os_linux.cpp revision 4592:e12c9b3740db
11541Srgrimes/* 21541Srgrimes * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved. 31541Srgrimes * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 41541Srgrimes * 535938Sdyson * This code is free software; you can redistribute it and/or modify it 61541Srgrimes * under the terms of the GNU General Public License version 2 only, as 71541Srgrimes * published by the Free Software Foundation. 81541Srgrimes * 91541Srgrimes * This code is distributed in the hope that it will be useful, but WITHOUT 101541Srgrimes * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 111541Srgrimes * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 121541Srgrimes * version 2 for more details (a copy is included in the LICENSE file that 131541Srgrimes * accompanied this code). 141541Srgrimes * 151541Srgrimes * You should have received a copy of the GNU General Public License version 161541Srgrimes * 2 along with this work; if not, write to the Free Software Foundation, 171541Srgrimes * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 181541Srgrimes * 191541Srgrimes * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 201541Srgrimes * or visit www.oracle.com if you need additional information or have any 211541Srgrimes * questions. 221541Srgrimes * 231541Srgrimes */ 241541Srgrimes 251541Srgrimes// no precompiled headers 261541Srgrimes#include "classfile/classLoader.hpp" 271541Srgrimes#include "classfile/systemDictionary.hpp" 281541Srgrimes#include "classfile/vmSymbols.hpp" 291541Srgrimes#include "code/icBuffer.hpp" 301541Srgrimes#include "code/vtableStubs.hpp" 311541Srgrimes#include "compiler/compileBroker.hpp" 321541Srgrimes#include "compiler/disassembler.hpp" 331541Srgrimes#include "interpreter/interpreter.hpp" 341541Srgrimes#include "jvm_linux.h" 351541Srgrimes#include "memory/allocation.inline.hpp" 361541Srgrimes#include "memory/filemap.hpp" 371541Srgrimes#include "mutex_linux.inline.hpp" 381541Srgrimes#include "oops/oop.inline.hpp" 391541Srgrimes#include "os_share_linux.hpp" 401541Srgrimes#include "prims/jniFastGetField.hpp" 411541Srgrimes#include "prims/jvm.h" 421541Srgrimes#include "prims/jvm_misc.hpp" 431541Srgrimes#include "runtime/arguments.hpp" 441541Srgrimes#include "runtime/extendedPC.hpp" 451541Srgrimes#include "runtime/globals.hpp" 461541Srgrimes#include "runtime/interfaceSupport.hpp" 471541Srgrimes#include "runtime/init.hpp" 481541Srgrimes#include "runtime/java.hpp" 491541Srgrimes#include "runtime/javaCalls.hpp" 501541Srgrimes#include "runtime/mutexLocker.hpp" 511541Srgrimes#include "runtime/objectMonitor.hpp" 521541Srgrimes#include "runtime/osThread.hpp" 531541Srgrimes#include "runtime/perfMemory.hpp" 541541Srgrimes#include "runtime/sharedRuntime.hpp" 551541Srgrimes#include "runtime/statSampler.hpp" 561541Srgrimes#include "runtime/stubRoutines.hpp" 571541Srgrimes#include "runtime/thread.inline.hpp" 581541Srgrimes#include "runtime/threadCritical.hpp" 591541Srgrimes#include "runtime/timer.hpp" 601541Srgrimes#include "services/attachListener.hpp" 611541Srgrimes#include "services/memTracker.hpp" 621541Srgrimes#include "services/runtimeService.hpp" 631541Srgrimes#include "utilities/decoder.hpp" 641541Srgrimes#include "utilities/defaultStream.hpp" 651541Srgrimes#include "utilities/events.hpp" 661541Srgrimes#include "utilities/elfFile.hpp" 671541Srgrimes#include "utilities/growableArray.hpp" 681541Srgrimes#include "utilities/vmError.hpp" 691541Srgrimes 701541Srgrimes// put OS-includes here 711541Srgrimes# include <sys/types.h> 721541Srgrimes# include <sys/mman.h> 731541Srgrimes# include <sys/stat.h> 741541Srgrimes# include <sys/select.h> 751541Srgrimes# include <pthread.h> 761541Srgrimes# include <signal.h> 771541Srgrimes# include <errno.h> 781541Srgrimes# include <dlfcn.h> 791541Srgrimes# include <stdio.h> 801541Srgrimes# include <unistd.h> 811541Srgrimes# include <sys/resource.h> 821541Srgrimes# include <pthread.h> 831541Srgrimes# include <sys/stat.h> 841541Srgrimes# include <sys/time.h> 851541Srgrimes# include <sys/times.h> 861541Srgrimes# include <sys/utsname.h> 871541Srgrimes# include <sys/socket.h> 881541Srgrimes# include <sys/wait.h> 891541Srgrimes# include <pwd.h> 901541Srgrimes# include <poll.h> 911541Srgrimes# include <semaphore.h> 921541Srgrimes# include <fcntl.h> 931541Srgrimes# include <string.h> 941541Srgrimes# include <syscall.h> 951541Srgrimes# include <sys/sysinfo.h> 961541Srgrimes# include <gnu/libc-version.h> 971541Srgrimes# include <sys/ipc.h> 981541Srgrimes# include <sys/shm.h> 991541Srgrimes# include <link.h> 1001541Srgrimes# include <stdint.h> 1011541Srgrimes# include <inttypes.h> 1021541Srgrimes# include <sys/ioctl.h> 1031541Srgrimes 1041541Srgrimes#define MAX_PATH (2 * K) 1051541Srgrimes 1061541Srgrimes// for timer info max values which include all bits 1071541Srgrimes#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 1081541Srgrimes 1091541Srgrimes#define LARGEPAGES_BIT (1 << 6) 1101541Srgrimes//////////////////////////////////////////////////////////////////////////////// 1111541Srgrimes// global variables 1121541Srgrimesjulong os::Linux::_physical_memory = 0; 1131541Srgrimes 1141541Srgrimesaddress os::Linux::_initial_thread_stack_bottom = NULL; 1151541Srgrimesuintptr_t os::Linux::_initial_thread_stack_size = 0; 1161541Srgrimes 1171541Srgrimesint (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 1181541Srgrimesint (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 1191541SrgrimesMutex* os::Linux::_createThread_lock = NULL; 1201541Srgrimespthread_t os::Linux::_main_thread; 1211541Srgrimesint os::Linux::_page_size = -1; 1221541Srgrimesbool os::Linux::_is_floating_stack = false; 1231541Srgrimesbool os::Linux::_is_NPTL = false; 1241541Srgrimesbool os::Linux::_supports_fast_thread_cpu_time = false; 1251541Srgrimesconst char * os::Linux::_glibc_version = NULL; 1261541Srgrimesconst char * os::Linux::_libpthread_version = NULL; 1271541Srgrimes 1281541Srgrimesstatic jlong initial_time_count=0; 1291541Srgrimes 1301541Srgrimesstatic int clock_tics_per_sec = 100; 1318019Sache 1328019Sache// For diagnostics to print a message once. see run_periodic_checks 1331541Srgrimesstatic sigset_t check_signal_done; 1341541Srgrimesstatic bool check_signals = true;; 1351541Srgrimes 1361541Srgrimesstatic pid_t _initial_pid = 0; 1371541Srgrimes 1381541Srgrimes/* Signal number used to suspend/resume a thread */ 1391541Srgrimes 1401541Srgrimes/* do not use any signal number less than SIGSEGV, see 4355769 */ 1411541Srgrimesstatic int SR_signum = SIGUSR2; 1421541Srgrimessigset_t SR_sigset; 1431541Srgrimes 1441541Srgrimes/* Used to protect dlsym() calls */ 1451541Srgrimesstatic pthread_mutex_t dl_mutex; 1461541Srgrimes 1471541Srgrimes#ifdef JAVASE_EMBEDDED 1481541Srgrimesclass MemNotifyThread: public Thread { 1491541Srgrimes friend class VMStructs; 1501541Srgrimes public: 1511541Srgrimes virtual void run(); 1521541Srgrimes 1531541Srgrimes private: 1541541Srgrimes static MemNotifyThread* _memnotify_thread; 1551541Srgrimes int _fd; 1561541Srgrimes 1571541Srgrimes public: 1581541Srgrimes 1591541Srgrimes // Constructor 1601541Srgrimes MemNotifyThread(int fd); 1611549Srgrimes 1621549Srgrimes // Tester 1631549Srgrimes bool is_memnotify_thread() const { return true; } 1641549Srgrimes 1652442Sdg // Printing 1662729Sdfr char* name() const { return (char*)"Linux MemNotify Thread"; } 1672729Sdfr 1681541Srgrimes // Returns the single instance of the MemNotifyThread 1692297Swollman static MemNotifyThread* memnotify_thread() { return _memnotify_thread; } 1701541Srgrimes 1711541Srgrimes // Create and start the single instance of MemNotifyThread 1721541Srgrimes static void start(); 1731541Srgrimes}; 1741541Srgrimes#endif // JAVASE_EMBEDDED 1751541Srgrimes 1761541Srgrimes// utility functions 1771541Srgrimes 1781541Srgrimesstatic int SR_initialize(); 1791541Srgrimes 1801541Srgrimesjulong os::available_memory() { 1811541Srgrimes return Linux::available_memory(); 1821541Srgrimes} 1831541Srgrimes 1841541Srgrimesjulong os::Linux::available_memory() { 1851541Srgrimes // values in struct sysinfo are "unsigned long" 1861541Srgrimes struct sysinfo si; 1871541Srgrimes sysinfo(&si); 1881541Srgrimes 18935938Sdyson return (julong)si.freeram * si.mem_unit; 19035938Sdyson} 19128400Speter 19229349Speterjulong os::physical_memory() { 19312865Speter return Linux::physical_memory(); 19412865Speter} 19512865Speter 19612865Speter//////////////////////////////////////////////////////////////////////////////// 19712865Speter// environment support 19812865Speter 19912865Speterbool os::getenv(const char* name, char* buf, int len) { 20012865Speter const char* val = ::getenv(name); 20112865Speter if (val != NULL && strlen(val) < (size_t)len) { 20212865Speter strcpy(buf, val); 20312865Speter return true; 20412865Speter } 20525582Speter if (len > 0) buf[0] = 0; // return a null string 20625582Speter return false; 20725582Speter} 20825582Speter 20914220Speter 21014220Speter// Return true if user is running as root. 21129349Speter 21224452Speterbool os::have_special_privileges() { 21324440Speter static bool init = false; 21435938Sdyson static bool privileges = false; 21535938Sdyson if (!init) { 21635938Sdyson privileges = (getuid() != geteuid()) || (getgid() != getegid()); 21735938Sdyson init = true; 21835938Sdyson } 21935938Sdyson return privileges; 22035938Sdyson} 22135938Sdyson 22225537Sdfr 22325537Sdfr#ifndef SYS_gettid 22425537Sdfr// i386: 224, ia64: 1105, amd64: 186, sparc 143 22525537Sdfr#ifdef __ia64__ 22625537Sdfr#define SYS_gettid 1105 22725537Sdfr#elif __i386__ 22825537Sdfr#define SYS_gettid 224 22925537Sdfr#elif __amd64__ 23025537Sdfr#define SYS_gettid 186 23125537Sdfr#elif __sparc__ 23228400Speter#define SYS_gettid 143 23326334Speter#else 23426671Sdyson#error define gettid for the arch 23526671Sdyson#endif 23626671Sdyson#endif 23726671Sdyson 23826671Sdyson// Cpu architecture string 23926671Sdyson#if defined(ZERO) 24026671Sdysonstatic char cpu_arch[] = ZERO_LIBARCH; 24126671Sdyson#elif defined(IA64) 24226671Sdysonstatic char cpu_arch[] = "ia64"; 24326671Sdyson#elif defined(IA32) 24426671Sdysonstatic char cpu_arch[] = "i386"; 24526671Sdyson#elif defined(AMD64) 24629391Sphkstatic char cpu_arch[] = "amd64"; 24734925Sdufault#elif defined(ARM) 24834925Sdufaultstatic char cpu_arch[] = "arm"; 24934925Sdufault#elif defined(PPC) 25034925Sdufaultstatic char cpu_arch[] = "ppc"; 25134925Sdufault#elif defined(SPARC) 25234925Sdufault# ifdef _LP64 25334925Sdufaultstatic char cpu_arch[] = "sparcv9"; 25434925Sdufault# else 25535938Sdysonstatic char cpu_arch[] = "sparc"; 25635938Sdyson# endif 257#else 258#error Add appropriate cpu_arch setting 259#endif 260 261 262// pid_t gettid() 263// 264// Returns the kernel thread id of the currently running thread. Kernel 265// thread id is used to access /proc. 266// 267// (Note that getpid() on LinuxThreads returns kernel thread id too; but 268// on NPTL, it returns the same pid for all threads, as required by POSIX.) 269// 270pid_t os::Linux::gettid() { 271 int rslt = syscall(SYS_gettid); 272 if (rslt == -1) { 273 // old kernel, no NPTL support 274 return getpid(); 275 } else { 276 return (pid_t)rslt; 277 } 278} 279 280// Most versions of linux have a bug where the number of processors are 281// determined by looking at the /proc file system. In a chroot environment, 282// the system call returns 1. This causes the VM to act as if it is 283// a single processor and elide locking (see is_MP() call). 284static bool unsafe_chroot_detected = false; 285static const char *unstable_chroot_error = "/proc file system not found.