os_linux.cpp revision 4454:cc32ccaaf47f
1232366Sdavide/* 2232366Sdavide * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved. 3232366Sdavide * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4232366Sdavide * 5232366Sdavide * This code is free software; you can redistribute it and/or modify it 6232366Sdavide * under the terms of the GNU General Public License version 2 only, as 7232366Sdavide * published by the Free Software Foundation. 8232366Sdavide * 9232366Sdavide * This code is distributed in the hope that it will be useful, but WITHOUT 10232366Sdavide * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11232366Sdavide * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12232366Sdavide * version 2 for more details (a copy is included in the LICENSE file that 13232366Sdavide * accompanied this code). 14232366Sdavide * 15232366Sdavide * You should have received a copy of the GNU General Public License version 16232366Sdavide * 2 along with this work; if not, write to the Free Software Foundation, 17232366Sdavide * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18232366Sdavide * 19232366Sdavide * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20232366Sdavide * or visit www.oracle.com if you need additional information or have any 21232366Sdavide * questions. 22232366Sdavide * 23232366Sdavide */ 24232366Sdavide 25232366Sdavide// no precompiled headers 26232366Sdavide#include "classfile/classLoader.hpp" 27241741Ssbruno#include "classfile/systemDictionary.hpp" 28232366Sdavide#include "classfile/vmSymbols.hpp" 29232366Sdavide#include "code/icBuffer.hpp" 30232366Sdavide#include "code/vtableStubs.hpp" 31232366Sdavide#include "compiler/compileBroker.hpp" 32232366Sdavide#include "compiler/disassembler.hpp" 33232366Sdavide#include "interpreter/interpreter.hpp" 34232366Sdavide#include "jvm_linux.h" 35232366Sdavide#include "memory/allocation.inline.hpp" 36232366Sdavide#include "memory/filemap.hpp" 37232366Sdavide#include "mutex_linux.inline.hpp" 38232366Sdavide#include "oops/oop.inline.hpp" 39232366Sdavide#include "os_share_linux.hpp" 40232366Sdavide#include "prims/jniFastGetField.hpp" 41232366Sdavide#include "prims/jvm.h" 42232366Sdavide#include "prims/jvm_misc.hpp" 43232366Sdavide#include "runtime/arguments.hpp" 44232366Sdavide#include "runtime/extendedPC.hpp" 45232366Sdavide#include "runtime/globals.hpp" 46232366Sdavide#include "runtime/interfaceSupport.hpp" 47232366Sdavide#include "runtime/init.hpp" 48232366Sdavide#include "runtime/java.hpp" 49232366Sdavide#include "runtime/javaCalls.hpp" 50232366Sdavide#include "runtime/mutexLocker.hpp" 51232366Sdavide#include "runtime/objectMonitor.hpp" 52232366Sdavide#include "runtime/osThread.hpp" 53232366Sdavide#include "runtime/perfMemory.hpp" 54232366Sdavide#include "runtime/sharedRuntime.hpp" 55232366Sdavide#include "runtime/statSampler.hpp" 56232366Sdavide#include "runtime/stubRoutines.hpp" 57232366Sdavide#include "runtime/thread.inline.hpp" 58232366Sdavide#include "runtime/threadCritical.hpp" 59232366Sdavide#include "runtime/timer.hpp" 60232366Sdavide#include "services/attachListener.hpp" 61232366Sdavide#include "services/memTracker.hpp" 62232366Sdavide#include "services/runtimeService.hpp" 63232366Sdavide#include "utilities/decoder.hpp" 64232366Sdavide#include "utilities/defaultStream.hpp" 65232366Sdavide#include "utilities/events.hpp" 66232366Sdavide#include "utilities/elfFile.hpp" 67232366Sdavide#include "utilities/growableArray.hpp" 68232366Sdavide#include "utilities/vmError.hpp" 69232366Sdavide 70232366Sdavide// put OS-includes here 71232366Sdavide# include <sys/types.h> 72232366Sdavide# include <sys/mman.h> 73232366Sdavide# include <sys/stat.h> 74232366Sdavide# include <sys/select.h> 75232366Sdavide# include <pthread.h> 76232366Sdavide# include <signal.h> 77232366Sdavide# include <errno.h> 78232366Sdavide# include <dlfcn.h> 79232366Sdavide# include <stdio.h> 80232366Sdavide# include <unistd.h> 81232366Sdavide# include <sys/resource.h> 82232366Sdavide# include <pthread.h> 83232366Sdavide# include <sys/stat.h> 84232366Sdavide# include <sys/time.h> 85232366Sdavide# include <sys/times.h> 86232366Sdavide# include <sys/utsname.h> 87232366Sdavide# include <sys/socket.h> 88232366Sdavide# include <sys/wait.h> 89232366Sdavide# include <pwd.h> 90232366Sdavide# include <poll.h> 91232366Sdavide# include <semaphore.h> 92232366Sdavide# include <fcntl.h> 93232366Sdavide# include <string.h> 94232366Sdavide# include <syscall.h> 95232366Sdavide# include <sys/sysinfo.h> 96232366Sdavide# include <gnu/libc-version.h> 97232366Sdavide# include <sys/ipc.h> 98232366Sdavide# include <sys/shm.h> 99232366Sdavide# include <link.h> 100232366Sdavide# include <stdint.h> 101232366Sdavide# include <inttypes.h> 102232366Sdavide# include <sys/ioctl.h> 103232366Sdavide 104232366Sdavide#define MAX_PATH (2 * K) 105232366Sdavide 106232366Sdavide// for timer info max values which include all bits 107232366Sdavide#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 108232366Sdavide 109232366Sdavide#define LARGEPAGES_BIT (1 << 6) 110232366Sdavide//////////////////////////////////////////////////////////////////////////////// 111232366Sdavide// global variables 112232366Sdavidejulong os::Linux::_physical_memory = 0; 113232366Sdavide 114232377Spluknetaddress os::Linux::_initial_thread_stack_bottom = NULL; 115232366Sdavideuintptr_t os::Linux::_initial_thread_stack_size = 0; 116232366Sdavide 117232366Sdavideint (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 118232366Sdavideint (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 119232366SdavideMutex* os::Linux::_createThread_lock = NULL; 120232366Sdavidepthread_t os::Linux::_main_thread; 121232366Sdavideint os::Linux::_page_size = -1; 122232366Sdavidebool os::Linux::_is_floating_stack = false; 123232366Sdavidebool os::Linux::_is_NPTL = false; 124232366Sdavidebool os::Linux::_supports_fast_thread_cpu_time = false; 125232366Sdavideconst char * os::Linux::_glibc_version = NULL; 126232366Sdavideconst char * os::Linux::_libpthread_version = NULL; 127232366Sdavide 128232366Sdavidestatic jlong initial_time_count=0; 129232377Spluknet 130232366Sdavidestatic int clock_tics_per_sec = 100; 131232366Sdavide 132232366Sdavide// For diagnostics to print a message once. see run_periodic_checks 133232377Spluknetstatic sigset_t check_signal_done; 134232366Sdavidestatic bool check_signals = true;; 135232366Sdavide 136232366Sdavidestatic pid_t _initial_pid = 0; 137232377Spluknet 138232366Sdavide/* Signal number used to suspend/resume a thread */ 139232366Sdavide 140232366Sdavide/* do not use any signal number less than SIGSEGV, see 4355769 */ 141232366Sdavidestatic int SR_signum = SIGUSR2; 142232366Sdavidesigset_t SR_sigset; 143232366Sdavide 144232377Spluknet/* Used to protect dlsym() calls */ 145232366Sdavidestatic pthread_mutex_t dl_mutex; 146232366Sdavide 147232366Sdavide#ifdef JAVASE_EMBEDDED 148232377Spluknetclass MemNotifyThread: public Thread { 149232366Sdavide friend class VMStructs; 150232366Sdavide public: 151232366Sdavide virtual void run(); 152232377Spluknet 153232366Sdavide private: 154232366Sdavide static MemNotifyThread* _memnotify_thread; 155232366Sdavide int _fd; 156232366Sdavide 157232377Spluknet public: 158232366Sdavide 159232366Sdavide // Constructor 160232366Sdavide MemNotifyThread(int fd); 161232377Spluknet 162232366Sdavide // Tester 163232366Sdavide bool is_memnotify_thread() const { return true; } 164232366Sdavide 165232377Spluknet // Printing 166232366Sdavide char* name() const { return (char*)"Linux MemNotify Thread"; } 167232366Sdavide 168232366Sdavide // Returns the single instance of the MemNotifyThread 169232377Spluknet static MemNotifyThread* memnotify_thread() { return _memnotify_thread; } 170232377Spluknet 171232366Sdavide // Create and start the single instance of MemNotifyThread 172232366Sdavide static void start(); 173232366Sdavide}; 174232377Spluknet#endif // JAVASE_EMBEDDED 175232377Spluknet 176232366Sdavide// utility functions 177232366Sdavide 178232366Sdavidestatic int SR_initialize(); 179232366Sdavide 180232377Spluknetjulong os::available_memory() { 181232366Sdavide return Linux::available_memory(); 182232366Sdavide} 183232366Sdavide 184232377Spluknetjulong os::Linux::available_memory() { 185232366Sdavide // values in struct sysinfo are "unsigned long" 186232366Sdavide struct sysinfo si; 187232366Sdavide sysinfo(&si); 188232366Sdavide 189232366Sdavide return (julong)si.freeram * si.mem_unit; 190232366Sdavide} 191232366Sdavide 192232366Sdavidejulong os::physical_memory() { 193232366Sdavide return Linux::physical_memory(); 194232366Sdavide} 195232366Sdavide 196232366Sdavidejulong os::allocatable_physical_memory(julong size) { 197232366Sdavide#ifdef _LP64 198232366Sdavide return size; 199232366Sdavide#else 200232366Sdavide julong result = MIN2(size, (julong)3800*M); 201232366Sdavide if (!is_allocatable(result)) { 202232366Sdavide // See comments under solaris for alignment considerations 203232366Sdavide julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); 204232366Sdavide result = MIN2(size, reasonable_size); 205232366Sdavide } 206232366Sdavide return result; 207232366Sdavide#endif // _LP64 208232366Sdavide} 209232366Sdavide 210232366Sdavide//////////////////////////////////////////////////////////////////////////////// 211241741Ssbruno// environment support 212233628Sfabient 213232366Sdavidebool os::getenv(const char* name, char* buf, int len) { 214232366Sdavide const char* val = ::getenv(name); 215232366Sdavide if (val != NULL && strlen(val) < (size_t)len) { 216232366Sdavide strcpy(buf, val); 217232366Sdavide return true; 218232366Sdavide } 219232366Sdavide if (len > 0) buf[0] = 0; // return a null string 220232366Sdavide return false; 221232366Sdavide} 222232366Sdavide 223232366Sdavide 224232366Sdavide// Return true if user is running as root. 225232366Sdavide 226232366Sdavidebool os::have_special_privileges() { 227232366Sdavide static bool init = false; 228232366Sdavide static bool privileges = false; 229232366Sdavide if (!init) { 230232366Sdavide privileges = (getuid() != geteuid()) || (getgid() != getegid()); 231232377Spluknet init = true; 232232366Sdavide } 233232366Sdavide return privileges; 234232366Sdavide} 235 236 237#ifndef SYS_gettid 238// i386: 224, ia64: 1105, amd64: 186, sparc 143 239#ifdef __ia64__ 240#define SYS_gettid 1105 241#elif __i386__ 242#define SYS_gettid 224 243#elif __amd64__ 244#define SYS_gettid 186 245#elif __sparc__ 246#define SYS_gettid 143 247#else 248#error define gettid for the arch 249#endif 250#endif 251 252// Cpu architecture string 253#if defined(ZERO) 254static char cpu_arch[] = ZERO_LIBARCH; 255#elif defined(IA64) 256static char cpu_arch[] = "ia64"; 257#elif defined(IA32) 258static char cpu_arch[] = "i386"; 259#elif defined(AMD64) 260static char cpu_arch[] = "amd64"; 261#elif defined(ARM) 262static char cpu_arch[] = "arm"; 263#elif defined(PPC) 264static char cpu_arch[] = "ppc"; 265#elif defined(SPARC) 266# ifdef _LP64 267static char cpu_arch[] = "sparcv9"; 268# else 269static char cpu_arch[] = "sparc"; 270# endif 271#else 272#error Add appropriate cpu_arch setting 273#endif 274 275 276// pid_t gettid() 277// 278// Returns the kernel thread id of the currently running thread. Kernel 279// thread id is used to access /proc. 280// 281// (Note that getpid() on LinuxThreads returns kernel thread id too; but 282// on NPTL, it returns the same pid for all threads, as required by POSIX.) 283// 284pid_t os::Linux::gettid() { 285 int rslt = syscall(SYS_gettid); 286 if (rslt == -1) { 287 // old kernel, no NPTL support 288 return getpid(); 289 } else { 290 return (pid_t)rslt; 291 } 292} 293 294// Most versions of linux have a bug where the number of processors are 295// determined by looking at the /proc file system. In a chroot environment, 296// the system call returns 1. This causes the VM to act as if it is 297// a single processor and elide locking (see is_MP() call). 298static bool unsafe_chroot_detected = false; 299static const char *unstable_chroot_error = "/proc file system not found.\n" 300 "Java may be unstable running multithreaded in a chroot " 301 "environment on Linux when /proc filesystem is not mounted."; 302 303void os::Linux::initialize_system_info() { 304 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 305 if (processor_count() == 1) { 306 pid_t pid = os::Linux::gettid(); 307 char fname[32]; 308 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 309 FILE *fp = fopen(fname, "r"); 310 if (fp == NULL) { 311 unsafe_chroot_detected = true; 312 } else { 313 fclose(fp); 314 } 315 } 316 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 317 assert(processor_count() > 0, "linux error"); 318} 319 320void os::init_system_properties_values() { 321// char arch[12]; 322// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 323 324 // The next steps are taken in the product version: 325 // 326 // Obtain the JAVA_HOME value from the location of libjvm.so. 327 // This library should be located at: 328 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 329 // 330 // If "/jre/lib/" appears at the right place in the path, then we 331 // assume libjvm.so is installed in a JDK and we use this path. 332 // 333 // Otherwise exit with message: "Could not create the Java virtual machine." 334 // 335 // The following extra steps are taken in the debugging version: 336 // 337 // If "/jre/lib/" does NOT appear at the right place in the path 338 // instead of exit check for $JAVA_HOME environment variable. 339 // 340 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 341 // then we append a fake suffix "hotspot/libjvm.so" to this path so 342 // it looks like libjvm.so is installed there 343 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 344 // 345 // Otherwise exit. 346 // 347 // Important note: if the location of libjvm.so changes this 348 // code needs to be changed accordingly. 349 350 // The next few definitions allow the code to be verbatim: 351#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal) 352#define getenv(n) ::getenv(n) 353 354/* 355 * See ld(1): 356 * The linker uses the following search paths to locate required 357 * shared libraries: 358 * 1: ... 359 * ... 360 * 7: The default directories, normally /lib and /usr/lib. 361 */ 362#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 363#define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 364#else 365#define DEFAULT_LIBPATH "/lib:/usr/lib" 366#endif 367 368#define EXTENSIONS_DIR "/lib/ext" 369#define ENDORSED_DIR "/lib/endorsed" 370#define REG_DIR "/usr/java/packages" 371 372 { 373 /* sysclasspath, java_home, dll_dir */ 374 { 375 char *home_path; 376 char *dll_path; 377 char *pslash; 378 char buf[MAXPATHLEN]; 379 os::jvm_path(buf, sizeof(buf)); 380 381 // Found the full path to libjvm.so. 382 // Now cut the path to <java_home>/jre if we can. 383 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 384 pslash = strrchr(buf, '/'); 385 if (pslash != NULL) 386 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 387 dll_path = malloc(strlen(buf) + 1); 388 if (dll_path == NULL) 389 return; 390 strcpy(dll_path, buf); 391 Arguments::set_dll_dir(dll_path); 392 393 if (pslash != NULL) { 394 pslash = strrchr(buf, '/'); 395 if (pslash != NULL) { 396 *pslash = '\0'; /* get rid of /<arch> */ 397 pslash = strrchr(buf, '/'); 398 if (pslash != NULL) 399 *pslash = '\0'; /* get rid of /lib */ 400 } 401 } 402 403 home_path = malloc(strlen(buf) + 1); 404 if (home_path == NULL) 405 return; 406 strcpy(home_path, buf); 407 Arguments::set_java_home(home_path); 408 409 if (!set_boot_path('/', ':')) 410 return; 411 } 412 413 /* 414 * Where to look for native libraries 415 * 416 * Note: Due to a legacy implementation, most of the library path 417 * is set in the launcher. This was to accomodate linking restrictions 418 * on legacy Linux implementations (which are no longer supported). 419 * Eventually, all the library path setting will be done here. 420 * 421 * However, to prevent the proliferation of improperly built native 422 * libraries, the new path component /usr/java/packages is added here. 423 * Eventually, all the library path setting will be done here. 424 */ 425 { 426 char *ld_library_path; 427 428 /* 429 * Construct the invariant part of ld_library_path. Note that the 430 * space for the colon and the trailing null are provided by the 431 * nulls included by the sizeof operator (so actually we allocate 432 * a byte more than necessary). 433 */ 434 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 435 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 436 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 437 438 /* 439 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 440 * should always exist (until the legacy problem cited above is 441 * addressed). 442 */ 443 char *v = getenv("LD_LIBRARY_PATH"); 444 if (v != NULL) { 445 char *t = ld_library_path; 446 /* That's +1 for the colon and +1 for the trailing '\0' */ 447 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 448 sprintf(ld_library_path, "%s:%s", v, t); 449 } 450 Arguments::set_library_path(ld_library_path); 451 } 452 453 /* 454 * Extensions directories. 455 * 456 * Note that the space for the colon and the trailing null are provided 457 * by the nulls included by the sizeof operator (so actually one byte more 458 * than necessary is allocated). 459 */ 460 { 461 char *buf = malloc(strlen(Arguments::get_java_home()) + 462 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 463 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 464 Arguments::get_java_home()); 465 Arguments::set_ext_dirs(buf); 466 } 467 468 /* Endorsed standards default directory. */ 469 { 470 char * buf; 471 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 472 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 473 Arguments::set_endorsed_dirs(buf); 474 } 475 } 476 477#undef malloc 478#undef getenv 479#undef EXTENSIONS_DIR 480#undef ENDORSED_DIR 481 482 // Done 483 return; 484} 485 486//////////////////////////////////////////////////////////////////////////////// 487// breakpoint support 488 489void os::breakpoint() { 490 BREAKPOINT; 491} 492 493extern "C" void breakpoint() { 494 // use debugger to set breakpoint here 495} 496 497//////////////////////////////////////////////////////////////////////////////// 498// signal support 499 500debug_only(static bool signal_sets_initialized = false); 501static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 502 503bool os::Linux::is_sig_ignored(int sig) { 504 struct sigaction oact; 505 sigaction(sig, (struct sigaction*)NULL, &oact); 506 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 507 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 508 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 509 return true; 510 else 511 return false; 512} 513 514void os::Linux::signal_sets_init() { 515 // Should also have an assertion stating we are still single-threaded. 516 assert(!signal_sets_initialized, "Already initialized"); 517 // Fill in signals that are necessarily unblocked for all threads in 518 // the VM. Currently, we unblock the following signals: 519 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 520 // by -Xrs (=ReduceSignalUsage)); 521 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 522 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 523 // the dispositions or masks wrt these signals. 524 // Programs embedding the VM that want to use the above signals for their 525 // own purposes must, at this time, use the "-Xrs" option to prevent 526 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 527 // (See bug 4345157, and other related bugs). 528 // In reality, though, unblocking these signals is really a nop, since 529 // these signals are not blocked by default. 530 sigemptyset(&unblocked_sigs); 531 sigemptyset(&allowdebug_blocked_sigs); 532 sigaddset(&unblocked_sigs, SIGILL); 533 sigaddset(&unblocked_sigs, SIGSEGV); 534 sigaddset(&unblocked_sigs, SIGBUS); 535 sigaddset(&unblocked_sigs, SIGFPE); 536 sigaddset(&unblocked_sigs, SR_signum); 537 538 if (!ReduceSignalUsage) { 539 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 540 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 541 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 542 } 543 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 544 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 545 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 546 } 547 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 548 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 549 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 550 } 551 } 552 // Fill in signals that are blocked by all but the VM thread. 553 sigemptyset(&vm_sigs); 554 if (!ReduceSignalUsage) 555 sigaddset(&vm_sigs, BREAK_SIGNAL); 556 debug_only(signal_sets_initialized = true); 557 558} 559 560// These are signals that are unblocked while a thread is running Java. 561// (For some reason, they get blocked by default.) 562sigset_t* os::Linux::unblocked_signals() { 563 assert(signal_sets_initialized, "Not initialized"); 564 return &unblocked_sigs; 565} 566 567// These are the signals that are blocked while a (non-VM) thread is 568// running Java. Only the VM thread handles these signals. 569sigset_t* os::Linux::vm_signals() { 570 assert(signal_sets_initialized, "Not initialized"); 571 return &vm_sigs; 572} 573 574// These are signals that are blocked during cond_wait to allow debugger in 575sigset_t* os::Linux::allowdebug_blocked_signals() { 576 assert(signal_sets_initialized, "Not initialized"); 577 return &allowdebug_blocked_sigs; 578} 579 580void os::Linux::hotspot_sigmask(Thread* thread) { 581 582 //Save caller's signal mask before setting VM signal mask 583 sigset_t caller_sigmask; 584 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 585 586 OSThread* osthread = thread->osthread(); 587 osthread->set_caller_sigmask(caller_sigmask); 588 589 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 590 591 if (!ReduceSignalUsage) { 592 if (thread->is_VM_thread()) { 593 // Only the VM thread handles BREAK_SIGNAL ... 594 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 595 } else { 596 // ... all other threads block BREAK_SIGNAL 597 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 598 } 599 } 600} 601 602////////////////////////////////////////////////////////////////////////////// 603// detecting pthread library 604 605void os::Linux::libpthread_init() { 606 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 607 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 608 // generic name for earlier versions. 609 // Define macros here so we can build HotSpot on old systems. 610# ifndef _CS_GNU_LIBC_VERSION 611# define _CS_GNU_LIBC_VERSION 2 612# endif 613# ifndef _CS_GNU_LIBPTHREAD_VERSION 614# define _CS_GNU_LIBPTHREAD_VERSION 3 615# endif 616 617 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 618 if (n > 0) { 619 char *str = (char *)malloc(n, mtInternal); 620 confstr(_CS_GNU_LIBC_VERSION, str, n); 621 os::Linux::set_glibc_version(str); 622 } else { 623 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 624 static char _gnu_libc_version[32]; 625 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 626 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 627 os::Linux::set_glibc_version(_gnu_libc_version); 628 } 629 630 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 631 if (n > 0) { 632 char *str = (char *)malloc(n, mtInternal); 633 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 634 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 635 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 636 // is the case. LinuxThreads has a hard limit on max number of threads. 637 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 638 // On the other hand, NPTL does not have such a limit, sysconf() 639 // will return -1 and errno is not changed. Check if it is really NPTL. 640 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 641 strstr(str, "NPTL") && 642 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 643 free(str); 644 os::Linux::set_libpthread_version("linuxthreads"); 645 } else { 646 os::Linux::set_libpthread_version(str); 647 } 648 } else { 649 // glibc before 2.3.2 only has LinuxThreads. 