif
duke@0 1338
duke@0 1339 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
duke@0 1340
duke@0 1341 void os::Linux::fast_thread_clock_init() {
duke@0 1342 if (!UseLinuxPosixThreadCPUClocks) {
duke@0 1343 return;
duke@0 1344 }
duke@0 1345 clockid_t clockid;
duke@0 1346 struct timespec tp;
duke@0 1347 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
duke@0 1348 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
duke@0 1349
duke@0 1350 // Switch to using fast clocks for thread cpu time if
duke@0 1351 // the sys_clock_getres() returns 0 error code.
duke@0 1352 // Note, that some kernels may support the current thread
duke@0 1353 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
duke@0 1354 // returned by the pthread_getcpuclockid().
duke@0 1355 // If the fast Posix clocks are supported then the sys_clock_getres()
duke@0 1356 // must return at least tp.tv_sec == 0 which means a resolution
duke@0 1357 // better than 1 sec. This is extra check for reliability.
duke@0 1358
duke@0 1359 if(pthread_getcpuclockid_func &&
duke@0 1360 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
duke@0 1361 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
duke@0 1362
duke@0 1363 _supports_fast_thread_cpu_time = true;
duke@0 1364 _pthread_getcpuclockid = pthread_getcpuclockid_func;
duke@0 1365 }
duke@0 1366 }
duke@0 1367
duke@0 1368 jlong os::javaTimeNanos() {
duke@0 1369 if (Linux::supports_monotonic_clock()) {
duke@0 1370 struct timespec tp;
duke@0 1371 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
duke@0 1372 assert(status == 0, "gettime error");
duke@0 1373 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
duke@0 1374 return result;
duke@0 1375 } else {
duke@0 1376 timeval time;
duke@0 1377 int status = gettimeofday(&time, NULL);
duke@0 1378 assert(status != -1, "linux error");
duke@0 1379 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
duke@0 1380 return 1000 * usecs;
duke@0 1381 }
duke@0 1382 }
duke@0 1383
duke@0 1384 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
duke@0 1385 if (Linux::supports_monotonic_clock()) {
duke@0 1386 info_ptr->max_value = ALL_64_BITS;
duke@0 1387
duke@0 1388 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
duke@0 1389 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
duke@0 1390 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
duke@0 1391 } else {
duke@0 1392 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
duke@0 1393 info_ptr->max_value = ALL_64_BITS;
duke@0 1394
duke@0 1395 // gettimeofday is a real time clock so it skips
duke@0 1396 info_ptr->may_skip_backward = true;
duke@0 1397 info_ptr->may_skip_forward = true;
duke@0 1398 }
duke@0 1399
duke@0 1400 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
duke@0 1401 }
duke@0 1402
duke@0 1403 // Return the real, user, and system times in seconds from an
duke@0 1404 // arbitrary fixed point in the past.
duke@0 1405 bool os::getTimesSecs(double* process_real_time,
duke@0 1406 double* process_user_time,
duke@0 1407 double* process_system_time) {
duke@0 1408 struct tms ticks;
duke@0 1409 clock_t real_ticks = times(&ticks);
duke@0 1410
duke@0 1411 if (real_ticks == (clock_t) (-1)) {
duke@0 1412 return false;
duke@0 1413 } else {
duke@0 1414 double ticks_per_second = (double) clock_tics_per_sec;
duke@0 1415 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
duke@0 1416 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
duke@0 1417 *process_real_time = ((double) real_ticks) / ticks_per_second;
duke@0 1418
duke@0 1419 return true;
duke@0 1420 }
duke@0 1421 }
duke@0 1422
duke@0 1423
duke@0 1424 char * os::local_time_string(char *buf, size_t buflen) {
duke@0 1425 struct tm t;
duke@0 1426 time_t long_time;
duke@0 1427 time(&long_time);
duke@0 1428 localtime_r(&long_time, &t);
duke@0 1429 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
duke@0 1430 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
duke@0 1431 t.tm_hour, t.tm_min, t.tm_sec);
duke@0 1432 return buf;
duke@0 1433 }
duke@0 1434
ysr@547 1435 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
ysr@547 1436 return localtime_r(clock, res);
ysr@547 1437 }
ysr@547 1438
duke@0 1439 ////////////////////////////////////////////////////////////////////////////////
duke@0 1440 // runtime exit support
duke@0 1441
duke@0 1442 // Note: os::shutdown() might be called very early during initialization, or
duke@0 1443 // called from signal handler. Before adding something to os::shutdown(), make
duke@0 1444 // sure it is async-safe and can handle partially initialized VM.
duke@0 1445 void os::shutdown() {
duke@0 1446
duke@0 1447 // allow PerfMemory to attempt cleanup of any persistent resources
duke@0 1448 perfMemory_exit();
duke@0 1449
duke@0 1450 // needs to remove object in file system
duke@0 1451 AttachListener::abort();
duke@0 1452
duke@0 1453 // flush buffered output, finish log files
duke@0 1454 ostream_abort();
duke@0 1455
duke@0 1456 // Check for abort hook
duke@0 1457 abort_hook_t abort_hook = Arguments::abort_hook();
duke@0 1458 if (abort_hook != NULL) {
duke@0 1459 abort_hook();
duke@0 1460 }
duke@0 1461
duke@0 1462 }
duke@0 1463
duke@0 1464 // Note: os::abort() might be called very early during initialization, or
duke@0 1465 // called from signal handler. Before adding something to os::abort(), make
duke@0 1466 // sure it is async-safe and can handle partially initialized VM.
duke@0 1467 void os::abort(bool dump_core) {
duke@0 1468 os::shutdown();
duke@0 1469 if (dump_core) {
duke@0 1470 #ifndef PRODUCT
duke@0 1471 fdStream out(defaultStream::output_fd());
duke@0 1472 out.print_raw("Current thread is ");
duke@0 1473 char buf[16];
duke@0 1474 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
duke@0 1475 out.print_raw_cr(buf);
duke@0 1476 out.print_raw_cr("Dumping core ...");
duke@0 1477 #endif
duke@0 1478 ::abort(); // dump core
duke@0 1479 }
duke@0 1480
duke@0 1481 ::exit(1);
duke@0 1482 }
duke@0 1483
duke@0 1484 // Die immediately, no exit hook, no abort hook, no cleanup.
duke@0 1485 void os::die() {
duke@0 1486 // _exit() on LinuxThreads only kills current thread
duke@0 1487 ::abort();
duke@0 1488 }
duke@0 1489
duke@0 1490 // unused on linux for now.
duke@0 1491 void os::set_error_file(const char *logfile) {}
duke@0 1492
duke@0 1493 intx os::current_thread_id() { return (intx)pthread_self(); }
duke@0 1494 int os::current_process_id() {
duke@0 1495
duke@0 1496 // Under the old linux thread library, linux gives each thread
duke@0 1497 // its own process id. Because of this each thread will return
duke@0 1498 // a different pid if this method were to return the result
duke@0 1499 // of getpid(2). Linux provides no api that returns the pid
duke@0 1500 // of the launcher thread for the vm. This implementation
duke@0 1501 // returns a unique pid, the pid of the launcher thread
duke@0 1502 // that starts the vm 'process'.
duke@0 1503
duke@0 1504 // Under the NPTL, getpid() returns the same pid as the
duke@0 1505 // launcher thread rather than a unique pid per thread.
duke@0 1506 // Use gettid() if you want the old pre NPTL behaviour.
duke@0 1507
duke@0 1508 // if you are looking for the result of a call to getpid() that
duke@0 1509 // returns a unique pid for the calling thread, then look at the
duke@0 1510 // OSThread::thread_id() method in osThread_linux.hpp file
duke@0 1511
duke@0 1512 return (int)(_initial_pid ? _initial_pid : getpid());
duke@0 1513 }
duke@0 1514
duke@0 1515 // DLL functions
duke@0 1516
duke@0 1517 const char* os::dll_file_extension() { return ".so"; }
duke@0 1518
duke@0 1519 const char* os::get_temp_directory() { return "/tmp/"; }
duke@0 1520
phh@691 1521 static bool file_exists(const char* filename) {
phh@691 1522 struct stat statbuf;
phh@691 1523 if (filename == NULL || strlen(filename) == 0) {
phh@691 1524 return false;
phh@691 1525 }
phh@691 1526 return os::stat(filename, &statbuf) == 0;
phh@691 1527 }
phh@691 1528
phh@691 1529 void os::dll_build_name(char* buffer, size_t buflen,
phh@691 1530 const char* pname, const char* fname) {
phh@691 1531 // Copied from libhpi
kamg@241 1532 const size_t pnamelen = pname ? strlen(pname) : 0;
kamg@241 1533
phh@691 1534 // Quietly truncate on buffer overflow. Should be an error.
kamg@241 1535 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
kamg@241 1536 *buffer = '\0';
kamg@241 1537 return;
kamg@241 1538 }
kamg@241 1539
kamg@241 1540 if (pnamelen == 0) {
phh@691 1541 snprintf(buffer, buflen, "lib%s.so", fname);
phh@691 1542 } else if (strchr(pname, *os::path_separator()) != NULL) {
phh@691 1543 int n;
phh@691 1544 char** pelements = split_path(pname, &n);
phh@691 1545 for (int i = 0 ; i < n ; i++) {
phh@691 1546 // Really shouldn't be NULL, but check can't hurt
phh@691 1547 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
phh@691 1548 continue; // skip the empty path values
phh@691 1549 }
phh@691 1550 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
phh@691 1551 if (file_exists(buffer)) {
phh@691 1552 break;
phh@691 1553 }
phh@691 1554 }
phh@691 1555 // release the storage
phh@691 1556 for (int i = 0 ; i < n ; i++) {
phh@691 1557 if (pelements[i] != NULL) {
phh@691 1558 FREE_C_HEAP_ARRAY(char, pelements[i]);
phh@691 1559 }
phh@691 1560 }
phh@691 1561 if (pelements != NULL) {
phh@691 1562 FREE_C_HEAP_ARRAY(char*, pelements);
phh@691 1563 }
kamg@241 1564 } else {
phh@691 1565 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
kamg@241 1566 }
kamg@241 1567 }
kamg@241 1568
duke@0 1569 const char* os::get_current_directory(char *buf, int buflen) {
duke@0 1570 return getcwd(buf, buflen);
duke@0 1571 }
duke@0 1572
duke@0 1573 // check if addr is inside libjvm[_g].so
duke@0 1574 bool os::address_is_in_vm(address addr) {
duke@0 1575 static address libjvm_base_addr;
duke@0 1576 Dl_info dlinfo;
duke@0 1577
duke@0 1578 if (libjvm_base_addr == NULL) {
duke@0 1579 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
duke@0 1580 libjvm_base_addr = (address)dlinfo.dli_fbase;
duke@0 1581 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
duke@0 1582 }
duke@0 1583
duke@0 1584 if (dladdr((void *)addr, &dlinfo)) {
duke@0 1585 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
duke@0 1586 }
duke@0 1587
duke@0 1588 return false;
duke@0 1589 }
duke@0 1590
duke@0 1591 bool os::dll_address_to_function_name(address addr, char *buf,
duke@0 1592 int buflen, int *offset) {
duke@0 1593 Dl_info dlinfo;
duke@0 1594
duke@0 1595 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
duke@0 1596 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
duke@0 1597 if (offset) *offset = addr - (address)dlinfo.dli_saddr;
duke@0 1598 return true;
duke@0 1599 } else {
duke@0 1600 if (buf) buf[0] = '\0';
duke@0 1601 if (offset) *offset = -1;
duke@0 1602 return false;
duke@0 1603 }
duke@0 1604 }
duke@0 1605
duke@0 1606 struct _address_to_library_name {
duke@0 1607 address addr; // input : memory address
duke@0 1608 size_t buflen; // size of fname
duke@0 1609 char* fname; // output: library name
duke@0 1610 address base; // library base addr
duke@0 1611 };
duke@0 1612
duke@0 1613 static int address_to_library_name_callback(struct dl_phdr_info *info,
duke@0 1614 size_t size, void *data) {
duke@0 1615 int i;
duke@0 1616 bool found = false;
duke@0 1617 address libbase = NULL;
duke@0 1618 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
duke@0 1619
duke@0 1620 // iterate through all loadable segments
duke@0 1621 for (i = 0; i < info->dlpi_phnum; i++) {
duke@0 1622 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
duke@0 1623 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
duke@0 1624 // base address of a library is the lowest address of its loaded
duke@0 1625 // segments.
duke@0 1626 if (libbase == NULL || libbase > segbase) {
duke@0 1627 libbase = segbase;
duke@0 1628 }
duke@0 1629 // see if 'addr' is within current segment
duke@0 1630 if (segbase <= d->addr &&
duke@0 1631 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
duke@0 1632 found = true;
duke@0 1633 }
duke@0 1634 }
duke@0 1635 }
duke@0 1636
duke@0 1637 // dlpi_name is NULL or empty if the ELF file is executable, return 0
duke@0 1638 // so dll_address_to_library_name() can fall through to use dladdr() which
duke@0 1639 // can figure out executable name from argv[0].
duke@0 1640 if (found && info->dlpi_name && info->dlpi_name[0]) {
duke@0 1641 d->base = libbase;
duke@0 1642 if (d->fname) {
duke@0 1643 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
duke@0 1644 }
duke@0 1645 return 1;
duke@0 1646 }
duke@0 1647 return 0;
duke@0 1648 }
duke@0 1649
duke@0 1650 bool os::dll_address_to_library_name(address addr, char* buf,
duke@0 1651 int buflen, int* offset) {
duke@0 1652 Dl_info dlinfo;
duke@0 1653 struct _address_to_library_name data;
duke@0 1654
duke@0 1655 // There is a bug in old glibc dladdr() implementation that it could resolve
duke@0 1656 // to wrong library name if the .so file has a base address != NULL. Here
duke@0 1657 // we iterate through the program headers of all loaded libraries to find
duke@0 1658 // out which library 'addr' really belongs to. This workaround can be
duke@0 1659 // removed once the minimum requirement for glibc is moved to 2.3.x.