\n" 286 "Java may be unstable running multithreaded in a chroot " 287 "environment on Linux when /proc filesystem is not mounted."; 288 289void os::Linux::initialize_system_info() { 290 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 291 if (processor_count() == 1) { 292 pid_t pid = os::Linux::gettid(); 293 char fname[32]; 294 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 295 FILE *fp = fopen(fname, "r"); 296 if (fp == NULL) { 297 unsafe_chroot_detected = true; 298 } else { 299 fclose(fp); 300 } 301 } 302 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 303 assert(processor_count() > 0, "linux error"); 304} 305 306void os::init_system_properties_values() { 307// char arch[12]; 308// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 309 310 // The next steps are taken in the product version: 311 // 312 // Obtain the JAVA_HOME value from the location of libjvm.so. 313 // This library should be located at: 314 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 315 // 316 // If "/jre/lib/" appears at the right place in the path, then we 317 // assume libjvm.so is installed in a JDK and we use this path. 318 // 319 // Otherwise exit with message: "Could not create the Java virtual machine." 320 // 321 // The following extra steps are taken in the debugging version: 322 // 323 // If "/jre/lib/" does NOT appear at the right place in the path 324 // instead of exit check for $JAVA_HOME environment variable. 325 // 326 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 327 // then we append a fake suffix "hotspot/libjvm.so" to this path so 328 // it looks like libjvm.so is installed there 329 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 330 // 331 // Otherwise exit. 332 // 333 // Important note: if the location of libjvm.so changes this 334 // code needs to be changed accordingly. 335 336 // The next few definitions allow the code to be verbatim: 337#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal) 338#define getenv(n) ::getenv(n) 339 340/* 341 * See ld(1): 342 * The linker uses the following search paths to locate required 343 * shared libraries: 344 * 1: ... 345 * ... 346 * 7: The default directories, normally /lib and /usr/lib. 347 */ 348#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 349#define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 350#else 351#define DEFAULT_LIBPATH "/lib:/usr/lib" 352#endif 353 354#define EXTENSIONS_DIR "/lib/ext" 355#define ENDORSED_DIR "/lib/endorsed" 356#define REG_DIR "/usr/java/packages" 357 358 { 359 /* sysclasspath, java_home, dll_dir */ 360 { 361 char *home_path; 362 char *dll_path; 363 char *pslash; 364 char buf[MAXPATHLEN]; 365 os::jvm_path(buf, sizeof(buf)); 366 367 // Found the full path to libjvm.so. 368 // Now cut the path to <java_home>/jre if we can. 369 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 370 pslash = strrchr(buf, '/'); 371 if (pslash != NULL) 372 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 373 dll_path = malloc(strlen(buf) + 1); 374 if (dll_path == NULL) 375 return; 376 strcpy(dll_path, buf); 377 Arguments::set_dll_dir(dll_path); 378 379 if (pslash != NULL) { 380 pslash = strrchr(buf, '/'); 381 if (pslash != NULL) { 382 *pslash = '\0'; /* get rid of /<arch> */ 383 pslash = strrchr(buf, '/'); 384 if (pslash != NULL) 385 *pslash = '\0'; /* get rid of /lib */ 386 } 387 } 388 389 home_path = malloc(strlen(buf) + 1); 390 if (home_path == NULL) 391 return; 392 strcpy(home_path, buf); 393 Arguments::set_java_home(home_path); 394 395 if (!set_boot_path('/', ':')) 396 return; 397 } 398 399 /* 400 * Where to look for native libraries 401 * 402 * Note: Due to a legacy implementation, most of the library path 403 * is set in the launcher. This was to accomodate linking restrictions 404 * on legacy Linux implementations (which are no longer supported). 405 * Eventually, all the library path setting will be done here. 406 * 407 * However, to prevent the proliferation of improperly built native 408 * libraries, the new path component /usr/java/packages is added here. 409 * Eventually, all the library path setting will be done here. 410 */ 411 { 412 char *ld_library_path; 413 414 /* 415 * Construct the invariant part of ld_library_path. Note that the 416 * space for the colon and the trailing null are provided by the 417 * nulls included by the sizeof operator (so actually we allocate 418 * a byte more than necessary). 419 */ 420 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 421 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 422 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 423 424 /* 425 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 426 * should always exist (until the legacy problem cited above is 427 * addressed). 428 */ 429 char *v = getenv("LD_LIBRARY_PATH"); 430 if (v != NULL) { 431 char *t = ld_library_path; 432 /* That's +1 for the colon and +1 for the trailing '\0' */ 433 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 434 sprintf(ld_library_path, "%s:%s", v, t); 435 } 436 Arguments::set_library_path(ld_library_path); 437 } 438 439 /* 440 * Extensions directories. 441 * 442 * Note that the space for the colon and the trailing null are provided 443 * by the nulls included by the sizeof operator (so actually one byte more 444 * than necessary is allocated). 445 */ 446 { 447 char *buf = malloc(strlen(Arguments::get_java_home()) + 448 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 449 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 450 Arguments::get_java_home()); 451 Arguments::set_ext_dirs(buf); 452 } 453 454 /* Endorsed standards default directory. */ 455 { 456 char * buf; 457 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 458 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 459 Arguments::set_endorsed_dirs(buf); 460 } 461 } 462 463#undef malloc 464#undef getenv 465#undef EXTENSIONS_DIR 466#undef ENDORSED_DIR 467 468 // Done 469 return; 470} 471 472//////////////////////////////////////////////////////////////////////////////// 473// breakpoint support 474 475void os::breakpoint() { 476 BREAKPOINT; 477} 478 479extern "C" void breakpoint() { 480 // use debugger to set breakpoint here 481} 482 483//////////////////////////////////////////////////////////////////////////////// 484// signal support 485 486debug_only(static bool signal_sets_initialized = false); 487static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 488 489bool os::Linux::is_sig_ignored(int sig) { 490 struct sigaction oact; 491 sigaction(sig, (struct sigaction*)NULL, &oact); 492 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 493 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 494 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 495 return true; 496 else 497 return false; 498} 499 500void os::Linux::signal_sets_init() { 501 // Should also have an assertion stating we are still single-threaded. 502 assert(!signal_sets_initialized, "Already initialized"); 503 // Fill in signals that are necessarily unblocked for all threads in 504 // the VM. Currently, we unblock the following signals: 505 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 506 // by -Xrs (=ReduceSignalUsage)); 507 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 508 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 509 // the dispositions or masks wrt these signals. 510 // Programs embedding the VM that want to use the above signals for their 511 // own purposes must, at this time, use the "-Xrs" option to prevent 512 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 513 // (See bug 4345157, and other related bugs). 514 // In reality, though, unblocking these signals is really a nop, since 515 // these signals are not blocked by default. 516 sigemptyset(&unblocked_sigs); 517 sigemptyset(&allowdebug_blocked_sigs); 518 sigaddset(&unblocked_sigs, SIGILL); 519 sigaddset(&unblocked_sigs, SIGSEGV); 520 sigaddset(&unblocked_sigs, SIGBUS); 521 sigaddset(&unblocked_sigs, SIGFPE); 522 sigaddset(&unblocked_sigs, SR_signum); 523 524 if (!ReduceSignalUsage) { 525 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 526 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 527 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 528 } 529 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 530 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 531 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 532 } 533 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 534 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 535 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 536 } 537 } 538 // Fill in signals that are blocked by all but the VM thread. 539 sigemptyset(&vm_sigs); 540 if (!ReduceSignalUsage) 541 sigaddset(&vm_sigs, BREAK_SIGNAL); 542 debug_only(signal_sets_initialized = true); 543 544} 545 546// These are signals that are unblocked while a thread is running Java. 547// (For some reason, they get blocked by default.) 548sigset_t* os::Linux::unblocked_signals() { 549 assert(signal_sets_initialized, "Not initialized"); 550 return &unblocked_sigs; 551} 552 553// These are the signals that are blocked while a (non-VM) thread is 554// running Java. Only the VM thread handles these signals. 555sigset_t* os::Linux::vm_signals() { 556 assert(signal_sets_initialized, "Not initialized"); 557 return &vm_sigs; 558} 559 560// These are signals that are blocked during cond_wait to allow debugger in 561sigset_t* os::Linux::allowdebug_blocked_signals() { 562 assert(signal_sets_initialized, "Not initialized"); 563 return &allowdebug_blocked_sigs; 564} 565 566void os::Linux::hotspot_sigmask(Thread* thread) { 567 568 //Save caller's signal mask before setting VM signal mask 569 sigset_t caller_sigmask; 570 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 571 572 OSThread* osthread = thread->osthread(); 573 osthread->set_caller_sigmask(caller_sigmask); 574 575 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 576 577 if (!ReduceSignalUsage) { 578 if (thread->is_VM_thread()) { 579 // Only the VM thread handles BREAK_SIGNAL ... 580 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 581 } else { 582 // ... all other threads block BREAK_SIGNAL 583 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 584 } 585 } 586} 587 588////////////////////////////////////////////////////////////////////////////// 589// detecting pthread library 590 591void os::Linux::libpthread_init() { 592 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 593 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 594 // generic name for earlier versions. 595 // Define macros here so we can build HotSpot on old systems. 596# ifndef _CS_GNU_LIBC_VERSION 597# define _CS_GNU_LIBC_VERSION 2 598# endif 599# ifndef _CS_GNU_LIBPTHREAD_VERSION 600# define _CS_GNU_LIBPTHREAD_VERSION 3 601# endif 602 603 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 604 if (n > 0) { 605 char *str = (char *)malloc(n, mtInternal); 606 confstr(_CS_GNU_LIBC_VERSION, str, n); 607 os::Linux::set_glibc_version(str); 608 } else { 609 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 610 static char _gnu_libc_version[32]; 611 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 612 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 613 os::Linux::set_glibc_version(_gnu_libc_version); 614 } 615 616 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 617 if (n > 0) { 618 char *str = (char *)malloc(n, mtInternal); 619 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 620 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 621 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 622 // is the case. LinuxThreads has a hard limit on max number of threads. 623 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 624 // On the other hand, NPTL does not have such a limit, sysconf() 625 // will return -1 and errno is not changed. Check if it is really NPTL. 626 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 627 strstr(str, "NPTL") && 628 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 629 free(str); 630 os::Linux::set_libpthread_version("linuxthreads"); 631 } else { 632 os::Linux::set_libpthread_version(str); 633 } 634 } else { 635 // glibc before 2.3.2 only has LinuxThreads. 636 os::Linux::set_libpthread_version("linuxthreads"); 637 } 638 639 if (strstr(libpthread_version(), "NPTL")) { 640 os::Linux::set_is_NPTL(); 641 } else { 642 os::Linux::set_is_LinuxThreads(); 643 } 644 645 // LinuxThreads have two flavors: floating-stack mode, which allows variable 646 // stack size; and fixed-stack mode. NPTL is always floating-stack. 