650 os::Linux::set_libpthread_version("linuxthreads"); 651 } 652 653 if (strstr(libpthread_version(), "NPTL")) { 654 os::Linux::set_is_NPTL(); 655 } else { 656 os::Linux::set_is_LinuxThreads(); 657 } 658 659 // LinuxThreads have two flavors: floating-stack mode, which allows variable 660 // stack size; and fixed-stack mode. NPTL is always floating-stack. 661 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 662 os::Linux::set_is_floating_stack(); 663 } 664} 665 666///////////////////////////////////////////////////////////////////////////// 667// thread stack 668 669// Force Linux kernel to expand current thread stack. If "bottom" is close 670// to the stack guard, caller should block all signals. 671// 672// MAP_GROWSDOWN: 673// A special mmap() flag that is used to implement thread stacks. It tells 674// kernel that the memory region should extend downwards when needed. This 675// allows early versions of LinuxThreads to only mmap the first few pages 676// when creating a new thread. Linux kernel will automatically expand thread 677// stack as needed (on page faults). 678// 679// However, because the memory region of a MAP_GROWSDOWN stack can grow on 680// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 681// region, it's hard to tell if the fault is due to a legitimate stack 682// access or because of reading/writing non-exist memory (e.g. buffer 683// overrun). As a rule, if the fault happens below current stack pointer, 684// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 685// application (see Linux kernel fault.c). 686// 687// This Linux feature can cause SIGSEGV when VM bangs thread stack for 688// stack overflow detection. 689// 690// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 691// not use this flag. However, the stack of initial thread is not created 692// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 693// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 694// and then attach the thread to JVM. 695// 696// To get around the problem and allow stack banging on Linux, we need to 697// manually expand thread stack after receiving the SIGSEGV. 698// 699// There are two ways to expand thread stack to address "bottom", we used 700// both of them in JVM before 1.5: 701// 1. adjust stack pointer first so that it is below "bottom", and then 702// touch "bottom" 703// 2. mmap() the page in question 704// 705// Now alternate signal stack is gone, it's harder to use 2. For instance, 706// if current sp is already near the lower end of page 101, and we need to 707// call mmap() to map page 100, it is possible that part of the mmap() frame 708// will be placed in page 100. When page 100 is mapped, it is zero-filled. 709// That will destroy the mmap() frame and cause VM to crash. 710// 711// The following code works by adjusting sp first, then accessing the "bottom" 712// page to force a page fault. Linux kernel will then automatically expand the 713// stack mapping. 714// 715// _expand_stack_to() assumes its frame size is less than page size, which 716// should always be true if the function is not inlined. 717 718#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 719#define NOINLINE 720#else 721#define NOINLINE __attribute__ ((noinline)) 722#endif 723 724static void _expand_stack_to(address bottom) NOINLINE; 725 726static void _expand_stack_to(address bottom) { 727 address sp; 728 size_t size; 729 volatile char *p; 730 731 // Adjust bottom to point to the largest address within the same page, it 732 // gives us a one-page buffer if alloca() allocates slightly more memory. 733 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 734 bottom += os::Linux::page_size() - 1; 735 736 // sp might be slightly above current stack pointer; if that's the case, we 737 // will alloca() a little more space than necessary, which is OK. Don't use 738 // os::current_stack_pointer(), as its result can be slightly below current 739 // stack pointer, causing us to not alloca enough to reach "bottom". 740 sp = (address)&sp; 741 742 if (sp > bottom) { 743 size = sp - bottom; 744 p = (volatile char *)alloca(size); 745 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 746 p[0] = '\0'; 747 } 748} 749 750bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 751 assert(t!=NULL, "just checking"); 752 assert(t->osthread()->expanding_stack(), "expand should be set"); 753 assert(t->stack_base() != NULL, "stack_base was not initialized"); 754 755 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 756 sigset_t mask_all, old_sigset; 757 sigfillset(&mask_all); 758 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 759 _expand_stack_to(addr); 760 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 761 return true; 762 } 763 return false; 764} 765 766////////////////////////////////////////////////////////////////////////////// 767// create new thread 768 769static address highest_vm_reserved_address(); 770 771// check if it's safe to start a new thread 772static bool _thread_safety_check(Thread* thread) { 773 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 774 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 775 // Heap is mmap'ed at lower end of memory space. Thread stacks are 776 // allocated (MAP_FIXED) from high address space. Every thread stack 777 // occupies a fixed size slot (usually 2Mbytes, but user can change 778 // it to other values if they rebuild LinuxThreads). 779 // 780 // Problem with MAP_FIXED is that mmap() can still succeed even part of 781 // the memory region has already been mmap'ed. That means if we have too 782 // many threads and/or very large heap, eventually thread stack will 783 // collide with heap. 784 // 785 // Here we try to prevent heap/stack collision by comparing current 786 // stack bottom with the highest address that has been mmap'ed by JVM 787 // plus a safety margin for memory maps created by native code. 788 // 789 // This feature can be disabled by setting ThreadSafetyMargin to 0 790 // 791 if (ThreadSafetyMargin > 0) { 792 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 793 794 // not safe if our stack extends below the safety margin 795 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 796 } else { 797 return true; 798 } 799 } else { 800 // Floating stack LinuxThreads or NPTL: 801 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 802 // there's not enough space left, pthread_create() will fail. If we come 803 // here, that means enough space has been reserved for stack. 804 return true; 805 } 806} 807 808// Thread start routine for all newly created threads 809static void *java_start(Thread *thread) { 810 // Try to randomize the cache line index of hot stack frames. 811 // This helps when threads of the same stack traces evict each other's 812 // cache lines. The threads can be either from the same JVM instance, or 813 // from different JVM instances. The benefit is especially true for 814 // processors with hyperthreading technology. 815 static int counter = 0; 816 int pid = os::current_process_id(); 817 alloca(((pid ^ counter++) & 7) * 128); 818 819 ThreadLocalStorage::set_thread(thread); 820 821 OSThread* osthread = thread->osthread(); 822 Monitor* sync = osthread->startThread_lock(); 823 824 // non floating stack LinuxThreads needs extra check, see above 825 if (!_thread_safety_check(thread)) { 826 // notify parent thread 827 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 828 osthread->set_state(ZOMBIE); 829 sync->notify_all(); 830 return NULL; 831 } 832 833 // thread_id is kernel thread id (similar to Solaris LWP id) 834 osthread->set_thread_id(os::Linux::gettid()); 835 836 if (UseNUMA) { 837 int lgrp_id = os::numa_get_group_id(); 838 if (lgrp_id != -1) { 839 thread->set_lgrp_id(lgrp_id); 840 } 841 } 842 // initialize signal mask for this thread 843 os::Linux::hotspot_sigmask(thread); 844 845 // initialize floating point control register 846 os::Linux::init_thread_fpu_state(); 847 848 // handshaking with parent thread 849 { 850 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 851 852 // notify parent thread 853 osthread->set_state(INITIALIZED); 854 sync->notify_all(); 855 856 // wait until os::start_thread() 857 while (osthread->get_state() == INITIALIZED) { 858 sync->wait(Mutex::_no_safepoint_check_flag); 859 } 860 } 861 862 // call one more level start routine 863 thread->run(); 864 865 return 0; 866} 867 868bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 869 assert(thread->osthread() == NULL, "caller responsible"); 870 871 // Allocate the OSThread object 872 OSThread* osthread = new OSThread(NULL, NULL); 873 if (osthread == NULL) { 874 return false; 875 } 876 877 // set the correct thread state 878 osthread->set_thread_type(thr_type); 879 880 // Initial state is ALLOCATED but not INITIALIZED 881 osthread->set_state(ALLOCATED); 882 883 thread->set_osthread(osthread); 884 885 // init thread attributes 886 pthread_attr_t attr; 887 pthread_attr_init(&attr); 888 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 889 890 // stack size 891 if (os::Linux::supports_variable_stack_size()) { 892 // calculate stack size if it's not specified by caller 893 if (stack_size == 0) { 894 stack_size = os::Linux::default_stack_size(thr_type); 895 896 switch (thr_type) { 897 case os::java_thread: 898 // Java threads use ThreadStackSize which default value can be 899 // changed with the flag -Xss 900 assert (JavaThread::stack_size_at_create() > 0, "this should be set"); 901 stack_size = JavaThread::stack_size_at_create(); 902 break; 903 case os::compiler_thread: 904 if (CompilerThreadStackSize > 0) { 905 stack_size = (size_t)(CompilerThreadStackSize * K); 906 break; 907 } // else fall through: 908 // use VMThreadStackSize if CompilerThreadStackSize is not defined 909 case os::vm_thread: 910 case os::pgc_thread: 911 case os::cgc_thread: 912 case os::watcher_thread: 913 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 914 break; 915 } 916 } 917 918 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 919 pthread_attr_setstacksize(&attr, stack_size); 920 } else { 921 // let pthread_create() pick the default value. 922 } 923 924 // glibc guard page 925 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 926 927 ThreadState state; 928 929 { 930 // Serialize thread creation if we are running with fixed stack LinuxThreads 931 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 932 if (lock) { 933 os::Linux::createThread_lock()->lock_without_safepoint_check(); 934 } 935 936 pthread_t tid; 937 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 938 939 pthread_attr_destroy(&attr); 940 941 if (ret != 0) { 942 if (PrintMiscellaneous && (Verbose || WizardMode)) { 943 perror("pthread_create()"); 944 } 945 // Need to clean up stuff we've allocated so far 946 thread->set_osthread(NULL); 947 delete osthread; 948 if (lock) os::Linux::createThread_lock()->unlock(); 949 return false; 950 } 951 952 // Store pthread info into the OSThread 953 osthread->set_pthread_id(tid); 954 955 // Wait until child thread is either initialized or aborted 956 { 957 Monitor* sync_with_child = osthread->startThread_lock(); 958 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 959 while ((state = osthread->get_state()) == ALLOCATED) { 960 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 961 } 962 } 963 964 if (lock) { 965 os::Linux::createThread_lock()->unlock(); 966 } 967 } 968 969 // Aborted due to thread limit being reached 970 if (state == ZOMBIE) { 971 thread->set_osthread(NULL); 972 delete osthread; 973 return false; 974 } 975 976 // The thread is returned suspended (in state INITIALIZED), 977 // and is started higher up in the call chain 978 assert(state == INITIALIZED, "race condition"); 979 return true; 980} 981 982///////////////////////////////////////////////////////////////////////////// 983// attach existing thread 984 985// bootstrap the main thread 986bool os::create_main_thread(JavaThread* thread) { 987 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 988 return create_attached_thread(thread); 989} 990 991bool os::create_attached_thread(JavaThread* thread) { 992#ifdef ASSERT 993 thread->verify_not_published(); 994#endif 995 996 // Allocate the OSThread object 997 OSThread* osthread = new OSThread(NULL, NULL); 998 999 if (osthread == NULL) { 1000 return false; 1001 } 1002 1003 // Store pthread info into the OSThread 1004 osthread->set_thread_id(os::Linux::gettid()); 1005 osthread->set_pthread_id(::pthread_self()); 1006 1007 // initialize floating point control register 1008 os::Linux::init_thread_fpu_state(); 1009 1010 // Initial thread state is RUNNABLE 1011 osthread->set_state(RUNNABLE); 1012 1013 thread->set_osthread(osthread); 1014 1015 if (UseNUMA) { 1016 int lgrp_id = os::numa_get_group_id(); 1017 if (lgrp_id != -1) { 1018 thread->set_lgrp_id(lgrp_id); 1019 } 1020 } 1021 1022 if (os::Linux::is_initial_thread()) { 1023 // If current thread is initial thread, its stack is mapped on demand, 1024 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 1025 // the entire stack region to avoid SEGV in stack banging. 1026 // It is also useful to get around the heap-stack-gap problem on SuSE 1027 // kernel (see 4821821 for details). We first expand stack to the top 1028 // of yellow zone, then enable stack yellow zone (order is significant, 1029 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1030 // is no gap between the last two virtual memory regions. 1031 1032 JavaThread *jt = (JavaThread *)thread; 1033 address addr = jt->stack_yellow_zone_base(); 1034 assert(addr != NULL, "initialization problem?"); 1035 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1036 1037 osthread->set_expanding_stack(); 1038 os::Linux::manually_expand_stack(jt, addr); 1039 osthread->clear_expanding_stack(); 1040 } 1041 1042 // initialize signal mask for this thread 1043 // and save the caller's signal mask 1044 os::Linux::hotspot_sigmask(thread); 1045 1046 return true; 1047} 1048 1049void os::pd_start_thread(Thread* thread) { 1050 OSThread * osthread = thread->osthread(); 1051 assert(osthread->get_state() != INITIALIZED, "just checking"); 1052 Monitor* sync_with_child = osthread->startThread_lock(); 1053 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1054 sync_with_child->notify(); 1055} 1056 1057// Free Linux resources related to the OSThread 1058void os::free_thread(OSThread* osthread) { 1059 assert(osthread != NULL, "osthread not set"); 1060 1061 if (Thread::current()->osthread() == osthread) { 1062 // Restore caller's signal mask 1063 sigset_t sigmask = osthread->caller_sigmask(); 1064 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1065 } 1066 1067 delete osthread; 1068} 1069 1070////////////////////////////////////////////////////////////////////////////// 1071// thread local storage 1072 1073int os::allocate_thread_local_storage() { 1074 pthread_key_t key; 1075 int rslt = pthread_key_create(&key, NULL); 1076 assert(rslt == 0, "cannot allocate thread local storage"); 1077 return (int)key; 1078} 1079 1080// Note: This is currently not used by VM, as we don't destroy TLS key 1081// on VM exit. 1082void os::free_thread_local_storage(int index) { 1083 int rslt = pthread_key_delete((pthread_key_t)index); 1084 assert(rslt == 0, "invalid index"); 1085} 1086 1087void os::thread_local_storage_at_put(int index, void* value) { 1088 int rslt = pthread_setspecific((pthread_key_t)index, value); 1089 assert(rslt == 0, "pthread_setspecific failed"); 1090} 1091 1092extern "C" Thread* get_thread() { 1093 return ThreadLocalStorage::thread(); 1094} 1095 1096////////////////////////////////////////////////////////////////////////////// 1097// initial thread 1098 1099// Check if current thread is the initial thread, similar to Solaris thr_main. 1100bool os::Linux::is_initial_thread(void) { 1101 char dummy; 1102 // If called before init complete, thread stack bottom will be null. 1103 // Can be called if fatal error occurs before initialization. 1104 if (initial_thread_stack_bottom() == NULL) return false; 1105 assert(initial_thread_stack_bottom() != NULL && 1106 initial_thread_stack_size() != 0, 1107 "os::init did not locate initial thread's stack region"); 1108 if ((address)&dummy >= initial_thread_stack_bottom() && 1109 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1110 return true; 1111 else return false; 1112} 1113 1114// Find the virtual memory area that contains addr 1115static bool find_vma(address addr, address* vma_low, address* vma_high) { 1116 FILE *fp = fopen("/proc/self/maps", "r"); 1117 if (fp) { 1118 address low, high; 1119 while (!feof(fp)) { 1120 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1121 if (low <= addr && addr < high) { 1122 if (vma_low) *vma_low = low; 1123 if (vma_high) *vma_high = high; 1124 fclose (fp); 1125 return true; 1126 } 1127 } 1128 for (;;) { 1129 int ch = fgetc(fp); 1130 if (ch == EOF || ch == (int)'\n') break; 1131 } 1132 } 1133 fclose(fp); 1134 } 1135 return false; 1136} 1137 1138// Locate initial thread stack. This special handling of initial thread stack 1139// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1140// bogus value for initial thread. 1141void os::Linux::capture_initial_stack(size_t max_size) { 1142 // stack size is the easy part, get it from RLIMIT_STACK 1143 size_t stack_size; 1144 struct rlimit rlim; 1145 getrlimit(RLIMIT_STACK, &rlim); 1146 stack_size = rlim.rlim_cur; 1147 1148 // 6308388: a bug in ld.so will relocate its own .data section to the 1149 // lower end of primordial stack; reduce ulimit -s value a little bit 1150 // so we won't install guard page on ld.so's data section. 1151 stack_size -= 2 * page_size(); 1152 1153 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1154 // 7.1, in both cases we will get 2G in return value. 1155 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1156 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1157 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1158 // in case other parts in glibc still assumes 2M max stack size. 1159 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1160 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1161 if (stack_size > 2 * K * K IA64_ONLY(*2)) 1162 stack_size = 2 * K * K IA64_ONLY(*2); 1163 // Try to figure out where the stack base (top) is. This is harder. 1164 // 1165 // When an application is started, glibc saves the initial stack pointer in 1166 // a global variable "__libc_stack_end", which is then used by system 1167 // libraries. __libc_stack_end should be pretty close to stack top. The 1168 // variable is available since the very early days. However, because it is 1169 // a private interface, it could disappear in the future. 1170 // 1171 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1172 // to __libc_stack_end, it is very close to stack top, but isn't the real 1173 // stack top. Note that /proc may not exist if VM is running as a chroot 1174 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1175 // /proc/<pid>/stat could change in the future (though unlikely). 1176 // 1177 // We try __libc_stack_end first. If that doesn't work, look for 1178 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1179 // as a hint, which should work well in most cases. 1180 1181 uintptr_t stack_start; 1182 1183 // try __libc_stack_end first 1184 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1185 if (p && *p) { 1186 stack_start = *p; 1187 } else { 1188 // see if we can get the start_stack field from /proc/self/stat 1189 FILE *fp; 1190 int pid; 1191 char state; 1192 int ppid; 1193 int pgrp; 1194 int session; 1195 int nr; 1196 int tpgrp; 1197 unsigned long flags; 1198 unsigned long minflt; 1199 unsigned long cminflt; 1200 unsigned long majflt; 1201 unsigned long cmajflt; 1202 unsigned long utime; 1203 unsigned long stime; 1204 long cutime; 1205 long cstime; 1206 long prio; 1207 long nice; 1208 long junk; 1209 long it_real; 1210 uintptr_t start; 1211 uintptr_t vsize; 1212 intptr_t rss; 1213 uintptr_t rsslim; 1214 uintptr_t scodes; 1215 uintptr_t ecode; 1216 int i; 1217 1218 // Figure what the primordial thread stack base is. Code is inspired 1219 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1220 // followed by command name surrounded by parentheses, state, etc. 1221 char stat[2048]; 1222 int statlen; 1223 1224 fp = fopen("/proc/self/stat", "r"); 1225 if (fp) { 1226 statlen = fread(stat, 1, 2047, fp); 1227 stat[statlen] = '\0'; 1228 fclose(fp); 1229 1230 // Skip pid and the command string. Note that we could be dealing with 1231 // weird command names, e.g. user could decide to rename java launcher 1232 // to "java 1.4.2 :)", then the stat file would look like 1233 // 1234 (java 1.4.2 :)) R ... ... 1234 // We don't really need to know the command string, just find the last 1235 // occurrence of ")" and then start parsing from there. See bug 4726580. 1236 char * s = strrchr(stat, ')'); 1237 1238 i = 0; 1239 if (s) { 1240 // Skip blank chars 1241 do s++; while (isspace(*s)); 1242 1243#define _UFM UINTX_FORMAT 1244#define _DFM INTX_FORMAT 1245 1246 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1247 /* 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 */ 1248 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM, 1249 &state, /* 3 %c */ 1250 &ppid, /* 4 %d */ 1251 &pgrp, /* 5 %d */ 1252 &session, /* 6 %d */ 1253 &nr, /* 7 %d */ 1254 &tpgrp, /* 8 %d */ 1255 &flags, /* 9 %lu */ 1256 &minflt, /* 10 %lu */ 1257 &cminflt, /* 11 %lu */ 1258 &majflt, /* 12 %lu */ 1259 &cmajflt, /* 13 %lu */ 1260 &utime, /* 14 %lu */ 1261 &stime, /* 15 %lu */ 1262 &cutime, /* 16 %ld */ 1263 &cstime, /* 17 %ld */ 1264 &prio, /* 18 %ld */ 1265 &nice, /* 19 %ld */ 1266 &junk, /* 20 %ld */ 1267 &it_real, /* 21 %ld */ 1268 &start, /* 22 UINTX_FORMAT */ 1269 &vsize, /* 23 UINTX_FORMAT */ 1270 &rss, /* 24 INTX_FORMAT */ 1271 &rsslim, /* 25 UINTX_FORMAT */ 1272 &scodes, /* 26 UINTX_FORMAT */ 1273 &ecode, /* 27 UINTX_FORMAT */ 1274 &stack_start); /* 28 UINTX_FORMAT */ 1275 } 1276 1277#undef _UFM 1278#undef _DFM 1279 1280 if (i != 28 - 2) { 1281 assert(false, "Bad conversion from /proc/self/stat"); 1282 // product mode - assume we are the initial thread, good luck in the 1283 // embedded case. 1284 warning("Can't detect initial thread stack location - bad conversion"); 1285 stack_start = (uintptr_t) &rlim; 1286 } 1287 } else { 1288 // For some reason we can't open /proc/self/stat (for example, running on 1289 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1290 // most cases, so don't abort: 1291 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1292 stack_start = (uintptr_t) &rlim; 1293 } 1294 } 1295 1296 // Now we have a pointer (stack_start) very close to the stack top, the 1297 // next thing to do is to figure out the exact location of stack top. We 1298 // can find out the virtual memory area that contains stack_start by 1299 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1300 // and its upper limit is the real stack top. (again, this would fail if 1301 // running inside chroot, because /proc may not exist.) 1302 1303 uintptr_t stack_top; 1304 address low, high; 1305 if (find_vma((address)stack_start, &low, &high)) { 1306 // success, "high" is the true stack top. (ignore "low", because initial 1307 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1308 stack_top = (uintptr_t)high; 1309 } else { 1310 // failed, likely because /proc/self/maps does not exist 1311 warning("Can't detect initial thread stack location - find_vma failed"); 1312 // best effort: stack_start is normally within a few pages below the real 1313 // stack top, use it as stack top, and reduce stack size so we won't put 1314 // guard page outside stack. 