duke@0 1660 data.addr = addr;
duke@0 1661 data.fname = buf;
duke@0 1662 data.buflen = buflen;
duke@0 1663 data.base = NULL;
duke@0 1664 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
duke@0 1665
duke@0 1666 if (rslt) {
duke@0 1667 // buf already contains library name
duke@0 1668 if (offset) *offset = addr - data.base;
duke@0 1669 return true;
duke@0 1670 } else if (dladdr((void*)addr, &dlinfo)){
duke@0 1671 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
duke@0 1672 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
duke@0 1673 return true;
duke@0 1674 } else {
duke@0 1675 if (buf) buf[0] = '\0';
duke@0 1676 if (offset) *offset = -1;
duke@0 1677 return false;
duke@0 1678 }
duke@0 1679 }
duke@0 1680
duke@0 1681 // Loads .dll/.so and
duke@0 1682 // in case of error it checks if .dll/.so was built for the
duke@0 1683 // same architecture as Hotspot is running on
duke@0 1684
duke@0 1685 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
duke@0 1686 {
duke@0 1687 void * result= ::dlopen(filename, RTLD_LAZY);
duke@0 1688 if (result != NULL) {
duke@0 1689 // Successful loading
duke@0 1690 return result;
duke@0 1691 }
duke@0 1692
duke@0 1693 Elf32_Ehdr elf_head;
duke@0 1694
duke@0 1695 // Read system error message into ebuf
duke@0 1696 // It may or may not be overwritten below
duke@0 1697 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
duke@0 1698 ebuf[ebuflen-1]='\0';
duke@0 1699 int diag_msg_max_length=ebuflen-strlen(ebuf);
duke@0 1700 char* diag_msg_buf=ebuf+strlen(ebuf);
duke@0 1701
duke@0 1702 if (diag_msg_max_length==0) {
duke@0 1703 // No more space in ebuf for additional diagnostics message
duke@0 1704 return NULL;
duke@0 1705 }
duke@0 1706
duke@0 1707
duke@0 1708 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
duke@0 1709
duke@0 1710 if (file_descriptor < 0) {
duke@0 1711 // Can't open library, report dlerror() message
duke@0 1712 return NULL;
duke@0 1713 }
duke@0 1714
duke@0 1715 bool failed_to_read_elf_head=
duke@0 1716 (sizeof(elf_head)!=
duke@0 1717 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
duke@0 1718
duke@0 1719 ::close(file_descriptor);
duke@0 1720 if (failed_to_read_elf_head) {
duke@0 1721 // file i/o error - report dlerror() msg
duke@0 1722 return NULL;
duke@0 1723 }
duke@0 1724
duke@0 1725 typedef struct {
duke@0 1726 Elf32_Half code; // Actual value as defined in elf.h
duke@0 1727 Elf32_Half compat_class; // Compatibility of archs at VM's sense
duke@0 1728 char elf_class; // 32 or 64 bit
duke@0 1729 char endianess; // MSB or LSB
duke@0 1730 char* name; // String representation
duke@0 1731 } arch_t;
duke@0 1732
duke@0 1733 #ifndef EM_486
duke@0 1734 #define EM_486 6 /* Intel 80486 */
duke@0 1735 #endif
duke@0 1736
duke@0 1737 static const arch_t arch_array[]={
duke@0 1738 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
duke@0 1739 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
duke@0 1740 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
duke@0 1741 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
duke@0 1742 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
duke@0 1743 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
duke@0 1744 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
duke@0 1745 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
duke@0 1746 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}
duke@0 1747 };
duke@0 1748
duke@0 1749 #if (defined IA32)
duke@0 1750 static Elf32_Half running_arch_code=EM_386;
duke@0 1751 #elif (defined AMD64)
duke@0 1752 static Elf32_Half running_arch_code=EM_X86_64;
duke@0 1753 #elif (defined IA64)
duke@0 1754 static Elf32_Half running_arch_code=EM_IA_64;
duke@0 1755 #elif (defined __sparc) && (defined _LP64)
duke@0 1756 static Elf32_Half running_arch_code=EM_SPARCV9;
duke@0 1757 #elif (defined __sparc) && (!defined _LP64)
duke@0 1758 static Elf32_Half running_arch_code=EM_SPARC;
duke@0 1759 #elif (defined __powerpc64__)
duke@0 1760 static Elf32_Half running_arch_code=EM_PPC64;
duke@0 1761 #elif (defined __powerpc__)
duke@0 1762 static Elf32_Half running_arch_code=EM_PPC;
duke@0 1763 #else
duke@0 1764 #error Method os::dll_load requires that one of following is defined:\
duke@0 1765 IA32, AMD64, IA64, __sparc, __powerpc__
duke@0 1766 #endif
duke@0 1767
duke@0 1768 // Identify compatability class for VM's architecture and library's architecture
duke@0 1769 // Obtain string descriptions for architectures
duke@0 1770
duke@0 1771 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
duke@0 1772 int running_arch_index=-1;
duke@0 1773
duke@0 1774 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
duke@0 1775 if (running_arch_code == arch_array[i].code) {
duke@0 1776 running_arch_index = i;
duke@0 1777 }
duke@0 1778 if (lib_arch.code == arch_array[i].code) {
duke@0 1779 lib_arch.compat_class = arch_array[i].compat_class;
duke@0 1780 lib_arch.name = arch_array[i].name;
duke@0 1781 }
duke@0 1782 }
duke@0 1783
duke@0 1784 assert(running_arch_index != -1,
duke@0 1785 "Didn't find running architecture code (running_arch_code) in arch_array");
duke@0 1786 if (running_arch_index == -1) {
duke@0 1787 // Even though running architecture detection failed
duke@0 1788 // we may still continue with reporting dlerror() message
duke@0 1789 return NULL;
duke@0 1790 }
duke@0 1791
duke@0 1792 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
duke@0 1793 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
duke@0 1794 return NULL;
duke@0 1795 }
duke@0 1796
duke@0 1797 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
duke@0 1798 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
duke@0 1799 return NULL;
duke@0 1800 }
duke@0 1801
duke@0 1802 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
duke@0 1803 if ( lib_arch.name!=NULL ) {
duke@0 1804 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
duke@0 1805 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
duke@0 1806 lib_arch.name, arch_array[running_arch_index].name);
duke@0 1807 } else {
duke@0 1808 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
duke@0 1809 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
duke@0 1810 lib_arch.code,
duke@0 1811 arch_array[running_arch_index].name);
duke@0 1812 }
duke@0 1813 }
duke@0 1814
duke@0 1815 return NULL;
duke@0 1816 }
duke@0 1817
kamg@241 1818 /*
kamg@241 1819 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
kamg@241 1820 * chances are you might want to run the generated bits against glibc-2.0
kamg@241 1821 * libdl.so, so always use locking for any version of glibc.
kamg@241 1822 */
kamg@241 1823 void* os::dll_lookup(void* handle, const char* name) {
kamg@241 1824 pthread_mutex_lock(&dl_mutex);
kamg@241 1825 void* res = dlsym(handle, name);
kamg@241 1826 pthread_mutex_unlock(&dl_mutex);
kamg@241 1827 return res;
kamg@241 1828 }
duke@0 1829
duke@0 1830
duke@0 1831 bool _print_ascii_file(const char* filename, outputStream* st) {
duke@0 1832 int fd = open(filename, O_RDONLY);
duke@0 1833 if (fd == -1) {
duke@0 1834 return false;
duke@0 1835 }
duke@0 1836
duke@0 1837 char buf[32];
duke@0 1838 int bytes;
duke@0 1839 while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
duke@0 1840 st->print_raw(buf, bytes);
duke@0 1841 }
duke@0 1842
duke@0 1843 close(fd);
duke@0 1844
duke@0 1845 return true;
duke@0 1846 }
duke@0 1847
duke@0 1848 void os::print_dll_info(outputStream *st) {
duke@0 1849 st->print_cr("Dynamic libraries:");
duke@0 1850
duke@0 1851 char fname[32];
duke@0 1852 pid_t pid = os::Linux::gettid();
duke@0 1853
duke@0 1854 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
duke@0 1855
duke@0 1856 if (!_print_ascii_file(fname, st)) {
duke@0 1857 st->print("Can not get library information for pid = %d\n", pid);
duke@0 1858 }
duke@0 1859 }
duke@0 1860
duke@0 1861
duke@0 1862 void os::print_os_info(outputStream* st) {
duke@0 1863 st->print("OS:");
duke@0 1864
duke@0 1865 // Try to identify popular distros.
duke@0 1866 // Most Linux distributions have /etc/XXX-release file, which contains
duke@0 1867 // the OS version string. Some have more than one /etc/XXX-release file
duke@0 1868 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
duke@0 1869 // so the order is important.
duke@0 1870 if (!_print_ascii_file("/etc/mandrake-release", st) &&
duke@0 1871 !_print_ascii_file("/etc/sun-release", st) &&
duke@0 1872 !_print_ascii_file("/etc/redhat-release", st) &&
duke@0 1873 !_print_ascii_file("/etc/SuSE-release", st) &&
duke@0 1874 !_print_ascii_file("/etc/turbolinux-release", st) &&
duke@0 1875 !_print_ascii_file("/etc/gentoo-release", st) &&
duke@0 1876 !_print_ascii_file("/etc/debian_version", st)) {
duke@0 1877 st->print("Linux");
duke@0 1878 }
duke@0 1879 st->cr();
duke@0 1880
duke@0 1881 // kernel
duke@0 1882 st->print("uname:");
duke@0 1883 struct utsname name;
duke@0 1884 uname(&name);
duke@0 1885 st->print(name.sysname); st->print(" ");
duke@0 1886 st->print(name.release); st->print(" ");
duke@0 1887 st->print(name.version); st->print(" ");
duke@0 1888 st->print(name.machine);
duke@0 1889 st->cr();
duke@0 1890
duke@0 1891 // Print warning if unsafe chroot environment detected
duke@0 1892 if (unsafe_chroot_detected) {
duke@0 1893 st->print("WARNING!! ");
duke@0 1894 st->print_cr(unstable_chroot_error);
duke@0 1895 }
duke@0 1896
duke@0 1897 // libc, pthread
duke@0 1898 st->print("libc:");
duke@0 1899 st->print(os::Linux::glibc_version()); st->print(" ");
duke@0 1900 st->print(os::Linux::libpthread_version()); st->print(" ");
duke@0 1901 if (os::Linux::is_LinuxThreads()) {
duke@0 1902 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
duke@0 1903 }
duke@0 1904 st->cr();
duke@0 1905
duke@0 1906 // rlimit
duke@0 1907 st->print("rlimit:");
duke@0 1908 struct rlimit rlim;
duke@0 1909
duke@0 1910 st->print(" STACK ");
duke@0 1911 getrlimit(RLIMIT_STACK, &rlim);
duke@0 1912 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
duke@0 1913 else st->print("%uk", rlim.rlim_cur >> 10);
duke@0 1914
duke@0 1915 st->print(", CORE ");
duke@0 1916 getrlimit(RLIMIT_CORE, &rlim);
duke@0 1917 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
duke@0 1918 else st->print("%uk", rlim.rlim_cur >> 10);
duke@0 1919
duke@0 1920 st->print(", NPROC ");
duke@0 1921 getrlimit(RLIMIT_NPROC, &rlim);
duke@0 1922 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
duke@0 1923 else st->print("%d", rlim.rlim_cur);
duke@0 1924
duke@0 1925 st->print(", NOFILE ");
duke@0 1926 getrlimit(RLIMIT_NOFILE, &rlim);
duke@0 1927 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
duke@0 1928 else st->print("%d", rlim.rlim_cur);
duke@0 1929
duke@0 1930 st->print(", AS ");
duke@0 1931 getrlimit(RLIMIT_AS, &rlim);
duke@0 1932 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
duke@0 1933 else st->print("%uk", rlim.rlim_cur >> 10);
duke@0 1934 st->cr();
duke@0 1935
duke@0 1936 // load average
duke@0 1937 st->print("load average:");
duke@0 1938 double loadavg[3];
duke@0 1939 os::loadavg(loadavg, 3);
duke@0 1940 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
duke@0 1941 st->cr();
duke@0 1942 }
duke@0 1943
duke@0 1944 void os::print_memory_info(outputStream* st) {
duke@0 1945
duke@0 1946 st->print("Memory:");
duke@0 1947 st->print(" %dk page", os::vm_page_size()>>10);
duke@0 1948
duke@0 1949 // values in struct sysinfo are "unsigned long"
duke@0 1950 struct sysinfo si;
duke@0 1951 sysinfo(&si);
duke@0 1952
duke@0 1953 st->print(", physical " UINT64_FORMAT "k",
duke@0 1954 os::physical_memory() >> 10);
duke@0 1955 st->print("(" UINT64_FORMAT "k free)",
duke@0 1956 os::available_memory() >> 10);
duke@0 1957 st->print(", swap " UINT64_FORMAT "k",
duke@0 1958 ((jlong)si.totalswap * si.mem_unit) >> 10);
duke@0 1959 st->print("(" UINT64_FORMAT "k free)",
duke@0 1960 ((jlong)si.freeswap * si.mem_unit) >> 10);
duke@0 1961 st->cr();
duke@0 1962 }
duke@0 1963
duke@0 1964 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
duke@0 1965 // but they're the same for all the linux arch that we support
duke@0 1966 // and they're the same for solaris but there's no common place to put this.