647 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 648 os::Linux::set_is_floating_stack(); 649 } 650} 651 652///////////////////////////////////////////////////////////////////////////// 653// thread stack 654 655// Force Linux kernel to expand current thread stack. If "bottom" is close 656// to the stack guard, caller should block all signals. 657// 658// MAP_GROWSDOWN: 659// A special mmap() flag that is used to implement thread stacks. It tells 660// kernel that the memory region should extend downwards when needed. This 661// allows early versions of LinuxThreads to only mmap the first few pages 662// when creating a new thread. Linux kernel will automatically expand thread 663// stack as needed (on page faults). 664// 665// However, because the memory region of a MAP_GROWSDOWN stack can grow on 666// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 667// region, it's hard to tell if the fault is due to a legitimate stack 668// access or because of reading/writing non-exist memory (e.g. buffer 669// overrun). As a rule, if the fault happens below current stack pointer, 670// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 671// application (see Linux kernel fault.c). 672// 673// This Linux feature can cause SIGSEGV when VM bangs thread stack for 674// stack overflow detection. 675// 676// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 677// not use this flag. However, the stack of initial thread is not created 678// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 679// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 680// and then attach the thread to JVM. 681// 682// To get around the problem and allow stack banging on Linux, we need to 683// manually expand thread stack after receiving the SIGSEGV. 684// 685// There are two ways to expand thread stack to address "bottom", we used 686// both of them in JVM before 1.5: 687// 1. adjust stack pointer first so that it is below "bottom", and then 688// touch "bottom" 689// 2. mmap() the page in question 690// 691// Now alternate signal stack is gone, it's harder to use 2. For instance, 692// if current sp is already near the lower end of page 101, and we need to 693// call mmap() to map page 100, it is possible that part of the mmap() frame 694// will be placed in page 100. When page 100 is mapped, it is zero-filled. 695// That will destroy the mmap() frame and cause VM to crash. 696// 697// The following code works by adjusting sp first, then accessing the "bottom" 698// page to force a page fault. Linux kernel will then automatically expand the 699// stack mapping. 700// 701// _expand_stack_to() assumes its frame size is less than page size, which 702// should always be true if the function is not inlined. 703 704#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 705#define NOINLINE 706#else 707#define NOINLINE __attribute__ ((noinline)) 708#endif 709 710static void _expand_stack_to(address bottom) NOINLINE; 711 712static void _expand_stack_to(address bottom) { 713 address sp; 714 size_t size; 715 volatile char *p; 716 717 // Adjust bottom to point to the largest address within the same page, it 718 // gives us a one-page buffer if alloca() allocates slightly more memory. 719 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 720 bottom += os::Linux::page_size() - 1; 721 722 // sp might be slightly above current stack pointer; if that's the case, we 723 // will alloca() a little more space than necessary, which is OK. Don't use 724 // os::current_stack_pointer(), as its result can be slightly below current 725 // stack pointer, causing us to not alloca enough to reach "bottom". 726 sp = (address)&sp; 727 728 if (sp > bottom) { 729 size = sp - bottom; 730 p = (volatile char *)alloca(size); 731 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 732 p[0] = '\0'; 733 } 734} 735 736bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 737 assert(t!=NULL, "just checking"); 738 assert(t->osthread()->expanding_stack(), "expand should be set"); 739 assert(t->stack_base() != NULL, "stack_base was not initialized"); 740 741 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 742 sigset_t mask_all, old_sigset; 743 sigfillset(&mask_all); 744 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 745 _expand_stack_to(addr); 746 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 747 return true; 748 } 749 return false; 750} 751 752////////////////////////////////////////////////////////////////////////////// 753// create new thread 754 755static address highest_vm_reserved_address(); 756 757// check if it's safe to start a new thread 758static bool _thread_safety_check(Thread* thread) { 759 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 760 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 761 // Heap is mmap'ed at lower end of memory space. Thread stacks are 762 // allocated (MAP_FIXED) from high address space. Every thread stack 763 // occupies a fixed size slot (usually 2Mbytes, but user can change 764 // it to other values if they rebuild LinuxThreads). 765 // 766 // Problem with MAP_FIXED is that mmap() can still succeed even part of 767 // the memory region has already been mmap'ed. That means if we have too 768 // many threads and/or very large heap, eventually thread stack will 769 // collide with heap. 770 // 771 // Here we try to prevent heap/stack collision by comparing current 772 // stack bottom with the highest address that has been mmap'ed by JVM 773 // plus a safety margin for memory maps created by native code. 774 // 775 // This feature can be disabled by setting ThreadSafetyMargin to 0 776 // 777 if (ThreadSafetyMargin > 0) { 778 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 779 780 // not safe if our stack extends below the safety margin 781 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 782 } else { 783 return true; 784 } 785 } else { 786 // Floating stack LinuxThreads or NPTL: 787 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 788 // there's not enough space left, pthread_create() will fail. If we come 789 // here, that means enough space has been reserved for stack. 790 return true; 791 } 792} 793 794// Thread start routine for all newly created threads 795static void *java_start(Thread *thread) { 796 // Try to randomize the cache line index of hot stack frames. 797 // This helps when threads of the same stack traces evict each other's 798 // cache lines. The threads can be either from the same JVM instance, or 799 // from different JVM instances. The benefit is especially true for 800 // processors with hyperthreading technology. 801 static int counter = 0; 802 int pid = os::current_process_id(); 803 alloca(((pid ^ counter++) & 7) * 128); 804 805 ThreadLocalStorage::set_thread(thread); 806 807 OSThread* osthread = thread->osthread(); 808 Monitor* sync = osthread->startThread_lock(); 809 810 // non floating stack LinuxThreads needs extra check, see above 811 if (!_thread_safety_check(thread)) { 812 // notify parent thread 813 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 814 osthread->set_state(ZOMBIE); 815 sync->notify_all(); 816 return NULL; 817 } 818 819 // thread_id is kernel thread id (similar to Solaris LWP id) 820 osthread->set_thread_id(os::Linux::gettid()); 821 822 if (UseNUMA) { 823 int lgrp_id = os::numa_get_group_id(); 824 if (lgrp_id != -1) { 825 thread->set_lgrp_id(lgrp_id); 826 } 827 } 828 // initialize signal mask for this thread 829 os::Linux::hotspot_sigmask(thread); 830 831 // initialize floating point control register 832 os::Linux::init_thread_fpu_state(); 833 834 // handshaking with parent thread 835 { 836 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 837 838 // notify parent thread 839 osthread->set_state(INITIALIZED); 840 sync->notify_all(); 841 842 // wait until os::start_thread() 843 while (osthread->get_state() == INITIALIZED) { 844 sync->wait(Mutex::_no_safepoint_check_flag); 845 } 846 } 847 848 // call one more level start routine 849 thread->run(); 850 851 return 0; 852} 853 854bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 855 assert(thread->osthread() == NULL, "caller responsible"); 856 857 // Allocate the OSThread object 858 OSThread* osthread = new OSThread(NULL, NULL); 859 if (osthread == NULL) { 860 return false; 861 } 862 863 // set the correct thread state 864 osthread->set_thread_type(thr_type); 865 866 // Initial state is ALLOCATED but not INITIALIZED 867 osthread->set_state(ALLOCATED); 868 869 thread->set_osthread(osthread); 870 871 // init thread attributes 872 pthread_attr_t attr; 873 pthread_attr_init(&attr); 874 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 875 876 // stack size 877 if (os::Linux::supports_variable_stack_size()) { 878 // calculate stack size if it's not specified by caller 879 if (stack_size == 0) { 880 stack_size = os::Linux::default_stack_size(thr_type); 881 882 switch (thr_type) { 883 case os::java_thread: 884 // Java threads use ThreadStackSize which default value can be 885 // changed with the flag -Xss 886 assert (JavaThread::stack_size_at_create() > 0, "this should be set"); 887 stack_size = JavaThread::stack_size_at_create(); 888 break; 889 case os::compiler_thread: 890 if (CompilerThreadStackSize > 0) { 891 stack_size = (size_t)(CompilerThreadStackSize * K); 892 break; 893 } // else fall through: 894 // use VMThreadStackSize if CompilerThreadStackSize is not defined 895 case os::vm_thread: 896 case os::pgc_thread: 897 case os::cgc_thread: 898 case os::watcher_thread: 899 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 900 break; 901 } 902 } 903 904 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 905 pthread_attr_setstacksize(&attr, stack_size); 906 } else { 907 // let pthread_create() pick the default value. 908 } 909 910 // glibc guard page 911 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 912 913 ThreadState state; 914 915 { 916 // Serialize thread creation if we are running with fixed stack LinuxThreads 917 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 918 if (lock) { 919 os::Linux::createThread_lock()->lock_without_safepoint_check(); 920 } 921 922 pthread_t tid; 923 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 924 925 pthread_attr_destroy(&attr); 926 927 if (ret != 0) { 928 if (PrintMiscellaneous && (Verbose || WizardMode)) { 929 perror("pthread_create()"); 930 } 931 // Need to clean up stuff we've allocated so far 932 thread->set_osthread(NULL); 933 delete osthread; 934 if (lock) os::Linux::createThread_lock()->unlock(); 935 return false; 936 } 937 938 // Store pthread info into the OSThread 939 osthread->set_pthread_id(tid); 940 941 // Wait until child thread is either initialized or aborted 942 { 943 Monitor* sync_with_child = osthread->startThread_lock(); 944 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 945 while ((state = osthread->get_state()) == ALLOCATED) { 946 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 947 } 948 } 949 950 if (lock) { 951 os::Linux::createThread_lock()->unlock(); 952 } 953 } 954 955 // Aborted due to thread limit being reached 956 if (state == ZOMBIE) { 957 thread->set_osthread(NULL); 958 delete osthread; 959 return false; 960 } 961 962 // The thread is returned suspended (in state INITIALIZED), 963 // and is started higher up in the call chain 964 assert(state == INITIALIZED, "race condition"); 965 return true; 966} 967 968///////////////////////////////////////////////////////////////////////////// 969// attach existing thread 970 971// bootstrap the main thread 972bool os::create_main_thread(JavaThread* thread) { 973 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 974 return create_attached_thread(thread); 975} 976 977bool os::create_attached_thread(JavaThread* thread) { 978#ifdef ASSERT 979 thread->verify_not_published(); 980#endif 981 982 // Allocate the OSThread object 983 OSThread* osthread = new OSThread(NULL, NULL); 984 985 if (osthread == NULL) { 986 return false; 987 } 988 989 // Store pthread info into the OSThread 990 osthread->set_thread_id(os::Linux::gettid()); 991 osthread->set_pthread_id(::pthread_self()); 992 993 // initialize floating point control register 994 os::Linux::init_thread_fpu_state(); 995 996 // Initial thread state is RUNNABLE 997 osthread->set_state(RUNNABLE); 998 999 thread->set_osthread(osthread); 1000 1001 if (UseNUMA) { 1002 int lgrp_id = os::numa_get_group_id(); 1003 if (lgrp_id != -1) { 1004 thread->set_lgrp_id(lgrp_id); 1005 } 1006 } 1007 1008 if (os::Linux::is_initial_thread()) { 1009 // If current thread is initial thread, its stack is mapped on demand, 1010 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 1011 // the entire stack region to avoid SEGV in stack banging. 