1315 stack_top = stack_start; 1316 stack_size -= 16 * page_size(); 1317 } 1318 1319 // stack_top could be partially down the page so align it 1320 stack_top = align_size_up(stack_top, page_size()); 1321 1322 if (max_size && stack_size > max_size) { 1323 _initial_thread_stack_size = max_size; 1324 } else { 1325 _initial_thread_stack_size = stack_size; 1326 } 1327 1328 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1329 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1330} 1331 1332//////////////////////////////////////////////////////////////////////////////// 1333// time support 1334 1335// Time since start-up in seconds to a fine granularity. 1336// Used by VMSelfDestructTimer and the MemProfiler. 1337double os::elapsedTime() { 1338 1339 return (double)(os::elapsed_counter()) * 0.000001; 1340} 1341 1342jlong os::elapsed_counter() { 1343 timeval time; 1344 int status = gettimeofday(&time, NULL); 1345 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1346} 1347 1348jlong os::elapsed_frequency() { 1349 return (1000 * 1000); 1350} 1351 1352// For now, we say that linux does not support vtime. I have no idea 1353// whether it can actually be made to (DLD, 9/13/05). 1354 1355bool os::supports_vtime() { return false; } 1356bool os::enable_vtime() { return false; } 1357bool os::vtime_enabled() { return false; } 1358double os::elapsedVTime() { 1359 // better than nothing, but not much 1360 return elapsedTime(); 1361} 1362 1363jlong os::javaTimeMillis() { 1364 timeval time; 1365 int status = gettimeofday(&time, NULL); 1366 assert(status != -1, "linux error"); 1367 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1368} 1369 1370#ifndef CLOCK_MONOTONIC 1371#define CLOCK_MONOTONIC (1) 1372#endif 1373 1374void os::Linux::clock_init() { 1375 // we do dlopen's in this particular order due to bug in linux 1376 // dynamical loader (see 6348968) leading to crash on exit 1377 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1378 if (handle == NULL) { 1379 handle = dlopen("librt.so", RTLD_LAZY); 1380 } 1381 1382 if (handle) { 1383 int (*clock_getres_func)(clockid_t, struct timespec*) = 1384 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1385 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1386 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1387 if (clock_getres_func && clock_gettime_func) { 1388 // See if monotonic clock is supported by the kernel. Note that some 1389 // early implementations simply return kernel jiffies (updated every 1390 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1391 // for nano time (though the monotonic property is still nice to have). 1392 // It's fixed in newer kernels, however clock_getres() still returns 1393 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1394 // resolution for now. Hopefully as people move to new kernels, this 1395 // won't be a problem. 1396 struct timespec res; 1397 struct timespec tp; 1398 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1399 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1400 // yes, monotonic clock is supported 1401 _clock_gettime = clock_gettime_func; 1402 } else { 1403 // close librt if there is no monotonic clock 1404 dlclose(handle); 1405 } 1406 } 1407 } 1408} 1409 1410#ifndef SYS_clock_getres 1411 1412#if defined(IA32) || defined(AMD64) 1413#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1414#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1415#else 1416#warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1417#define sys_clock_getres(x,y) -1 1418#endif 1419 1420#else 1421#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1422#endif 1423 1424void os::Linux::fast_thread_clock_init() { 1425 if (!UseLinuxPosixThreadCPUClocks) { 1426 return; 1427 } 1428 clockid_t clockid; 1429 struct timespec tp; 1430 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1431 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1432 1433 // Switch to using fast clocks for thread cpu time if 1434 // the sys_clock_getres() returns 0 error code. 1435 // Note, that some kernels may support the current thread 1436 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1437 // returned by the pthread_getcpuclockid(). 1438 // If the fast Posix clocks are supported then the sys_clock_getres() 1439 // must return at least tp.tv_sec == 0 which means a resolution 1440 // better than 1 sec. This is extra check for reliability. 1441 1442 if(pthread_getcpuclockid_func && 1443 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1444 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1445 1446 _supports_fast_thread_cpu_time = true; 1447 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1448 } 1449} 1450 1451jlong os::javaTimeNanos() { 1452 if (Linux::supports_monotonic_clock()) { 1453 struct timespec tp; 1454 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1455 assert(status == 0, "gettime error"); 1456 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1457 return result; 1458 } else { 1459 timeval time; 1460 int status = gettimeofday(&time, NULL); 1461 assert(status != -1, "linux error"); 1462 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1463 return 1000 * usecs; 1464 } 1465} 1466 1467void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1468 if (Linux::supports_monotonic_clock()) { 1469 info_ptr->max_value = ALL_64_BITS; 1470 1471 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1472 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1473 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1474 } else { 1475 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1476 info_ptr->max_value = ALL_64_BITS; 1477 1478 // gettimeofday is a real time clock so it skips 1479 info_ptr->may_skip_backward = true; 1480 info_ptr->may_skip_forward = true; 1481 } 1482 1483 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1484} 1485 1486// Return the real, user, and system times in seconds from an 1487// arbitrary fixed point in the past. 1488bool os::getTimesSecs(double* process_real_time, 1489 double* process_user_time, 1490 double* process_system_time) { 1491 struct tms ticks; 1492 clock_t real_ticks = times(&ticks); 1493 1494 if (real_ticks == (clock_t) (-1)) { 1495 return false; 1496 } else { 1497 double ticks_per_second = (double) clock_tics_per_sec; 1498 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1499 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1500 *process_real_time = ((double) real_ticks) / ticks_per_second; 1501 1502 return true; 1503 } 1504} 1505 1506 1507char * os::local_time_string(char *buf, size_t buflen) { 1508 struct tm t; 1509 time_t long_time; 1510 time(&long_time); 1511 localtime_r(&long_time, &t); 1512 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1513 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1514 t.tm_hour, t.tm_min, t.tm_sec); 1515 return buf; 1516} 1517 1518struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1519 return localtime_r(clock, res); 1520} 1521 1522//////////////////////////////////////////////////////////////////////////////// 1523// runtime exit support 1524 1525// Note: os::shutdown() might be called very early during initialization, or 1526// called from signal handler. Before adding something to os::shutdown(), make 1527// sure it is async-safe and can handle partially initialized VM. 1528void os::shutdown() { 1529 1530 // allow PerfMemory to attempt cleanup of any persistent resources 1531 perfMemory_exit(); 1532 1533 // needs to remove object in file system 1534 AttachListener::abort(); 1535 1536 // flush buffered output, finish log files 1537 ostream_abort(); 1538 1539 // Check for abort hook 1540 abort_hook_t abort_hook = Arguments::abort_hook(); 1541 if (abort_hook != NULL) { 1542 abort_hook(); 1543 } 1544 1545} 1546 1547// Note: os::abort() might be called very early during initialization, or 1548// called from signal handler. Before adding something to os::abort(), make 1549// sure it is async-safe and can handle partially initialized VM. 1550void os::abort(bool dump_core) { 1551 os::shutdown(); 1552 if (dump_core) { 1553#ifndef PRODUCT 1554 fdStream out(defaultStream::output_fd()); 1555 out.print_raw("Current thread is "); 1556 char buf[16]; 1557 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1558 out.print_raw_cr(buf); 1559 out.print_raw_cr("Dumping core ..."); 1560#endif 1561 ::abort(); // dump core 1562 } 1563 1564 ::exit(1); 1565} 1566 1567// Die immediately, no exit hook, no abort hook, no cleanup. 1568void os::die() { 1569 // _exit() on LinuxThreads only kills current thread 1570 ::abort(); 1571} 1572 1573// unused on linux for now. 1574void os::set_error_file(const char *logfile) {} 1575 1576 1577// This method is a copy of JDK's sysGetLastErrorString 1578// from src/solaris/hpi/src/system_md.c 1579 1580size_t os::lasterror(char *buf, size_t len) { 1581 1582 if (errno == 0) return 0; 1583 1584 const char *s = ::strerror(errno); 1585 size_t n = ::strlen(s); 1586 if (n >= len) { 1587 n = len - 1; 1588 } 1589 ::strncpy(buf, s, n); 1590 buf[n] = '\0'; 1591 return n; 1592} 1593 1594intx os::current_thread_id() { return (intx)pthread_self(); } 1595int os::current_process_id() { 1596 1597 // Under the old linux thread library, linux gives each thread 1598 // its own process id. Because of this each thread will return 1599 // a different pid if this method were to return the result 1600 // of getpid(2). Linux provides no api that returns the pid 1601 // of the launcher thread for the vm. This implementation 1602 // returns a unique pid, the pid of the launcher thread 1603 // that starts the vm 'process'. 1604 1605 // Under the NPTL, getpid() returns the same pid as the 1606 // launcher thread rather than a unique pid per thread. 1607 // Use gettid() if you want the old pre NPTL behaviour. 1608 1609 // if you are looking for the result of a call to getpid() that 1610 // returns a unique pid for the calling thread, then look at the 1611 // OSThread::thread_id() method in osThread_linux.hpp file 1612 1613 return (int)(_initial_pid ? _initial_pid : getpid()); 1614} 1615 1616// DLL functions 1617 1618const char* os::dll_file_extension() { return ".so"; } 1619 1620// This must be hard coded because it's the system's temporary 1621// directory not the java application's temp directory, ala java.io.tmpdir. 1622const char* os::get_temp_directory() { return "/tmp"; } 1623 1624static bool file_exists(const char* filename) { 1625 struct stat statbuf; 1626 if (filename == NULL || strlen(filename) == 0) { 1627 return false; 1628 } 1629 return os::stat(filename, &statbuf) == 0; 1630} 1631 1632bool os::dll_build_name(char* buffer, size_t buflen, 1633 const char* pname, const char* fname) { 1634 bool retval = false; 1635 // Copied from libhpi 1636 const size_t pnamelen = pname ? strlen(pname) : 0; 1637 1638 // Return error on buffer overflow. 1639 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1640 return retval; 1641 } 1642 1643 if (pnamelen == 0) { 1644 snprintf(buffer, buflen, "lib%s.so", fname); 1645 retval = true; 1646 } else if (strchr(pname, *os::path_separator()) != NULL) { 1647 int n; 1648 char** pelements = split_path(pname, &n); 1649 if (pelements == NULL) { 1650 return false; 1651 } 1652 for (int i = 0 ; i < n ; i++) { 1653 // Really shouldn't be NULL, but check can't hurt 1654 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1655 continue; // skip the empty path values 1656 } 1657 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1658 if (file_exists(buffer)) { 1659 retval = true; 1660 break; 1661 } 1662 } 1663 // release the storage 1664 for (int i = 0 ; i < n ; i++) { 1665 if (pelements[i] != NULL) { 1666 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1667 } 1668 } 1669 if (pelements != NULL) { 1670 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1671 } 1672 } else { 1673 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1674 retval = true; 1675 } 1676 return retval; 1677} 1678 1679const char* os::get_current_directory(char *buf, int buflen) { 1680 return getcwd(buf, buflen); 1681} 1682 1683// check if addr is inside libjvm.so 1684bool os::address_is_in_vm(address addr) { 1685 static address libjvm_base_addr; 1686 Dl_info dlinfo; 1687 1688 if (libjvm_base_addr == NULL) { 1689 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); 1690 libjvm_base_addr = (address)dlinfo.dli_fbase; 1691 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1692 } 1693 1694 if (dladdr((void *)addr, &dlinfo)) { 1695 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1696 } 1697 1698 return false; 1699} 1700 1701bool os::dll_address_to_function_name(address addr, char *buf, 1702 int buflen, int *offset) { 1703 Dl_info dlinfo; 1704 1705 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { 1706 if (buf != NULL) { 1707 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1708 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1709 } 1710 } 1711 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1712 return true; 1713 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) { 1714 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1715 buf, buflen, offset, dlinfo.dli_fname)) { 1716 return true; 1717 } 1718 } 1719 1720 if (buf != NULL) buf[0] = '\0'; 1721 if (offset != NULL) *offset = -1; 1722 return false; 1723} 1724 1725struct _address_to_library_name { 1726 address addr; // input : memory address 1727 size_t buflen; // size of fname 1728 char* fname; // output: library name 1729 address base; // library base addr 1730}; 1731 1732static int address_to_library_name_callback(struct dl_phdr_info *info, 1733 size_t size, void *data) { 1734 int i; 1735 bool found = false; 1736 address libbase = NULL; 1737 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1738 1739 // iterate through all loadable segments 1740 for (i = 0; i < info->dlpi_phnum; i++) { 1741 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1742 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1743 // base address of a library is the lowest address of its loaded 1744 // segments. 1745 if (libbase == NULL || libbase > segbase) { 1746 libbase = segbase; 1747 } 1748 // see if 'addr' is within current segment 1749 if (segbase <= d->addr && 1750 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1751 found = true; 1752 } 1753 } 1754 } 1755 1756 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1757 // so dll_address_to_library_name() can fall through to use dladdr() which 1758 // can figure out executable name from argv[0]. 1759 if (found && info->dlpi_name && info->dlpi_name[0]) { 1760 d->base = libbase; 1761 if (d->fname) { 1762 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1763 } 1764 return 1; 1765 } 1766 return 0; 1767} 1768 1769bool os::dll_address_to_library_name(address addr, char* buf, 1770 int buflen, int* offset) { 1771 Dl_info dlinfo; 1772 struct _address_to_library_name data; 1773 1774 // There is a bug in old glibc dladdr() implementation that it could resolve 1775 // to wrong library name if the .so file has a base address != NULL. Here 1776 // we iterate through the program headers of all loaded libraries to find 1777 // out which library 'addr' really belongs to. This workaround can be 1778 // removed once the minimum requirement for glibc is moved to 2.3.x. 1779 data.addr = addr; 1780 data.fname = buf; 1781 data.buflen = buflen; 1782 data.base = NULL; 1783 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1784 1785 if (rslt) { 1786 // buf already contains library name 1787 if (offset) *offset = addr - data.base; 1788 return true; 1789 } else if (dladdr((void*)addr, &dlinfo)){ 1790 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1791 if (offset) *offset = addr - (address)dlinfo.dli_fbase; 1792 return true; 1793 } else { 1794 if (buf) buf[0] = '\0'; 1795 if (offset) *offset = -1; 1796 return false; 1797 } 1798} 1799 1800 // Loads .dll/.so and 1801 // in case of error it checks if .dll/.so was built for the 1802 // same architecture as Hotspot is running on 1803 1804 1805// Remember the stack's state. The Linux dynamic linker will change 1806// the stack to 'executable' at most once, so we must safepoint only once. 1807bool os::Linux::_stack_is_executable = false; 1808 1809// VM operation that loads a library. This is necessary if stack protection 1810// of the Java stacks can be lost during loading the library. If we 1811// do not stop the Java threads, they can stack overflow before the stacks 1812// are protected again. 1813class VM_LinuxDllLoad: public VM_Operation { 1814 private: 1815 const char *_filename; 1816 char *_ebuf; 1817 int _ebuflen; 1818 void *_lib; 1819 public: 1820 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1821 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1822 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1823 void doit() { 1824 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1825 os::Linux::_stack_is_executable = true; 1826 } 1827 void* loaded_library() { return _lib; } 1828}; 1829 1830void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1831{ 1832 void * result = NULL; 1833 bool load_attempted = false; 1834 1835 // Check whether the library to load might change execution rights 1836 // of the stack. If they are changed, the protection of the stack 1837 // guard pages will be lost. We need a safepoint to fix this. 1838 // 1839 // See Linux man page execstack(8) for more info. 1840 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1841 ElfFile ef(filename); 1842 if (!ef.specifies_noexecstack()) { 1843 if (!is_init_completed()) { 1844 os::Linux::_stack_is_executable = true; 1845 // This is OK - No Java threads have been created yet, and hence no 1846 // stack guard pages to fix. 1847 // 1848 // This should happen only when you are building JDK7 using a very 1849 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1850 // 1851 // Dynamic loader will make all stacks executable after 1852 // this function returns, and will not do that again. 1853 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1854 } else { 1855 warning("You have loaded library %s which might have disabled stack guard. " 1856 "The VM will try to fix the stack guard now.\n" 1857 "It's highly recommended that you fix the library with " 1858 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1859 filename); 1860 1861 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1862 JavaThread *jt = JavaThread::current(); 1863 if (jt->thread_state() != _thread_in_native) { 1864 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1865 // that requires ExecStack. Cannot enter safe point. Let's give up. 1866 warning("Unable to fix stack guard. Giving up."); 1867 } else { 1868 if (!LoadExecStackDllInVMThread) { 1869 // This is for the case where the DLL has an static 1870 // constructor function that executes JNI code. We cannot 1871 // load such DLLs in the VMThread. 1872 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1873 } 1874 1875 ThreadInVMfromNative tiv(jt); 1876 debug_only(VMNativeEntryWrapper vew;) 1877 1878 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1879 VMThread::execute(&op); 1880 if (LoadExecStackDllInVMThread) { 1881 result = op.loaded_library(); 1882 } 1883 load_attempted = true; 1884 } 1885 } 1886 } 1887 } 1888 1889 if (!load_attempted) { 1890 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1891 } 1892 1893 if (result != NULL) { 1894 // Successful loading 1895 return result; 1896 } 1897 1898 Elf32_Ehdr elf_head; 1899 int diag_msg_max_length=ebuflen-strlen(ebuf); 1900 char* diag_msg_buf=ebuf+strlen(ebuf); 1901 1902 if (diag_msg_max_length==0) { 1903 // No more space in ebuf for additional diagnostics message 1904 return NULL; 1905 } 1906 1907 1908 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1909 1910 if (file_descriptor < 0) { 1911 // Can't open library, report dlerror() message 1912 return NULL; 1913 } 1914 1915 bool failed_to_read_elf_head= 1916 (sizeof(elf_head)!= 1917 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1918 1919 ::close(file_descriptor); 1920 if (failed_to_read_elf_head) { 1921 // file i/o error - report dlerror() msg 1922 return NULL; 1923 } 1924 1925 typedef struct { 1926 Elf32_Half code; // Actual value as defined in elf.h 1927 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1928 char elf_class; // 32 or 64 bit 1929 char endianess; // MSB or LSB 1930 char* name; // String representation 1931 } arch_t; 1932 1933 #ifndef EM_486 1934 #define EM_486 6 /* Intel 80486 */ 1935 #endif 1936 1937 static const arch_t arch_array[]={ 1938 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1939 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1940 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1941 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1942 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1943 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1944 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1945 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1946 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1947 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1948 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1949 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1950 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1951 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1952 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1953 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1954 }; 1955 1956 #if (defined IA32) 1957 static Elf32_Half running_arch_code=EM_386; 1958 #elif (defined AMD64) 1959 static Elf32_Half running_arch_code=EM_X86_64; 1960 #elif (defined IA64) 1961 static Elf32_Half running_arch_code=EM_IA_64; 1962 #elif (defined __sparc) && (defined _LP64) 1963 static Elf32_Half running_arch_code=EM_SPARCV9; 1964 #elif (defined __sparc) && (!defined _LP64) 1965 static Elf32_Half running_arch_code=EM_SPARC; 1966 #elif (defined __powerpc64__) 1967 static Elf32_Half running_arch_code=EM_PPC64; 1968 #elif (defined __powerpc__) 1969 static Elf32_Half running_arch_code=EM_PPC; 1970 #elif (defined ARM) 1971 static Elf32_Half running_arch_code=EM_ARM; 1972 #elif (defined S390) 1973 static Elf32_Half running_arch_code=EM_S390; 1974 #elif (defined ALPHA) 1975 static Elf32_Half running_arch_code=EM_ALPHA; 1976 #elif (defined MIPSEL) 1977 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1978 #elif (defined PARISC) 1979 static Elf32_Half running_arch_code=EM_PARISC; 1980 #elif (defined MIPS) 1981 static Elf32_Half running_arch_code=EM_MIPS; 1982 #elif (defined M68K) 1983 static Elf32_Half running_arch_code=EM_68K; 1984 #else 1985 #error Method os::dll_load requires that one of following is defined:\ 1986 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 1987 #endif 1988 1989 // Identify compatability class for VM's architecture and library's architecture 1990 // Obtain string descriptions for architectures 1991 1992 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1993 int running_arch_index=-1; 1994 1995 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1996 if (running_arch_code == arch_array[i].code) { 1997 running_arch_index = i; 1998 } 1999 if (lib_arch.code == arch_array[i].code) { 2000 lib_arch.compat_class = arch_array[i].compat_class; 2001 lib_arch.name = arch_array[i].name; 2002 } 2003 } 2004 2005 assert(running_arch_index != -1, 2006 "Didn't find running architecture code (running_arch_code) in arch_array"); 2007 if (running_arch_index == -1) { 2008 // Even though running architecture detection failed 2009 // we may still continue with reporting dlerror() message 2010 return NULL; 2011 } 2012 2013 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 2014 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 2015 return NULL; 2016 } 2017 2018#ifndef S390 2019 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 2020 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 2021 return NULL; 2022 } 2023#endif // !S390 2024 2025 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2026 if ( lib_arch.name!