duke@0 1967 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
duke@0 1968 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
duke@0 1969 "ILL_COPROC", "ILL_BADSTK" };
duke@0 1970
duke@0 1971 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
duke@0 1972 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
duke@0 1973 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
duke@0 1974
duke@0 1975 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
duke@0 1976
duke@0 1977 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
duke@0 1978
duke@0 1979 void os::print_siginfo(outputStream* st, void* siginfo) {
duke@0 1980 st->print("siginfo:");
duke@0 1981
duke@0 1982 const int buflen = 100;
duke@0 1983 char buf[buflen];
duke@0 1984 siginfo_t *si = (siginfo_t*)siginfo;
duke@0 1985 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
duke@0 1986 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
duke@0 1987 st->print("si_errno=%s", buf);
duke@0 1988 } else {
duke@0 1989 st->print("si_errno=%d", si->si_errno);
duke@0 1990 }
duke@0 1991 const int c = si->si_code;
duke@0 1992 assert(c > 0, "unexpected si_code");
duke@0 1993 switch (si->si_signo) {
duke@0 1994 case SIGILL:
duke@0 1995 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
duke@0 1996 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
duke@0 1997 break;
duke@0 1998 case SIGFPE:
duke@0 1999 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
duke@0 2000 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
duke@0 2001 break;
duke@0 2002 case SIGSEGV:
duke@0 2003 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
duke@0 2004 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
duke@0 2005 break;
duke@0 2006 case SIGBUS:
duke@0 2007 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
duke@0 2008 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
duke@0 2009 break;
duke@0 2010 default:
duke@0 2011 st->print(", si_code=%d", si->si_code);
duke@0 2012 // no si_addr
duke@0 2013 }
duke@0 2014
duke@0 2015 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
duke@0 2016 UseSharedSpaces) {
duke@0 2017 FileMapInfo* mapinfo = FileMapInfo::current_info();
duke@0 2018 if (mapinfo->is_in_shared_space(si->si_addr)) {
duke@0 2019 st->print("\n\nError accessing class data sharing archive." \
duke@0 2020 " Mapped file inaccessible during execution, " \
duke@0 2021 " possible disk/network problem.");
duke@0 2022 }
duke@0 2023 }
duke@0 2024 st->cr();
duke@0 2025 }
duke@0 2026
duke@0 2027
duke@0 2028 static void print_signal_handler(outputStream* st, int sig,
duke@0 2029 char* buf, size_t buflen);
duke@0 2030
duke@0 2031 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
duke@0 2032 st->print_cr("Signal Handlers:");
duke@0 2033 print_signal_handler(st, SIGSEGV, buf, buflen);
duke@0 2034 print_signal_handler(st, SIGBUS , buf, buflen);
duke@0 2035 print_signal_handler(st, SIGFPE , buf, buflen);
duke@0 2036 print_signal_handler(st, SIGPIPE, buf, buflen);
duke@0 2037 print_signal_handler(st, SIGXFSZ, buf, buflen);
duke@0 2038 print_signal_handler(st, SIGILL , buf, buflen);
duke@0 2039 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
duke@0 2040 print_signal_handler(st, SR_signum, buf, buflen);
duke@0 2041 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
duke@0 2042 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
duke@0 2043 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
duke@0 2044 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
duke@0 2045 }
duke@0 2046
duke@0 2047 static char saved_jvm_path[MAXPATHLEN] = {0};
duke@0 2048
duke@0 2049 // Find the full path to the current module, libjvm.so or libjvm_g.so
duke@0 2050 void os::jvm_path(char *buf, jint len) {
duke@0 2051 // Error checking.
duke@0 2052 if (len < MAXPATHLEN) {
duke@0 2053 assert(false, "must use a large-enough buffer");
duke@0 2054 buf[0] = '\0';
duke@0 2055 return;
duke@0 2056 }
duke@0 2057 // Lazy resolve the path to current module.
duke@0 2058 if (saved_jvm_path[0] != 0) {
duke@0 2059 strcpy(buf, saved_jvm_path);
duke@0 2060 return;
duke@0 2061 }
duke@0 2062
duke@0 2063 char dli_fname[MAXPATHLEN];
duke@0 2064 bool ret = dll_address_to_library_name(
duke@0 2065 CAST_FROM_FN_PTR(address, os::jvm_path),
duke@0 2066 dli_fname, sizeof(dli_fname), NULL);
duke@0 2067 assert(ret != 0, "cannot locate libjvm");
xlu@512 2068 if (realpath(dli_fname, buf) == NULL)
xlu@512 2069 return;
duke@0 2070
duke@0 2071 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
duke@0 2072 // Support for the gamma launcher. Typical value for buf is
duke@0 2073 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
duke@0 2074 // the right place in the string, then assume we are installed in a JDK and
duke@0 2075 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
duke@0 2076 // up the path so it looks like libjvm.so is installed there (append a
duke@0 2077 // fake suffix hotspot/libjvm.so).
duke@0 2078 const char *p = buf + strlen(buf) - 1;
duke@0 2079 for (int count = 0; p > buf && count < 5; ++count) {
duke@0 2080 for (--p; p > buf && *p != '/'; --p)
duke@0 2081 /* empty */ ;
duke@0 2082 }
duke@0 2083
duke@0 2084 if (strncmp(p, "/jre/lib/", 9) != 0) {
duke@0 2085 // Look for JAVA_HOME in the environment.
duke@0 2086 char* java_home_var = ::getenv("JAVA_HOME");
duke@0 2087 if (java_home_var != NULL && java_home_var[0] != 0) {
duke@0 2088 // Check the current module name "libjvm.so" or "libjvm_g.so".
duke@0 2089 p = strrchr(buf, '/');
duke@0 2090 assert(strstr(p, "/libjvm") == p, "invalid library name");
duke@0 2091 p = strstr(p, "_g") ? "_g" : "";
duke@0 2092
xlu@512 2093 if (realpath(java_home_var, buf) == NULL)
xlu@512 2094 return;
duke@0 2095 sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
duke@0 2096 if (0 == access(buf, F_OK)) {
duke@0 2097 // Use current module name "libjvm[_g].so" instead of
duke@0 2098 // "libjvm"debug_only("_g")".so" since for fastdebug version
duke@0 2099 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
duke@0 2100 // It is used when we are choosing the HPI library's name
duke@0 2101 // "libhpi[_g].so" in hpi::initialize_get_interface().
duke@0 2102 sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
duke@0 2103 } else {
duke@0 2104 // Go back to path of .so
xlu@512 2105 if (realpath(dli_fname, buf) == NULL)
xlu@512 2106 return;
duke@0 2107 }
duke@0 2108 }
duke@0 2109 }
duke@0 2110 }
duke@0 2111
duke@0 2112 strcpy(saved_jvm_path, buf);
duke@0 2113 }
duke@0 2114
duke@0 2115 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
duke@0 2116 // no prefix required, not even "_"
duke@0 2117 }
duke@0 2118
duke@0 2119 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
duke@0 2120 // no suffix required
duke@0 2121 }
duke@0 2122
duke@0 2123 ////////////////////////////////////////////////////////////////////////////////
duke@0 2124 // sun.misc.Signal support
duke@0 2125
duke@0 2126 static volatile jint sigint_count = 0;
duke@0 2127
duke@0 2128 static void
duke@0 2129 UserHandler(int sig, void *siginfo, void *context) {
duke@0 2130 // 4511530 - sem_post is serialized and handled by the manager thread. When
duke@0 2131 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
duke@0 2132 // don't want to flood the manager thread with sem_post requests.
duke@0 2133 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
duke@0 2134 return;
duke@0 2135
duke@0 2136 // Ctrl-C is pressed during error reporting, likely because the error
duke@0 2137 // handler fails to abort. Let VM die immediately.
duke@0 2138 if (sig == SIGINT && is_error_reported()) {
duke@0 2139 os::die();
duke@0 2140 }
duke@0 2141
duke@0 2142 os::signal_notify(sig);
duke@0 2143 }
duke@0 2144
duke@0 2145 void* os::user_handler() {
duke@0 2146 return CAST_FROM_FN_PTR(void*, UserHandler);
duke@0 2147 }
duke@0 2148
duke@0 2149 extern "C" {
duke@0 2150 typedef void (*sa_handler_t)(int);
duke@0 2151 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
duke@0 2152 }
duke@0 2153
duke@0 2154 void* os::signal(int signal_number, void* handler) {
duke@0 2155 struct sigaction sigAct, oldSigAct;
duke@0 2156
duke@0 2157 sigfillset(&(sigAct.sa_mask));
duke@0 2158 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
duke@0 2159 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
duke@0 2160
duke@0 2161 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
duke@0 2162 // -1 means registration failed
duke@0 2163 return (void *)-1;
duke@0 2164 }
duke@0 2165
duke@0 2166 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
duke@0 2167 }
duke@0 2168
duke@0 2169 void os::signal_raise(int signal_number) {
duke@0 2170 ::raise(signal_number);
duke@0 2171 }
duke@0 2172
duke@0 2173 /*
duke@0 2174 * The following code is moved from os.cpp for making this
duke@0 2175 * code platform specific, which it is by its very nature.
duke@0 2176 */
duke@0 2177
duke@0 2178 // Will be modified when max signal is changed to be dynamic
duke@0 2179 int os::sigexitnum_pd() {
duke@0 2180 return NSIG;
duke@0 2181 }
duke@0 2182
duke@0 2183 // a counter for each possible signal value
duke@0 2184 static volatile jint pending_signals[NSIG+1] = { 0 };
duke@0 2185
duke@0 2186 // Linux(POSIX) specific hand shaking semaphore.
duke@0 2187 static sem_t sig_sem;
duke@0 2188
duke@0 2189 void os::signal_init_pd() {
duke@0 2190 // Initialize signal structures
duke@0 2191 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
duke@0 2192
duke@0 2193 // Initialize signal semaphore
duke@0 2194 ::sem_init(&sig_sem, 0, 0);
duke@0 2195 }
duke@0 2196
duke@0 2197 void os::signal_notify(int sig) {
duke@0 2198 Atomic::inc(&pending_signals[sig]);
duke@0 2199 ::sem_post(&sig_sem);
duke@0 2200 }
duke@0 2201
duke@0 2202 static int check_pending_signals(bool wait) {
duke@0 2203 Atomic::store(0, &sigint_count);
duke@0 2204 for (;;) {
duke@0 2205 for (int i = 0; i < NSIG + 1; i++) {
duke@0 2206 jint n = pending_signals[i];
duke@0 2207 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
duke@0 2208 return i;
duke@0 2209 }
duke@0 2210 }
duke@0 2211 if (!wait) {
duke@0 2212 return -1;
duke@0 2213 }
duke@0 2214 JavaThread *thread = JavaThread::current();
duke@0 2215 ThreadBlockInVM tbivm(thread);
duke@0 2216
duke@0 2217 bool threadIsSuspended;
duke@0 2218 do {
duke@0 2219 thread->set_suspend_equivalent();
duke@0 2220 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
duke@0 2221 ::sem_wait(&sig_sem);
duke@0 2222
duke@0 2223 // were we externally suspended while we were waiting?
duke@0 2224 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
duke@0 2225 if (threadIsSuspended) {
duke@0 2226 //
duke@0 2227 // The semaphore has been incremented, but while we were waiting
duke@0 2228 // another thread suspended us. We don't want to continue running
duke@0 2229 // while suspended because that would surprise the thread that
duke@0 2230 // suspended us.
duke@0 2231 //
duke@0 2232 ::sem_post(&sig_sem);
duke@0 2233
duke@0 2234 thread->java_suspend_self();
duke@0 2235 }
duke@0 2236 } while (threadIsSuspended);
duke@0 2237 }
duke@0 2238 }
duke@0 2239
duke@0 2240 int os::signal_lookup() {
duke@0 2241 return check_pending_signals(false);
duke@0 2242 }
duke@0 2243
duke@0 2244 int os::signal_wait() {
duke@0 2245 return check_pending_signals(true);
duke@0 2246 }
duke@0 2247
duke@0 2248 ////////////////////////////////////////////////////////////////////////////////
duke@0 2249 // Virtual Memory
duke@0 2250
duke@0 2251 int os::vm_page_size() {
duke@0 2252 // Seems redundant as all get out
duke@0 2253 assert(os::Linux::page_size() != -1, "must call os::init");
duke@0 2254 return os::Linux::page_size();
duke@0 2255 }
duke@0 2256
duke@0 2257 // Solaris allocates memory by pages.
duke@0 2258 int os::vm_allocation_granularity() {
duke@0 2259 assert(os::Linux::page_size() != -1, "must call os::init");
duke@0 2260 return os::Linux::page_size();
duke@0 2261 }
duke@0 2262
duke@0 2263 // Rationale behind this function:
duke@0 2264 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
duke@0 2265 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
duke@0 2266 // samples for JITted code. Here we create private executable mapping over the code cache
duke@0 2267 // and then we can use standard (well, almost, as mapping can change) way to provide
duke@0 2268 // info for the reporting script by storing timestamp and location of symbol
duke@0 2269 void linux_wrap_code(char* base, size_t size) {
duke@0 2270 static volatile jint cnt = 0;
duke@0 2271
duke@0 2272 if (!UseOprofile) {
duke@0 2273 return;
duke@0 2274 }
duke@0 2275
duke@0 2276 char buf[40];
duke@0 2277 int num = Atomic::add(1, &cnt);
duke@0 2278
duke@0 2279 sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num);
duke@0 2280 unlink(buf);
duke@0 2281
duke@0 2282 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU);
duke@0 2283
duke@0 2284 if (fd != -1) {
duke@0 2285 off_t rv = lseek(fd, size-2, SEEK_SET);
duke@0 2286 if (rv != (off_t)-1) {
duke@0 2287 if (write(fd, "", 1) == 1) {
duke@0 2288 mmap(base, size,
duke@0 2289 PROT_READ|PROT_WRITE|PROT_EXEC,
duke@0 2290 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
duke@0 2291 }
duke@0 2292 }
duke@0 2293 close(fd);
duke@0 2294 unlink(buf);
duke@0 2295 }
duke@0 2296 }
duke@0 2297
duke@0 2298 // NOTE: Linux kernel does not really reserve the pages for us.