1012 // It is also useful to get around the heap-stack-gap problem on SuSE 1013 // kernel (see 4821821 for details). We first expand stack to the top 1014 // of yellow zone, then enable stack yellow zone (order is significant, 1015 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1016 // is no gap between the last two virtual memory regions. 1017 1018 JavaThread *jt = (JavaThread *)thread; 1019 address addr = jt->stack_yellow_zone_base(); 1020 assert(addr != NULL, "initialization problem?"); 1021 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1022 1023 osthread->set_expanding_stack(); 1024 os::Linux::manually_expand_stack(jt, addr); 1025 osthread->clear_expanding_stack(); 1026 } 1027 1028 // initialize signal mask for this thread 1029 // and save the caller's signal mask 1030 os::Linux::hotspot_sigmask(thread); 1031 1032 return true; 1033} 1034 1035void os::pd_start_thread(Thread* thread) { 1036 OSThread * osthread = thread->osthread(); 1037 assert(osthread->get_state() != INITIALIZED, "just checking"); 1038 Monitor* sync_with_child = osthread->startThread_lock(); 1039 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1040 sync_with_child->notify(); 1041} 1042 1043// Free Linux resources related to the OSThread 1044void os::free_thread(OSThread* osthread) { 1045 assert(osthread != NULL, "osthread not set"); 1046 1047 if (Thread::current()->osthread() == osthread) { 1048 // Restore caller's signal mask 1049 sigset_t sigmask = osthread->caller_sigmask(); 1050 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1051 } 1052 1053 delete osthread; 1054} 1055 1056////////////////////////////////////////////////////////////////////////////// 1057// thread local storage 1058 1059int os::allocate_thread_local_storage() { 1060 pthread_key_t key; 1061 int rslt = pthread_key_create(&key, NULL); 1062 assert(rslt == 0, "cannot allocate thread local storage"); 1063 return (int)key; 1064} 1065 1066// Note: This is currently not used by VM, as we don't destroy TLS key 1067// on VM exit. 1068void os::free_thread_local_storage(int index) { 1069 int rslt = pthread_key_delete((pthread_key_t)index); 1070 assert(rslt == 0, "invalid index"); 1071} 1072 1073void os::thread_local_storage_at_put(int index, void* value) { 1074 int rslt = pthread_setspecific((pthread_key_t)index, value); 1075 assert(rslt == 0, "pthread_setspecific failed"); 1076} 1077 1078extern "C" Thread* get_thread() { 1079 return ThreadLocalStorage::thread(); 1080} 1081 1082////////////////////////////////////////////////////////////////////////////// 1083// initial thread 1084 1085// Check if current thread is the initial thread, similar to Solaris thr_main. 1086bool os::Linux::is_initial_thread(void) { 1087 char dummy; 1088 // If called before init complete, thread stack bottom will be null. 1089 // Can be called if fatal error occurs before initialization. 1090 if (initial_thread_stack_bottom() == NULL) return false; 1091 assert(initial_thread_stack_bottom() != NULL && 1092 initial_thread_stack_size() != 0, 1093 "os::init did not locate initial thread's stack region"); 1094 if ((address)&dummy >= initial_thread_stack_bottom() && 1095 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1096 return true; 1097 else return false; 1098} 1099 1100// Find the virtual memory area that contains addr 1101static bool find_vma(address addr, address* vma_low, address* vma_high) { 1102 FILE *fp = fopen("/proc/self/maps", "r"); 1103 if (fp) { 1104 address low, high; 1105 while (!feof(fp)) { 1106 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1107 if (low <= addr && addr < high) { 1108 if (vma_low) *vma_low = low; 1109 if (vma_high) *vma_high = high; 1110 fclose (fp); 1111 return true; 1112 } 1113 } 1114 for (;;) { 1115 int ch = fgetc(fp); 1116 if (ch == EOF || ch == (int)'\n') break; 1117 } 1118 } 1119 fclose(fp); 1120 } 1121 return false; 1122} 1123 1124// Locate initial thread stack. This special handling of initial thread stack 1125// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1126// bogus value for initial thread. 1127void os::Linux::capture_initial_stack(size_t max_size) { 1128 // stack size is the easy part, get it from RLIMIT_STACK 1129 size_t stack_size; 1130 struct rlimit rlim; 1131 getrlimit(RLIMIT_STACK, &rlim); 1132 stack_size = rlim.rlim_cur; 1133 1134 // 6308388: a bug in ld.so will relocate its own .data section to the 1135 // lower end of primordial stack; reduce ulimit -s value a little bit 1136 // so we won't install guard page on ld.so's data section. 1137 stack_size -= 2 * page_size(); 1138 1139 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1140 // 7.1, in both cases we will get 2G in return value. 1141 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1142 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1143 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1144 // in case other parts in glibc still assumes 2M max stack size. 1145 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1146 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1147 if (stack_size > 2 * K * K IA64_ONLY(*2)) 1148 stack_size = 2 * K * K IA64_ONLY(*2); 1149 // Try to figure out where the stack base (top) is. This is harder. 1150 // 1151 // When an application is started, glibc saves the initial stack pointer in 1152 // a global variable "__libc_stack_end", which is then used by system 1153 // libraries. __libc_stack_end should be pretty close to stack top. The 1154 // variable is available since the very early days. However, because it is 1155 // a private interface, it could disappear in the future. 1156 // 1157 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1158 // to __libc_stack_end, it is very close to stack top, but isn't the real 1159 // stack top. Note that /proc may not exist if VM is running as a chroot 1160 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1161 // /proc/<pid>/stat could change in the future (though unlikely). 1162 // 1163 // We try __libc_stack_end first. If that doesn't work, look for 1164 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1165 // as a hint, which should work well in most cases. 1166 1167 uintptr_t stack_start; 1168 1169 // try __libc_stack_end first 1170 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1171 if (p && *p) { 1172 stack_start = *p; 1173 } else { 1174 // see if we can get the start_stack field from /proc/self/stat 1175 FILE *fp; 1176 int pid; 1177 char state; 1178 int ppid; 1179 int pgrp; 1180 int session; 1181 int nr; 1182 int tpgrp; 1183 unsigned long flags; 1184 unsigned long minflt; 1185 unsigned long cminflt; 1186 unsigned long majflt; 1187 unsigned long cmajflt; 1188 unsigned long utime; 1189 unsigned long stime; 1190 long cutime; 1191 long cstime; 1192 long prio; 1193 long nice; 1194 long junk; 1195 long it_real; 1196 uintptr_t start; 1197 uintptr_t vsize; 1198 intptr_t rss; 1199 uintptr_t rsslim; 1200 uintptr_t scodes; 1201 uintptr_t ecode; 1202 int i; 1203 1204 // Figure what the primordial thread stack base is. Code is inspired 1205 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1206 // followed by command name surrounded by parentheses, state, etc. 1207 char stat[2048]; 1208 int statlen; 1209 1210 fp = fopen("/proc/self/stat", "r"); 1211 if (fp) { 1212 statlen = fread(stat, 1, 2047, fp); 1213 stat[statlen] = '\0'; 1214 fclose(fp); 1215 1216 // Skip pid and the command string. Note that we could be dealing with 1217 // weird command names, e.g. user could decide to rename java launcher 1218 // to "java 1.4.2 :)", then the stat file would look like 1219 // 1234 (java 1.4.2 :)) R ... ... 1220 // We don't really need to know the command string, just find the last 1221 // occurrence of ")" and then start parsing from there. See bug 4726580. 1222 char * s = strrchr(stat, ')'); 1223 1224 i = 0; 1225 if (s) { 1226 // Skip blank chars 1227 do s++; while (isspace(*s)); 1228 1229#define _UFM UINTX_FORMAT 1230#define _DFM INTX_FORMAT 1231 1232 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1233 /* 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 */ 1234 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, 1235 &state, /* 3 %c */ 1236 &ppid, /* 4 %d */ 1237 &pgrp, /* 5 %d */ 1238 &session, /* 6 %d */ 1239 &nr, /* 7 %d */ 1240 &tpgrp, /* 8 %d */ 1241 &flags, /* 9 %lu */ 1242 &minflt, /* 10 %lu */ 1243 &cminflt, /* 11 %lu */ 1244 &majflt, /* 12 %lu */ 1245 &cmajflt, /* 13 %lu */ 1246 &utime, /* 14 %lu */ 1247 &stime, /* 15 %lu */ 1248 &cutime, /* 16 %ld */ 1249 &cstime, /* 17 %ld */ 1250 &prio, /* 18 %ld */ 1251 &nice, /* 19 %ld */ 1252 &junk, /* 20 %ld */ 1253 &it_real, /* 21 %ld */ 1254 &start, /* 22 UINTX_FORMAT */ 1255 &vsize, /* 23 UINTX_FORMAT */ 1256 &rss, /* 24 INTX_FORMAT */ 1257 &rsslim, /* 25 UINTX_FORMAT */ 1258 &scodes, /* 26 UINTX_FORMAT */ 1259 &ecode, /* 27 UINTX_FORMAT */ 1260 &stack_start); /* 28 UINTX_FORMAT */ 1261 } 1262 1263#undef _UFM 1264#undef _DFM 1265 1266 if (i != 28 - 2) { 1267 assert(false, "Bad conversion from /proc/self/stat"); 1268 // product mode - assume we are the initial thread, good luck in the 1269 // embedded case. 1270 warning("Can't detect initial thread stack location - bad conversion"); 1271 stack_start = (uintptr_t) &rlim; 1272 } 1273 } else { 1274 // For some reason we can't open /proc/self/stat (for example, running on 1275 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1276 // most cases, so don't abort: 1277 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1278 stack_start = (uintptr_t) &rlim; 1279 } 1280 } 1281 1282 // Now we have a pointer (stack_start) very close to the stack top, the 1283 // next thing to do is to figure out the exact location of stack top. We 1284 // can find out the virtual memory area that contains stack_start by 1285 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1286 // and its upper limit is the real stack top. (again, this would fail if 1287 // running inside chroot, because /proc may not exist.) 1288 1289 uintptr_t stack_top; 1290 address low, high; 1291 if (find_vma((address)stack_start, &low, &high)) { 1292 // success, "high" is the true stack top. (ignore "low", because initial 1293 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1294 stack_top = (uintptr_t)high; 1295 } else { 1296 // failed, likely because /proc/self/maps does not exist 1297 warning("Can't detect initial thread stack location - find_vma failed"); 1298 // best effort: stack_start is normally within a few pages below the real 1299 // stack top, use it as stack top, and reduce stack size so we won't put 1300 // guard page outside stack. 1301 stack_top = stack_start; 1302 stack_size -= 16 * page_size(); 1303 } 1304 1305 // stack_top could be partially down the page so align it 1306 stack_top = align_size_up(stack_top, page_size()); 1307 1308 if (max_size && stack_size > max_size) { 1309 _initial_thread_stack_size = max_size; 1310 } else { 1311 _initial_thread_stack_size = stack_size; 1312 } 1313 1314 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1315 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1316} 1317 1318//////////////////////////////////////////////////////////////////////////////// 1319// time support 1320 1321// Time since start-up in seconds to a fine granularity. 1322// Used by VMSelfDestructTimer and the MemProfiler. 