=NULL ) { 2027 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2028 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2029 lib_arch.name, arch_array[running_arch_index].name); 2030 } else { 2031 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2032 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2033 lib_arch.code, 2034 arch_array[running_arch_index].name); 2035 } 2036 } 2037 2038 return NULL; 2039} 2040 2041void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) { 2042 void * result = ::dlopen(filename, RTLD_LAZY); 2043 if (result == NULL) { 2044 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2045 ebuf[ebuflen-1] = '\0'; 2046 } 2047 return result; 2048} 2049 2050void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) { 2051 void * result = NULL; 2052 if (LoadExecStackDllInVMThread) { 2053 result = dlopen_helper(filename, ebuf, ebuflen); 2054 } 2055 2056 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2057 // library that requires an executable stack, or which does not have this 2058 // stack attribute set, dlopen changes the stack attribute to executable. The 2059 // read protection of the guard pages gets lost. 2060 // 2061 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2062 // may have been queued at the same time. 2063 2064 if (!_stack_is_executable) { 2065 JavaThread *jt = Threads::first(); 2066 2067 while (jt) { 2068 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2069 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2070 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2071 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2072 warning("Attempt to reguard stack yellow zone failed."); 2073 } 2074 } 2075 jt = jt->next(); 2076 } 2077 } 2078 2079 return result; 2080} 2081 2082/* 2083 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2084 * chances are you might want to run the generated bits against glibc-2.0 2085 * libdl.so, so always use locking for any version of glibc. 2086 */ 2087void* os::dll_lookup(void* handle, const char* name) { 2088 pthread_mutex_lock(&dl_mutex); 2089 void* res = dlsym(handle, name); 2090 pthread_mutex_unlock(&dl_mutex); 2091 return res; 2092} 2093 2094 2095static bool _print_ascii_file(const char* filename, outputStream* st) { 2096 int fd = ::open(filename, O_RDONLY); 2097 if (fd == -1) { 2098 return false; 2099 } 2100 2101 char buf[32]; 2102 int bytes; 2103 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2104 st->print_raw(buf, bytes); 2105 } 2106 2107 ::close(fd); 2108 2109 return true; 2110} 2111 2112void os::print_dll_info(outputStream *st) { 2113 st->print_cr("Dynamic libraries:"); 2114 2115 char fname[32]; 2116 pid_t pid = os::Linux::gettid(); 2117 2118 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2119 2120 if (!_print_ascii_file(fname, st)) { 2121 st->print("Can not get library information for pid = %d\n", pid); 2122 } 2123} 2124 2125void os::print_os_info_brief(outputStream* st) { 2126 os::Linux::print_distro_info(st); 2127 2128 os::Posix::print_uname_info(st); 2129 2130 os::Linux::print_libversion_info(st); 2131 2132} 2133 2134void os::print_os_info(outputStream* st) { 2135 st->print("OS:"); 2136 2137 os::Linux::print_distro_info(st); 2138 2139 os::Posix::print_uname_info(st); 2140 2141 // Print warning if unsafe chroot environment detected 2142 if (unsafe_chroot_detected) { 2143 st->print("WARNING!! "); 2144 st->print_cr(unstable_chroot_error); 2145 } 2146 2147 os::Linux::print_libversion_info(st); 2148 2149 os::Posix::print_rlimit_info(st); 2150 2151 os::Posix::print_load_average(st); 2152 2153 os::Linux::print_full_memory_info(st); 2154} 2155 2156// Try to identify popular distros. 2157// Most Linux distributions have /etc/XXX-release file, which contains 2158// the OS version string. Some have more than one /etc/XXX-release file 2159// (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 2160// so the order is important. 2161void os::Linux::print_distro_info(outputStream* st) { 2162 if (!_print_ascii_file("/etc/mandrake-release", st) && 2163 !_print_ascii_file("/etc/sun-release", st) && 2164 !_print_ascii_file("/etc/redhat-release", st) && 2165 !_print_ascii_file("/etc/SuSE-release", st) && 2166 !_print_ascii_file("/etc/turbolinux-release", st) && 2167 !_print_ascii_file("/etc/gentoo-release", st) && 2168 !_print_ascii_file("/etc/debian_version", st) && 2169 !_print_ascii_file("/etc/ltib-release", st) && 2170 !_print_ascii_file("/etc/angstrom-version", st)) { 2171 st->print("Linux"); 2172 } 2173 st->cr(); 2174} 2175 2176void os::Linux::print_libversion_info(outputStream* st) { 2177 // libc, pthread 2178 st->print("libc:"); 2179 st->print(os::Linux::glibc_version()); st->print(" "); 2180 st->print(os::Linux::libpthread_version()); st->print(" "); 2181 if (os::Linux::is_LinuxThreads()) { 2182 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2183 } 2184 st->cr(); 2185} 2186 2187void os::Linux::print_full_memory_info(outputStream* st) { 2188 st->print("\n/proc/meminfo:\n"); 2189 _print_ascii_file("/proc/meminfo", st); 2190 st->cr(); 2191} 2192 2193void os::print_memory_info(outputStream* st) { 2194 2195 st->print("Memory:"); 2196 st->print(" %dk page", os::vm_page_size()>>10); 2197 2198 // values in struct sysinfo are "unsigned long" 2199 struct sysinfo si; 2200 sysinfo(&si); 2201 2202 st->print(", physical " UINT64_FORMAT "k", 2203 os::physical_memory() >> 10); 2204 st->print("(" UINT64_FORMAT "k free)", 2205 os::available_memory() >> 10); 2206 st->print(", swap " UINT64_FORMAT "k", 2207 ((jlong)si.totalswap * si.mem_unit) >> 10); 2208 st->print("(" UINT64_FORMAT "k free)", 2209 ((jlong)si.freeswap * si.mem_unit) >> 10); 2210 st->cr(); 2211} 2212 2213void os::pd_print_cpu_info(outputStream* st) { 2214 st->print("\n/proc/cpuinfo:\n"); 2215 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2216 st->print(" <Not Available>"); 2217 } 2218 st->cr(); 2219} 2220 2221// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 2222// but they're the same for all the linux arch that we support 2223// and they're the same for solaris but there's no common place to put this. 2224const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 2225 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 2226 "ILL_COPROC", "ILL_BADSTK" }; 2227 2228const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 2229 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 2230 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 2231 2232const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 2233 2234const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 2235 2236void os::print_siginfo(outputStream* st, void* siginfo) { 2237 st->print("siginfo:"); 2238 2239 const int buflen = 100; 2240 char buf[buflen]; 2241 siginfo_t *si = (siginfo_t*)siginfo; 2242 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 2243 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 2244 st->print("si_errno=%s", buf); 2245 } else { 2246 st->print("si_errno=%d", si->si_errno); 2247 } 2248 const int c = si->si_code; 2249 assert(c > 0, "unexpected si_code"); 2250 switch (si->si_signo) { 2251 case SIGILL: 2252 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 2253 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2254 break; 2255 case SIGFPE: 2256 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 2257 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2258 break; 2259 case SIGSEGV: 2260 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 2261 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2262 break; 2263 case SIGBUS: 2264 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 2265 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2266 break; 2267 default: 2268 st->print(", si_code=%d", si->si_code); 2269 // no si_addr 2270 } 2271 2272 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2273 UseSharedSpaces) { 2274 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2275 if (mapinfo->is_in_shared_space(si->si_addr)) { 2276 st->print("\n\nError accessing class data sharing archive." \ 2277 " Mapped file inaccessible during execution, " \ 2278 " possible disk/network problem."); 2279 } 2280 } 2281 st->cr(); 2282} 2283 2284 2285static void print_signal_handler(outputStream* st, int sig, 2286 char* buf, size_t buflen); 2287 2288void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2289 st->print_cr("Signal Handlers:"); 2290 print_signal_handler(st, SIGSEGV, buf, buflen); 2291 print_signal_handler(st, SIGBUS , buf, buflen); 2292 print_signal_handler(st, SIGFPE , buf, buflen); 2293 print_signal_handler(st, SIGPIPE, buf, buflen); 2294 print_signal_handler(st, SIGXFSZ, buf, buflen); 2295 print_signal_handler(st, SIGILL , buf, buflen); 2296 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2297 print_signal_handler(st, SR_signum, buf, buflen); 2298 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2299 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2300 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2301 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2302} 2303 2304static char saved_jvm_path[MAXPATHLEN] = {0}; 2305 2306// Find the full path to the current module, libjvm.so 2307void os::jvm_path(char *buf, jint buflen) { 2308 // Error checking. 2309 if (buflen < MAXPATHLEN) { 2310 assert(false, "must use a large-enough buffer"); 2311 buf[0] = '\0'; 2312 return; 2313 } 2314 // Lazy resolve the path to current module. 2315 if (saved_jvm_path[0] != 0) { 2316 strcpy(buf, saved_jvm_path); 2317 return; 2318 } 2319 2320 char dli_fname[MAXPATHLEN]; 2321 bool ret = dll_address_to_library_name( 2322 CAST_FROM_FN_PTR(address, os::jvm_path), 2323 dli_fname, sizeof(dli_fname), NULL); 2324 assert(ret != 0, "cannot locate libjvm"); 2325 char *rp = realpath(dli_fname, buf); 2326 if (rp == NULL) 2327 return; 2328 2329 if (Arguments::created_by_gamma_launcher()) { 2330 // Support for the gamma launcher. Typical value for buf is 2331 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 2332 // the right place in the string, then assume we are installed in a JDK and 2333 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2334 // up the path so it looks like libjvm.so is installed there (append a 2335 // fake suffix hotspot/libjvm.so). 2336 const char *p = buf + strlen(buf) - 1; 2337 for (int count = 0; p > buf && count < 5; ++count) { 2338 for (--p; p > buf && *p != '/'; --p) 2339 /* empty */ ; 2340 } 2341 2342 if (strncmp(p, "/jre/lib/", 9) != 0) { 2343 // Look for JAVA_HOME in the environment. 2344 char* java_home_var = ::getenv("JAVA_HOME"); 2345 if (java_home_var != NULL && java_home_var[0] != 0) { 2346 char* jrelib_p; 2347 int len; 2348 2349 // Check the current module name "libjvm.so". 2350 p = strrchr(buf, '/'); 2351 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2352 2353 rp = realpath(java_home_var, buf); 2354 if (rp == NULL) 2355 return; 2356 2357 // determine if this is a legacy image or modules image 2358 // modules image doesn't have "jre" subdirectory 2359 len = strlen(buf); 2360 jrelib_p = buf + len; 2361 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2362 if (0 != access(buf, F_OK)) { 2363 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2364 } 2365 2366 if (0 == access(buf, F_OK)) { 2367 // Use current module name "libjvm.so" 2368 len = strlen(buf); 2369 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2370 } else { 2371 // Go back to path of .so 2372 rp = realpath(dli_fname, buf); 2373 if (rp == NULL) 2374 return; 2375 } 2376 } 2377 } 2378 } 2379 2380 strcpy(saved_jvm_path, buf); 2381} 2382 2383void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2384 // no prefix required, not even "_" 2385} 2386 2387void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2388 // no suffix required 2389} 2390 2391//////////////////////////////////////////////////////////////////////////////// 2392// sun.misc.Signal support 2393 2394static volatile jint sigint_count = 0; 2395 2396static void 2397UserHandler(int sig, void *siginfo, void *context) { 2398 // 4511530 - sem_post is serialized and handled by the manager thread. When 2399 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2400 // don't want to flood the manager thread with sem_post requests. 2401 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2402 return; 2403 2404 // Ctrl-C is pressed during error reporting, likely because the error 2405 // handler fails to abort. Let VM die immediately. 2406 if (sig == SIGINT && is_error_reported()) { 2407 os::die(); 2408 } 2409 2410 os::signal_notify(sig); 2411} 2412 2413void* os::user_handler() { 2414 return CAST_FROM_FN_PTR(void*, UserHandler); 2415} 2416 2417extern "C" { 2418 typedef void (*sa_handler_t)(int); 2419 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2420} 2421 2422void* os::signal(int signal_number, void* handler) { 2423 struct sigaction sigAct, oldSigAct; 2424 2425 sigfillset(&(sigAct.sa_mask)); 2426 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2427 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2428 2429 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2430 // -1 means registration failed 2431 return (void *)-1; 2432 } 2433 2434 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2435} 2436 2437void os::signal_raise(int signal_number) { 2438 ::raise(signal_number); 2439} 2440 2441/* 2442 * The following code is moved from os.cpp for making this 2443 * code platform specific, which it is by its very nature. 2444 */ 2445 2446// Will be modified when max signal is changed to be dynamic 2447int os::sigexitnum_pd() { 2448 return NSIG; 2449} 2450 2451// a counter for each possible signal value 2452static volatile jint pending_signals[NSIG+1] = { 0 }; 2453 2454// Linux(POSIX) specific hand shaking semaphore. 2455static sem_t sig_sem; 2456 2457void os::signal_init_pd() { 2458 // Initialize signal structures 2459 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2460 2461 // Initialize signal semaphore 2462 ::sem_init(&sig_sem, 0, 0); 2463} 2464 2465void os::signal_notify(int sig) { 2466 Atomic::inc(&pending_signals[sig]); 2467 ::sem_post(&sig_sem); 2468} 2469 2470static int check_pending_signals(bool wait) { 2471 Atomic::store(0, &sigint_count); 2472 for (;;) { 2473 for (int i = 0; i < NSIG + 1; i++) { 2474 jint n = pending_signals[i]; 2475 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2476 return i; 2477 } 2478 } 2479 if (!wait) { 2480 return -1; 2481 } 2482 JavaThread *thread = JavaThread::current(); 2483 ThreadBlockInVM tbivm(thread); 2484 2485 bool threadIsSuspended; 2486 do { 2487 thread->set_suspend_equivalent(); 2488 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2489 ::sem_wait(&sig_sem); 2490 2491 // were we externally suspended while we were waiting? 2492 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2493 if (threadIsSuspended) { 2494 // 2495 // The semaphore has been incremented, but while we were waiting 2496 // another thread suspended us. We don't want to continue running 2497 // while suspended because that would surprise the thread that 2498 // suspended us. 2499 // 2500 ::sem_post(&sig_sem); 2501 2502 thread->java_suspend_self(); 2503 } 2504 } while (threadIsSuspended); 2505 } 2506} 2507 2508int os::signal_lookup() { 2509 return check_pending_signals(false); 2510} 2511 2512int os::signal_wait() { 2513 return check_pending_signals(true); 2514} 2515 2516//////////////////////////////////////////////////////////////////////////////// 2517// Virtual Memory 2518 2519int os::vm_page_size() { 2520 // Seems redundant as all get out 2521 assert(os::Linux::page_size() != -1, "must call os::init"); 2522 return os::Linux::page_size(); 2523} 2524 2525// Solaris allocates memory by pages. 2526int os::vm_allocation_granularity() { 2527 assert(os::Linux::page_size() != -1, "must call os::init"); 2528 return os::Linux::page_size(); 2529} 2530 2531// Rationale behind this function: 2532// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2533// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2534// samples for JITted code. Here we create private executable mapping over the code cache 2535// and then we can use standard (well, almost, as mapping can change) way to provide 2536// info for the reporting script by storing timestamp and location of symbol 2537void linux_wrap_code(char* base, size_t size) { 2538 static volatile jint cnt = 0; 2539 2540 if (!UseOprofile) { 2541 return; 2542 } 2543 2544 char buf[PATH_MAX+1]; 2545 int num = Atomic::add(1, &cnt); 2546 2547 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2548 os::get_temp_directory(), os::current_process_id(), num); 2549 unlink(buf); 2550 2551 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2552 2553 if (fd != -1) { 2554 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2555 if (rv != (off_t)-1) { 2556 if (::write(fd, "", 1) == 1) { 2557 mmap(base, size, 2558 PROT_READ|PROT_WRITE|PROT_EXEC, 2559 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2560 } 2561 } 2562 ::close(fd); 2563 unlink(buf); 2564 } 2565} 2566 2567// NOTE: Linux kernel does not really reserve the pages for us. 2568// All it does is to check if there are enough free pages 2569// left at the time of mmap(). This could be a potential 2570// problem. 2571bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2572 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2573 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2574 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2575 if (res != (uintptr_t) MAP_FAILED) { 2576 if (UseNUMAInterleaving) { 2577 numa_make_global(addr, size); 2578 } 2579 return true; 2580 } 2581 return false; 2582} 2583 2584// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2585#ifndef MAP_HUGETLB 2586#define MAP_HUGETLB 0x40000 2587#endif 2588 2589// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2590#ifndef MADV_HUGEPAGE 2591#define MADV_HUGEPAGE 14 2592#endif 2593 2594bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2595 bool exec) { 2596 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) { 2597 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2598 uintptr_t res = 2599 (uintptr_t) ::mmap(addr, size, prot, 2600 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB, 2601 -1, 0); 2602 if (res != (uintptr_t) MAP_FAILED) { 2603 if (UseNUMAInterleaving) { 2604 numa_make_global(addr, size); 2605 } 2606 return true; 2607 } 2608 // Fall through and try to use small pages 2609 } 2610 2611 if (commit_memory(addr, size, exec)) { 2612 realign_memory(addr, size, alignment_hint); 2613 return true; 2614 } 2615 return false; 2616} 2617 2618void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2619 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) { 2620 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2621 // be supported or the memory may already be backed by huge pages. 2622 ::madvise(addr, bytes, MADV_HUGEPAGE); 2623 } 2624} 2625 2626void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2627 // This method works by doing an mmap over an existing mmaping and effectively discarding 2628 // the existing pages. However it won't work for SHM-based large pages that cannot be 2629 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2630 // small pages on top of the SHM segment. This method always works for small pages, so we 2631 // allow that in any case. 2632 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) { 2633 commit_memory(addr, bytes, alignment_hint, false); 2634 } 2635} 2636 2637void os::numa_make_global(char *addr, size_t bytes) { 2638 Linux::numa_interleave_memory(addr, bytes); 2639} 2640 2641void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2642 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2643} 2644 2645bool os::numa_topology_changed() { return false; } 2646 2647size_t os::numa_get_groups_num() { 2648 int max_node = Linux::numa_max_node(); 2649 return max_node > 0 ? max_node + 1 : 1; 2650} 2651 2652int os::numa_get_group_id() { 2653 int cpu_id = Linux::sched_getcpu(); 2654 if (cpu_id != -1) { 2655 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2656 if (lgrp_id != -1) { 2657 return lgrp_id; 2658 } 2659 } 2660 return 0; 2661} 2662 2663size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2664 for (size_t i = 0; i < size; i++) { 2665 ids[i] = i; 2666 } 2667 return size; 2668} 2669 2670bool os::get_page_info(char *start, page_info* info) { 2671 return false; 2672} 2673 2674char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2675 return end; 2676} 2677 2678 2679int os::Linux::sched_getcpu_syscall(void) { 2680 unsigned int cpu; 2681 int retval = -1; 2682 2683#if defined(IA32) 2684# ifndef SYS_getcpu 2685# define SYS_getcpu 318 2686# endif 2687 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2688#elif defined(AMD64) 2689// Unfortunately we have to bring all these macros here from vsyscall.h 2690// to be able to compile on old linuxes. 2691# define __NR_vgetcpu 2 2692# define VSYSCALL_START (-10UL << 20) 2693# define VSYSCALL_SIZE 1024 2694# define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2695 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2696 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2697 retval = vgetcpu(&cpu, NULL, NULL); 2698#endif 2699 2700 return (retval == -1) ? retval : cpu; 2701} 2702 2703// Something to do with the numa-aware allocator needs these symbols 2704extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2705extern "C" JNIEXPORT void numa_error(char *where) { } 2706extern "C" JNIEXPORT int fork1() { return fork(); } 2707 2708 2709// If we are running with libnuma version > 2, then we should 2710// be trying to use symbols with versions 1.1 2711// If we are running with earlier version, which did not have symbol versions, 2712// we should use the base version. 2713void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2714 void *f = dlvsym(handle, name, "libnuma_1.1"); 2715 if (f == NULL) { 2716 f = dlsym(handle, name); 2717 } 2718 return f; 2719} 2720 2721bool os::Linux::libnuma_init() { 2722 // sched_getcpu() should be in libc. 2723 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2724 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2725 2726 // If it's not, try a direct syscall. 2727 if (sched_getcpu() == -1) 2728 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall)); 2729 2730 if (sched_getcpu() != -1) { // Does it work? 2731 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2732 if (handle != NULL) { 2733 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2734 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2735 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2736 libnuma_dlsym(handle, "numa_max_node"))); 2737 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2738 libnuma_dlsym(handle, "numa_available"))); 2739 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2740 libnuma_dlsym(handle, "numa_tonode_memory"))); 2741 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2742 libnuma_dlsym(handle, "numa_interleave_memory"))); 2743 2744 2745 if (numa_available() != -1) { 2746 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2747 // Create a cpu -> node mapping 2748 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2749 rebuild_cpu_to_node_map(); 2750 return true; 2751 } 2752 } 2753 } 2754 return false; 2755} 2756 2757// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2758// The table is later used in get_node_by_cpu(). 2759void os::Linux::rebuild_cpu_to_node_map() { 2760 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2761 // in libnuma (possible values are starting from 16, 2762 // and continuing up with every other power of 2, but less 2763 // than the maximum number of CPUs supported by kernel), and 2764 // is a subject to change (in libnuma version 2 the requirements 2765 // are more reasonable) we'll just hardcode the number they use 2766 // in the library. 