duke@0 2299 // All it does is to check if there are enough free pages
duke@0 2300 // left at the time of mmap(). This could be a potential
duke@0 2301 // problem.
coleenp@655 2302 bool os::commit_memory(char* addr, size_t size, bool exec) {
coleenp@655 2303 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
coleenp@655 2304 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
duke@0 2305 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
duke@0 2306 return res != (uintptr_t) MAP_FAILED;
duke@0 2307 }
duke@0 2308
coleenp@655 2309 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
coleenp@655 2310 bool exec) {
coleenp@655 2311 return commit_memory(addr, size, exec);
duke@0 2312 }
duke@0 2313
duke@0 2314 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { }
iveresov@140 2315
iveresov@140 2316 void os::free_memory(char *addr, size_t bytes) {
iveresov@760 2317 ::mmap(addr, bytes, PROT_READ | PROT_WRITE,
iveresov@760 2318 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
iveresov@140 2319 }
iveresov@140 2320
iveresov@462 2321 void os::numa_make_global(char *addr, size_t bytes) {
iveresov@462 2322 Linux::numa_interleave_memory(addr, bytes);
iveresov@462 2323 }
iveresov@140 2324
iveresov@140 2325 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
iveresov@140 2326 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
iveresov@140 2327 }
iveresov@140 2328
iveresov@140 2329 bool os::numa_topology_changed() { return false; }
iveresov@140 2330
iveresov@140 2331 size_t os::numa_get_groups_num() {
iveresov@140 2332 int max_node = Linux::numa_max_node();
iveresov@140 2333 return max_node > 0 ? max_node + 1 : 1;
iveresov@140 2334 }
iveresov@140 2335
iveresov@140 2336 int os::numa_get_group_id() {
iveresov@140 2337 int cpu_id = Linux::sched_getcpu();
iveresov@140 2338 if (cpu_id != -1) {
iveresov@140 2339 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
iveresov@140 2340 if (lgrp_id != -1) {
iveresov@140 2341 return lgrp_id;
iveresov@140 2342 }
duke@0 2343 }
duke@0 2344 return 0;
duke@0 2345 }
duke@0 2346
iveresov@140 2347 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
iveresov@140 2348 for (size_t i = 0; i < size; i++) {
iveresov@140 2349 ids[i] = i;
iveresov@140 2350 }
iveresov@140 2351 return size;
iveresov@140 2352 }
iveresov@140 2353
duke@0 2354 bool os::get_page_info(char *start, page_info* info) {
duke@0 2355 return false;
duke@0 2356 }
duke@0 2357
duke@0 2358 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
duke@0 2359 return end;
duke@0 2360 }
duke@0 2361
iveresov@140 2362 extern "C" void numa_warn(int number, char *where, ...) { }
iveresov@140 2363 extern "C" void numa_error(char *where) { }
iveresov@140 2364
iveresov@762 2365
iveresov@762 2366 // If we are running with libnuma version > 2, then we should
iveresov@762 2367 // be trying to use symbols with versions 1.1
iveresov@762 2368 // If we are running with earlier version, which did not have symbol versions,
iveresov@762 2369 // we should use the base version.
iveresov@762 2370 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
iveresov@762 2371 void *f = dlvsym(handle, name, "libnuma_1.1");
iveresov@762 2372 if (f == NULL) {
iveresov@762 2373 f = dlsym(handle, name);
iveresov@762 2374 }
iveresov@762 2375 return f;
iveresov@762 2376 }
iveresov@762 2377
iveresov@462 2378 bool os::Linux::libnuma_init() {
iveresov@140 2379 // sched_getcpu() should be in libc.
iveresov@140 2380 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
iveresov@140 2381 dlsym(RTLD_DEFAULT, "sched_getcpu")));
iveresov@140 2382
iveresov@140 2383 if (sched_getcpu() != -1) { // Does it work?
iveresov@266 2384 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
iveresov@140 2385 if (handle != NULL) {
iveresov@140 2386 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
iveresov@762 2387 libnuma_dlsym(handle, "numa_node_to_cpus")));
iveresov@140 2388 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
iveresov@762 2389 libnuma_dlsym(handle, "numa_max_node")));
iveresov@140 2390 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
iveresov@762 2391 libnuma_dlsym(handle, "numa_available")));
iveresov@140 2392 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
iveresov@762 2393 libnuma_dlsym(handle, "numa_tonode_memory")));
iveresov@462 2394 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
iveresov@762 2395 libnuma_dlsym(handle, "numa_interleave_memory")));
iveresov@462 2396
iveresov@462 2397
iveresov@140 2398 if (numa_available() != -1) {
iveresov@762 2399 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
iveresov@140 2400 // Create a cpu -> node mapping
iveresov@140 2401 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
iveresov@140 2402 rebuild_cpu_to_node_map();
iveresov@462 2403 return true;
iveresov@140 2404 }
iveresov@140 2405 }
iveresov@140 2406 }
iveresov@462 2407 return false;
iveresov@140 2408 }
iveresov@140 2409
iveresov@140 2410 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
iveresov@140 2411 // The table is later used in get_node_by_cpu().
iveresov@140 2412 void os::Linux::rebuild_cpu_to_node_map() {
iveresov@462 2413 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
iveresov@462 2414 // in libnuma (possible values are starting from 16,
iveresov@462 2415 // and continuing up with every other power of 2, but less
iveresov@462 2416 // than the maximum number of CPUs supported by kernel), and
iveresov@462 2417 // is a subject to change (in libnuma version 2 the requirements
iveresov@462 2418 // are more reasonable) we'll just hardcode the number they use
iveresov@462 2419 // in the library.
iveresov@462 2420 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
iveresov@462 2421
iveresov@462 2422 size_t cpu_num = os::active_processor_count();
iveresov@462 2423 size_t cpu_map_size = NCPUS / BitsPerCLong;
iveresov@462 2424 size_t cpu_map_valid_size =
iveresov@462 2425 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
iveresov@462 2426
iveresov@140 2427 cpu_to_node()->clear();
iveresov@140 2428 cpu_to_node()->at_grow(cpu_num - 1);
iveresov@462 2429 size_t node_num = numa_get_groups_num();
iveresov@462 2430
iveresov@140 2431 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
iveresov@462 2432 for (size_t i = 0; i < node_num; i++) {
iveresov@140 2433 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
iveresov@462 2434 for (size_t j = 0; j < cpu_map_valid_size; j++) {
iveresov@140 2435 if (cpu_map[j] != 0) {
iveresov@462 2436 for (size_t k = 0; k < BitsPerCLong; k++) {
iveresov@140 2437 if (cpu_map[j] & (1UL << k)) {
iveresov@462 2438 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
iveresov@140 2439 }
iveresov@140 2440 }
iveresov@140 2441 }
iveresov@140 2442 }
iveresov@140 2443 }
iveresov@140 2444 }
iveresov@140 2445 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
iveresov@140 2446 }
iveresov@140 2447
iveresov@140 2448 int os::Linux::get_node_by_cpu(int cpu_id) {
iveresov@140 2449 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
iveresov@140 2450 return cpu_to_node()->at(cpu_id);
iveresov@140 2451 }
iveresov@140 2452 return -1;
iveresov@140 2453 }
iveresov@140 2454
iveresov@140 2455 GrowableArray<int>* os::Linux::_cpu_to_node;
iveresov@140 2456 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
iveresov@140 2457 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
iveresov@140 2458 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
iveresov@140 2459 os::Linux::numa_available_func_t os::Linux::_numa_available;
iveresov@140 2460 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
iveresov@462 2461 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
iveresov@462 2462 unsigned long* os::Linux::_numa_all_nodes;
iveresov@140 2463
duke@0 2464 bool os::uncommit_memory(char* addr, size_t size) {
coleenp@655 2465 return ::mmap(addr, size, PROT_NONE,
duke@0 2466 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0)
duke@0 2467 != MAP_FAILED;
duke@0 2468 }
duke@0 2469
duke@0 2470 static address _highest_vm_reserved_address = NULL;
duke@0 2471
duke@0 2472 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
duke@0 2473 // at 'requested_addr'. If there are existing memory mappings at the same
duke@0 2474 // location, however, they will be overwritten. If 'fixed' is false,
duke@0 2475 // 'requested_addr' is only treated as a hint, the return value may or
duke@0 2476 // may not start from the requested address. Unlike Linux mmap(), this
duke@0 2477 // function returns NULL to indicate failure.
duke@0 2478 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
duke@0 2479 char * addr;
duke@0 2480 int flags;
duke@0 2481
duke@0 2482 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
duke@0 2483 if (fixed) {
duke@0 2484 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
duke@0 2485 flags |= MAP_FIXED;
duke@0 2486 }
duke@0 2487
coleenp@655 2488 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
coleenp@655 2489 // to PROT_EXEC if executable when we commit the page.
coleenp@655 2490 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
duke@0 2491 flags, -1, 0);
duke@0 2492
duke@0 2493 if (addr != MAP_FAILED) {
duke@0 2494 // anon_mmap() should only get called during VM initialization,
duke@0 2495 // don't need lock (actually we can skip locking even it can be called
duke@0 2496 // from multiple threads, because _highest_vm_reserved_address is just a
duke@0 2497 // hint about the upper limit of non-stack memory regions.)
duke@0 2498 if ((address)addr + bytes > _highest_vm_reserved_address) {
duke@0 2499 _highest_vm_reserved_address = (address)addr + bytes;
duke@0 2500 }
duke@0 2501 }
duke@0 2502
duke@0 2503 return addr == MAP_FAILED ? NULL : addr;
duke@0 2504 }
duke@0 2505
duke@0 2506 // Don't update _highest_vm_reserved_address, because there might be memory
duke@0 2507 // regions above addr + size. If so, releasing a memory region only creates
duke@0 2508 // a hole in the address space, it doesn't help prevent heap-stack collision.
duke@0 2509 //
duke@0 2510 static int anon_munmap(char * addr, size_t size) {
duke@0 2511 return ::munmap(addr, size) == 0;
duke@0 2512 }
duke@0 2513
duke@0 2514 char* os::reserve_memory(size_t bytes, char* requested_addr,
duke@0 2515 size_t alignment_hint) {
duke@0 2516 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
duke@0 2517 }
duke@0 2518
duke@0 2519 bool os::release_memory(char* addr, size_t size) {
duke@0 2520 return anon_munmap(addr, size);
duke@0 2521 }
duke@0 2522
duke@0 2523 static address highest_vm_reserved_address() {
duke@0 2524 return _highest_vm_reserved_address;
duke@0 2525 }
duke@0 2526
duke@0 2527 static bool linux_mprotect(char* addr, size_t size, int prot) {
duke@0 2528 // Linux wants the mprotect address argument to be page aligned.
duke@0 2529 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
duke@0 2530
duke@0 2531 // According to SUSv3, mprotect() should only be used with mappings
duke@0 2532 // established by mmap(), and mmap() always maps whole pages. Unaligned
duke@0 2533 // 'addr' likely indicates problem in the VM (e.g. trying to change
duke@0 2534 // protection of malloc'ed or statically allocated memory). Check the
duke@0 2535 // caller if you hit this assert.
duke@0 2536 assert(addr == bottom, "sanity check");
duke@0 2537
duke@0 2538 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
duke@0 2539 return ::mprotect(bottom, size, prot) == 0;
duke@0 2540 }
duke@0 2541
coleenp@237 2542 // Set protections specified
coleenp@237 2543 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
coleenp@237 2544 bool is_committed) {
coleenp@237 2545 unsigned int p = 0;
coleenp@237 2546 switch (prot) {
coleenp@237 2547 case MEM_PROT_NONE: p = PROT_NONE; break;
coleenp@237 2548 case MEM_PROT_READ: p = PROT_READ; break;
coleenp@237 2549 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
coleenp@237 2550 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
coleenp@237 2551 default:
coleenp@237 2552 ShouldNotReachHere();
coleenp@237 2553 }
coleenp@237 2554 // is_committed is unused.
coleenp@237 2555 return linux_mprotect(addr, bytes, p);
duke@0 2556 }
duke@0 2557
duke@0 2558 bool os::guard_memory(char* addr, size_t size) {
duke@0 2559 return linux_mprotect(addr, size, PROT_NONE);
duke@0 2560 }
duke@0 2561
duke@0 2562 bool os::unguard_memory(char* addr, size_t size) {
coleenp@475 2563 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
duke@0 2564 }
duke@0 2565
duke@0 2566 // Large page support
duke@0 2567
duke@0 2568 static size_t _large_page_size = 0;
duke@0 2569
duke@0 2570 bool os::large_page_init() {
duke@0 2571 if (!UseLargePages) return false;
duke@0 2572
duke@0 2573 if (LargePageSizeInBytes) {
duke@0 2574 _large_page_size = LargePageSizeInBytes;
duke@0 2575 } else {
duke@0 2576 // large_page_size on Linux is used to round up heap size. x86 uses either
duke@0 2577 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
duke@0 2578 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
duke@0 2579 // page as large as 256M.
duke@0 2580 //
duke@0 2581 // Here we try to figure out page size by parsing /proc/meminfo and looking
duke@0 2582 // for a line with the following format:
duke@0 2583 // Hugepagesize: 2048 kB
duke@0 2584 //
duke@0 2585 // If we can't determine the value (e.g. /proc is not mounted, or the text
duke@0 2586 // format has been changed), we'll use the largest page size supported by
duke@0 2587 // the processor.