1323double os::elapsedTime() { 1324 1325 return (double)(os::elapsed_counter()) * 0.000001; 1326} 1327 1328jlong os::elapsed_counter() { 1329 timeval time; 1330 int status = gettimeofday(&time, NULL); 1331 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1332} 1333 1334jlong os::elapsed_frequency() { 1335 return (1000 * 1000); 1336} 1337 1338// For now, we say that linux does not support vtime. I have no idea 1339// whether it can actually be made to (DLD, 9/13/05). 1340 1341bool os::supports_vtime() { return false; } 1342bool os::enable_vtime() { return false; } 1343bool os::vtime_enabled() { return false; } 1344double os::elapsedVTime() { 1345 // better than nothing, but not much 1346 return elapsedTime(); 1347} 1348 1349jlong os::javaTimeMillis() { 1350 timeval time; 1351 int status = gettimeofday(&time, NULL); 1352 assert(status != -1, "linux error"); 1353 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1354} 1355 1356#ifndef CLOCK_MONOTONIC 1357#define CLOCK_MONOTONIC (1) 1358#endif 1359 1360void os::Linux::clock_init() { 1361 // we do dlopen's in this particular order due to bug in linux 1362 // dynamical loader (see 6348968) leading to crash on exit 1363 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1364 if (handle == NULL) { 1365 handle = dlopen("librt.so", RTLD_LAZY); 1366 } 1367 1368 if (handle) { 1369 int (*clock_getres_func)(clockid_t, struct timespec*) = 1370 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1371 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1372 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1373 if (clock_getres_func && clock_gettime_func) { 1374 // See if monotonic clock is supported by the kernel. Note that some 1375 // early implementations simply return kernel jiffies (updated every 1376 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1377 // for nano time (though the monotonic property is still nice to have). 1378 // It's fixed in newer kernels, however clock_getres() still returns 1379 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1380 // resolution for now. Hopefully as people move to new kernels, this 1381 // won't be a problem. 1382 struct timespec res; 1383 struct timespec tp; 1384 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1385 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1386 // yes, monotonic clock is supported 1387 _clock_gettime = clock_gettime_func; 1388 } else { 1389 // close librt if there is no monotonic clock 1390 dlclose(handle); 1391 } 1392 } 1393 } 1394} 1395 1396#ifndef SYS_clock_getres 1397 1398#if defined(IA32) || defined(AMD64) 1399#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1400#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1401#else 1402#warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1403#define sys_clock_getres(x,y) -1 1404#endif 1405 1406#else 1407#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1408#endif 1409 1410void os::Linux::fast_thread_clock_init() { 1411 if (!UseLinuxPosixThreadCPUClocks) { 1412 return; 1413 } 1414 clockid_t clockid; 1415 struct timespec tp; 1416 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1417 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1418 1419 // Switch to using fast clocks for thread cpu time if 1420 // the sys_clock_getres() returns 0 error code. 1421 // Note, that some kernels may support the current thread 1422 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1423 // returned by the pthread_getcpuclockid(). 1424 // If the fast Posix clocks are supported then the sys_clock_getres() 1425 // must return at least tp.tv_sec == 0 which means a resolution 1426 // better than 1 sec. This is extra check for reliability. 1427 1428 if(pthread_getcpuclockid_func && 1429 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1430 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1431 1432 _supports_fast_thread_cpu_time = true; 1433 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1434 } 1435} 1436 1437jlong os::javaTimeNanos() { 1438 if (Linux::supports_monotonic_clock()) { 1439 struct timespec tp; 1440 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1441 assert(status == 0, "gettime error"); 1442 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1443 return result; 1444 } else { 1445 timeval time; 1446 int status = gettimeofday(&time, NULL); 1447 assert(status != -1, "linux error"); 1448 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1449 return 1000 * usecs; 1450 } 1451} 1452 1453void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1454 if (Linux::supports_monotonic_clock()) { 1455 info_ptr->max_value = ALL_64_BITS; 1456 1457 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1458 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1459 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1460 } else { 1461 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1462 info_ptr->max_value = ALL_64_BITS; 1463 1464 // gettimeofday is a real time clock so it skips 1465 info_ptr->may_skip_backward = true; 1466 info_ptr->may_skip_forward = true; 1467 } 1468 1469 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1470} 1471 1472// Return the real, user, and system times in seconds from an 1473// arbitrary fixed point in the past. 1474bool os::getTimesSecs(double* process_real_time, 1475 double* process_user_time, 1476 double* process_system_time) { 1477 struct tms ticks; 1478 clock_t real_ticks = times(&ticks); 1479 1480 if (real_ticks == (clock_t) (-1)) { 1481 return false; 1482 } else { 1483 double ticks_per_second = (double) clock_tics_per_sec; 1484 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1485 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1486 *process_real_time = ((double) real_ticks) / ticks_per_second; 1487 1488 return true; 1489 } 1490} 1491 1492 1493char * os::local_time_string(char *buf, size_t buflen) { 1494 struct tm t; 1495 time_t long_time; 1496 time(&long_time); 1497 localtime_r(&long_time, &t); 1498 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1499 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1500 t.tm_hour, t.tm_min, t.tm_sec); 1501 return buf; 1502} 1503 1504struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1505 return localtime_r(clock, res); 1506} 1507 1508//////////////////////////////////////////////////////////////////////////////// 1509// runtime exit support 1510 1511// Note: os::shutdown() might be called very early during initialization, or 1512// called from signal handler. Before adding something to os::shutdown(), make 1513// sure it is async-safe and can handle partially initialized VM. 1514void os::shutdown() { 1515 1516 // allow PerfMemory to attempt cleanup of any persistent resources 1517 perfMemory_exit(); 1518 1519 // needs to remove object in file system 1520 AttachListener::abort(); 1521 1522 // flush buffered output, finish log files 1523 ostream_abort(); 1524 1525 // Check for abort hook 1526 abort_hook_t abort_hook = Arguments::abort_hook(); 1527 if (abort_hook != NULL) { 1528 abort_hook(); 1529 } 1530 1531} 1532 1533// Note: os::abort() might be called very early during initialization, or 1534// called from signal handler. Before adding something to os::abort(), make 1535// sure it is async-safe and can handle partially initialized VM. 1536void os::abort(bool dump_core) { 1537 os::shutdown(); 1538 if (dump_core) { 1539#ifndef PRODUCT 1540 fdStream out(defaultStream::output_fd()); 1541 out.print_raw("Current thread is "); 1542 char buf[16]; 1543 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1544 out.print_raw_cr(buf); 1545 out.print_raw_cr("Dumping core ..."); 1546#endif 1547 ::abort(); // dump core 1548 } 1549 1550 ::exit(1); 1551} 1552 1553// Die immediately, no exit hook, no abort hook, no cleanup. 1554void os::die() { 1555 // _exit() on LinuxThreads only kills current thread 1556 ::abort(); 1557} 1558 1559// unused on linux for now. 1560void os::set_error_file(const char *logfile) {} 1561 1562 1563// This method is a copy of JDK's sysGetLastErrorString 1564// from src/solaris/hpi/src/system_md.c 1565 1566size_t os::lasterror(char *buf, size_t len) { 1567 1568 if (errno == 0) return 0; 1569 1570 const char *s = ::strerror(errno); 1571 size_t n = ::strlen(s); 1572 if (n >= len) { 1573 n = len - 1; 1574 } 1575 ::strncpy(buf, s, n); 1576 buf[n] = '\0'; 1577 return n; 1578} 1579 1580intx os::current_thread_id() { return (intx)pthread_self(); } 1581int os::current_process_id() { 1582 1583 // Under the old linux thread library, linux gives each thread 1584 // its own process id. Because of this each thread will return 1585 // a different pid if this method were to return the result 1586 // of getpid(2). Linux provides no api that returns the pid 1587 // of the launcher thread for the vm. This implementation 1588 // returns a unique pid, the pid of the launcher thread 1589 // that starts the vm 'process'. 1590 1591 // Under the NPTL, getpid() returns the same pid as the 1592 // launcher thread rather than a unique pid per thread. 1593 // Use gettid() if you want the old pre NPTL behaviour. 1594 1595 // if you are looking for the result of a call to getpid() that 1596 // returns a unique pid for the calling thread, then look at the 1597 // OSThread::thread_id() method in osThread_linux.hpp file 1598 1599 return (int)(_initial_pid ? _initial_pid : getpid()); 1600} 1601 1602// DLL functions 1603 1604const char* os::dll_file_extension() { return ".so"; } 1605 1606// This must be hard coded because it's the system's temporary 1607// directory not the java application's temp directory, ala java.io.tmpdir. 1608const char* os::get_temp_directory() { return "/tmp"; } 1609 1610static bool file_exists(const char* filename) { 1611 struct stat statbuf; 1612 if (filename == NULL || strlen(filename) == 0) { 1613 return false; 1614 } 1615 return os::stat(filename, &statbuf) == 0; 1616} 1617 1618bool os::dll_build_name(char* buffer, size_t buflen, 1619 const char* pname, const char* fname) { 1620 bool retval = false; 1621 // Copied from libhpi 1622 const size_t pnamelen = pname ? strlen(pname) : 0; 1623 1624 // Return error on buffer overflow. 1625 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1626 return retval; 1627 } 1628 1629 if (pnamelen == 0) { 1630 snprintf(buffer, buflen, "lib%s.so", fname); 1631 retval = true; 1632 } else if (strchr(pname, *os::path_separator()) != NULL) { 1633 int n; 1634 char** pelements = split_path(pname, &n); 1635 if (pelements == NULL) { 1636 return false; 1637 } 1638 for (int i = 0 ; i < n ; i++) { 1639 // Really shouldn't be NULL, but check can't hurt 1640 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1641 continue; // skip the empty path values 1642 } 1643 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1644 if (file_exists(buffer)) { 1645 retval = true; 1646 break; 1647 } 1648 } 1649 // release the storage 1650 for (int i = 0 ; i < n ; i++) { 1651 if (pelements[i] != NULL) { 1652 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1653 } 1654 } 1655 if (pelements != NULL) { 1656 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1657 } 1658 } else { 1659 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1660 retval = true; 1661 } 1662 return retval; 1663} 1664 1665// check if addr is inside libjvm.so 1666bool os::address_is_in_vm(address addr) { 1667 static address libjvm_base_addr; 1668 Dl_info dlinfo; 1669 1670 if (libjvm_base_addr == NULL) { 1671 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); 1672 libjvm_base_addr = (address)dlinfo.dli_fbase; 1673 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1674 } 1675 1676 if (dladdr((void *)addr, &dlinfo)) { 1677 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1678 } 1679 1680 return false; 1681} 1682 1683bool os::dll_address_to_function_name(address addr, char *buf, 1684 int buflen, int *offset) { 1685 Dl_info dlinfo; 1686 1687 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { 1688 if (buf != NULL) { 1689 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1690 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1691 } 1692 } 1693 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1694 return true; 1695 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) { 1696 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1697 buf, buflen, offset, dlinfo.dli_fname)) { 1698 return true; 1699 } 1700 } 1701 1702 if (buf != NULL) buf[0] = '\0'; 1703 if (offset != NULL) *offset = -1; 1704 return false; 1705} 1706 1707struct _address_to_library_name { 1708 address addr; // input : memory address 1709 size_t buflen; // size of fname 1710 char* fname; // output: library name 1711 address base; // library base addr 1712}; 1713 1714static int address_to_library_name_callback(struct dl_phdr_info *info, 1715 size_t size, void *data) { 1716 int i; 1717 bool found = false; 1718 address libbase = NULL; 1719 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1720 1721 // iterate through all loadable segments 1722 for (i = 0; i < info->dlpi_phnum; i++) { 1723 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1724 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1725 // base address of a library is the lowest address of its loaded 1726 // segments. 