2767 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2768 2769 size_t cpu_num = os::active_processor_count(); 2770 size_t cpu_map_size = NCPUS / BitsPerCLong; 2771 size_t cpu_map_valid_size = 2772 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2773 2774 cpu_to_node()->clear(); 2775 cpu_to_node()->at_grow(cpu_num - 1); 2776 size_t node_num = numa_get_groups_num(); 2777 2778 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2779 for (size_t i = 0; i < node_num; i++) { 2780 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2781 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2782 if (cpu_map[j] != 0) { 2783 for (size_t k = 0; k < BitsPerCLong; k++) { 2784 if (cpu_map[j] & (1UL << k)) { 2785 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2786 } 2787 } 2788 } 2789 } 2790 } 2791 } 2792 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2793} 2794 2795int os::Linux::get_node_by_cpu(int cpu_id) { 2796 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2797 return cpu_to_node()->at(cpu_id); 2798 } 2799 return -1; 2800} 2801 2802GrowableArray<int>* os::Linux::_cpu_to_node; 2803os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2804os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2805os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2806os::Linux::numa_available_func_t os::Linux::_numa_available; 2807os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2808os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2809unsigned long* os::Linux::_numa_all_nodes; 2810 2811bool os::pd_uncommit_memory(char* addr, size_t size) { 2812 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2813 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2814 return res != (uintptr_t) MAP_FAILED; 2815} 2816 2817// Linux uses a growable mapping for the stack, and if the mapping for 2818// the stack guard pages is not removed when we detach a thread the 2819// stack cannot grow beyond the pages where the stack guard was 2820// mapped. If at some point later in the process the stack expands to 2821// that point, the Linux kernel cannot expand the stack any further 2822// because the guard pages are in the way, and a segfault occurs. 2823// 2824// However, it's essential not to split the stack region by unmapping 2825// a region (leaving a hole) that's already part of the stack mapping, 2826// so if the stack mapping has already grown beyond the guard pages at 2827// the time we create them, we have to truncate the stack mapping. 2828// So, we need to know the extent of the stack mapping when 2829// create_stack_guard_pages() is called. 2830 2831// Find the bounds of the stack mapping. Return true for success. 2832// 2833// We only need this for stacks that are growable: at the time of 2834// writing thread stacks don't use growable mappings (i.e. those 2835// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 2836// only applies to the main thread. 2837 2838static 2839bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) { 2840 2841 char buf[128]; 2842 int fd, sz; 2843 2844 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) { 2845 return false; 2846 } 2847 2848 const char kw[] = "[stack]"; 2849 const int kwlen = sizeof(kw)-1; 2850 2851 // Address part of /proc/self/maps couldn't be more than 128 bytes 2852 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) { 2853 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) { 2854 // Extract addresses 2855 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) { 2856 uintptr_t sp = (uintptr_t) __builtin_frame_address(0); 2857 if (sp >= *bottom && sp <= *top) { 2858 ::close(fd); 2859 return true; 2860 } 2861 } 2862 } 2863 } 2864 2865 ::close(fd); 2866 return false; 2867} 2868 2869 2870// If the (growable) stack mapping already extends beyond the point 2871// where we're going to put our guard pages, truncate the mapping at 2872// that point by munmap()ping it. This ensures that when we later 2873// munmap() the guard pages we don't leave a hole in the stack 2874// mapping. This only affects the main/initial thread, but guard 2875// against future OS changes 2876bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 2877 uintptr_t stack_extent, stack_base; 2878 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2879 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2880 assert(os::Linux::is_initial_thread(), 2881 "growable stack in non-initial thread"); 2882 if (stack_extent < (uintptr_t)addr) 2883 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent); 2884 } 2885 2886 return os::commit_memory(addr, size); 2887} 2888 2889// If this is a growable mapping, remove the guard pages entirely by 2890// munmap()ping them. If not, just call uncommit_memory(). This only 2891// affects the main/initial thread, but guard against future OS changes 2892bool os::remove_stack_guard_pages(char* addr, size_t size) { 2893 uintptr_t stack_extent, stack_base; 2894 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2895 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2896 assert(os::Linux::is_initial_thread(), 2897 "growable stack in non-initial thread"); 2898 2899 return ::munmap(addr, size) == 0; 2900 } 2901 2902 return os::uncommit_memory(addr, size); 2903} 2904 2905static address _highest_vm_reserved_address = NULL; 2906 2907// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2908// at 'requested_addr'. If there are existing memory mappings at the same 2909// location, however, they will be overwritten. If 'fixed' is false, 2910// 'requested_addr' is only treated as a hint, the return value may or 2911// may not start from the requested address. Unlike Linux mmap(), this 2912// function returns NULL to indicate failure. 2913static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2914 char * addr; 2915 int flags; 2916 2917 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2918 if (fixed) { 2919 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2920 flags |= MAP_FIXED; 2921 } 2922 2923 // Map uncommitted pages PROT_READ and PROT_WRITE, change access 2924 // to PROT_EXEC if executable when we commit the page. 2925 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE, 2926 flags, -1, 0); 2927 2928 if (addr != MAP_FAILED) { 2929 // anon_mmap() should only get called during VM initialization, 2930 // don't need lock (actually we can skip locking even it can be called 2931 // from multiple threads, because _highest_vm_reserved_address is just a 2932 // hint about the upper limit of non-stack memory regions.) 2933 if ((address)addr + bytes > _highest_vm_reserved_address) { 2934 _highest_vm_reserved_address = (address)addr + bytes; 2935 } 2936 } 2937 2938 return addr == MAP_FAILED ? NULL : addr; 2939} 2940 2941// Don't update _highest_vm_reserved_address, because there might be memory 2942// regions above addr + size. If so, releasing a memory region only creates 2943// a hole in the address space, it doesn't help prevent heap-stack collision. 2944// 2945static int anon_munmap(char * addr, size_t size) { 2946 return ::munmap(addr, size) == 0; 2947} 2948 2949char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 2950 size_t alignment_hint) { 2951 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 2952} 2953 2954bool os::pd_release_memory(char* addr, size_t size) { 2955 return anon_munmap(addr, size); 2956} 2957 2958static address highest_vm_reserved_address() { 2959 return _highest_vm_reserved_address; 2960} 2961 2962static bool linux_mprotect(char* addr, size_t size, int prot) { 2963 // Linux wants the mprotect address argument to be page aligned. 2964 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 2965 2966 // According to SUSv3, mprotect() should only be used with mappings 2967 // established by mmap(), and mmap() always maps whole pages. Unaligned 2968 // 'addr' likely indicates problem in the VM (e.g. trying to change 2969 // protection of malloc'ed or statically allocated memory). Check the 2970 // caller if you hit this assert. 2971 assert(addr == bottom, "sanity check"); 2972 2973 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 2974 return ::mprotect(bottom, size, prot) == 0; 2975} 2976 2977// Set protections specified 2978bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 2979 bool is_committed) { 2980 unsigned int p = 0; 2981 switch (prot) { 2982 case MEM_PROT_NONE: p = PROT_NONE; break; 2983 case MEM_PROT_READ: p = PROT_READ; break; 2984 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 2985 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 2986 default: 2987 ShouldNotReachHere(); 2988 } 2989 // is_committed is unused. 2990 return linux_mprotect(addr, bytes, p); 2991} 2992 2993bool os::guard_memory(char* addr, size_t size) { 2994 return linux_mprotect(addr, size, PROT_NONE); 2995} 2996 2997bool os::unguard_memory(char* addr, size_t size) { 2998 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 2999} 3000 3001bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3002 bool result = false; 3003 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE, 3004 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3005 -1, 0); 3006 3007 if (p != (void *) -1) { 3008 // We don't know if this really is a huge page or not. 3009 FILE *fp = fopen("/proc/self/maps", "r"); 3010 if (fp) { 3011 while (!feof(fp)) { 3012 char chars[257]; 3013 long x = 0; 3014 if (fgets(chars, sizeof(chars), fp)) { 3015 if (sscanf(chars, "%lx-%*x", &x) == 1 3016 && x == (long)p) { 3017 if (strstr (chars, "hugepage")) { 3018 result = true; 3019 break; 3020 } 3021 } 3022 } 3023 } 3024 fclose(fp); 3025 } 3026 munmap (p, page_size); 3027 if (result) 3028 return true; 3029 } 3030 3031 if (warn) { 3032 warning("HugeTLBFS is not supported by the operating system."); 3033 } 3034 3035 return result; 3036} 3037 3038/* 3039* Set the coredump_filter bits to include largepages in core dump (bit 6) 3040* 3041* From the coredump_filter documentation: 3042* 3043* - (bit 0) anonymous private memory 3044* - (bit 1) anonymous shared memory 3045* - (bit 2) file-backed private memory 3046* - (bit 3) file-backed shared memory 3047* - (bit 4) ELF header pages in file-backed private memory areas (it is 3048* effective only if the bit 2 is cleared) 3049* - (bit 5) hugetlb private memory 3050* - (bit 6) hugetlb shared memory 3051*/ 3052static void set_coredump_filter(void) { 3053 FILE *f; 3054 long cdm; 3055 3056 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3057 return; 3058 } 3059 3060 if (fscanf(f, "%lx", &cdm) != 1) { 3061 fclose(f); 3062 return; 3063 } 3064 3065 rewind(f); 3066 3067 if ((cdm & LARGEPAGES_BIT) == 0) { 3068 cdm |= LARGEPAGES_BIT; 3069 fprintf(f, "%#lx", cdm); 3070 } 3071 3072 fclose(f); 3073} 3074 3075// Large page support 3076 3077static size_t _large_page_size = 0; 3078 3079void os::large_page_init() { 3080 if (!UseLargePages) { 3081 UseHugeTLBFS = false; 3082 UseSHM = false; 3083 return; 3084 } 3085 3086 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) { 3087 // If UseLargePages is specified on the command line try both methods, 3088 // if it's default, then try only HugeTLBFS. 3089 if (FLAG_IS_DEFAULT(UseLargePages)) { 3090 UseHugeTLBFS = true; 3091 } else { 3092 UseHugeTLBFS = UseSHM = true; 3093 } 3094 } 3095 3096 if (LargePageSizeInBytes) { 3097 _large_page_size = LargePageSizeInBytes; 3098 } else { 3099 // large_page_size on Linux is used to round up heap size. x86 uses either 3100 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3101 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3102 // page as large as 256M. 3103 // 3104 // Here we try to figure out page size by parsing /proc/meminfo and looking 3105 // for a line with the following format: 3106 // Hugepagesize: 2048 kB 3107 // 3108 // If we can't determine the value (e.g. /proc is not mounted, or the text 3109 // format has been changed), we'll use the largest page size supported by 3110 // the processor. 3111 3112#ifndef ZERO 3113 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3114 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3115#endif // ZERO 3116 3117 FILE *fp = fopen("/proc/meminfo", "r"); 3118 if (fp) { 3119 while (!feof(fp)) { 3120 int x = 0; 3121 char buf[16]; 3122 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3123 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3124 _large_page_size = x * K; 3125 break; 3126 } 3127 } else { 3128 // skip to next line 3129 for (;;) { 3130 int ch = fgetc(fp); 3131 if (ch == EOF || ch == (int)'\n') break; 3132 } 3133 } 3134 } 3135 fclose(fp); 3136 } 3137 } 3138 3139 // print a warning if any large page related flag is specified on command line 3140 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3141 3142 const size_t default_page_size = (size_t)Linux::page_size(); 3143 if (_large_page_size > default_page_size) { 3144 _page_sizes[0] = _large_page_size; 3145 _page_sizes[1] = default_page_size; 3146 _page_sizes[2] = 0; 3147 } 3148 UseHugeTLBFS = UseHugeTLBFS && 3149 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size); 3150 3151 if (UseHugeTLBFS) 3152 UseSHM = false; 3153 3154 UseLargePages = UseHugeTLBFS || UseSHM; 3155 3156 set_coredump_filter(); 3157} 3158 3159#ifndef SHM_HUGETLB 3160#define SHM_HUGETLB 04000 3161#endif 3162 3163char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) { 3164 // "exec" is passed in but not used. Creating the shared image for 3165 // the code cache doesn't have an SHM_X executable permission to check. 3166 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3167 3168 key_t key = IPC_PRIVATE; 3169 char *addr; 3170 3171 bool warn_on_failure = UseLargePages && 3172 (!FLAG_IS_DEFAULT(UseLargePages) || 3173 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 3174 ); 3175 char msg[128]; 3176 3177 // Create a large shared memory region to attach to based on size. 3178 // Currently, size is the total size of the heap 3179 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3180 if (shmid == -1) { 3181 // Possible reasons for shmget failure: 3182 // 1. shmmax is too small for Java heap. 3183 // > check shmmax value: cat /proc/sys/kernel/shmmax 3184 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3185 // 2. not enough large page memory. 3186 // > check available large pages: cat /proc/meminfo 3187 // > increase amount of large pages: 3188 // echo new_value > /proc/sys/vm/nr_hugepages 3189 // Note 1: different Linux may use different name for this property, 3190 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3191 // Note 2: it's possible there's enough physical memory available but 3192 // they are so fragmented after a long run that they can't 3193 // coalesce into large pages. Try to reserve large pages when 3194 // the system is still "fresh". 3195 if (warn_on_failure) { 3196 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3197 warning(msg); 3198 } 3199 return NULL; 3200 } 3201 3202 // attach to the region 3203 addr = (char*)shmat(shmid, req_addr, 0); 3204 int err = errno; 3205 3206 // Remove shmid. If shmat() is successful, the actual shared memory segment 3207 // will be deleted when it's detached by shmdt() or when the process 3208 // terminates. If shmat() is not successful this will remove the shared 3209 // segment immediately. 3210 shmctl(shmid, IPC_RMID, NULL); 3211 3212 if ((intptr_t)addr == -1) { 3213 if (warn_on_failure) { 3214 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3215 warning(msg); 3216 } 3217 return NULL; 3218 } 3219 3220 if ((addr != NULL) && UseNUMAInterleaving) { 3221 numa_make_global(addr, bytes); 3222 } 3223 3224 // The memory is committed 3225 address pc = CALLER_PC; 3226 MemTracker::record_virtual_memory_reserve((address)addr, bytes, pc); 3227 MemTracker::record_virtual_memory_commit((address)addr, bytes, pc); 3228 3229 return addr; 3230} 3231 3232bool os::release_memory_special(char* base, size_t bytes) { 3233 // detaching the SHM segment will also delete it, see reserve_memory_special() 3234 int rslt = shmdt(base); 3235 if (rslt == 0) { 3236 MemTracker::record_virtual_memory_uncommit((address)base, bytes); 3237 MemTracker::record_virtual_memory_release((address)base, bytes); 3238 return true; 3239 } else { 3240 return false; 3241 } 3242} 3243 3244size_t os::large_page_size() { 3245 return _large_page_size; 3246} 3247 3248// HugeTLBFS allows application to commit large page memory on demand; 3249// with SysV SHM the entire memory region must be allocated as shared 3250// memory. 3251bool os::can_commit_large_page_memory() { 3252 return UseHugeTLBFS; 3253} 3254 3255bool os::can_execute_large_page_memory() { 3256 return UseHugeTLBFS; 3257} 3258 3259// Reserve memory at an arbitrary address, only if that area is 3260// available (and not reserved for something else). 3261 3262char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3263 const int max_tries = 10; 3264 char* base[max_tries]; 3265 size_t size[max_tries]; 3266 const size_t gap = 0x000000; 3267 3268 // Assert only that the size is a multiple of the page size, since 3269 // that's all that mmap requires, and since that's all we really know 3270 // about at this low abstraction level. If we need higher alignment, 3271 // we can either pass an alignment to this method or verify alignment 3272 // in one of the methods further up the call chain. See bug 5044738. 3273 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3274 3275 // Repeatedly allocate blocks until the block is allocated at the 3276 // right spot. Give up after max_tries. Note that reserve_memory() will 3277 // automatically update _highest_vm_reserved_address if the call is 3278 // successful. The variable tracks the highest memory address every reserved 3279 // by JVM. It is used to detect heap-stack collision if running with 3280 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3281 // space than needed, it could confuse the collision detecting code. To 3282 // solve the problem, save current _highest_vm_reserved_address and 3283 // calculate the correct value before return. 3284 address old_highest = _highest_vm_reserved_address; 3285 3286 // Linux mmap allows caller to pass an address as hint; give it a try first, 3287 // if kernel honors the hint then we can return immediately. 3288 char * addr = anon_mmap(requested_addr, bytes, false); 3289 if (addr == requested_addr) { 3290 return requested_addr; 3291 } 3292 3293 if (addr != NULL) { 3294 // mmap() is successful but it fails to reserve at the requested address 3295 anon_munmap(addr, bytes); 3296 } 3297 3298 int i; 3299 for (i = 0; i < max_tries; ++i) { 3300 base[i] = reserve_memory(bytes); 3301 3302 if (base[i] != NULL) { 3303 // Is this the block we wanted? 3304 if (base[i] == requested_addr) { 3305 size[i] = bytes; 3306 break; 3307 } 3308 3309 // Does this overlap the block we wanted? Give back the overlapped 3310 // parts and try again. 3311 3312 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3313 if (top_overlap >= 0 && top_overlap < bytes) { 3314 unmap_memory(base[i], top_overlap); 3315 base[i] += top_overlap; 3316 size[i] = bytes - top_overlap; 3317 } else { 3318 size_t bottom_overlap = base[i] + bytes - requested_addr; 3319 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3320 unmap_memory(requested_addr, bottom_overlap); 3321 size[i] = bytes - bottom_overlap; 3322 } else { 3323 size[i] = bytes; 3324 } 3325 } 3326 } 3327 } 3328 3329 // Give back the unused reserved pieces. 3330 3331 for (int j = 0; j < i; ++j) { 3332 if (base[j] != NULL) { 3333 unmap_memory(base[j], size[j]); 3334 } 3335 } 3336 3337 if (i < max_tries) { 3338 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3339 return requested_addr; 3340 } else { 3341 _highest_vm_reserved_address = old_highest; 3342 return NULL; 3343 } 3344} 3345 3346size_t os::read(int fd, void *buf, unsigned int nBytes) { 3347 return ::read(fd, buf, nBytes); 3348} 3349 3350// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 3351// Solaris uses poll(), linux uses park(). 3352// Poll() is likely a better choice, assuming that Thread.interrupt() 3353// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 3354// SIGSEGV, see 4355769. 3355 3356int os::sleep(Thread* thread, jlong millis, bool interruptible) { 3357 assert(thread == Thread::current(), "thread consistency check"); 3358 3359 ParkEvent * const slp = thread->_SleepEvent ; 3360 slp->reset() ; 3361 OrderAccess::fence() ; 3362 3363 if (interruptible) { 3364 jlong prevtime = javaTimeNanos(); 3365 3366 for (;;) { 3367 if (os::is_interrupted(thread, true)) { 3368 return OS_INTRPT; 3369 } 3370 3371 jlong newtime = javaTimeNanos(); 3372 3373 if (newtime - prevtime < 0) { 3374 // time moving backwards, should only happen if no monotonic clock 3375 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3376 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3377 } else { 3378 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3379 } 3380 3381 if(millis <= 0) { 3382 return OS_OK; 3383 } 3384 3385 prevtime = newtime; 3386 3387 { 3388 assert(thread->is_Java_thread(), "sanity check"); 3389 JavaThread *jt = (JavaThread *) thread; 3390 ThreadBlockInVM tbivm(jt); 3391 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 3392 3393 jt->set_suspend_equivalent(); 3394 // cleared by handle_special_suspend_equivalent_condition() or 3395 // java_suspend_self() via check_and_wait_while_suspended() 3396 3397 slp->park(millis); 3398 3399 // were we externally suspended while we were waiting? 3400 jt->check_and_wait_while_suspended(); 3401 } 3402 } 3403 } else { 3404 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 3405 jlong prevtime = javaTimeNanos(); 3406 3407 for (;;) { 3408 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 3409 // the 1st iteration ... 3410 jlong newtime = javaTimeNanos(); 3411 3412 if (newtime - prevtime < 0) { 3413 // time moving backwards, should only happen if no monotonic clock 3414 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3415 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3416 } else { 3417 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3418 } 3419 3420 if(millis <= 0) break ; 3421 3422 prevtime = newtime; 3423 slp->park(millis); 3424 } 3425 return OS_OK ; 3426 } 3427} 3428 3429int os::naked_sleep() { 3430 // %% make the sleep time an integer flag. for now use 1 millisec. 3431 return os::sleep(Thread::current(), 1, false); 3432} 3433 3434// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3435void os::infinite_sleep() { 3436 while (true) { // sleep forever ... 3437 ::sleep(100); // ... 100 seconds at a time 3438 } 3439} 3440 3441// Used to convert frequent JVM_Yield() to nops 3442bool os::dont_yield() { 3443 return DontYieldALot; 3444} 3445 3446void os::yield() { 3447 sched_yield(); 3448} 3449 3450os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3451 3452void os::yield_all(int attempts) { 3453 // Yields to all threads, including threads with lower priorities 3454 // Threads on Linux are all with same priority. The Solaris style 3455 // os::yield_all() with nanosleep(1ms) is not necessary. 3456 sched_yield(); 3457} 3458 3459// Called from the tight loops to possibly influence time-sharing heuristics 3460void os::loop_breaker(int attempts) { 3461 os::yield_all(attempts); 3462} 3463 3464//////////////////////////////////////////////////////////////////////////////// 3465// thread priority support 3466 3467// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3468// only supports dynamic priority, static priority must be zero. For real-time 3469// applications, Linux supports SCHED_RR which allows static priority (1-99). 3470// However, for large multi-threaded applications, SCHED_RR is not only slower 3471// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3472// of 5 runs - Sep 2005). 