duke@0 2588
duke@0 2589 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M);
duke@0 2590
duke@0 2591 FILE *fp = fopen("/proc/meminfo", "r");
duke@0 2592 if (fp) {
duke@0 2593 while (!feof(fp)) {
duke@0 2594 int x = 0;
duke@0 2595 char buf[16];
duke@0 2596 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
duke@0 2597 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
duke@0 2598 _large_page_size = x * K;
duke@0 2599 break;
duke@0 2600 }
duke@0 2601 } else {
duke@0 2602 // skip to next line
duke@0 2603 for (;;) {
duke@0 2604 int ch = fgetc(fp);
duke@0 2605 if (ch == EOF || ch == (int)'\n') break;
duke@0 2606 }
duke@0 2607 }
duke@0 2608 }
duke@0 2609 fclose(fp);
duke@0 2610 }
duke@0 2611 }
duke@0 2612
duke@0 2613 const size_t default_page_size = (size_t)Linux::page_size();
duke@0 2614 if (_large_page_size > default_page_size) {
duke@0 2615 _page_sizes[0] = _large_page_size;
duke@0 2616 _page_sizes[1] = default_page_size;
duke@0 2617 _page_sizes[2] = 0;
duke@0 2618 }
duke@0 2619
duke@0 2620 // Large page support is available on 2.6 or newer kernel, some vendors
duke@0 2621 // (e.g. Redhat) have backported it to their 2.4 based distributions.
duke@0 2622 // We optimistically assume the support is available. If later it turns out
duke@0 2623 // not true, VM will automatically switch to use regular page size.
duke@0 2624 return true;
duke@0 2625 }
duke@0 2626
duke@0 2627 #ifndef SHM_HUGETLB
duke@0 2628 #define SHM_HUGETLB 04000
duke@0 2629 #endif
duke@0 2630
coleenp@655 2631 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
coleenp@655 2632 // "exec" is passed in but not used. Creating the shared image for
coleenp@655 2633 // the code cache doesn't have an SHM_X executable permission to check.
duke@0 2634 assert(UseLargePages, "only for large pages");
duke@0 2635
duke@0 2636 key_t key = IPC_PRIVATE;
duke@0 2637 char *addr;
duke@0 2638
duke@0 2639 bool warn_on_failure = UseLargePages &&
duke@0 2640 (!FLAG_IS_DEFAULT(UseLargePages) ||
duke@0 2641 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
duke@0 2642 );
duke@0 2643 char msg[128];
duke@0 2644
duke@0 2645 // Create a large shared memory region to attach to based on size.
duke@0 2646 // Currently, size is the total size of the heap
duke@0 2647 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
duke@0 2648 if (shmid == -1) {
duke@0 2649 // Possible reasons for shmget failure:
duke@0 2650 // 1. shmmax is too small for Java heap.
duke@0 2651 // > check shmmax value: cat /proc/sys/kernel/shmmax
duke@0 2652 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
duke@0 2653 // 2. not enough large page memory.
duke@0 2654 // > check available large pages: cat /proc/meminfo
duke@0 2655 // > increase amount of large pages:
duke@0 2656 // echo new_value > /proc/sys/vm/nr_hugepages
duke@0 2657 // Note 1: different Linux may use different name for this property,
duke@0 2658 // e.g. on Redhat AS-3 it is "hugetlb_pool".
duke@0 2659 // Note 2: it's possible there's enough physical memory available but
duke@0 2660 // they are so fragmented after a long run that they can't
duke@0 2661 // coalesce into large pages. Try to reserve large pages when
duke@0 2662 // the system is still "fresh".
duke@0 2663 if (warn_on_failure) {
duke@0 2664 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
duke@0 2665 warning(msg);
duke@0 2666 }
duke@0 2667 return NULL;
duke@0 2668 }
duke@0 2669
duke@0 2670 // attach to the region
duke@0 2671 addr = (char*)shmat(shmid, NULL, 0);
duke@0 2672 int err = errno;
duke@0 2673
duke@0 2674 // Remove shmid. If shmat() is successful, the actual shared memory segment
duke@0 2675 // will be deleted when it's detached by shmdt() or when the process
duke@0 2676 // terminates. If shmat() is not successful this will remove the shared
duke@0 2677 // segment immediately.
duke@0 2678 shmctl(shmid, IPC_RMID, NULL);
duke@0 2679
duke@0 2680 if ((intptr_t)addr == -1) {
duke@0 2681 if (warn_on_failure) {
duke@0 2682 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
duke@0 2683 warning(msg);
duke@0 2684 }
duke@0 2685 return NULL;
duke@0 2686 }
duke@0 2687
duke@0 2688 return addr;
duke@0 2689 }
duke@0 2690
duke@0 2691 bool os::release_memory_special(char* base, size_t bytes) {
duke@0 2692 // detaching the SHM segment will also delete it, see reserve_memory_special()
duke@0 2693 int rslt = shmdt(base);
duke@0 2694 return rslt == 0;
duke@0 2695 }
duke@0 2696
duke@0 2697 size_t os::large_page_size() {
duke@0 2698 return _large_page_size;
duke@0 2699 }
duke@0 2700
duke@0 2701 // Linux does not support anonymous mmap with large page memory. The only way
duke@0 2702 // to reserve large page memory without file backing is through SysV shared
duke@0 2703 // memory API. The entire memory region is committed and pinned upfront.
duke@0 2704 // Hopefully this will change in the future...
duke@0 2705 bool os::can_commit_large_page_memory() {
duke@0 2706 return false;
duke@0 2707 }
duke@0 2708
jcoomes@79 2709 bool os::can_execute_large_page_memory() {
jcoomes@79 2710 return false;
jcoomes@79 2711 }
jcoomes@79 2712
duke@0 2713 // Reserve memory at an arbitrary address, only if that area is
duke@0 2714 // available (and not reserved for something else).
duke@0 2715
duke@0 2716 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
duke@0 2717 const int max_tries = 10;
duke@0 2718 char* base[max_tries];
duke@0 2719 size_t size[max_tries];
duke@0 2720 const size_t gap = 0x000000;
duke@0 2721
duke@0 2722 // Assert only that the size is a multiple of the page size, since
duke@0 2723 // that's all that mmap requires, and since that's all we really know
duke@0 2724 // about at this low abstraction level. If we need higher alignment,
duke@0 2725 // we can either pass an alignment to this method or verify alignment
duke@0 2726 // in one of the methods further up the call chain. See bug 5044738.
duke@0 2727 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
duke@0 2728
duke@0 2729 // Repeatedly allocate blocks until the block is allocated at the
duke@0 2730 // right spot. Give up after max_tries. Note that reserve_memory() will
duke@0 2731 // automatically update _highest_vm_reserved_address if the call is
duke@0 2732 // successful. The variable tracks the highest memory address every reserved
duke@0 2733 // by JVM. It is used to detect heap-stack collision if running with
duke@0 2734 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
duke@0 2735 // space than needed, it could confuse the collision detecting code. To
duke@0 2736 // solve the problem, save current _highest_vm_reserved_address and
duke@0 2737 // calculate the correct value before return.
duke@0 2738 address old_highest = _highest_vm_reserved_address;
duke@0 2739
duke@0 2740 // Linux mmap allows caller to pass an address as hint; give it a try first,
duke@0 2741 // if kernel honors the hint then we can return immediately.
duke@0 2742 char * addr = anon_mmap(requested_addr, bytes, false);
duke@0 2743 if (addr == requested_addr) {
duke@0 2744 return requested_addr;
duke@0 2745 }
duke@0 2746
duke@0 2747 if (addr != NULL) {
duke@0 2748 // mmap() is successful but it fails to reserve at the requested address
duke@0 2749 anon_munmap(addr, bytes);
duke@0 2750 }
duke@0 2751
duke@0 2752 int i;
duke@0 2753 for (i = 0; i < max_tries; ++i) {
duke@0 2754 base[i] = reserve_memory(bytes);
duke@0 2755
duke@0 2756 if (base[i] != NULL) {
duke@0 2757 // Is this the block we wanted?
duke@0 2758 if (base[i] == requested_addr) {
duke@0 2759 size[i] = bytes;
duke@0 2760 break;
duke@0 2761 }
duke@0 2762
duke@0 2763 // Does this overlap the block we wanted? Give back the overlapped
duke@0 2764 // parts and try again.
duke@0 2765
duke@0 2766 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
duke@0 2767 if (top_overlap >= 0 && top_overlap < bytes) {
duke@0 2768 unmap_memory(base[i], top_overlap);
duke@0 2769 base[i] += top_overlap;
duke@0 2770 size[i] = bytes - top_overlap;
duke@0 2771 } else {
duke@0 2772 size_t bottom_overlap = base[i] + bytes - requested_addr;
duke@0 2773 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
duke@0 2774 unmap_memory(requested_addr, bottom_overlap);
duke@0 2775 size[i] = bytes - bottom_overlap;
duke@0 2776 } else {
duke@0 2777 size[i] = bytes;
duke@0 2778 }
duke@0 2779 }
duke@0 2780 }
duke@0 2781 }
duke@0 2782
duke@0 2783 // Give back the unused reserved pieces.
duke@0 2784
duke@0 2785 for (int j = 0; j < i; ++j) {
duke@0 2786 if (base[j] != NULL) {
duke@0 2787 unmap_memory(base[j], size[j]);
duke@0 2788 }
duke@0 2789 }
duke@0 2790
duke@0 2791 if (i < max_tries) {
duke@0 2792 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
duke@0 2793 return requested_addr;
duke@0 2794 } else {
duke@0 2795 _highest_vm_reserved_address = old_highest;
duke@0 2796 return NULL;
duke@0 2797 }
duke@0 2798 }
duke@0 2799
duke@0 2800 size_t os::read(int fd, void *buf, unsigned int nBytes) {
duke@0 2801 return ::read(fd, buf, nBytes);
duke@0 2802 }
duke@0 2803
duke@0 2804 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
duke@0 2805 // Solaris uses poll(), linux uses park().
duke@0 2806 // Poll() is likely a better choice, assuming that Thread.interrupt()
duke@0 2807 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
duke@0 2808 // SIGSEGV, see 4355769.
duke@0 2809
duke@0 2810 const int NANOSECS_PER_MILLISECS = 1000000;
duke@0 2811
duke@0 2812 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
duke@0 2813 assert(thread == Thread::current(), "thread consistency check");
duke@0 2814
duke@0 2815 ParkEvent * const slp = thread->_SleepEvent ;
duke@0 2816 slp->reset() ;
duke@0 2817 OrderAccess::fence() ;
duke@0 2818
duke@0 2819 if (interruptible) {
duke@0 2820 jlong prevtime = javaTimeNanos();
duke@0 2821
duke@0 2822 for (;;) {
duke@0 2823 if (os::is_interrupted(thread, true)) {
duke@0 2824 return OS_INTRPT;
duke@0 2825 }
duke@0 2826
duke@0 2827 jlong newtime = javaTimeNanos();
duke@0 2828
duke@0 2829 if (newtime - prevtime < 0) {
duke@0 2830 // time moving backwards, should only happen if no monotonic clock
duke@0 2831 // not a guarantee() because JVM should not abort on kernel/glibc bugs
duke@0 2832 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
duke@0 2833 } else {
duke@0 2834 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
duke@0 2835 }
duke@0 2836
duke@0 2837 if(millis <= 0) {
duke@0 2838 return OS_OK;
duke@0 2839 }
duke@0 2840
duke@0 2841 prevtime = newtime;
duke@0 2842
duke@0 2843 {
duke@0 2844 assert(thread->is_Java_thread(), "sanity check");
duke@0 2845 JavaThread *jt = (JavaThread *) thread;
duke@0 2846 ThreadBlockInVM tbivm(jt);
duke@0 2847 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
duke@0 2848
duke@0 2849 jt->set_suspend_equivalent();
duke@0 2850 // cleared by handle_special_suspend_equivalent_condition() or
duke@0 2851 // java_suspend_self() via check_and_wait_while_suspended()
duke@0 2852
duke@0 2853 slp->park(millis);
duke@0 2854
duke@0 2855 // were we externally suspended while we were waiting?
duke@0 2856 jt->check_and_wait_while_suspended();
duke@0 2857 }
duke@0 2858 }
duke@0 2859 } else {
duke@0 2860 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
duke@0 2861 jlong prevtime = javaTimeNanos();
duke@0 2862
duke@0 2863 for (;;) {
duke@0 2864 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
duke@0 2865 // the 1st iteration ...
duke@0 2866 jlong newtime = javaTimeNanos();
duke@0 2867
duke@0 2868 if (newtime - prevtime < 0) {
duke@0 2869 // time moving backwards, should only happen if no monotonic clock
duke@0 2870 // not a guarantee() because JVM should not abort on kernel/glibc bugs
duke@0 2871 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
duke@0 2872 } else {
duke@0 2873 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
duke@0 2874 }
duke@0 2875
duke@0 2876 if(millis <= 0) break ;
duke@0 2877
duke@0 2878 prevtime = newtime;
duke@0 2879 slp->park(millis);
duke@0 2880 }
duke@0 2881 return OS_OK ;
duke@0 2882 }
duke@0 2883 }
duke@0 2884
duke@0 2885 int os::naked_sleep() {
duke@0 2886 // %% make the sleep time an integer flag. for now use 1 millisec.
duke@0 2887 return os::sleep(Thread::current(), 1, false);
duke@0 2888 }
duke@0 2889
duke@0 2890 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
duke@0 2891 void os::infinite_sleep() {
duke@0 2892 while (true) { // sleep forever ...
duke@0 2893 ::sleep(100); // ... 100 seconds at a time
duke@0 2894 }
duke@0 2895 }
duke@0 2896
duke@0 2897 // Used to convert frequent JVM_Yield() to nops
duke@0 2898 bool os::dont_yield() {
duke@0 2899 return DontYieldALot;
duke@0 2900 }
duke@0 2901
duke@0 2902 void os::yield() {
duke@0 2903 sched_yield();
duke@0 2904 }
duke@0 2905
duke@0 2906 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
duke@0 2907
duke@0 2908 void os::yield_all(int attempts) {
duke@0 2909 // Yields to all threads, including threads with lower priorities
duke@0 2910 // Threads on Linux are all with same priority. The Solaris style
duke@0 2911 // os::yield_all() with nanosleep(1ms) is not necessary.