1727 if (libbase == NULL || libbase > segbase) { 1728 libbase = segbase; 1729 } 1730 // see if 'addr' is within current segment 1731 if (segbase <= d->addr && 1732 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1733 found = true; 1734 } 1735 } 1736 } 1737 1738 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1739 // so dll_address_to_library_name() can fall through to use dladdr() which 1740 // can figure out executable name from argv[0]. 1741 if (found && info->dlpi_name && info->dlpi_name[0]) { 1742 d->base = libbase; 1743 if (d->fname) { 1744 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1745 } 1746 return 1; 1747 } 1748 return 0; 1749} 1750 1751bool os::dll_address_to_library_name(address addr, char* buf, 1752 int buflen, int* offset) { 1753 Dl_info dlinfo; 1754 struct _address_to_library_name data; 1755 1756 // There is a bug in old glibc dladdr() implementation that it could resolve 1757 // to wrong library name if the .so file has a base address != NULL. Here 1758 // we iterate through the program headers of all loaded libraries to find 1759 // out which library 'addr' really belongs to. This workaround can be 1760 // removed once the minimum requirement for glibc is moved to 2.3.x. 1761 data.addr = addr; 1762 data.fname = buf; 1763 data.buflen = buflen; 1764 data.base = NULL; 1765 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1766 1767 if (rslt) { 1768 // buf already contains library name 1769 if (offset) *offset = addr - data.base; 1770 return true; 1771 } else if (dladdr((void*)addr, &dlinfo)){ 1772 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1773 if (offset) *offset = addr - (address)dlinfo.dli_fbase; 1774 return true; 1775 } else { 1776 if (buf) buf[0] = '\0'; 1777 if (offset) *offset = -1; 1778 return false; 1779 } 1780} 1781 1782 // Loads .dll/.so and 1783 // in case of error it checks if .dll/.so was built for the 1784 // same architecture as Hotspot is running on 1785 1786 1787// Remember the stack's state. The Linux dynamic linker will change 1788// the stack to 'executable' at most once, so we must safepoint only once. 1789bool os::Linux::_stack_is_executable = false; 1790 1791// VM operation that loads a library. This is necessary if stack protection 1792// of the Java stacks can be lost during loading the library. If we 1793// do not stop the Java threads, they can stack overflow before the stacks 1794// are protected again. 1795class VM_LinuxDllLoad: public VM_Operation { 1796 private: 1797 const char *_filename; 1798 char *_ebuf; 1799 int _ebuflen; 1800 void *_lib; 1801 public: 1802 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1803 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1804 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1805 void doit() { 1806 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1807 os::Linux::_stack_is_executable = true; 1808 } 1809 void* loaded_library() { return _lib; } 1810}; 1811 1812void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1813{ 1814 void * result = NULL; 1815 bool load_attempted = false; 1816 1817 // Check whether the library to load might change execution rights 1818 // of the stack. If they are changed, the protection of the stack 1819 // guard pages will be lost. We need a safepoint to fix this. 1820 // 1821 // See Linux man page execstack(8) for more info. 1822 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1823 ElfFile ef(filename); 1824 if (!ef.specifies_noexecstack()) { 1825 if (!is_init_completed()) { 1826 os::Linux::_stack_is_executable = true; 1827 // This is OK - No Java threads have been created yet, and hence no 1828 // stack guard pages to fix. 1829 // 1830 // This should happen only when you are building JDK7 using a very 1831 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1832 // 1833 // Dynamic loader will make all stacks executable after 1834 // this function returns, and will not do that again. 1835 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1836 } else { 1837 warning("You have loaded library %s which might have disabled stack guard. " 1838 "The VM will try to fix the stack guard now.\n" 1839 "It's highly recommended that you fix the library with " 1840 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1841 filename); 1842 1843 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1844 JavaThread *jt = JavaThread::current(); 1845 if (jt->thread_state() != _thread_in_native) { 1846 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1847 // that requires ExecStack. Cannot enter safe point. Let's give up. 1848 warning("Unable to fix stack guard. Giving up."); 1849 } else { 1850 if (!LoadExecStackDllInVMThread) { 1851 // This is for the case where the DLL has an static 1852 // constructor function that executes JNI code. We cannot 1853 // load such DLLs in the VMThread. 1854 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1855 } 1856 1857 ThreadInVMfromNative tiv(jt); 1858 debug_only(VMNativeEntryWrapper vew;) 1859 1860 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1861 VMThread::execute(&op); 1862 if (LoadExecStackDllInVMThread) { 1863 result = op.loaded_library(); 1864 } 1865 load_attempted = true; 1866 } 1867 } 1868 } 1869 } 1870 1871 if (!load_attempted) { 1872 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1873 } 1874 1875 if (result != NULL) { 1876 // Successful loading 1877 return result; 1878 } 1879 1880 Elf32_Ehdr elf_head; 1881 int diag_msg_max_length=ebuflen-strlen(ebuf); 1882 char* diag_msg_buf=ebuf+strlen(ebuf); 1883 1884 if (diag_msg_max_length==0) { 1885 // No more space in ebuf for additional diagnostics message 1886 return NULL; 1887 } 1888 1889 1890 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1891 1892 if (file_descriptor < 0) { 1893 // Can't open library, report dlerror() message 1894 return NULL; 1895 } 1896 1897 bool failed_to_read_elf_head= 1898 (sizeof(elf_head)!= 1899 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1900 1901 ::close(file_descriptor); 1902 if (failed_to_read_elf_head) { 1903 // file i/o error - report dlerror() msg 1904 return NULL; 1905 } 1906 1907 typedef struct { 1908 Elf32_Half code; // Actual value as defined in elf.h 1909 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1910 char elf_class; // 32 or 64 bit 1911 char endianess; // MSB or LSB 1912 char* name; // String representation 1913 } arch_t; 1914 1915 #ifndef EM_486 1916 #define EM_486 6 /* Intel 80486 */ 1917 #endif 1918 1919 static const arch_t arch_array[]={ 1920 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1921 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1922 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1923 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1924 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1925 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1926 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1927 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1928 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1929 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1930 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1931 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1932 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1933 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1934 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1935 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1936 }; 1937 1938 #if (defined IA32) 1939 static Elf32_Half running_arch_code=EM_386; 1940 #elif (defined AMD64) 1941 static Elf32_Half running_arch_code=EM_X86_64; 1942 #elif (defined IA64) 1943 static Elf32_Half running_arch_code=EM_IA_64; 1944 #elif (defined __sparc) && (defined _LP64) 1945 static Elf32_Half running_arch_code=EM_SPARCV9; 1946 #elif (defined __sparc) && (!defined _LP64) 1947 static Elf32_Half running_arch_code=EM_SPARC; 1948 #elif (defined __powerpc64__) 1949 static Elf32_Half running_arch_code=EM_PPC64; 1950 #elif (defined __powerpc__) 1951 static Elf32_Half running_arch_code=EM_PPC; 1952 #elif (defined ARM) 1953 static Elf32_Half running_arch_code=EM_ARM; 1954 #elif (defined S390) 1955 static Elf32_Half running_arch_code=EM_S390; 1956 #elif (defined ALPHA) 1957 static Elf32_Half running_arch_code=EM_ALPHA; 1958 #elif (defined MIPSEL) 1959 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1960 #elif (defined PARISC) 1961 static Elf32_Half running_arch_code=EM_PARISC; 1962 #elif (defined MIPS) 1963 static Elf32_Half running_arch_code=EM_MIPS; 1964 #elif (defined M68K) 1965 static Elf32_Half running_arch_code=EM_68K; 1966 #else 1967 #error Method os::dll_load requires that one of following is defined:\ 1968 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 1969 #endif 1970 1971 // Identify compatability class for VM's architecture and library's architecture 1972 // Obtain string descriptions for architectures 1973 1974 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1975 int running_arch_index=-1; 1976 1977 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1978 if (running_arch_code == arch_array[i].code) { 1979 running_arch_index = i; 1980 } 1981 if (lib_arch.code == arch_array[i].code) { 1982 lib_arch.compat_class = arch_array[i].compat_class; 1983 lib_arch.name = arch_array[i].name; 1984 } 1985 } 1986 1987 assert(running_arch_index != -1, 1988 "Didn't find running architecture code (running_arch_code) in arch_array"); 1989 if (running_arch_index == -1) { 1990 // Even though running architecture detection failed 1991 // we may still continue with reporting dlerror() message 1992 return NULL; 1993 } 1994 1995 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1996 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1997 return NULL; 1998 } 1999 2000#ifndef S390 2001 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 2002 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 2003 return NULL; 2004 } 2005#endif // !S390 2006 2007 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2008 if ( lib_arch.name!=NULL ) { 2009 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2010 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2011 lib_arch.name, arch_array[running_arch_index].name); 2012 } else { 2013 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2014 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2015 lib_arch.code, 2016 arch_array[running_arch_index].name); 2017 } 2018 } 2019 2020 return NULL; 2021} 2022 2023void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) { 2024 void * result = ::dlopen(filename, RTLD_LAZY); 2025 if (result == NULL) { 2026 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2027 ebuf[ebuflen-1] = '\0'; 2028 } 2029 return result; 2030} 2031 2032void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) { 2033 void * result = NULL; 2034 if (LoadExecStackDllInVMThread) { 2035 result = dlopen_helper(filename, ebuf, ebuflen); 2036 } 2037 2038 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2039 // library that requires an executable stack, or which does not have this 2040 // stack attribute set, dlopen changes the stack attribute to executable. The 2041 // read protection of the guard pages gets lost. 2042 // 2043 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2044 // may have been queued at the same time. 2045 2046 if (!