3473// 3474// The following code actually changes the niceness of kernel-thread/LWP. It 3475// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3476// not the entire user process, and user level threads are 1:1 mapped to kernel 3477// threads. It has always been the case, but could change in the future. For 3478// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3479// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3480 3481int os::java_to_os_priority[CriticalPriority + 1] = { 3482 19, // 0 Entry should never be used 3483 3484 4, // 1 MinPriority 3485 3, // 2 3486 2, // 3 3487 3488 1, // 4 3489 0, // 5 NormPriority 3490 -1, // 6 3491 3492 -2, // 7 3493 -3, // 8 3494 -4, // 9 NearMaxPriority 3495 3496 -5, // 10 MaxPriority 3497 3498 -5 // 11 CriticalPriority 3499}; 3500 3501static int prio_init() { 3502 if (ThreadPriorityPolicy == 1) { 3503 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3504 // if effective uid is not root. Perhaps, a more elegant way of doing 3505 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3506 if (geteuid() != 0) { 3507 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3508 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3509 } 3510 ThreadPriorityPolicy = 0; 3511 } 3512 } 3513 if (UseCriticalJavaThreadPriority) { 3514 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3515 } 3516 return 0; 3517} 3518 3519OSReturn os::set_native_priority(Thread* thread, int newpri) { 3520 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3521 3522 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3523 return (ret == 0) ? OS_OK : OS_ERR; 3524} 3525 3526OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3527 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3528 *priority_ptr = java_to_os_priority[NormPriority]; 3529 return OS_OK; 3530 } 3531 3532 errno = 0; 3533 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3534 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3535} 3536 3537// Hint to the underlying OS that a task switch would not be good. 3538// Void return because it's a hint and can fail. 3539void os::hint_no_preempt() {} 3540 3541//////////////////////////////////////////////////////////////////////////////// 3542// suspend/resume support 3543 3544// the low-level signal-based suspend/resume support is a remnant from the 3545// old VM-suspension that used to be for java-suspension, safepoints etc, 3546// within hotspot. Now there is a single use-case for this: 3547// - calling get_thread_pc() on the VMThread by the flat-profiler task 3548// that runs in the watcher thread. 3549// The remaining code is greatly simplified from the more general suspension 3550// code that used to be used. 3551// 3552// The protocol is quite simple: 3553// - suspend: 3554// - sends a signal to the target thread 3555// - polls the suspend state of the osthread using a yield loop 3556// - target thread signal handler (SR_handler) sets suspend state 3557// and blocks in sigsuspend until continued 3558// - resume: 3559// - sets target osthread state to continue 3560// - sends signal to end the sigsuspend loop in the SR_handler 3561// 3562// Note that the SR_lock plays no role in this suspend/resume protocol. 3563// 3564 3565static void resume_clear_context(OSThread *osthread) { 3566 osthread->set_ucontext(NULL); 3567 osthread->set_siginfo(NULL); 3568 3569 // notify the suspend action is completed, we have now resumed 3570 osthread->sr.clear_suspended(); 3571} 3572 3573static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3574 osthread->set_ucontext(context); 3575 osthread->set_siginfo(siginfo); 3576} 3577 3578// 3579// Handler function invoked when a thread's execution is suspended or 3580// resumed. We have to be careful that only async-safe functions are 3581// called here (Note: most pthread functions are not async safe and 3582// should be avoided.) 3583// 3584// Note: sigwait() is a more natural fit than sigsuspend() from an 3585// interface point of view, but sigwait() prevents the signal hander 3586// from being run. libpthread would get very confused by not having 3587// its signal handlers run and prevents sigwait()'s use with the 3588// mutex granting granting signal. 3589// 3590// Currently only ever called on the VMThread 3591// 3592static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3593 // Save and restore errno to avoid confusing native code with EINTR 3594 // after sigsuspend. 3595 int old_errno = errno; 3596 3597 Thread* thread = Thread::current(); 3598 OSThread* osthread = thread->osthread(); 3599 assert(thread->is_VM_thread(), "Must be VMThread"); 3600 // read current suspend action 3601 int action = osthread->sr.suspend_action(); 3602 if (action == os::Linux::SuspendResume::SR_SUSPEND) { 3603 suspend_save_context(osthread, siginfo, context); 3604 3605 // Notify the suspend action is about to be completed. do_suspend() 3606 // waits until SR_SUSPENDED is set and then returns. We will wait 3607 // here for a resume signal and that completes the suspend-other 3608 // action. do_suspend/do_resume is always called as a pair from 3609 // the same thread - so there are no races 3610 3611 // notify the caller 3612 osthread->sr.set_suspended(); 3613 3614 sigset_t suspend_set; // signals for sigsuspend() 3615 3616 // get current set of blocked signals and unblock resume signal 3617 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3618 sigdelset(&suspend_set, SR_signum); 3619 3620 // wait here until we are resumed 3621 do { 3622 sigsuspend(&suspend_set); 3623 // ignore all returns until we get a resume signal 3624 } while (osthread->sr.suspend_action() != os::Linux::SuspendResume::SR_CONTINUE); 3625 3626 resume_clear_context(osthread); 3627 3628 } else { 3629 assert(action == os::Linux::SuspendResume::SR_CONTINUE, "unexpected sr action"); 3630 // nothing special to do - just leave the handler 3631 } 3632 3633 errno = old_errno; 3634} 3635 3636 3637static int SR_initialize() { 3638 struct sigaction act; 3639 char *s; 3640 /* Get signal number to use for suspend/resume */ 3641 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3642 int sig = ::strtol(s, 0, 10); 3643 if (sig > 0 || sig < _NSIG) { 3644 SR_signum = sig; 3645 } 3646 } 3647 3648 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3649 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3650 3651 sigemptyset(&SR_sigset); 3652 sigaddset(&SR_sigset, SR_signum); 3653 3654 /* Set up signal handler for suspend/resume */ 3655 act.sa_flags = SA_RESTART|SA_SIGINFO; 3656 act.sa_handler = (void (*)(int)) SR_handler; 3657 3658 // SR_signum is blocked by default. 3659 // 4528190 - We also need to block pthread restart signal (32 on all 3660 // supported Linux platforms). Note that LinuxThreads need to block 3661 // this signal for all threads to work properly. So we don't have 3662 // to use hard-coded signal number when setting up the mask. 3663 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 3664 3665 if (sigaction(SR_signum, &act, 0) == -1) { 3666 return -1; 3667 } 3668 3669 // Save signal flag 3670 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 3671 return 0; 3672} 3673 3674 3675// returns true on success and false on error - really an error is fatal 3676// but this seems the normal response to library errors 3677static bool do_suspend(OSThread* osthread) { 3678 // mark as suspended and send signal 3679 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_SUSPEND); 3680 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3681 assert_status(status == 0, status, "pthread_kill"); 3682 3683 // check status and wait until notified of suspension 3684 if (status == 0) { 3685 for (int i = 0; !osthread->sr.is_suspended(); i++) { 3686 os::yield_all(i); 3687 } 3688 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE); 3689 return true; 3690 } 3691 else { 3692 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE); 3693 return false; 3694 } 3695} 3696 3697static void do_resume(OSThread* osthread) { 3698 assert(osthread->sr.is_suspended(), "thread should be suspended"); 3699 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_CONTINUE); 3700 3701 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3702 assert_status(status == 0, status, "pthread_kill"); 3703 // check status and wait unit notified of resumption 3704 if (status == 0) { 3705 for (int i = 0; osthread->sr.is_suspended(); i++) { 3706 os::yield_all(i); 3707 } 3708 } 3709 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE); 3710} 3711 3712//////////////////////////////////////////////////////////////////////////////// 3713// interrupt support 3714 3715void os::interrupt(Thread* thread) { 3716 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3717 "possibility of dangling Thread pointer"); 3718 3719 OSThread* osthread = thread->osthread(); 3720 3721 if (!osthread->interrupted()) { 3722 osthread->set_interrupted(true); 3723 // More than one thread can get here with the same value of osthread, 3724 // resulting in multiple notifications. We do, however, want the store 3725 // to interrupted() to be visible to other threads before we execute unpark(). 3726 OrderAccess::fence(); 3727 ParkEvent * const slp = thread->_SleepEvent ; 3728 if (slp != NULL) slp->unpark() ; 3729 } 3730 3731 // For JSR166. Unpark even if interrupt status already was set 3732 if (thread->is_Java_thread()) 3733 ((JavaThread*)thread)->parker()->unpark(); 3734 3735 ParkEvent * ev = thread->_ParkEvent ; 3736 if (ev != NULL) ev->unpark() ; 3737 3738} 3739 3740bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 3741 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3742 "possibility of dangling Thread pointer"); 3743 3744 OSThread* osthread = thread->osthread(); 3745 3746 bool interrupted = osthread->interrupted(); 3747 3748 if (interrupted && clear_interrupted) { 3749 osthread->set_interrupted(false); 3750 // consider thread->_SleepEvent->reset() ... optional optimization 3751 } 3752 3753 return interrupted; 3754} 3755 3756/////////////////////////////////////////////////////////////////////////////////// 3757// signal handling (except suspend/resume) 3758 3759// This routine may be used by user applications as a "hook" to catch signals. 3760// The user-defined signal handler must pass unrecognized signals to this 3761// routine, and if it returns true (non-zero), then the signal handler must 3762// return immediately. If the flag "abort_if_unrecognized" is true, then this 3763// routine will never retun false (zero), but instead will execute a VM panic 3764// routine kill the process. 3765// 3766// If this routine returns false, it is OK to call it again. This allows 3767// the user-defined signal handler to perform checks either before or after 3768// the VM performs its own checks. Naturally, the user code would be making 3769// a serious error if it tried to handle an exception (such as a null check 3770// or breakpoint) that the VM was generating for its own correct operation. 3771// 3772// This routine may recognize any of the following kinds of signals: 3773// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 3774// It should be consulted by handlers for any of those signals. 3775// 3776// The caller of this routine must pass in the three arguments supplied 3777// to the function referred to in the "sa_sigaction" (not the "sa_handler") 3778// field of the structure passed to sigaction(). This routine assumes that 3779// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 3780// 3781// Note that the VM will print warnings if it detects conflicting signal 3782// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 3783// 3784extern "C" JNIEXPORT int 3785JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 3786 void* ucontext, int abort_if_unrecognized); 3787 3788void signalHandler(int sig, siginfo_t* info, void* uc) { 3789 assert(info != NULL && uc != NULL, "it must be old kernel"); 3790 int orig_errno = errno; // Preserve errno value over signal handler. 3791 JVM_handle_linux_signal(sig, info, uc, true); 3792 errno = orig_errno; 3793} 3794 3795 3796// This boolean allows users to forward their own non-matching signals 3797// to JVM_handle_linux_signal, harmlessly. 3798bool os::Linux::signal_handlers_are_installed = false; 3799 3800// For signal-chaining 3801struct sigaction os::Linux::sigact[MAXSIGNUM]; 3802unsigned int os::Linux::sigs = 0; 3803bool os::Linux::libjsig_is_loaded = false; 3804typedef struct sigaction *(*get_signal_t)(int); 3805get_signal_t os::Linux::get_signal_action = NULL; 3806 3807struct sigaction* os::Linux::get_chained_signal_action(int sig) { 3808 struct sigaction *actp = NULL; 3809 3810 if (libjsig_is_loaded) { 3811 // Retrieve the old signal handler from libjsig 3812 actp = (*get_signal_action)(sig); 3813 } 3814 if (actp == NULL) { 3815 // Retrieve the preinstalled signal handler from jvm 3816 actp = get_preinstalled_handler(sig); 3817 } 3818 3819 return actp; 3820} 3821 3822static bool call_chained_handler(struct sigaction *actp, int sig, 3823 siginfo_t *siginfo, void *context) { 3824 // Call the old signal handler 3825 if (actp->sa_handler == SIG_DFL) { 3826 // It's more reasonable to let jvm treat it as an unexpected exception 3827 // instead of taking the default action. 3828 return false; 3829 } else if (actp->sa_handler != SIG_IGN) { 3830 if ((actp->sa_flags & SA_NODEFER) == 0) { 3831 // automaticlly block the signal 3832 sigaddset(&(actp->sa_mask), sig); 3833 } 3834 3835 sa_handler_t hand; 3836 sa_sigaction_t sa; 3837 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 3838 // retrieve the chained handler 3839 if (siginfo_flag_set) { 3840 sa = actp->sa_sigaction; 3841 } else { 3842 hand = actp->sa_handler; 3843 } 3844 3845 if ((actp->sa_flags & SA_RESETHAND) != 0) { 3846 actp->sa_handler = SIG_DFL; 3847 } 3848 3849 // try to honor the signal mask 3850 sigset_t oset; 3851 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 3852 3853 // call into the chained handler 3854 if (siginfo_flag_set) { 3855 (*sa)(sig, siginfo, context); 3856 } else { 3857 (*hand)(sig); 3858 } 3859 3860 // restore the signal mask 3861 pthread_sigmask(SIG_SETMASK, &oset, 0); 3862 } 3863 // Tell jvm's signal handler the signal is taken care of. 3864 return true; 3865} 3866 3867bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 3868 bool chained = false; 3869 // signal-chaining 3870 if (UseSignalChaining) { 3871 struct sigaction *actp = get_chained_signal_action(sig); 3872 if (actp != NULL) { 3873 chained = call_chained_handler(actp, sig, siginfo, context); 3874 } 3875 } 3876 return chained; 3877} 3878 3879struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 3880 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 3881 return &sigact[sig]; 3882 } 3883 return NULL; 3884} 3885 3886void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 3887 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3888 sigact[sig] = oldAct; 3889 sigs |= (unsigned int)1 << sig; 3890} 3891 3892// for diagnostic 3893int os::Linux::sigflags[MAXSIGNUM]; 3894 3895int os::Linux::get_our_sigflags(int sig) { 3896 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3897 return sigflags[sig]; 3898} 3899 3900void os::Linux::set_our_sigflags(int sig, int flags) { 3901 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3902 sigflags[sig] = flags; 3903} 3904 3905void os::Linux::set_signal_handler(int sig, bool set_installed) { 3906 // Check for overwrite. 3907 struct sigaction oldAct; 3908 sigaction(sig, (struct sigaction*)NULL, &oldAct); 3909 3910 void* oldhand = oldAct.sa_sigaction 3911 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3912 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3913 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 3914 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 3915 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 3916 if (AllowUserSignalHandlers || !set_installed) { 3917 // Do not overwrite; user takes responsibility to forward to us. 3918 return; 3919 } else if (UseSignalChaining) { 3920 // save the old handler in jvm 3921 save_preinstalled_handler(sig, oldAct); 3922 // libjsig also interposes the sigaction() call below and saves the 3923 // old sigaction on it own. 3924 } else { 3925 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 3926 "%#lx for signal %d.", (long)oldhand, sig)); 3927 } 3928 } 3929 3930 struct sigaction sigAct; 3931 sigfillset(&(sigAct.sa_mask)); 3932 sigAct.sa_handler = SIG_DFL; 3933 if (!set_installed) { 3934 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3935 } else { 3936 sigAct.sa_sigaction = signalHandler; 3937 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3938 } 3939 // Save flags, which are set by ours 3940 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3941 sigflags[sig] = sigAct.sa_flags; 3942 3943 int ret = sigaction(sig, &sigAct, &oldAct); 3944 assert(ret == 0, "check"); 3945 3946 void* oldhand2 = oldAct.sa_sigaction 3947 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3948 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3949 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 3950} 3951 3952// install signal handlers for signals that HotSpot needs to 3953// handle in order to support Java-level exception handling. 3954 3955void os::Linux::install_signal_handlers() { 3956 if (!signal_handlers_are_installed) { 3957 signal_handlers_are_installed = true; 3958 3959 // signal-chaining 3960 typedef void (*signal_setting_t)(); 3961 signal_setting_t begin_signal_setting = NULL; 3962 signal_setting_t end_signal_setting = NULL; 3963 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3964 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 3965 if (begin_signal_setting != NULL) { 3966 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3967 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 3968 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 3969 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 3970 libjsig_is_loaded = true; 3971 assert(UseSignalChaining, "should enable signal-chaining"); 3972 } 3973 if (libjsig_is_loaded) { 3974 // Tell libjsig jvm is setting signal handlers 3975 (*begin_signal_setting)(); 3976 } 3977 3978 set_signal_handler(SIGSEGV, true); 3979 set_signal_handler(SIGPIPE, true); 3980 set_signal_handler(SIGBUS, true); 3981 set_signal_handler(SIGILL, true); 3982 set_signal_handler(SIGFPE, true); 3983 set_signal_handler(SIGXFSZ, true); 3984 3985 if (libjsig_is_loaded) { 3986 // Tell libjsig jvm finishes setting signal handlers 3987 (*end_signal_setting)(); 3988 } 3989 3990 // We don't activate signal checker if libjsig is in place, we trust ourselves 3991 // and if UserSignalHandler is installed all bets are off. 3992 // Log that signal checking is off only if -verbose:jni is specified. 3993 if (CheckJNICalls) { 3994 if (libjsig_is_loaded) { 3995 if (PrintJNIResolving) { 3996 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 3997 } 3998 check_signals = false; 3999 } 4000 if (AllowUserSignalHandlers) { 4001 if (PrintJNIResolving) { 4002 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4003 } 4004 check_signals = false; 4005 } 4006 } 4007 } 4008} 4009 4010// This is the fastest way to get thread cpu time on Linux. 4011// Returns cpu time (user+sys) for any thread, not only for current. 4012// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4013// It might work on 2.6.10+ with a special kernel/glibc patch. 4014// For reference, please, see IEEE Std 1003.1-2004: 4015// http://www.unix.org/single_unix_specification 4016 4017jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4018 struct timespec tp; 4019 int rc = os::Linux::clock_gettime(clockid, &tp); 4020 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4021 4022 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4023} 4024 4025///// 4026// glibc on Linux platform uses non-documented flag 4027// to indicate, that some special sort of signal 4028// trampoline is used. 4029// We will never set this flag, and we should 4030// ignore this flag in our diagnostic 4031#ifdef SIGNIFICANT_SIGNAL_MASK 4032#undef SIGNIFICANT_SIGNAL_MASK 4033#endif 4034#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4035 4036static const char* get_signal_handler_name(address handler, 4037 char* buf, int buflen) { 4038 int offset; 4039 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4040 if (found) { 4041 // skip directory names 4042 const char *p1, *p2; 4043 p1 = buf; 4044 size_t len = strlen(os::file_separator()); 4045 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4046 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4047 } else { 4048 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4049 } 4050 return buf; 4051} 4052 4053static void print_signal_handler(outputStream* st, int sig, 4054 char* buf, size_t buflen) { 4055 struct sigaction sa; 4056 4057 sigaction(sig, NULL, &sa); 4058 4059 // See comment for SIGNIFICANT_SIGNAL_MASK define 4060 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4061 4062 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4063 4064 address handler = (sa.sa_flags & SA_SIGINFO) 4065 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4066 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4067 4068 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4069 st->print("SIG_DFL"); 4070 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4071 st->print("SIG_IGN"); 4072 } else { 4073 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4074 } 4075 4076 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 4077 4078 address rh = VMError::get_resetted_sighandler(sig); 4079 // May be, handler was resetted by VMError? 4080 if(rh != NULL) { 4081 handler = rh; 4082 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4083 } 4084 4085 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 4086 4087 // Check: is it our handler? 4088 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4089 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4090 // It is our signal handler 4091 // check for flags, reset system-used one! 4092 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4093 st->print( 4094 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4095 os::Linux::get_our_sigflags(sig)); 4096 } 4097 } 4098 st->cr(); 4099} 4100 4101 4102#define DO_SIGNAL_CHECK(sig) \ 4103 if (!sigismember(&check_signal_done, sig)) \ 4104 os::Linux::check_signal_handler(sig) 4105 4106// This method is a periodic task to check for misbehaving JNI applications 4107// under CheckJNI, we can add any periodic checks here 4108 4109void os::run_periodic_checks() { 4110 4111 if (check_signals == false) return; 4112 4113 // SEGV and BUS if overridden could potentially prevent 4114 // generation of hs*.log in the event of a crash, debugging 4115 // such a case can be very challenging, so we absolutely 4116 // check the following for a good measure: 4117 DO_SIGNAL_CHECK(SIGSEGV); 4118 DO_SIGNAL_CHECK(SIGILL); 4119 DO_SIGNAL_CHECK(SIGFPE); 4120 DO_SIGNAL_CHECK(SIGBUS); 4121 DO_SIGNAL_CHECK(SIGPIPE); 4122 DO_SIGNAL_CHECK(SIGXFSZ); 4123 4124 4125 // ReduceSignalUsage allows the user to override these handlers 4126 // see comments at the very top and jvm_solaris.h 4127 if (!ReduceSignalUsage) { 4128 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4129 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4130 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4131 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4132 } 4133 4134 DO_SIGNAL_CHECK(SR_signum); 4135 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4136} 4137 4138typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4139 4140static os_sigaction_t os_sigaction = NULL; 4141 4142void os::Linux::check_signal_handler(int sig) { 4143 char buf[O_BUFLEN]; 4144 address jvmHandler = NULL; 4145 4146 4147 struct sigaction act; 4148 if (os_sigaction == NULL) { 4149 // only trust the default sigaction, in case it has been interposed 4150 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4151 if (os_sigaction == NULL) return; 4152 } 4153 4154 os_sigaction(sig, (struct sigaction*)NULL, &act); 4155 4156 4157 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4158 4159 address thisHandler = (act.sa_flags & SA_SIGINFO) 4160 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4161 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 4162 4163 4164 switch(sig) { 4165 case SIGSEGV: 4166 case SIGBUS: 4167 case SIGFPE: 4168 case SIGPIPE: 4169 case SIGILL: 4170 case SIGXFSZ: 4171 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4172 break; 4173 4174 case SHUTDOWN1_SIGNAL: 4175 case SHUTDOWN2_SIGNAL: 4176 case SHUTDOWN3_SIGNAL: 4177 case BREAK_SIGNAL: 4178 jvmHandler = (address)user_handler(); 4179 break; 4180 4181 case INTERRUPT_SIGNAL: 4182 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4183 break; 4184 4185 default: 4186 if (sig == SR_signum) { 4187 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4188 } else { 4189 return; 4190 } 4191 break; 4192 } 4193 4194 if (thisHandler != jvmHandler) { 4195 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4196 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4197 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4198 // No need to check this sig any longer 4199 sigaddset(&check_signal_done, sig); 4200 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4201 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4202 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4203 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4204 // No need to check this sig any longer 4205 sigaddset(&check_signal_done, sig); 4206 } 4207 4208 // Dump all the signal 4209 if (sigismember(&check_signal_done, sig)) { 4210 print_signal_handlers(tty, buf, O_BUFLEN); 4211 } 4212} 4213 4214extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 4215 4216extern bool signal_name(int signo, char* buf, size_t len); 4217 4218const char* os::exception_name(int exception_code, char* buf, size_t size) { 4219 if (0 < exception_code && exception_code <= SIGRTMAX) { 4220 // signal 4221 if (!signal_name(exception_code, buf, size)) { 4222 jio_snprintf(buf, size, "SIG%d", exception_code); 4223 } 4224 return buf; 4225 } else { 4226 return NULL; 4227 } 4228} 4229 4230// this is called _before_ the most of global arguments have been parsed 4231void os::init(void) { 4232 char dummy; /* used to get a guess on initial stack address */ 4233// first_hrtime = gethrtime(); 4234 4235 // With LinuxThreads the JavaMain thread pid (primordial thread) 4236 // is different than the pid of the java launcher thread. 