duke@0 2912 sched_yield();
duke@0 2913 }
duke@0 2914
duke@0 2915 // Called from the tight loops to possibly influence time-sharing heuristics
duke@0 2916 void os::loop_breaker(int attempts) {
duke@0 2917 os::yield_all(attempts);
duke@0 2918 }
duke@0 2919
duke@0 2920 ////////////////////////////////////////////////////////////////////////////////
duke@0 2921 // thread priority support
duke@0 2922
duke@0 2923 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
duke@0 2924 // only supports dynamic priority, static priority must be zero. For real-time
duke@0 2925 // applications, Linux supports SCHED_RR which allows static priority (1-99).
duke@0 2926 // However, for large multi-threaded applications, SCHED_RR is not only slower
duke@0 2927 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
duke@0 2928 // of 5 runs - Sep 2005).
duke@0 2929 //
duke@0 2930 // The following code actually changes the niceness of kernel-thread/LWP. It
duke@0 2931 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
duke@0 2932 // not the entire user process, and user level threads are 1:1 mapped to kernel
duke@0 2933 // threads. It has always been the case, but could change in the future. For
duke@0 2934 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
duke@0 2935 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
duke@0 2936
duke@0 2937 int os::java_to_os_priority[MaxPriority + 1] = {
duke@0 2938 19, // 0 Entry should never be used
duke@0 2939
duke@0 2940 4, // 1 MinPriority
duke@0 2941 3, // 2
duke@0 2942 2, // 3
duke@0 2943
duke@0 2944 1, // 4
duke@0 2945 0, // 5 NormPriority
duke@0 2946 -1, // 6
duke@0 2947
duke@0 2948 -2, // 7
duke@0 2949 -3, // 8
duke@0 2950 -4, // 9 NearMaxPriority
duke@0 2951
duke@0 2952 -5 // 10 MaxPriority
duke@0 2953 };
duke@0 2954
duke@0 2955 static int prio_init() {
duke@0 2956 if (ThreadPriorityPolicy == 1) {
duke@0 2957 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
duke@0 2958 // if effective uid is not root. Perhaps, a more elegant way of doing
duke@0 2959 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
duke@0 2960 if (geteuid() != 0) {
duke@0 2961 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
duke@0 2962 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
duke@0 2963 }
duke@0 2964 ThreadPriorityPolicy = 0;
duke@0 2965 }
duke@0 2966 }
duke@0 2967 return 0;
duke@0 2968 }
duke@0 2969
duke@0 2970 OSReturn os::set_native_priority(Thread* thread, int newpri) {
duke@0 2971 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
duke@0 2972
duke@0 2973 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
duke@0 2974 return (ret == 0) ? OS_OK : OS_ERR;
duke@0 2975 }
duke@0 2976
duke@0 2977 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
duke@0 2978 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
duke@0 2979 *priority_ptr = java_to_os_priority[NormPriority];
duke@0 2980 return OS_OK;
duke@0 2981 }
duke@0 2982
duke@0 2983 errno = 0;
duke@0 2984 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
duke@0 2985 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
duke@0 2986 }
duke@0 2987
duke@0 2988 // Hint to the underlying OS that a task switch would not be good.
duke@0 2989 // Void return because it's a hint and can fail.
duke@0 2990 void os::hint_no_preempt() {}
duke@0 2991
duke@0 2992 ////////////////////////////////////////////////////////////////////////////////
duke@0 2993 // suspend/resume support
duke@0 2994
duke@0 2995 // the low-level signal-based suspend/resume support is a remnant from the
duke@0 2996 // old VM-suspension that used to be for java-suspension, safepoints etc,
duke@0 2997 // within hotspot. Now there is a single use-case for this:
duke@0 2998 // - calling get_thread_pc() on the VMThread by the flat-profiler task
duke@0 2999 // that runs in the watcher thread.
duke@0 3000 // The remaining code is greatly simplified from the more general suspension
duke@0 3001 // code that used to be used.
duke@0 3002 //
duke@0 3003 // The protocol is quite simple:
duke@0 3004 // - suspend:
duke@0 3005 // - sends a signal to the target thread
duke@0 3006 // - polls the suspend state of the osthread using a yield loop
duke@0 3007 // - target thread signal handler (SR_handler) sets suspend state
duke@0 3008 // and blocks in sigsuspend until continued
duke@0 3009 // - resume:
duke@0 3010 // - sets target osthread state to continue
duke@0 3011 // - sends signal to end the sigsuspend loop in the SR_handler
duke@0 3012 //
duke@0 3013 // Note that the SR_lock plays no role in this suspend/resume protocol.
duke@0 3014 //
duke@0 3015
duke@0 3016 static void resume_clear_context(OSThread *osthread) {
duke@0 3017 osthread->set_ucontext(NULL);
duke@0 3018 osthread->set_siginfo(NULL);
duke@0 3019
duke@0 3020 // notify the suspend action is completed, we have now resumed
duke@0 3021 osthread->sr.clear_suspended();
duke@0 3022 }
duke@0 3023
duke@0 3024 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
duke@0 3025 osthread->set_ucontext(context);
duke@0 3026 osthread->set_siginfo(siginfo);
duke@0 3027 }
duke@0 3028
duke@0 3029 //
duke@0 3030 // Handler function invoked when a thread's execution is suspended or
duke@0 3031 // resumed. We have to be careful that only async-safe functions are
duke@0 3032 // called here (Note: most pthread functions are not async safe and
duke@0 3033 // should be avoided.)
duke@0 3034 //
duke@0 3035 // Note: sigwait() is a more natural fit than sigsuspend() from an
duke@0 3036 // interface point of view, but sigwait() prevents the signal hander
duke@0 3037 // from being run. libpthread would get very confused by not having
duke@0 3038 // its signal handlers run and prevents sigwait()'s use with the
duke@0 3039 // mutex granting granting signal.
duke@0 3040 //
duke@0 3041 // Currently only ever called on the VMThread
duke@0 3042 //
duke@0 3043 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
duke@0 3044 // Save and restore errno to avoid confusing native code with EINTR
duke@0 3045 // after sigsuspend.
duke@0 3046 int old_errno = errno;
duke@0 3047
duke@0 3048 Thread* thread = Thread::current();
duke@0 3049 OSThread* osthread = thread->osthread();
duke@0 3050 assert(thread->is_VM_thread(), "Must be VMThread");
duke@0 3051 // read current suspend action
duke@0 3052 int action = osthread->sr.suspend_action();
duke@0 3053 if (action == SR_SUSPEND) {
duke@0 3054 suspend_save_context(osthread, siginfo, context);
duke@0 3055
duke@0 3056 // Notify the suspend action is about to be completed. do_suspend()
duke@0 3057 // waits until SR_SUSPENDED is set and then returns. We will wait
duke@0 3058 // here for a resume signal and that completes the suspend-other
duke@0 3059 // action. do_suspend/do_resume is always called as a pair from
duke@0 3060 // the same thread - so there are no races
duke@0 3061
duke@0 3062 // notify the caller
duke@0 3063 osthread->sr.set_suspended();
duke@0 3064
duke@0 3065 sigset_t suspend_set; // signals for sigsuspend()
duke@0 3066
duke@0 3067 // get current set of blocked signals and unblock resume signal
duke@0 3068 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
duke@0 3069 sigdelset(&suspend_set, SR_signum);
duke@0 3070
duke@0 3071 // wait here until we are resumed
duke@0 3072 do {
duke@0 3073 sigsuspend(&suspend_set);
duke@0 3074 // ignore all returns until we get a resume signal
duke@0 3075 } while (osthread->sr.suspend_action() != SR_CONTINUE);
duke@0 3076
duke@0 3077 resume_clear_context(osthread);
duke@0 3078
duke@0 3079 } else {
duke@0 3080 assert(action == SR_CONTINUE, "unexpected sr action");
duke@0 3081 // nothing special to do - just leave the handler
duke@0 3082 }
duke@0 3083
duke@0 3084 errno = old_errno;
duke@0 3085 }
duke@0 3086
duke@0 3087
duke@0 3088 static int SR_initialize() {
duke@0 3089 struct sigaction act;
duke@0 3090 char *s;
duke@0 3091 /* Get signal number to use for suspend/resume */
duke@0 3092 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
duke@0 3093 int sig = ::strtol(s, 0, 10);
duke@0 3094 if (sig > 0 || sig < _NSIG) {
duke@0 3095 SR_signum = sig;
duke@0 3096 }
duke@0 3097 }
duke@0 3098
duke@0 3099 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
duke@0 3100 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
duke@0 3101
duke@0 3102 sigemptyset(&SR_sigset);
duke@0 3103 sigaddset(&SR_sigset, SR_signum);
duke@0 3104
duke@0 3105 /* Set up signal handler for suspend/resume */
duke@0 3106 act.sa_flags = SA_RESTART|SA_SIGINFO;
duke@0 3107 act.sa_handler = (void (*)(int)) SR_handler;
duke@0 3108
duke@0 3109 // SR_signum is blocked by default.
duke@0 3110 // 4528190 - We also need to block pthread restart signal (32 on all
duke@0 3111 // supported Linux platforms). Note that LinuxThreads need to block
duke@0 3112 // this signal for all threads to work properly. So we don't have
duke@0 3113 // to use hard-coded signal number when setting up the mask.
duke@0 3114 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
duke@0 3115
duke@0 3116 if (sigaction(SR_signum, &act, 0) == -1) {
duke@0 3117 return -1;
duke@0 3118 }
duke@0 3119
duke@0 3120 // Save signal flag
duke@0 3121 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
duke@0 3122 return 0;
duke@0 3123 }
duke@0 3124
duke@0 3125 static int SR_finalize() {
duke@0 3126 return 0;
duke@0 3127 }
duke@0 3128
duke@0 3129
duke@0 3130 // returns true on success and false on error - really an error is fatal
duke@0 3131 // but this seems the normal response to library errors
duke@0 3132 static bool do_suspend(OSThread* osthread) {
duke@0 3133 // mark as suspended and send signal
duke@0 3134 osthread->sr.set_suspend_action(SR_SUSPEND);
duke@0 3135 int status = pthread_kill(osthread->pthread_id(), SR_signum);
duke@0 3136 assert_status(status == 0, status, "pthread_kill");
duke@0 3137
duke@0 3138 // check status and wait until notified of suspension
duke@0 3139 if (status == 0) {
duke@0 3140 for (int i = 0; !osthread->sr.is_suspended(); i++) {
duke@0 3141 os::yield_all(i);
duke@0 3142 }
duke@0 3143 osthread->sr.set_suspend_action(SR_NONE);
duke@0 3144 return true;
duke@0 3145 }
duke@0 3146 else {
duke@0 3147 osthread->sr.set_suspend_action(SR_NONE);
duke@0 3148 return false;
duke@0 3149 }
duke@0 3150 }
duke@0 3151
duke@0 3152 static void do_resume(OSThread* osthread) {
duke@0 3153 assert(osthread->sr.is_suspended(), "thread should be suspended");
duke@0 3154 osthread->sr.set_suspend_action(SR_CONTINUE);
duke@0 3155
duke@0 3156 int status = pthread_kill(osthread->pthread_id(), SR_signum);
duke@0 3157 assert_status(status == 0, status, "pthread_kill");
duke@0 3158 // check status and wait unit notified of resumption
duke@0 3159 if (status == 0) {
duke@0 3160 for (int i = 0; osthread->sr.is_suspended(); i++) {
duke@0 3161 os::yield_all(i);
duke@0 3162 }
duke@0 3163 }
duke@0 3164 osthread->sr.set_suspend_action(SR_NONE);
duke@0 3165 }
duke@0 3166
duke@0 3167 ////////////////////////////////////////////////////////////////////////////////
duke@0 3168 // interrupt support
duke@0 3169
duke@0 3170 void os::interrupt(Thread* thread) {
duke@0 3171 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
duke@0 3172 "possibility of dangling Thread pointer");
duke@0 3173
duke@0 3174 OSThread* osthread = thread->osthread();
duke@0 3175
duke@0 3176 if (!osthread->interrupted()) {
duke@0 3177 osthread->set_interrupted(true);
duke@0 3178 // More than one thread can get here with the same value of osthread,
duke@0 3179 // resulting in multiple notifications. We do, however, want the store
duke@0 3180 // to interrupted() to be visible to other threads before we execute unpark().
duke@0 3181 OrderAccess::fence();
duke@0 3182 ParkEvent * const slp = thread->_SleepEvent ;
duke@0 3183 if (slp != NULL) slp->unpark() ;
duke@0 3184 }
duke@0 3185
duke@0 3186 // For JSR166. Unpark even if interrupt status already was set
duke@0 3187 if (thread->is_Java_thread())
duke@0 3188 ((JavaThread*)thread)->parker()->unpark();
duke@0 3189
duke@0 3190 ParkEvent * ev = thread->_ParkEvent ;
duke@0 3191 if (ev != NULL) ev->unpark() ;
duke@0 3192
duke@0 3193 }
duke@0 3194
duke@0 3195 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
duke@0 3196 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
duke@0 3197 "possibility of dangling Thread pointer");
duke@0 3198
duke@0 3199 OSThread* osthread = thread->osthread();
duke@0 3200
duke@0 3201 bool interrupted = osthread->interrupted();
duke@0 3202
duke@0 3203 if (interrupted && clear_interrupted) {
duke@0 3204 osthread->set_interrupted(false);
duke@0 3205 // consider thread->_SleepEvent->reset() ... optional optimization
duke@0 3206 }
duke@0 3207
duke@0 3208 return interrupted;
duke@0 3209 }
duke@0 3210
duke@0 3211 ///////////////////////////////////////////////////////////////////////////////////
duke@0 3212 // signal handling (except suspend/resume)
duke@0 3213
duke@0 3214 // This routine may be used by user applications as a "hook" to catch signals.