_stack_is_executable) { 2047 JavaThread *jt = Threads::first(); 2048 2049 while (jt) { 2050 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2051 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2052 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2053 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2054 warning("Attempt to reguard stack yellow zone failed."); 2055 } 2056 } 2057 jt = jt->next(); 2058 } 2059 } 2060 2061 return result; 2062} 2063 2064/* 2065 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2066 * chances are you might want to run the generated bits against glibc-2.0 2067 * libdl.so, so always use locking for any version of glibc. 2068 */ 2069void* os::dll_lookup(void* handle, const char* name) { 2070 pthread_mutex_lock(&dl_mutex); 2071 void* res = dlsym(handle, name); 2072 pthread_mutex_unlock(&dl_mutex); 2073 return res; 2074} 2075 2076 2077static bool _print_ascii_file(const char* filename, outputStream* st) { 2078 int fd = ::open(filename, O_RDONLY); 2079 if (fd == -1) { 2080 return false; 2081 } 2082 2083 char buf[32]; 2084 int bytes; 2085 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2086 st->print_raw(buf, bytes); 2087 } 2088 2089 ::close(fd); 2090 2091 return true; 2092} 2093 2094void os::print_dll_info(outputStream *st) { 2095 st->print_cr("Dynamic libraries:"); 2096 2097 char fname[32]; 2098 pid_t pid = os::Linux::gettid(); 2099 2100 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2101 2102 if (!_print_ascii_file(fname, st)) { 2103 st->print("Can not get library information for pid = %d\n", pid); 2104 } 2105} 2106 2107void os::print_os_info_brief(outputStream* st) { 2108 os::Linux::print_distro_info(st); 2109 2110 os::Posix::print_uname_info(st); 2111 2112 os::Linux::print_libversion_info(st); 2113 2114} 2115 2116void os::print_os_info(outputStream* st) { 2117 st->print("OS:"); 2118 2119 os::Linux::print_distro_info(st); 2120 2121 os::Posix::print_uname_info(st); 2122 2123 // Print warning if unsafe chroot environment detected 2124 if (unsafe_chroot_detected) { 2125 st->print("WARNING!! "); 2126 st->print_cr(unstable_chroot_error); 2127 } 2128 2129 os::Linux::print_libversion_info(st); 2130 2131 os::Posix::print_rlimit_info(st); 2132 2133 os::Posix::print_load_average(st); 2134 2135 os::Linux::print_full_memory_info(st); 2136} 2137 2138// Try to identify popular distros. 2139// Most Linux distributions have /etc/XXX-release file, which contains 2140// the OS version string. Some have more than one /etc/XXX-release file 2141// (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 2142// so the order is important. 2143void os::Linux::print_distro_info(outputStream* st) { 2144 if (!_print_ascii_file("/etc/mandrake-release", st) && 2145 !_print_ascii_file("/etc/sun-release", st) && 2146 !_print_ascii_file("/etc/redhat-release", st) && 2147 !_print_ascii_file("/etc/SuSE-release", st) && 2148 !_print_ascii_file("/etc/turbolinux-release", st) && 2149 !_print_ascii_file("/etc/gentoo-release", st) && 2150 !_print_ascii_file("/etc/debian_version", st) && 2151 !_print_ascii_file("/etc/ltib-release", st) && 2152 !_print_ascii_file("/etc/angstrom-version", st)) { 2153 st->print("Linux"); 2154 } 2155 st->cr(); 2156} 2157 2158void os::Linux::print_libversion_info(outputStream* st) { 2159 // libc, pthread 2160 st->print("libc:"); 2161 st->print(os::Linux::glibc_version()); st->print(" "); 2162 st->print(os::Linux::libpthread_version()); st->print(" "); 2163 if (os::Linux::is_LinuxThreads()) { 2164 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2165 } 2166 st->cr(); 2167} 2168 2169void os::Linux::print_full_memory_info(outputStream* st) { 2170 st->print("\n/proc/meminfo:\n"); 2171 _print_ascii_file("/proc/meminfo", st); 2172 st->cr(); 2173} 2174 2175void os::print_memory_info(outputStream* st) { 2176 2177 st->print("Memory:"); 2178 st->print(" %dk page", os::vm_page_size()>>10); 2179 2180 // values in struct sysinfo are "unsigned long" 2181 struct sysinfo si; 2182 sysinfo(&si); 2183 2184 st->print(", physical " UINT64_FORMAT "k", 2185 os::physical_memory() >> 10); 2186 st->print("(" UINT64_FORMAT "k free)", 2187 os::available_memory() >> 10); 2188 st->print(", swap " UINT64_FORMAT "k", 2189 ((jlong)si.totalswap * si.mem_unit) >> 10); 2190 st->print("(" UINT64_FORMAT "k free)", 2191 ((jlong)si.freeswap * si.mem_unit) >> 10); 2192 st->cr(); 2193} 2194 2195void os::pd_print_cpu_info(outputStream* st) { 2196 st->print("\n/proc/cpuinfo:\n"); 2197 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2198 st->print(" <Not Available>"); 2199 } 2200 st->cr(); 2201} 2202 2203// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 2204// but they're the same for all the linux arch that we support 2205// and they're the same for solaris but there's no common place to put this. 2206const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 2207 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 2208 "ILL_COPROC", "ILL_BADSTK" }; 2209 2210const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 2211 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 2212 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 2213 2214const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 2215 2216const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 2217 2218void os::print_siginfo(outputStream* st, void* siginfo) { 2219 st->print("siginfo:"); 2220 2221 const int buflen = 100; 2222 char buf[buflen]; 2223 siginfo_t *si = (siginfo_t*)siginfo; 2224 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 2225 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 2226 st->print("si_errno=%s", buf); 2227 } else { 2228 st->print("si_errno=%d", si->si_errno); 2229 } 2230 const int c = si->si_code; 2231 assert(c > 0, "unexpected si_code"); 2232 switch (si->si_signo) { 2233 case SIGILL: 2234 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 2235 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2236 break; 2237 case SIGFPE: 2238 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 2239 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2240 break; 2241 case SIGSEGV: 2242 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 2243 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2244 break; 2245 case SIGBUS: 2246 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 2247 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2248 break; 2249 default: 2250 st->print(", si_code=%d", si->si_code); 2251 // no si_addr 2252 } 2253 2254 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2255 UseSharedSpaces) { 2256 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2257 if (mapinfo->is_in_shared_space(si->si_addr)) { 2258 st->print("\n\nError accessing class data sharing archive." \ 2259 " Mapped file inaccessible during execution, " \ 2260 " possible disk/network problem."); 2261 } 2262 } 2263 st->cr(); 2264} 2265 2266 2267static void print_signal_handler(outputStream* st, int sig, 2268 char* buf, size_t buflen); 2269 2270void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2271 st->print_cr("Signal Handlers:"); 2272 print_signal_handler(st, SIGSEGV, buf, buflen); 2273 print_signal_handler(st, SIGBUS , buf, buflen); 2274 print_signal_handler(st, SIGFPE , buf, buflen); 2275 print_signal_handler(st, SIGPIPE, buf, buflen); 2276 print_signal_handler(st, SIGXFSZ, buf, buflen); 2277 print_signal_handler(st, SIGILL , buf, buflen); 2278 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2279 print_signal_handler(st, SR_signum, buf, buflen); 2280 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2281 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2282 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2283 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2284} 2285 2286static char saved_jvm_path[MAXPATHLEN] = {0}; 2287 2288// Find the full path to the current module, libjvm.so 2289void os::jvm_path(char *buf, jint buflen) { 2290 // Error checking. 2291 if (buflen < MAXPATHLEN) { 2292 assert(false, "must use a large-enough buffer"); 2293 buf[0] = '\0'; 2294 return; 2295 } 2296 // Lazy resolve the path to current module. 2297 if (saved_jvm_path[0] != 0) { 2298 strcpy(buf, saved_jvm_path); 2299 return; 2300 } 2301 2302 char dli_fname[MAXPATHLEN]; 2303 bool ret = dll_address_to_library_name( 2304 CAST_FROM_FN_PTR(address, os::jvm_path), 2305 dli_fname, sizeof(dli_fname), NULL); 2306 assert(ret != 0, "cannot locate libjvm"); 2307 char *rp = realpath(dli_fname, buf); 2308 if (rp == NULL) 2309 return; 2310 2311 if (Arguments::created_by_gamma_launcher()) { 2312 // Support for the gamma launcher. Typical value for buf is 2313 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 2314 // the right place in the string, then assume we are installed in a JDK and 2315 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2316 // up the path so it looks like libjvm.so is installed there (append a 2317 // fake suffix hotspot/libjvm.so). 2318 const char *p = buf + strlen(buf) - 1; 2319 for (int count = 0; p > buf && count < 5; ++count) { 2320 for (--p; p > buf && *p != '/'; --p) 2321 /* empty */ ; 2322 } 2323 2324 if (strncmp(p, "/jre/lib/", 9) != 0) { 2325 // Look for JAVA_HOME in the environment. 2326 char* java_home_var = ::getenv("JAVA_HOME"); 2327 if (java_home_var != NULL && java_home_var[0] != 0) { 2328 char* jrelib_p; 2329 int len; 2330 2331 // Check the current module name "libjvm.so". 2332 p = strrchr(buf, '/'); 2333 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2334 2335 rp = realpath(java_home_var, buf); 2336 if (rp == NULL) 2337 return; 2338 2339 // determine if this is a legacy image or modules image 2340 // modules image doesn't have "jre" subdirectory 2341 len = strlen(buf); 2342 jrelib_p = buf + len; 2343 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2344 if (0 != access(buf, F_OK)) { 2345 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2346 } 2347 2348 if (0 == access(buf, F_OK)) { 2349 // Use current module name "libjvm.so" 2350 len = strlen(buf); 2351 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2352 } else { 2353 // Go back to path of .so 2354 rp = realpath(dli_fname, buf); 2355 if (rp == NULL) 2356 return; 2357 } 2358 } 2359 } 2360 } 2361 2362 strcpy(saved_jvm_path, buf); 2363} 2364 2365void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2366 // no prefix required, not even "_" 2367} 2368 2369void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2370 // no suffix required 2371} 2372 2373//////////////////////////////////////////////////////////////////////////////// 2374// sun.misc.Signal support 2375 2376static volatile jint sigint_count = 0; 2377 2378static void 2379UserHandler(int sig, void *siginfo, void *context) { 2380 // 4511530 - sem_post is serialized and handled by the manager thread. When 2381 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2382 // don't want to flood the manager thread with sem_post requests. 2383 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2384 return; 2385 2386 // Ctrl-C is pressed during error reporting, likely because the error 2387 // handler fails to abort. Let VM die immediately. 