4237 // So, on Linux, the launcher thread pid is passed to the VM 4238 // via the sun.java.launcher.pid property. 4239 // Use this property instead of getpid() if it was correctly passed. 4240 // See bug 6351349. 4241 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4242 4243 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4244 4245 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4246 4247 init_random(1234567); 4248 4249 ThreadCritical::initialize(); 4250 4251 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4252 if (Linux::page_size() == -1) { 4253 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4254 strerror(errno))); 4255 } 4256 init_page_sizes((size_t) Linux::page_size()); 4257 4258 Linux::initialize_system_info(); 4259 4260 // main_thread points to the aboriginal thread 4261 Linux::_main_thread = pthread_self(); 4262 4263 Linux::clock_init(); 4264 initial_time_count = os::elapsed_counter(); 4265 pthread_mutex_init(&dl_mutex, NULL); 4266} 4267 4268// To install functions for atexit system call 4269extern "C" { 4270 static void perfMemory_exit_helper() { 4271 perfMemory_exit(); 4272 } 4273} 4274 4275// this is called _after_ the global arguments have been parsed 4276jint os::init_2(void) 4277{ 4278 Linux::fast_thread_clock_init(); 4279 4280 // Allocate a single page and mark it as readable for safepoint polling 4281 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4282 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 4283 4284 os::set_polling_page( polling_page ); 4285 4286#ifndef PRODUCT 4287 if(Verbose && PrintMiscellaneous) 4288 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 4289#endif 4290 4291 if (!UseMembar) { 4292 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4293 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); 4294 os::set_memory_serialize_page( mem_serialize_page ); 4295 4296#ifndef PRODUCT 4297 if(Verbose && PrintMiscellaneous) 4298 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 4299#endif 4300 } 4301 4302 os::large_page_init(); 4303 4304 // initialize suspend/resume support - must do this before signal_sets_init() 4305 if (SR_initialize() != 0) { 4306 perror("SR_initialize failed"); 4307 return JNI_ERR; 4308 } 4309 4310 Linux::signal_sets_init(); 4311 Linux::install_signal_handlers(); 4312 4313 // Check minimum allowable stack size for thread creation and to initialize 4314 // the java system classes, including StackOverflowError - depends on page 4315 // size. Add a page for compiler2 recursion in main thread. 4316 // Add in 2*BytesPerWord times page size to account for VM stack during 4317 // class initialization depending on 32 or 64 bit VM. 4318 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4319 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+ 4320 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size()); 4321 4322 size_t threadStackSizeInBytes = ThreadStackSize * K; 4323 if (threadStackSizeInBytes != 0 && 4324 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4325 tty->print_cr("\nThe stack size specified is too small, " 4326 "Specify at least %dk", 4327 os::Linux::min_stack_allowed/ K); 4328 return JNI_ERR; 4329 } 4330 4331 // Make the stack size a multiple of the page size so that 4332 // the yellow/red zones can be guarded. 4333 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4334 vm_page_size())); 4335 4336 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4337 4338 Linux::libpthread_init(); 4339 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4340 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4341 Linux::glibc_version(), Linux::libpthread_version(), 4342 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4343 } 4344 4345 if (UseNUMA) { 4346 if (!Linux::libnuma_init()) { 4347 UseNUMA = false; 4348 } else { 4349 if ((Linux::numa_max_node() < 1)) { 4350 // There's only one node(they start from 0), disable NUMA. 4351 UseNUMA = false; 4352 } 4353 } 4354 // With SHM large pages we cannot uncommit a page, so there's not way 4355 // we can make the adaptive lgrp chunk resizing work. If the user specified 4356 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and 4357 // disable adaptive resizing. 4358 if (UseNUMA && UseLargePages && UseSHM) { 4359 if (!FLAG_IS_DEFAULT(UseNUMA)) { 4360 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) { 4361 UseLargePages = false; 4362 } else { 4363 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing"); 4364 UseAdaptiveSizePolicy = false; 4365 UseAdaptiveNUMAChunkSizing = false; 4366 } 4367 } else { 4368 UseNUMA = false; 4369 } 4370 } 4371 if (!UseNUMA && ForceNUMA) { 4372 UseNUMA = true; 4373 } 4374 } 4375 4376 if (MaxFDLimit) { 4377 // set the number of file descriptors to max. print out error 4378 // if getrlimit/setrlimit fails but continue regardless. 4379 struct rlimit nbr_files; 4380 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4381 if (status != 0) { 4382 if (PrintMiscellaneous && (Verbose || WizardMode)) 4383 perror("os::init_2 getrlimit failed"); 4384 } else { 4385 nbr_files.rlim_cur = nbr_files.rlim_max; 4386 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4387 if (status != 0) { 4388 if (PrintMiscellaneous && (Verbose || WizardMode)) 4389 perror("os::init_2 setrlimit failed"); 4390 } 4391 } 4392 } 4393 4394 // Initialize lock used to serialize thread creation (see os::create_thread) 4395 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4396 4397 // at-exit methods are called in the reverse order of their registration. 4398 // atexit functions are called on return from main or as a result of a 4399 // call to exit(3C). There can be only 32 of these functions registered 4400 // and atexit() does not set errno. 4401 4402 if (PerfAllowAtExitRegistration) { 4403 // only register atexit functions if PerfAllowAtExitRegistration is set. 4404 // atexit functions can be delayed until process exit time, which 4405 // can be problematic for embedded VM situations. Embedded VMs should 4406 // call DestroyJavaVM() to assure that VM resources are released. 4407 4408 // note: perfMemory_exit_helper atexit function may be removed in 4409 // the future if the appropriate cleanup code can be added to the 4410 // VM_Exit VMOperation's doit method. 4411 if (atexit(perfMemory_exit_helper) != 0) { 4412 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 4413 } 4414 } 4415 4416 // initialize thread priority policy 4417 prio_init(); 4418 4419 return JNI_OK; 4420} 4421 4422// this is called at the end of vm_initialization 4423void os::init_3(void) 4424{ 4425#ifdef JAVASE_EMBEDDED 4426 // Start the MemNotifyThread 4427 if (LowMemoryProtection) { 4428 MemNotifyThread::start(); 4429 } 4430 return; 4431#endif 4432} 4433 4434// Mark the polling page as unreadable 4435void os::make_polling_page_unreadable(void) { 4436 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4437 fatal("Could not disable polling page"); 4438}; 4439 4440// Mark the polling page as readable 4441void os::make_polling_page_readable(void) { 4442 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4443 fatal("Could not enable polling page"); 4444 } 4445}; 4446 4447int os::active_processor_count() { 4448 // Linux doesn't yet have a (official) notion of processor sets, 4449 // so just return the number of online processors. 4450 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4451 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4452 return online_cpus; 4453} 4454 4455void os::set_native_thread_name(const char *name) { 4456 // Not yet implemented. 4457 return; 4458} 4459 4460bool os::distribute_processes(uint length, uint* distribution) { 4461 // Not yet implemented. 4462 return false; 4463} 4464 4465bool os::bind_to_processor(uint processor_id) { 4466 // Not yet implemented. 4467 return false; 4468} 4469 4470/// 4471 4472// Suspends the target using the signal mechanism and then grabs the PC before 4473// resuming the target. Used by the flat-profiler only 4474ExtendedPC os::get_thread_pc(Thread* thread) { 4475 // Make sure that it is called by the watcher for the VMThread 4476 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4477 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4478 4479 ExtendedPC epc; 4480 4481 OSThread* osthread = thread->osthread(); 4482 if (do_suspend(osthread)) { 4483 if (osthread->ucontext() != NULL) { 4484 epc = os::Linux::ucontext_get_pc(osthread->ucontext()); 4485 } else { 4486 // NULL context is unexpected, double-check this is the VMThread 4487 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4488 } 4489 do_resume(osthread); 4490 } 4491 // failure means pthread_kill failed for some reason - arguably this is 4492 // a fatal problem, but such problems are ignored elsewhere 4493 4494 return epc; 4495} 4496 4497int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4498{ 4499 if (is_NPTL()) { 4500 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4501 } else { 4502 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4503 // word back to default 64bit precision if condvar is signaled. Java 4504 // wants 53bit precision. Save and restore current value. 4505 int fpu = get_fpu_control_word(); 4506 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4507 set_fpu_control_word(fpu); 4508 return status; 4509 } 4510} 4511 4512//////////////////////////////////////////////////////////////////////////////// 4513// debug support 4514 4515bool os::find(address addr, outputStream* st) { 4516 Dl_info dlinfo; 4517 memset(&dlinfo, 0, sizeof(dlinfo)); 4518 if (dladdr(addr, &dlinfo)) { 4519 st->print(PTR_FORMAT ": ", addr); 4520 if (dlinfo.dli_sname != NULL) { 4521 st->print("%s+%#x", dlinfo.dli_sname, 4522 addr - (intptr_t)dlinfo.dli_saddr); 4523 } else if (dlinfo.dli_fname) { 4524 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4525 } else { 4526 st->print("<absolute address>"); 4527 } 4528 if (dlinfo.dli_fname) { 4529 st->print(" in %s", dlinfo.dli_fname); 4530 } 4531 if (dlinfo.dli_fbase) { 4532 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4533 } 4534 st->cr(); 4535 4536 if (Verbose) { 4537 // decode some bytes around the PC 4538 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 4539 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 4540 address lowest = (address) dlinfo.dli_sname; 4541 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4542 if (begin < lowest) begin = lowest; 4543 Dl_info dlinfo2; 4544 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr 4545 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 4546 end = (address) dlinfo2.dli_saddr; 4547 Disassembler::decode(begin, end, st); 4548 } 4549 return true; 4550 } 4551 return false; 4552} 4553 4554//////////////////////////////////////////////////////////////////////////////// 4555// misc 4556 4557// This does not do anything on Linux. This is basically a hook for being 4558// able to use structured exception handling (thread-local exception filters) 4559// on, e.g., Win32. 4560void 4561os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4562 JavaCallArguments* args, Thread* thread) { 4563 f(value, method, args, thread); 4564} 4565 4566void os::print_statistics() { 4567} 4568 4569int os::message_box(const char* title, const char* message) { 4570 int i; 4571 fdStream err(defaultStream::error_fd()); 4572 for (i = 0; i < 78; i++) err.print_raw("="); 4573 err.cr(); 4574 err.print_raw_cr(title); 4575 for (i = 0; i < 78; i++) err.print_raw("-"); 4576 err.cr(); 4577 err.print_raw_cr(message); 4578 for (i = 0; i < 78; i++) err.print_raw("="); 4579 err.cr(); 4580 4581 char buf[16]; 4582 // Prevent process from exiting upon "read error" without consuming all CPU 4583 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 4584 4585 return buf[0] == 'y' || buf[0] == 'Y'; 4586} 4587 4588int os::stat(const char *path, struct stat *sbuf) { 4589 char pathbuf[MAX_PATH]; 4590 if (strlen(path) > MAX_PATH - 1) { 4591 errno = ENAMETOOLONG; 4592 return -1; 4593 } 4594 os::native_path(strcpy(pathbuf, path)); 4595 return ::stat(pathbuf, sbuf); 4596} 4597 4598bool os::check_heap(bool force) { 4599 return true; 4600} 4601 4602int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 4603 return ::vsnprintf(buf, count, format, args); 4604} 4605 4606// Is a (classpath) directory empty? 4607bool os::dir_is_empty(const char* path) { 4608 DIR *dir = NULL; 4609 struct dirent *ptr; 4610 4611 dir = opendir(path); 4612 if (dir == NULL) return true; 4613 4614 /* Scan the directory */ 4615 bool result = true; 4616 char buf[sizeof(struct dirent) + MAX_PATH]; 4617 while (result && (ptr = ::readdir(dir)) != NULL) { 4618 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 4619 result = false; 4620 } 4621 } 4622 closedir(dir); 4623 return result; 4624} 4625 4626// This code originates from JDK's sysOpen and open64_w 4627// from src/solaris/hpi/src/system_md.c 4628 4629#ifndef O_DELETE 4630#define O_DELETE 0x10000 4631#endif 4632 4633// Open a file. Unlink the file immediately after open returns 4634// if the specified oflag has the O_DELETE flag set. 4635// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 4636 4637int os::open(const char *path, int oflag, int mode) { 4638 4639 if (strlen(path) > MAX_PATH - 1) { 4640 errno = ENAMETOOLONG; 4641 return -1; 4642 } 4643 int fd; 4644 int o_delete = (oflag & O_DELETE); 4645 oflag = oflag & ~O_DELETE; 4646 4647 fd = ::open64(path, oflag, mode); 4648 if (fd == -1) return -1; 4649 4650 //If the open succeeded, the file might still be a directory 4651 { 4652 struct stat64 buf64; 4653 int ret = ::fstat64(fd, &buf64); 4654 int st_mode = buf64.st_mode; 4655 4656 if (ret != -1) { 4657 if ((st_mode & S_IFMT) == S_IFDIR) { 4658 errno = EISDIR; 4659 ::close(fd); 4660 return -1; 4661 } 4662 } else { 4663 ::close(fd); 4664 return -1; 4665 } 4666 } 4667 4668 /* 4669 * All file descriptors that are opened in the JVM and not 4670 * specifically destined for a subprocess should have the 4671 * close-on-exec flag set. If we don't set it, then careless 3rd 4672 * party native code might fork and exec without closing all 4673 * appropriate file descriptors (e.g. as we do in closeDescriptors in 4674 * UNIXProcess.c), and this in turn might: 4675 * 4676 * - cause end-of-file to fail to be detected on some file 4677 * descriptors, resulting in mysterious hangs, or 4678 * 4679 * - might cause an fopen in the subprocess to fail on a system 4680 * suffering from bug 1085341. 4681 * 4682 * (Yes, the default setting of the close-on-exec flag is a Unix 4683 * design flaw) 4684 * 4685 * See: 4686 * 1085341: 32-bit stdio routines should support file descriptors >255 4687 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 4688 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 4689 */ 4690#ifdef FD_CLOEXEC 4691 { 4692 int flags = ::fcntl(fd, F_GETFD); 4693 if (flags != -1) 4694 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 4695 } 4696#endif 4697 4698 if (o_delete != 0) { 4699 ::unlink(path); 4700 } 4701 return fd; 4702} 4703 4704 4705// create binary file, rewriting existing file if required 4706int os::create_binary_file(const char* path, bool rewrite_existing) { 4707 int oflags = O_WRONLY | O_CREAT; 4708 if (!rewrite_existing) { 4709 oflags |= O_EXCL; 4710 } 4711 return ::open64(path, oflags, S_IREAD | S_IWRITE); 4712} 4713 4714// return current position of file pointer 4715jlong os::current_file_offset(int fd) { 4716 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 4717} 4718 4719// move file pointer to the specified offset 4720jlong os::seek_to_file_offset(int fd, jlong offset) { 4721 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 4722} 4723 4724// This code originates from JDK's sysAvailable 4725// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 4726 4727int os::available(int fd, jlong *bytes) { 4728 jlong cur, end; 4729 int mode; 4730 struct stat64 buf64; 4731 4732 if (::fstat64(fd, &buf64) >= 0) { 4733 mode = buf64.st_mode; 4734 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 4735 /* 4736 * XXX: is the following call interruptible? If so, this might 4737 * need to go through the INTERRUPT_IO() wrapper as for other 4738 * blocking, interruptible calls in this file. 4739 */ 4740 int n; 4741 if (::ioctl(fd, FIONREAD, &n) >= 0) { 4742 *bytes = n; 4743 return 1; 4744 } 4745 } 4746 } 4747 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 4748 return 0; 4749 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 4750 return 0; 4751 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 4752 return 0; 4753 } 4754 *bytes = end - cur; 4755 return 1; 4756} 4757 4758int os::socket_available(int fd, jint *pbytes) { 4759 // Linux doc says EINTR not returned, unlike Solaris 4760 int ret = ::ioctl(fd, FIONREAD, pbytes); 4761 4762 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 4763 // is expected to return 0 on failure and 1 on success to the jdk. 4764 return (ret < 0) ? 0 : 1; 4765} 4766 4767// Map a block of memory. 4768char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 4769 char *addr, size_t bytes, bool read_only, 4770 bool allow_exec) { 4771 int prot; 4772 int flags = MAP_PRIVATE; 4773 4774 if (read_only) { 4775 prot = PROT_READ; 4776 } else { 4777 prot = PROT_READ | PROT_WRITE; 4778 } 4779 4780 if (allow_exec) { 4781 prot |= PROT_EXEC; 4782 } 4783 4784 if (addr != NULL) { 4785 flags |= MAP_FIXED; 4786 } 4787 4788 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 4789 fd, file_offset); 4790 if (mapped_address == MAP_FAILED) { 4791 return NULL; 4792 } 4793 return mapped_address; 4794} 4795 4796 4797// Remap a block of memory. 4798char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 4799 char *addr, size_t bytes, bool read_only, 4800 bool allow_exec) { 4801 // same as map_memory() on this OS 4802 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 4803 allow_exec); 4804} 4805 4806 4807// Unmap a block of memory. 4808bool os::pd_unmap_memory(char* addr, size_t bytes) { 4809 return munmap(addr, bytes) == 0; 4810} 4811 4812static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 4813 4814static clockid_t thread_cpu_clockid(Thread* thread) { 4815 pthread_t tid = thread->osthread()->pthread_id(); 4816 clockid_t clockid; 4817 4818 // Get thread clockid 4819 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 4820 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 4821 return clockid; 4822} 4823 4824// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 4825// are used by JVM M&M and JVMTI to get user+sys or user CPU time 4826// of a thread. 4827// 4828// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 4829// the fast estimate available on the platform. 4830 4831jlong os::current_thread_cpu_time() { 4832 if (os::Linux::supports_fast_thread_cpu_time()) { 4833 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4834 } else { 4835 // return user + sys since the cost is the same 4836 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 4837 } 4838} 4839 4840jlong os::thread_cpu_time(Thread* thread) { 4841 // consistent with what current_thread_cpu_time() returns 4842 if (os::Linux::supports_fast_thread_cpu_time()) { 4843 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4844 } else { 4845 return slow_thread_cpu_time(thread, true /* user + sys */); 4846 } 4847} 4848 4849jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 4850 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4851 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4852 } else { 4853 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 4854 } 4855} 4856 4857jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4858 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4859 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4860 } else { 4861 return slow_thread_cpu_time(thread, user_sys_cpu_time); 4862 } 4863} 4864 4865// 4866// -1 on error. 4867// 4868 4869static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4870 static bool proc_task_unchecked = true; 4871 static const char *proc_stat_path = "/proc/%d/stat"; 4872 pid_t tid = thread->osthread()->thread_id(); 4873 char *s; 4874 char stat[2048]; 4875 int statlen; 4876 char proc_name[64]; 4877 int count; 4878 long sys_time, user_time; 4879 char cdummy; 4880 int idummy; 4881 long ldummy; 4882 FILE *fp; 4883 4884 // The /proc/<tid>/stat aggregates per-process usage on 4885 // new Linux kernels 2.6+ where NPTL is supported. 4886 // The /proc/self/task/<tid>/stat still has the per-thread usage. 4887 // See bug 6328462. 4888 // There possibly can be cases where there is no directory 4889 // /proc/self/task, so we check its availability. 4890 if (proc_task_unchecked && os::Linux::is_NPTL()) { 4891 // This is executed only once 4892 proc_task_unchecked = false; 4893 fp = fopen("/proc/self/task", "r"); 4894 if (fp != NULL) { 4895 proc_stat_path = "/proc/self/task/%d/stat"; 4896 fclose(fp); 4897 } 4898 } 4899 4900 sprintf(proc_name, proc_stat_path, tid); 4901 fp = fopen(proc_name, "r"); 4902 if ( fp == NULL ) return -1; 4903 statlen = fread(stat, 1, 2047, fp); 4904 stat[statlen] = '\0'; 4905 fclose(fp); 4906 4907 // Skip pid and the command string. Note that we could be dealing with 4908 // weird command names, e.g. user could decide to rename java launcher 4909 // to "java 1.4.2 :)", then the stat file would look like 4910 // 1234 (java 1.4.2 :)) R ... ... 4911 // We don't really need to know the command string, just find the last 4912 // occurrence of ")" and then start parsing from there. See bug 4726580. 4913 s = strrchr(stat, ')'); 4914 if (s == NULL ) return -1; 4915 4916 // Skip blank chars 4917 do s++; while (isspace(*s)); 4918 4919 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 4920 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 4921 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 4922 &user_time, &sys_time); 4923 if ( count != 13 ) return -1; 4924 if (user_sys_cpu_time) { 4925 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4926 } else { 4927 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4928 } 4929} 4930 4931void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4932 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4933 info_ptr->may_skip_backward = false; // elapsed time not wall time 4934 info_ptr->may_skip_forward = false; // elapsed time not wall time 4935 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4936} 4937 4938void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4939 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4940 info_ptr->may_skip_backward = false; // elapsed time not wall time 4941 info_ptr->may_skip_forward = false; // elapsed time not wall time 4942 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4943} 4944 4945bool os::is_thread_cpu_time_supported() { 4946 return true; 4947} 4948 4949// System loadavg support. Returns -1 if load average cannot be obtained. 