duke@0 3215 // The user-defined signal handler must pass unrecognized signals to this
duke@0 3216 // routine, and if it returns true (non-zero), then the signal handler must
duke@0 3217 // return immediately. If the flag "abort_if_unrecognized" is true, then this
duke@0 3218 // routine will never retun false (zero), but instead will execute a VM panic
duke@0 3219 // routine kill the process.
duke@0 3220 //
duke@0 3221 // If this routine returns false, it is OK to call it again. This allows
duke@0 3222 // the user-defined signal handler to perform checks either before or after
duke@0 3223 // the VM performs its own checks. Naturally, the user code would be making
duke@0 3224 // a serious error if it tried to handle an exception (such as a null check
duke@0 3225 // or breakpoint) that the VM was generating for its own correct operation.
duke@0 3226 //
duke@0 3227 // This routine may recognize any of the following kinds of signals:
duke@0 3228 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
duke@0 3229 // It should be consulted by handlers for any of those signals.
duke@0 3230 //
duke@0 3231 // The caller of this routine must pass in the three arguments supplied
duke@0 3232 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
duke@0 3233 // field of the structure passed to sigaction(). This routine assumes that
duke@0 3234 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
duke@0 3235 //
duke@0 3236 // Note that the VM will print warnings if it detects conflicting signal
duke@0 3237 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
duke@0 3238 //
duke@0 3239 extern "C" int
duke@0 3240 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
duke@0 3241 void* ucontext, int abort_if_unrecognized);
duke@0 3242
duke@0 3243 void signalHandler(int sig, siginfo_t* info, void* uc) {
duke@0 3244 assert(info != NULL && uc != NULL, "it must be old kernel");
duke@0 3245 JVM_handle_linux_signal(sig, info, uc, true);
duke@0 3246 }
duke@0 3247
duke@0 3248
duke@0 3249 // This boolean allows users to forward their own non-matching signals
duke@0 3250 // to JVM_handle_linux_signal, harmlessly.
duke@0 3251 bool os::Linux::signal_handlers_are_installed = false;
duke@0 3252
duke@0 3253 // For signal-chaining
duke@0 3254 struct sigaction os::Linux::sigact[MAXSIGNUM];
duke@0 3255 unsigned int os::Linux::sigs = 0;
duke@0 3256 bool os::Linux::libjsig_is_loaded = false;
duke@0 3257 typedef struct sigaction *(*get_signal_t)(int);
duke@0 3258 get_signal_t os::Linux::get_signal_action = NULL;
duke@0 3259
duke@0 3260 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
duke@0 3261 struct sigaction *actp = NULL;
duke@0 3262
duke@0 3263 if (libjsig_is_loaded) {
duke@0 3264 // Retrieve the old signal handler from libjsig
duke@0 3265 actp = (*get_signal_action)(sig);
duke@0 3266 }
duke@0 3267 if (actp == NULL) {
duke@0 3268 // Retrieve the preinstalled signal handler from jvm
duke@0 3269 actp = get_preinstalled_handler(sig);
duke@0 3270 }
duke@0 3271
duke@0 3272 return actp;
duke@0 3273 }
duke@0 3274
duke@0 3275 static bool call_chained_handler(struct sigaction *actp, int sig,
duke@0 3276 siginfo_t *siginfo, void *context) {
duke@0 3277 // Call the old signal handler
duke@0 3278 if (actp->sa_handler == SIG_DFL) {
duke@0 3279 // It's more reasonable to let jvm treat it as an unexpected exception
duke@0 3280 // instead of taking the default action.
duke@0 3281 return false;
duke@0 3282 } else if (actp->sa_handler != SIG_IGN) {
duke@0 3283 if ((actp->sa_flags & SA_NODEFER) == 0) {
duke@0 3284 // automaticlly block the signal
duke@0 3285 sigaddset(&(actp->sa_mask), sig);
duke@0 3286 }
duke@0 3287
duke@0 3288 sa_handler_t hand;
duke@0 3289 sa_sigaction_t sa;
duke@0 3290 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
duke@0 3291 // retrieve the chained handler
duke@0 3292 if (siginfo_flag_set) {
duke@0 3293 sa = actp->sa_sigaction;
duke@0 3294 } else {
duke@0 3295 hand = actp->sa_handler;
duke@0 3296 }
duke@0 3297
duke@0 3298 if ((actp->sa_flags & SA_RESETHAND) != 0) {
duke@0 3299 actp->sa_handler = SIG_DFL;
duke@0 3300 }
duke@0 3301
duke@0 3302 // try to honor the signal mask
duke@0 3303 sigset_t oset;
duke@0 3304 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
duke@0 3305
duke@0 3306 // call into the chained handler
duke@0 3307 if (siginfo_flag_set) {
duke@0 3308 (*sa)(sig, siginfo, context);
duke@0 3309 } else {
duke@0 3310 (*hand)(sig);
duke@0 3311 }
duke@0 3312
duke@0 3313 // restore the signal mask
duke@0 3314 pthread_sigmask(SIG_SETMASK, &oset, 0);
duke@0 3315 }
duke@0 3316 // Tell jvm's signal handler the signal is taken care of.
duke@0 3317 return true;
duke@0 3318 }
duke@0 3319
duke@0 3320 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
duke@0 3321 bool chained = false;
duke@0 3322 // signal-chaining
duke@0 3323 if (UseSignalChaining) {
duke@0 3324 struct sigaction *actp = get_chained_signal_action(sig);
duke@0 3325 if (actp != NULL) {
duke@0 3326 chained = call_chained_handler(actp, sig, siginfo, context);
duke@0 3327 }
duke@0 3328 }
duke@0 3329 return chained;
duke@0 3330 }
duke@0 3331
duke@0 3332 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
duke@0 3333 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
duke@0 3334 return &sigact[sig];
duke@0 3335 }
duke@0 3336 return NULL;
duke@0 3337 }
duke@0 3338
duke@0 3339 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
duke@0 3340 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
duke@0 3341 sigact[sig] = oldAct;
duke@0 3342 sigs |= (unsigned int)1 << sig;
duke@0 3343 }
duke@0 3344
duke@0 3345 // for diagnostic
duke@0 3346 int os::Linux::sigflags[MAXSIGNUM];
duke@0 3347
duke@0 3348 int os::Linux::get_our_sigflags(int sig) {
duke@0 3349 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
duke@0 3350 return sigflags[sig];
duke@0 3351 }
duke@0 3352
duke@0 3353 void os::Linux::set_our_sigflags(int sig, int flags) {
duke@0 3354 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
duke@0 3355 sigflags[sig] = flags;
duke@0 3356 }
duke@0 3357
duke@0 3358 void os::Linux::set_signal_handler(int sig, bool set_installed) {
duke@0 3359 // Check for overwrite.
duke@0 3360 struct sigaction oldAct;
duke@0 3361 sigaction(sig, (struct sigaction*)NULL, &oldAct);
duke@0 3362
duke@0 3363 void* oldhand = oldAct.sa_sigaction
duke@0 3364 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
duke@0 3365 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
duke@0 3366 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
duke@0 3367 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
duke@0 3368 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
duke@0 3369 if (AllowUserSignalHandlers || !set_installed) {
duke@0 3370 // Do not overwrite; user takes responsibility to forward to us.
duke@0 3371 return;
duke@0 3372 } else if (UseSignalChaining) {
duke@0 3373 // save the old handler in jvm
duke@0 3374 save_preinstalled_handler(sig, oldAct);
duke@0 3375 // libjsig also interposes the sigaction() call below and saves the
duke@0 3376 // old sigaction on it own.
duke@0 3377 } else {
duke@0 3378 fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig);
duke@0 3379 }
duke@0 3380 }
duke@0 3381
duke@0 3382 struct sigaction sigAct;
duke@0 3383 sigfillset(&(sigAct.sa_mask));
duke@0 3384 sigAct.sa_handler = SIG_DFL;
duke@0 3385 if (!set_installed) {
duke@0 3386 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
duke@0 3387 } else {
duke@0 3388 sigAct.sa_sigaction = signalHandler;
duke@0 3389 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
duke@0 3390 }
duke@0 3391 // Save flags, which are set by ours
duke@0 3392 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
duke@0 3393 sigflags[sig] = sigAct.sa_flags;
duke@0 3394
duke@0 3395 int ret = sigaction(sig, &sigAct, &oldAct);
duke@0 3396 assert(ret == 0, "check");
duke@0 3397
duke@0 3398 void* oldhand2 = oldAct.sa_sigaction
duke@0 3399 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
duke@0 3400 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
duke@0 3401 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
duke@0 3402 }
duke@0 3403
duke@0 3404 // install signal handlers for signals that HotSpot needs to
duke@0 3405 // handle in order to support Java-level exception handling.
duke@0 3406
duke@0 3407 void os::Linux::install_signal_handlers() {
duke@0 3408 if (!signal_handlers_are_installed) {
duke@0 3409 signal_handlers_are_installed = true;
duke@0 3410
duke@0 3411 // signal-chaining
duke@0 3412 typedef void (*signal_setting_t)();
duke@0 3413 signal_setting_t begin_signal_setting = NULL;
duke@0 3414 signal_setting_t end_signal_setting = NULL;
duke@0 3415 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
duke@0 3416 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
duke@0 3417 if (begin_signal_setting != NULL) {
duke@0 3418 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
duke@0 3419 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
duke@0 3420 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
duke@0 3421 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
duke@0 3422 libjsig_is_loaded = true;
duke@0 3423 assert(UseSignalChaining, "should enable signal-chaining");
duke@0 3424 }
duke@0 3425 if (libjsig_is_loaded) {
duke@0 3426 // Tell libjsig jvm is setting signal handlers
duke@0 3427 (*begin_signal_setting)();
duke@0 3428 }
duke@0 3429
duke@0 3430 set_signal_handler(SIGSEGV, true);
duke@0 3431 set_signal_handler(SIGPIPE, true);
duke@0 3432 set_signal_handler(SIGBUS, true);
duke@0 3433 set_signal_handler(SIGILL, true);
duke@0 3434 set_signal_handler(SIGFPE, true);
duke@0 3435 set_signal_handler(SIGXFSZ, true);
duke@0 3436
duke@0 3437 if (libjsig_is_loaded) {
duke@0 3438 // Tell libjsig jvm finishes setting signal handlers
duke@0 3439 (*end_signal_setting)();
duke@0 3440 }
duke@0 3441
duke@0 3442 // We don't activate signal checker if libjsig is in place, we trust ourselves
duke@0 3443 // and if UserSignalHandler is installed all bets are off
duke@0 3444 if (CheckJNICalls) {
duke@0 3445 if (libjsig_is_loaded) {
duke@0 3446 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
duke@0 3447 check_signals = false;
duke@0 3448 }
duke@0 3449 if (AllowUserSignalHandlers) {
duke@0 3450 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
duke@0 3451 check_signals = false;
duke@0 3452 }
duke@0 3453 }
duke@0 3454 }
duke@0 3455 }
duke@0 3456
duke@0 3457 // This is the fastest way to get thread cpu time on Linux.
duke@0 3458 // Returns cpu time (user+sys) for any thread, not only for current.
duke@0 3459 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
duke@0 3460 // It might work on 2.6.10+ with a special kernel/glibc patch.
duke@0 3461 // For reference, please, see IEEE Std 1003.1-2004:
duke@0 3462 // http://www.unix.org/single_unix_specification
duke@0 3463
duke@0 3464 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
duke@0 3465 struct timespec tp;
duke@0 3466 int rc = os::Linux::clock_gettime(clockid, &tp);
duke@0 3467 assert(rc == 0, "clock_gettime is expected to return 0 code");
duke@0 3468
duke@0 3469 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
duke@0 3470 }
duke@0 3471
duke@0 3472 /////
duke@0 3473 // glibc on Linux platform uses non-documented flag
duke@0 3474 // to indicate, that some special sort of signal
duke@0 3475 // trampoline is used.
duke@0 3476 // We will never set this flag, and we should
duke@0 3477 // ignore this flag in our diagnostic
duke@0 3478 #ifdef SIGNIFICANT_SIGNAL_MASK
duke@0 3479 #undef SIGNIFICANT_SIGNAL_MASK
duke@0 3480 #endif
duke@0 3481 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
duke@0 3482
duke@0 3483 static const char* get_signal_handler_name(address handler,
duke@0 3484 char* buf, int buflen) {
duke@0 3485 int offset;
duke@0 3486 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
duke@0 3487 if (found) {
duke@0 3488 // skip directory names
duke@0 3489 const char *p1, *p2;
duke@0 3490 p1 = buf;
duke@0 3491 size_t len = strlen(os::file_separator());
duke@0 3492 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
duke@0 3493 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
duke@0 3494 } else {
duke@0 3495 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
duke@0 3496 }
duke@0 3497 return buf;
duke@0 3498 }
duke@0 3499
duke@0 3500 static void print_signal_handler(outputStream* st, int sig,
duke@0 3501 char* buf, size_t buflen) {
duke@0 3502 struct sigaction sa;
duke@0 3503
duke@0 3504 sigaction(sig, NULL, &sa);
duke@0 3505
duke@0 3506 // See comment for SIGNIFICANT_SIGNAL_MASK define
duke@0 3507 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
duke@0 3508
duke@0 3509 st->print("%s: ", os::exception_name(sig, buf, buflen));
duke@0 3510
duke@0 3511 address handler = (sa.sa_flags & SA_SIGINFO)
duke@0 3512 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
duke@0 3513 : CAST_FROM_FN_PTR(address, sa.sa_handler);
duke@0 3514
duke@0 3515 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
duke@0 3516 st->print("SIG_DFL");
duke@0 3517 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
duke@0 3518 st->print("SIG_IGN");
duke@0 3519 } else {
duke@0 3520 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
duke@0 3521 }
duke@0 3522
duke@0 3523 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
duke@0 3524
duke@0 3525 address rh = VMError::get_resetted_sighandler(sig);
duke@0 3526 // May be, handler was resetted by VMError?
duke@0 3527 if(rh != NULL) {
duke@0 3528 handler = rh;
duke@0 3529 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
duke@0 3530 }
duke@0 3531
duke@0 3532 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
duke@0 3533
duke@0 3534 // Check: is it our handler?
duke@0 3535 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
duke@0 3536 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
duke@0 3537 // It is our signal handler
duke@0 3538 // check for flags, reset system-used one!
duke@0 3539 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
duke@0 3540 st->print(
duke@0 3541 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
duke@0 3542 os::Linux::get_our_sigflags(sig));
duke@0 3543 }
duke@0 3544 }
duke@0 3545 st->cr();
duke@0 3546 }
duke@0 3547
duke@0 3548
duke@0 3549 #define DO_SIGNAL_CHECK(sig) \
duke@0 3550 if (!sigismember(&check_signal_done, sig)) \
duke@0 3551 os::Linux::check_signal_handler(sig)
duke@0 3552
duke@0 3553 // This method is a periodic task to check for misbehaving JNI applications
duke@0 3554 // under CheckJNI, we can add any periodic checks here
duke@0 3555
duke@0 3556 void os::run_periodic_checks() {
duke@0 3557
duke@0 3558 if (check_signals == false) return;
duke@0 3559
duke@0 3560 // SEGV and BUS if overridden could potentially prevent
duke@0 3561 // generation of hs*.log in the event of a crash, debugging
duke@0 3562 // such a case can be very challenging, so we absolutely
duke@0 3563 // check the following for a good measure:
duke@0 3564 DO_SIGNAL_CHECK(SIGSEGV);
duke@0 3565 DO_SIGNAL_CHECK(SIGILL);
duke@0 3566 DO_SIGNAL_CHECK(SIGFPE);
duke@0 3567 DO_SIGNAL_CHECK(SIGBUS);
duke@0 3568 DO_SIGNAL_CHECK(SIGPIPE);
duke@0 3569 DO_SIGNAL_CHECK(SIGXFSZ);
duke@0 3570
duke@0 3571
duke@0 3572 // ReduceSignalUsage allows the user to override these handlers
duke@0 3573 // see comments at the very top and jvm_solaris.h
duke@0 3574 if (!ReduceSignalUsage) {
duke@0 3575 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
duke@0 3576 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
duke@0 3577 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
duke@0 3578 DO_SIGNAL_CHECK(BREAK_SIGNAL);
duke@0 3579 }
duke@0 3580
duke@0 3581 DO_SIGNAL_CHECK(SR_signum);
duke@0 3582 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
duke@0 3583 }
duke@0 3584
duke@0 3585 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
duke@0 3586
duke@0 3587 static os_sigaction_t os_sigaction = NULL;
duke@0 3588
duke@0 3589 void os::Linux::check_signal_handler(int sig) {
duke@0 3590 char buf[O_BUFLEN];
duke@0 3591 address jvmHandler = NULL;
duke@0 3592
duke@0 3593
duke@0 3594 struct sigaction act;
duke@0 3595 if (os_sigaction == NULL) {
duke@0 3596 // only trust the default sigaction, in case it has been interposed
duke@0 3597 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
duke@0 3598 if (os_sigaction == NULL) return;
duke@0 3599 }
duke@0 3600
duke@0 3601 os_sigaction(sig, (struct sigaction*)NULL, &act);
duke@0 3602
duke@0 3603
duke@0 3604 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
duke@0 3605
duke@0 3606 address thisHandler = (act.sa_flags & SA_SIGINFO)
duke@0 3607 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
duke@0 3608 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
duke@0 3609
duke@0 3610
duke@0 3611 switch(sig) {
duke@0 3612 case SIGSEGV:
duke@0 3613 case SIGBUS:
duke@0 3614 case SIGFPE:
duke@0 3615 case SIGPIPE:
duke@0 3616 case SIGILL:
duke@0 3617 case SIGXFSZ:
duke@0 3618 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
duke@0 3619 break;
duke@0 3620
duke@0 3621 case SHUTDOWN1_SIGNAL:
duke@0 3622 case SHUTDOWN2_SIGNAL:
duke@0 3623 case SHUTDOWN3_SIGNAL:
duke@0 3624 case BREAK_SIGNAL:
duke@0 3625 jvmHandler = (address)user_handler();
duke@0 3626 break;
duke@0 3627
duke@0 3628 case INTERRUPT_SIGNAL:
duke@0 3629 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
duke@0 3630 break;
duke@0 3631
duke@0 3632 default:
duke@0 3633 if (sig == SR_signum) {
duke@0 3634 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
duke@0 3635 } else {
duke@0 3636 return;
duke@0 3637 }
duke@0 3638 break;
duke@0 3639 }
duke@0 3640
duke@0 3641 if (thisHandler != jvmHandler) {
duke@0 3642 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
duke@0 3643 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
duke@0 3644 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
duke@0 3645 // No need to check this sig any longer
duke@0 3646 sigaddset(&check_signal_done, sig);
duke@0 3647 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
duke@0 3648 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
duke@0 3649 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
duke@0 3650 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
duke@0 3651 // No need to check this sig any longer
duke@0 3652 sigaddset(&check_signal_done, sig);
duke@0 3653 }
duke@0 3654
duke@0 3655 // Dump all the signal
duke@0 3656 if (sigismember(&check_signal_done, sig)) {
duke@0 3657 print_signal_handlers(tty, buf, O_BUFLEN);
duke@0 3658 }
duke@0 3659 }
duke@0 3660
duke@0 3661 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
duke@0 3662
duke@0 3663 extern bool signal_name(int signo, char* buf, size_t len);
duke@0 3664
duke@0 3665 const char* os::exception_name(int exception_code, char* buf, size_t size) {
duke@0 3666 if (0 < exception_code && exception_code <= SIGRTMAX) {
duke@0 3667 // signal
duke@0 3668 if (!signal_name(exception_code, buf, size)) {
duke@0 3669 jio_snprintf(buf, size, "SIG%d", exception_code);
duke@0 3670 }
duke@0 3671 return buf;
duke@0 3672 } else {
duke@0 3673 return NULL;
duke@0 3674 }
duke@0 3675 }
duke@0 3676
duke@0 3677 // this is called _before_ the most of global arguments have been parsed
duke@0 3678 void os::init(void) {
duke@0 3679 char dummy; /* used to get a guess on initial stack address */
duke@0 3680 // first_hrtime = gethrtime();
duke@0 3681
duke@0 3682 // With LinuxThreads the JavaMain thread pid (primordial thread)
duke@0 3683 // is different than the pid of the java launcher thread.
duke@0 3684 // So, on Linux, the launcher thread pid is passed to the VM
duke@0 3685 // via the sun.java.launcher.pid property.
duke@0 3686 // Use this property instead of getpid() if it was correctly passed.
duke@0 3687 // See bug 6351349.
duke@0 3688 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
duke@0 3689
duke@0 3690 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
duke@0 3691
duke@0 3692 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
duke@0 3693
duke@0 3694 init_random(1234567);
duke@0 3695
duke@0 3696 ThreadCritical::initialize();
duke@0 3697
duke@0 3698 Linux::set_page_size(sysconf(_SC_PAGESIZE));
duke@0 3699 if (Linux::page_size() == -1) {
duke@0 3700 fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno));
duke@0 3701 }
duke@0 3702 init_page_sizes((size_t) Linux::page_size());
duke@0 3703
duke@0 3704 Linux::initialize_system_info();
duke@0 3705
duke@0 3706 // main_thread points to the aboriginal thread
duke@0 3707 Linux::_main_thread = pthread_self();
duke@0 3708
duke@0 3709 Linux::clock_init();
duke@0 3710 initial_time_count = os::elapsed_counter();
kamg@241 3711 pthread_mutex_init(&dl_mutex, NULL);
duke@0 3712 }
duke@0 3713
duke@0 3714 // To install functions for atexit system call
duke@0 3715 extern "C" {
duke@0 3716 static void perfMemory_exit_helper() {
duke@0 3717 perfMemory_exit();
duke@0 3718 }
duke@0 3719 }
duke@0 3720
duke@0 3721 // this is called _after_ the global arguments have been parsed
duke@0 3722 jint os::init_2(void)
duke@0 3723 {
duke@0 3724 Linux::fast_thread_clock_init();
duke@0 3725
duke@0 3726 // Allocate a single page and mark it as readable for safepoint polling
duke@0 3727 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
duke@0 3728 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
duke@0 3729
duke@0 3730 os::set_polling_page( polling_page );
duke@0 3731
duke@0 3732 #ifndef PRODUCT
duke@0 3733 if(Verbose && PrintMiscellaneous)
duke@0 3734 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
duke@0 3735 #endif
duke@0 3736
duke@0 3737 if (!UseMembar) {
duke@0 3738 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
duke@0 3739 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
duke@0 3740 os::set_memory_serialize_page( mem_serialize_page );
duke@0 3741
duke@0 3742 #ifndef PRODUCT
duke@0 3743 if(Verbose && PrintMiscellaneous)
duke@0 3744 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
duke@0 3745 #endif
duke@0 3746 }
duke@0 3747
duke@0 3748 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
duke@0 3749
duke@0 3750 // initialize suspend/resume support - must do this before signal_sets_init()
duke@0 3751 if (SR_initialize() != 0) {
duke@0 3752 perror("SR_initialize failed");
duke@0 3753 return JNI_ERR;
duke@0 3754 }
duke@0 3755
duke@0 3756 Linux::signal_sets_init();
duke@0 3757 Linux::install_signal_handlers();
duke@0 3758
duke@0 3759 size_t threadStackSizeInBytes = ThreadStackSize * K;
duke@0 3760 if (threadStackSizeInBytes != 0 &&
duke@0 3761 threadStackSizeInBytes < Linux::min_stack_allowed) {
duke@0 3762 tty->print_cr("\nThe stack size specified is too small, "
duke@0 3763 "Specify at least %dk",
duke@0 3764 Linux::min_stack_allowed / K);
duke@0 3765 return JNI_ERR;
duke@0 3766 }
duke@0 3767
duke@0 3768 // Make the stack size a multiple of the page size so that
duke@0 3769 // the yellow/red zones can be guarded.
duke@0 3770 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
duke@0 3771 vm_page_size()));
duke@0 3772
duke@0 3773 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
duke@0 3774
duke@0 3775 Linux::libpthread_init();
duke@0 3776 if (PrintMiscellaneous && (Verbose || WizardMode)) {
duke@0 3777 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
duke@0 3778 Linux::glibc_version(), Linux::libpthread_version(),
duke@0 3779 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
duke@0 3780 }
duke@0 3781
iveresov@140 3782 if (UseNUMA) {
iveresov@462 3783 if (!Linux::libnuma_init()) {
iveresov@462 3784 UseNUMA = false;
iveresov@462 3785 } else {
iveresov@462 3786 if ((Linux::numa_max_node() < 1)) {
iveresov@462 3787 // There's only one node(they start from 0), disable NUMA.
iveresov@462 3788 UseNUMA = false;
iveresov@462 3789 }
iveresov@462 3790 }
iveresov@462 3791 if (!UseNUMA && ForceNUMA) {
iveresov@462 3792 UseNUMA = true;
iveresov@462 3793 }
iveresov@140 3794 }
iveresov@140 3795
duke@0 3796 if (MaxFDLimit) {
duke@0 3797 // set the number of file descriptors to max. print out error
duke@0 3798 // if getrlimit/setrlimit fails but continue regardless.
duke@0 3799 struct rlimit nbr_files;
duke@0 3800 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
duke@0 3801 if (status != 0) {
duke@0 3802 if (PrintMiscellaneous && (Verbose || WizardMode))
duke@0 3803 perror("os::init_2 getrlimit failed");
duke@0 3804 } else {
duke@0 3805 nbr_files.rlim_cur = nbr_files.rlim_max;
duke@0 3806 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
duke@0 3807 if (status != 0) {
duke@0 3808 if (PrintMiscellaneous && (Verbose || WizardMode))
duke@0 3809 perror("os::init_2 setrlimit failed");
duke@0 3810 }
duke@0 3811 }
duke@0 3812 }
duke@0 3813
duke@0 3814 // Initialize lock used to serialize thread creation (see os::create_thread)
duke@0 3815 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
duke@0 3816
duke@0 3817 // Initialize HPI.
duke@0 3818 jint hpi_result = hpi::initialize();
duke@0 3819 if (hpi_result != JNI_OK) {
duke@0 3820 tty->print_cr("There was an error trying to initialize the HPI library.");
duke@0 3821 return hpi_result;
duke@0 3822 }
duke@0 3823
duke@0 3824 // at-exit methods are called in the reverse order of their registration.
duke@0 3825 // atexit functions are called on return from main or as a result of a
duke@0 3826 // call to exit(3C). There can be only 32 of these functions registered
duke@0 3827 // and atexit() does not set errno.
duke@0 3828
duke@0 3829 if (PerfAllowAtExitRegistration) {
duke@0 3830 // only register atexit functions if PerfAllowAtExitRegistration is set.
duke@0 3831 // atexit functions can be delayed until process exit time, which
duke@0 3832 // can be problematic for embedded VM situations. Embedded VMs should
duke@0 3833 // call DestroyJavaVM() to assure that VM resources are released.
duke@0 3834
duke@0 3835 // note: perfMemory_exit_helper atexit function may be removed in
duke@0 3836 // the future if the appropriate cleanup code can be added to the