4950// Linux doesn't yet have a (official) notion of processor sets, 4951// so just return the system wide load average. 4952int os::loadavg(double loadavg[], int nelem) { 4953 return ::getloadavg(loadavg, nelem); 4954} 4955 4956void os::pause() { 4957 char filename[MAX_PATH]; 4958 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 4959 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 4960 } else { 4961 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 4962 } 4963 4964 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 4965 if (fd != -1) { 4966 struct stat buf; 4967 ::close(fd); 4968 while (::stat(filename, &buf) == 0) { 4969 (void)::poll(NULL, 0, 100); 4970 } 4971 } else { 4972 jio_fprintf(stderr, 4973 "Could not open pause file '%s', continuing immediately.\n", filename); 4974 } 4975} 4976 4977 4978// Refer to the comments in os_solaris.cpp park-unpark. 4979// 4980// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 4981// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 4982// For specifics regarding the bug see GLIBC BUGID 261237 : 4983// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 4984// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 4985// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 4986// is used. (The simple C test-case provided in the GLIBC bug report manifests the 4987// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 4988// and monitorenter when we're using 1-0 locking. All those operations may result in 4989// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 4990// of libpthread avoids the problem, but isn't practical. 4991// 4992// Possible remedies: 4993// 4994// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 4995// This is palliative and probabilistic, however. If the thread is preempted 4996// between the call to compute_abstime() and pthread_cond_timedwait(), more 4997// than the minimum period may have passed, and the abstime may be stale (in the 4998// past) resultin in a hang. Using this technique reduces the odds of a hang 4999// but the JVM is still vulnerable, particularly on heavily loaded systems. 5000// 5001// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5002// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5003// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5004// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5005// thread. 5006// 5007// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5008// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5009// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5010// This also works well. In fact it avoids kernel-level scalability impediments 5011// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5012// timers in a graceful fashion. 5013// 5014// 4. When the abstime value is in the past it appears that control returns 5015// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5016// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5017// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5018// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5019// It may be possible to avoid reinitialization by checking the return 5020// value from pthread_cond_timedwait(). In addition to reinitializing the 5021// condvar we must establish the invariant that cond_signal() is only called 5022// within critical sections protected by the adjunct mutex. This prevents 5023// cond_signal() from "seeing" a condvar that's in the midst of being 5024// reinitialized or that is corrupt. Sadly, this invariant obviates the 5025// desirable signal-after-unlock optimization that avoids futile context switching. 5026// 5027// I'm also concerned that some versions of NTPL might allocate an auxilliary 5028// structure when a condvar is used or initialized. cond_destroy() would 5029// release the helper structure. Our reinitialize-after-timedwait fix 5030// put excessive stress on malloc/free and locks protecting the c-heap. 5031// 5032// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5033// It may be possible to refine (4) by checking the kernel and NTPL verisons 5034// and only enabling the work-around for vulnerable environments. 5035 5036// utility to compute the abstime argument to timedwait: 5037// millis is the relative timeout time 5038// abstime will be the absolute timeout time 5039// TODO: replace compute_abstime() with unpackTime() 5040 5041static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5042 if (millis < 0) millis = 0; 5043 struct timeval now; 5044 int status = gettimeofday(&now, NULL); 5045 assert(status == 0, "gettimeofday"); 5046 jlong seconds = millis / 1000; 5047 millis %= 1000; 5048 if (seconds > 50000000) { // see man cond_timedwait(3T) 5049 seconds = 50000000; 5050 } 5051 abstime->tv_sec = now.tv_sec + seconds; 5052 long usec = now.tv_usec + millis * 1000; 5053 if (usec >= 1000000) { 5054 abstime->tv_sec += 1; 5055 usec -= 1000000; 5056 } 5057 abstime->tv_nsec = usec * 1000; 5058 return abstime; 5059} 5060 5061 5062// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 5063// Conceptually TryPark() should be equivalent to park(0). 5064 5065int os::PlatformEvent::TryPark() { 5066 for (;;) { 5067 const int v = _Event ; 5068 guarantee ((v == 0) || (v == 1), "invariant") ; 5069 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 5070 } 5071} 5072 5073void os::PlatformEvent::park() { // AKA "down()" 5074 // Invariant: Only the thread associated with the Event/PlatformEvent 5075 // may call park(). 5076 // TODO: assert that _Assoc != NULL or _Assoc == Self 5077 int v ; 5078 for (;;) { 5079 v = _Event ; 5080 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5081 } 5082 guarantee (v >= 0, "invariant") ; 5083 if (v == 0) { 5084 // Do this the hard way by blocking ... 5085 int status = pthread_mutex_lock(_mutex); 5086 assert_status(status == 0, status, "mutex_lock"); 5087 guarantee (_nParked == 0, "invariant") ; 5088 ++ _nParked ; 5089 while (_Event < 0) { 5090 status = pthread_cond_wait(_cond, _mutex); 5091 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5092 // Treat this the same as if the wait was interrupted 5093 if (status == ETIME) { status = EINTR; } 5094 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5095 } 5096 -- _nParked ; 5097 5098 _Event = 0 ; 5099 status = pthread_mutex_unlock(_mutex); 5100 assert_status(status == 0, status, "mutex_unlock"); 5101 // Paranoia to ensure our locked and lock-free paths interact 5102 // correctly with each other. 5103 OrderAccess::fence(); 5104 } 5105 guarantee (_Event >= 0, "invariant") ; 5106} 5107 5108int os::PlatformEvent::park(jlong millis) { 5109 guarantee (_nParked == 0, "invariant") ; 5110 5111 int v ; 5112 for (;;) { 5113 v = _Event ; 5114 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5115 } 5116 guarantee (v >= 0, "invariant") ; 5117 if (v != 0) return OS_OK ; 5118 5119 // We do this the hard way, by blocking the thread. 5120 // Consider enforcing a minimum timeout value. 5121 struct timespec abst; 5122 compute_abstime(&abst, millis); 5123 5124 int ret = OS_TIMEOUT; 5125 int status = pthread_mutex_lock(_mutex); 5126 assert_status(status == 0, status, "mutex_lock"); 5127 guarantee (_nParked == 0, "invariant") ; 5128 ++_nParked ; 5129 5130 // Object.wait(timo) will return because of 5131 // (a) notification 5132 // (b) timeout 5133 // (c) thread.interrupt 5134 // 5135 // Thread.interrupt and object.notify{All} both call Event::set. 5136 // That is, we treat thread.interrupt as a special case of notification. 5137 // The underlying Solaris implementation, cond_timedwait, admits 5138 // spurious/premature wakeups, but the JLS/JVM spec prevents the 5139 // JVM from making those visible to Java code. As such, we must 5140 // filter out spurious wakeups. We assume all ETIME returns are valid. 5141 // 5142 // TODO: properly differentiate simultaneous notify+interrupt. 5143 // In that case, we should propagate the notify to another waiter. 5144 5145 while (_Event < 0) { 5146 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5147 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5148 pthread_cond_destroy (_cond); 5149 pthread_cond_init (_cond, NULL) ; 5150 } 5151 assert_status(status == 0 || status == EINTR || 5152 status == ETIME || status == ETIMEDOUT, 5153 status, "cond_timedwait"); 5154 if (!FilterSpuriousWakeups) break ; // previous semantics 5155 if (status == ETIME || status == ETIMEDOUT) break ; 5156 // We consume and ignore EINTR and spurious wakeups. 5157 } 5158 --_nParked ; 5159 if (_Event >= 0) { 5160 ret = OS_OK; 5161 } 5162 _Event = 0 ; 5163 status = pthread_mutex_unlock(_mutex); 5164 assert_status(status == 0, status, "mutex_unlock"); 5165 assert (_nParked == 0, "invariant") ; 5166 // Paranoia to ensure our locked and lock-free paths interact 5167 // correctly with each other. 5168 OrderAccess::fence(); 5169 return ret; 5170} 5171 5172void os::PlatformEvent::unpark() { 5173 // Transitions for _Event: 5174 // 0 :=> 1 5175 // 1 :=> 1 5176 // -1 :=> either 0 or 1; must signal target thread 5177 // That is, we can safely transition _Event from -1 to either 5178 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back 5179 // unpark() calls. 5180 // See also: "Semaphores in Plan 9" by Mullender & Cox 5181 // 5182 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5183 // that it will take two back-to-back park() calls for the owning 5184 // thread to block. This has the benefit of forcing a spurious return 5185 // from the first park() call after an unpark() call which will help 5186 // shake out uses of park() and unpark() without condition variables. 5187 5188 if (Atomic::xchg(1, &_Event) >= 0) return; 5189 5190 // Wait for the thread associated with the event to vacate 5191 int status = pthread_mutex_lock(_mutex); 5192 assert_status(status == 0, status, "mutex_lock"); 5193 int AnyWaiters = _nParked; 5194 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5195 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5196 AnyWaiters = 0; 5197 pthread_cond_signal(_cond); 5198 } 5199 status = pthread_mutex_unlock(_mutex); 5200 assert_status(status == 0, status, "mutex_unlock"); 5201 if (AnyWaiters != 0) { 5202 status = pthread_cond_signal(_cond); 5203 assert_status(status == 0, status, "cond_signal"); 5204 } 5205 5206 // Note that we signal() _after dropping the lock for "immortal" Events. 5207 // This is safe and avoids a common class of futile wakeups. In rare 5208 // circumstances this can cause a thread to return prematurely from 5209 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5210 // simply re-test the condition and re-park itself. 5211} 5212 5213 5214// JSR166 5215// ------------------------------------------------------- 5216 5217/* 5218 * The solaris and linux implementations of park/unpark are fairly 5219 * conservative for now, but can be improved. They currently use a 5220 * mutex/condvar pair, plus a a count. 5221 * Park decrements count if > 0, else does a condvar wait. Unpark 5222 * sets count to 1 and signals condvar. Only one thread ever waits 5223 * on the condvar. Contention seen when trying to park implies that someone 5224 * is unparking you, so don't wait. And spurious returns are fine, so there 5225 * is no need to track notifications. 5226 */ 5227 5228#define MAX_SECS 100000000 5229/* 5230 * This code is common to linux and solaris and will be moved to a 5231 * common place in dolphin. 5232 * 5233 * The passed in time value is either a relative time in nanoseconds 5234 * or an absolute time in milliseconds. Either way it has to be unpacked 5235 * into suitable seconds and nanoseconds components and stored in the 5236 * given timespec structure. 5237 * Given time is a 64-bit value and the time_t used in the timespec is only 5238 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 5239 * overflow if times way in the future are given. Further on Solaris versions 5240 * prior to 10 there is a restriction (see cond_timedwait) that the specified 5241 * number of seconds, in abstime, is less than current_time + 100,000,000. 5242 * As it will be 28 years before "now + 100000000" will overflow we can 5243 * ignore overflow and just impose a hard-limit on seconds using the value 5244 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 5245 * years from "now". 5246 */ 5247 5248static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5249 assert (time > 0, "convertTime"); 5250 5251 struct timeval now; 5252 int status = gettimeofday(&now, NULL); 5253 assert(status == 0, "gettimeofday"); 5254 5255 time_t max_secs = now.tv_sec + MAX_SECS; 5256 5257 if (isAbsolute) { 5258 jlong secs = time / 1000; 5259 if (secs > max_secs) { 5260 absTime->tv_sec = max_secs; 5261 } 5262 else { 5263 absTime->tv_sec = secs; 5264 } 5265 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5266 } 5267 else { 5268 jlong secs = time / NANOSECS_PER_SEC; 5269 if (secs >= MAX_SECS) { 5270 absTime->tv_sec = max_secs; 5271 absTime->tv_nsec = 0; 5272 } 5273 else { 5274 absTime->tv_sec = now.tv_sec + secs; 5275 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5276 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5277 absTime->tv_nsec -= NANOSECS_PER_SEC; 5278 ++absTime->tv_sec; // note: this must be <= max_secs 5279 } 5280 } 5281 } 5282 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5283 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5284 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5285 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5286} 5287 5288void Parker::park(bool isAbsolute, jlong time) { 5289 // Ideally we'd do something useful while spinning, such 5290 // as calling unpackTime(). 5291 5292 // Optional fast-path check: 5293 // Return immediately if a permit is available. 5294 // We depend on Atomic::xchg() having full barrier semantics 5295 // since we are doing a lock-free update to _counter. 5296 if (Atomic::xchg(0, &_counter) > 0) return; 5297 5298 Thread* thread = Thread::current(); 5299 assert(thread->is_Java_thread(), "Must be JavaThread"); 5300 JavaThread *jt = (JavaThread *)thread; 5301 5302 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5303 // Check interrupt before trying to wait 5304 if (Thread::is_interrupted(thread, false)) { 5305 return; 5306 } 5307 5308 // Next, demultiplex/decode time arguments 5309 timespec absTime; 5310 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5311 return; 5312 } 5313 if (time > 0) { 5314 unpackTime(&absTime, isAbsolute, time); 5315 } 5316 5317 5318 // Enter safepoint region 5319 // Beware of deadlocks such as 6317397. 5320 // The per-thread Parker:: mutex is a classic leaf-lock. 5321 // In particular a thread must never block on the Threads_lock while 5322 // holding the Parker:: mutex. If safepoints are pending both the 5323 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5324 ThreadBlockInVM tbivm(jt); 5325 5326 // Don't wait if cannot get lock since interference arises from 5327 // unblocking. Also. check interrupt before trying wait 5328 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5329 return; 5330 } 5331 5332 int status ; 5333 if (_counter > 0) { // no wait needed 5334 _counter = 0; 5335 status = pthread_mutex_unlock(_mutex); 5336 assert (status == 0, "invariant") ; 5337 // Paranoia to ensure our locked and lock-free paths interact 5338 // correctly with each other and Java-level accesses. 5339 OrderAccess::fence(); 5340 return; 5341 } 5342 5343#ifdef ASSERT 5344 // Don't catch signals while blocked; let the running threads have the signals. 5345 // (This allows a debugger to break into the running thread.) 5346 sigset_t oldsigs; 5347 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5348 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5349#endif 5350 5351 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5352 jt->set_suspend_equivalent(); 5353 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5354 5355 if (time == 0) { 5356 status = pthread_cond_wait (_cond, _mutex) ; 5357 } else { 5358 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 5359 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5360 pthread_cond_destroy (_cond) ; 5361 pthread_cond_init (_cond, NULL); 5362 } 5363 } 5364 assert_status(status == 0 || status == EINTR || 5365 status == ETIME || status == ETIMEDOUT, 5366 status, "cond_timedwait"); 5367 5368#ifdef ASSERT 5369 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5370#endif 5371 5372 _counter = 0 ; 5373 status = pthread_mutex_unlock(_mutex) ; 5374 assert_status(status == 0, status, "invariant") ; 5375 // Paranoia to ensure our locked and lock-free paths interact 5376 // correctly with each other and Java-level accesses. 5377 OrderAccess::fence(); 5378 5379 // If externally suspended while waiting, re-suspend 5380 if (jt->handle_special_suspend_equivalent_condition()) { 5381 jt->java_suspend_self(); 5382 } 5383} 5384 5385void Parker::unpark() { 5386 int s, status ; 5387 status = pthread_mutex_lock(_mutex); 5388 assert (status == 0, "invariant") ; 5389 s = _counter; 5390 _counter = 1; 5391 if (s < 1) { 5392 if (WorkAroundNPTLTimedWaitHang) { 5393 status = pthread_cond_signal (_cond) ; 5394 assert (status == 0, "invariant") ; 5395 status = pthread_mutex_unlock(_mutex); 5396 assert (status == 0, "invariant") ; 5397 } else { 5398 status = pthread_mutex_unlock(_mutex); 5399 assert (status == 0, "invariant") ; 5400 status = pthread_cond_signal (_cond) ; 5401 assert (status == 0, "invariant") ; 5402 } 5403 } else { 5404 pthread_mutex_unlock(_mutex); 5405 assert (status == 0, "invariant") ; 5406 } 5407} 5408 5409 5410extern char** environ; 5411 5412#ifndef __NR_fork 5413#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5414#endif 5415 5416#ifndef __NR_execve 5417#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5418#endif 5419 5420// Run the specified command in a separate process. Return its exit value, 5421// or -1 on failure (e.g. can't fork a new process). 5422// Unlike system(), this function can be called from signal handler. It 5423// doesn't block SIGINT et al. 5424int os::fork_and_exec(char* cmd) { 5425 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5426 5427 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5428 // pthread_atfork handlers and reset pthread library. All we need is a 5429 // separate process to execve. Make a direct syscall to fork process. 5430 // On IA64 there's no fork syscall, we have to use fork() and hope for 5431 // the best... 5432 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5433 IA64_ONLY(fork();) 5434 5435 if (pid < 0) { 5436 // fork failed 5437 return -1; 5438 5439 } else if (pid == 0) { 5440 // child process 5441 5442 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5443 // first to kill every thread on the thread list. Because this list is 5444 // not reset by fork() (see notes above), execve() will instead kill 5445 // every thread in the parent process. We know this is the only thread 5446 // in the new process, so make a system call directly. 5447 // IA64 should use normal execve() from glibc to match the glibc fork() 5448 // above. 5449 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5450 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5451 5452 // execve failed 5453 _exit(-1); 5454 5455 } else { 5456 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5457 // care about the actual exit code, for now. 5458 5459 int status; 5460 5461 // Wait for the child process to exit. This returns immediately if 5462 // the child has already exited. */ 5463 while (waitpid(pid, &status, 0) < 0) { 5464 switch (errno) { 5465 case ECHILD: return 0; 5466 case EINTR: break; 5467 default: return -1; 5468 } 5469 } 5470 5471 if (WIFEXITED(status)) { 5472 // The child exited normally; get its exit code. 5473 return WEXITSTATUS(status); 5474 } else if (WIFSIGNALED(status)) { 5475 // The child exited because of a signal 5476 // The best value to return is 0x80 + signal number, 5477 // because that is what all Unix shells do, and because 5478 // it allows callers to distinguish between process exit and 5479 // process death by signal. 5480 return 0x80 + WTERMSIG(status); 5481 } else { 5482 // Unknown exit code; pass it through 5483 return status; 5484 } 5485 } 5486} 5487 5488// is_headless_jre() 5489// 5490// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5491// in order to report if we are running in a headless jre 5492// 5493// Since JDK8 xawt/libmawt.so was moved into the same directory 5494// as libawt.so, and renamed libawt_xawt.so 5495// 5496bool os::is_headless_jre() { 5497 struct stat statbuf; 5498 char buf[MAXPATHLEN]; 5499 char libmawtpath[MAXPATHLEN]; 5500 const char *xawtstr = "/xawt/libmawt.so"; 5501 const char *new_xawtstr = "/libawt_xawt.so"; 5502 char *p; 5503 5504 // Get path to libjvm.so 5505 os::jvm_path(buf, sizeof(buf)); 5506 5507 // Get rid of libjvm.so 5508 p = strrchr(buf, '/'); 5509 if (p == NULL) return false; 5510 else *p = '\0'; 5511 5512 // Get rid of client or server 5513 p = strrchr(buf, '/'); 5514 if (p == NULL) return false; 5515 else *p = '\0'; 5516 5517 // check xawt/libmawt.so 5518 strcpy(libmawtpath, buf); 5519 strcat(libmawtpath, xawtstr); 5520 if (::stat(libmawtpath, &statbuf) == 0) return false; 5521 5522 // check libawt_xawt.so 5523 strcpy(libmawtpath, buf); 5524 strcat(libmawtpath, new_xawtstr); 5525 if (::stat(libmawtpath, &statbuf) == 0) return false; 5526 5527 return true; 5528} 5529 5530// Get the default path to the core file 5531// Returns the length of the string 5532int os::get_core_path(char* buffer, size_t bufferSize) { 5533 const char* p = get_current_directory(buffer, bufferSize); 5534 5535 if (p == NULL) { 5536 assert(p != NULL, "failed to get current directory"); 5537 return 0; 5538 } 5539 5540 return strlen(buffer); 5541} 5542 5543#ifdef JAVASE_EMBEDDED 5544// 5545// A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 5546// 5547MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 5548 5549// ctor 5550// 5551MemNotifyThread::MemNotifyThread(int fd): Thread() { 5552 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 5553 _fd = fd; 5554 5555 if (os::create_thread(this, os::os_thread)) { 5556 _memnotify_thread = this; 5557 os::set_priority(this, NearMaxPriority); 5558 os::start_thread(this); 5559 } 5560} 5561 5562// Where all the work gets done 5563// 5564void MemNotifyThread::run() { 5565 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 5566 5567 // Set up the select arguments 5568 fd_set rfds; 5569 if (_fd != -1) { 5570 FD_ZERO(&rfds); 5571 FD_SET(_fd, &rfds); 5572 } 5573 5574 // Now wait for the mem_notify device to wake up 5575 while (1) { 5576 // Wait for the mem_notify device to signal us.. 5577 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 5578 if (rc == -1) { 5579 perror("select!\n"); 5580 break; 5581 } else if (rc) { 5582 //ssize_t free_before = os::available_memory(); 5583 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 5584 5585 // The kernel is telling us there is not much memory left... 5586 // try to do something about that 5587 5588 // If we are not already in a GC, try one. 5589 if (!Universe::heap()->is_gc_active()) { 5590 Universe::heap()->collect(GCCause::_allocation_failure); 5591 5592 //ssize_t free_after = os::available_memory(); 5593 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 5594 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 5595 } 5596 // We might want to do something like the following if we find the GC's are not helping... 5597 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 5598 } 5599 } 5600} 5601 5602// 5603// See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 5604// 5605void MemNotifyThread::start() { 5606 int fd; 5607 fd = open ("/dev/mem_notify", O_RDONLY, 0); 5608 if (fd < 0) { 5609 return; 5610 } 5611 5612 if (memnotify_thread() == NULL) { 5613 new MemNotifyThread(fd); 5614 } 5615} 5616#endif // JAVASE_EMBEDDED 5617