2388 if (sig == SIGINT && is_error_reported()) { 2389 os::die(); 2390 } 2391 2392 os::signal_notify(sig); 2393} 2394 2395void* os::user_handler() { 2396 return CAST_FROM_FN_PTR(void*, UserHandler); 2397} 2398 2399extern "C" { 2400 typedef void (*sa_handler_t)(int); 2401 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2402} 2403 2404void* os::signal(int signal_number, void* handler) { 2405 struct sigaction sigAct, oldSigAct; 2406 2407 sigfillset(&(sigAct.sa_mask)); 2408 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2409 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2410 2411 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2412 // -1 means registration failed 2413 return (void *)-1; 2414 } 2415 2416 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2417} 2418 2419void os::signal_raise(int signal_number) { 2420 ::raise(signal_number); 2421} 2422 2423/* 2424 * The following code is moved from os.cpp for making this 2425 * code platform specific, which it is by its very nature. 2426 */ 2427 2428// Will be modified when max signal is changed to be dynamic 2429int os::sigexitnum_pd() { 2430 return NSIG; 2431} 2432 2433// a counter for each possible signal value 2434static volatile jint pending_signals[NSIG+1] = { 0 }; 2435 2436// Linux(POSIX) specific hand shaking semaphore. 2437static sem_t sig_sem; 2438 2439void os::signal_init_pd() { 2440 // Initialize signal structures 2441 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2442 2443 // Initialize signal semaphore 2444 ::sem_init(&sig_sem, 0, 0); 2445} 2446 2447void os::signal_notify(int sig) { 2448 Atomic::inc(&pending_signals[sig]); 2449 ::sem_post(&sig_sem); 2450} 2451 2452static int check_pending_signals(bool wait) { 2453 Atomic::store(0, &sigint_count); 2454 for (;;) { 2455 for (int i = 0; i < NSIG + 1; i++) { 2456 jint n = pending_signals[i]; 2457 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2458 return i; 2459 } 2460 } 2461 if (!wait) { 2462 return -1; 2463 } 2464 JavaThread *thread = JavaThread::current(); 2465 ThreadBlockInVM tbivm(thread); 2466 2467 bool threadIsSuspended; 2468 do { 2469 thread->set_suspend_equivalent(); 2470 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2471 ::sem_wait(&sig_sem); 2472 2473 // were we externally suspended while we were waiting? 2474 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2475 if (threadIsSuspended) { 2476 // 2477 // The semaphore has been incremented, but while we were waiting 2478 // another thread suspended us. We don't want to continue running 2479 // while suspended because that would surprise the thread that 2480 // suspended us. 2481 // 2482 ::sem_post(&sig_sem); 2483 2484 thread->java_suspend_self(); 2485 } 2486 } while (threadIsSuspended); 2487 } 2488} 2489 2490int os::signal_lookup() { 2491 return check_pending_signals(false); 2492} 2493 2494int os::signal_wait() { 2495 return check_pending_signals(true); 2496} 2497 2498//////////////////////////////////////////////////////////////////////////////// 2499// Virtual Memory 2500 2501int os::vm_page_size() { 2502 // Seems redundant as all get out 2503 assert(os::Linux::page_size() != -1, "must call os::init"); 2504 return os::Linux::page_size(); 2505} 2506 2507// Solaris allocates memory by pages. 2508int os::vm_allocation_granularity() { 2509 assert(os::Linux::page_size() != -1, "must call os::init"); 2510 return os::Linux::page_size(); 2511} 2512 2513// Rationale behind this function: 2514// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2515// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2516// samples for JITted code. Here we create private executable mapping over the code cache 2517// and then we can use standard (well, almost, as mapping can change) way to provide 2518// info for the reporting script by storing timestamp and location of symbol 2519void linux_wrap_code(char* base, size_t size) { 2520 static volatile jint cnt = 0; 2521 2522 if (!UseOprofile) { 2523 return; 2524 } 2525 2526 char buf[PATH_MAX+1]; 2527 int num = Atomic::add(1, &cnt); 2528 2529 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2530 os::get_temp_directory(), os::current_process_id(), num); 2531 unlink(buf); 2532 2533 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2534 2535 if (fd != -1) { 2536 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2537 if (rv != (off_t)-1) { 2538 if (::write(fd, "", 1) == 1) { 2539 mmap(base, size, 2540 PROT_READ|PROT_WRITE|PROT_EXEC, 2541 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2542 } 2543 } 2544 ::close(fd); 2545 unlink(buf); 2546 } 2547} 2548 2549// NOTE: Linux kernel does not really reserve the pages for us. 2550// All it does is to check if there are enough free pages 2551// left at the time of mmap(). This could be a potential 2552// problem. 2553bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2554 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2555 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2556 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2557 if (res != (uintptr_t) MAP_FAILED) { 2558 if (UseNUMAInterleaving) { 2559 numa_make_global(addr, size); 2560 } 2561 return true; 2562 } 2563 return false; 2564} 2565 2566// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2567#ifndef MAP_HUGETLB 2568#define MAP_HUGETLB 0x40000 2569#endif 2570 2571// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2572#ifndef MADV_HUGEPAGE 2573#define MADV_HUGEPAGE 14 2574#endif 2575 2576bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2577 bool exec) { 2578 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) { 2579 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2580 uintptr_t res = 2581 (uintptr_t) ::mmap(addr, size, prot, 2582 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB, 2583 -1, 0); 2584 if (res != (uintptr_t) MAP_FAILED) { 2585 if (UseNUMAInterleaving) { 2586 numa_make_global(addr, size); 2587 } 2588 return true; 2589 } 2590 // Fall through and try to use small pages 2591 } 2592 2593 if (commit_memory(addr, size, exec)) { 2594 realign_memory(addr, size, alignment_hint); 2595 return true; 2596 } 2597 return false; 2598} 2599 2600void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2601 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) { 2602 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2603 // be supported or the memory may already be backed by huge pages. 2604 ::madvise(addr, bytes, MADV_HUGEPAGE); 2605 } 2606} 2607 2608void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2609 // This method works by doing an mmap over an existing mmaping and effectively discarding 2610 // the existing pages. However it won't work for SHM-based large pages that cannot be 2611 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2612 // small pages on top of the SHM segment. This method always works for small pages, so we 2613 // allow that in any case. 2614 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) { 2615 commit_memory(addr, bytes, alignment_hint, false); 2616 } 2617} 2618 2619void os::numa_make_global(char *addr, size_t bytes) { 2620 Linux::numa_interleave_memory(addr, bytes); 2621} 2622 2623void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2624 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2625} 2626 2627bool os::numa_topology_changed() { return false; } 2628 2629size_t os::numa_get_groups_num() { 2630 int max_node = Linux::numa_max_node(); 2631 return max_node > 0 ? max_node + 1 : 1; 2632} 2633 2634int os::numa_get_group_id() { 2635 int cpu_id = Linux::sched_getcpu(); 2636 if (cpu_id != -1) { 2637 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2638 if (lgrp_id != -1) { 2639 return lgrp_id; 2640 } 2641 } 2642 return 0; 2643} 2644 2645size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2646 for (size_t i = 0; i < size; i++) { 2647 ids[i] = i; 2648 } 2649 return size; 2650} 2651 2652bool os::get_page_info(char *start, page_info* info) { 2653 return false; 2654} 2655 2656char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2657 return end; 2658} 2659 2660 2661int os::Linux::sched_getcpu_syscall(void) { 2662 unsigned int cpu; 2663 int retval = -1; 2664 2665#if defined(IA32) 2666# ifndef SYS_getcpu 2667# define SYS_getcpu 318 2668# endif 2669 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2670#elif defined(AMD64) 2671// Unfortunately we have to bring all these macros here from vsyscall.h 2672// to be able to compile on old linuxes. 2673# define __NR_vgetcpu 2 2674# define VSYSCALL_START (-10UL << 20) 2675# define VSYSCALL_SIZE 1024 2676# define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2677 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2678 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2679 retval = vgetcpu(&cpu, NULL, NULL); 2680#endif 2681 2682 return (retval == -1) ? retval : cpu; 2683} 2684 2685// Something to do with the numa-aware allocator needs these symbols 2686extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2687extern "C" JNIEXPORT void numa_error(char *where) { } 2688extern "C" JNIEXPORT int fork1() { return fork(); } 2689 2690 2691// If we are running with libnuma version > 2, then we should 2692// be trying to use symbols with versions 1.1 2693// If we are running with earlier version, which did not have symbol versions, 2694// we should use the base version. 2695void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2696 void *f = dlvsym(handle, name, "libnuma_1.1"); 2697 if (f == NULL) { 2698 f = dlsym(handle, name); 2699 } 2700 return f; 2701} 2702 2703bool os::Linux::libnuma_init() { 2704 // sched_getcpu() should be in libc. 2705 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2706 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2707 2708 // If it's not, try a direct syscall. 2709 if (sched_getcpu() == -1) 2710 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall)); 2711 2712 if (sched_getcpu() != -1) { // Does it work? 2713 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2714 if (handle != NULL) { 2715 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2716 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2717 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2718 libnuma_dlsym(handle, "numa_max_node"))); 2719 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2720 libnuma_dlsym(handle, "numa_available"))); 2721 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2722 libnuma_dlsym(handle, "numa_tonode_memory"))); 2723 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2724 libnuma_dlsym(handle, "numa_interleave_memory"))); 2725 2726 2727 if (numa_available() != -1) { 2728 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2729 // Create a cpu -> node mapping 2730 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2731 rebuild_cpu_to_node_map(); 2732 return true; 2733 } 2734 } 2735 } 2736 return false; 2737} 2738 2739// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2740// The table is later used in get_node_by_cpu(). 2741void os::Linux::rebuild_cpu_to_node_map() { 2742 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2743 // in libnuma (possible values are starting from 16, 2744 // and continuing up with every other power of 2, but less 2745 // than the maximum number of CPUs supported by kernel), and 2746 // is a subject to change (in libnuma version 2 the requirements 2747 // are more reasonable) we'll just hardcode the number they use 2748 // in the library. 2749 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2750 2751 size_t cpu_num = os::active_processor_count(); 2752 size_t cpu_map_size = NCPUS / BitsPerCLong; 2753 size_t cpu_map_valid_size = 2754 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2755 2756 cpu_to_node()->clear(); 2757 cpu_to_node()->at_grow(cpu_num - 1); 2758 size_t node_num = numa_get_groups_num(); 2759 2760 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2761 for (size_t i = 0; i < node_num; i++) { 2762 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2763 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2764 if (cpu_map[j] != 0) { 2765 for (size_t k = 0; k < BitsPerCLong; k++) { 2766 if (cpu_map[j] & (1UL << k)) { 2767 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2768 } 2769 } 2770 } 2771 } 2772 } 2773 } 2774 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2775} 2776 2777int os::Linux::get_node_by_cpu(int cpu_id) { 2778 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2779 return cpu_to_node()->at(cpu_id); 2780 } 2781 return -1; 2782} 2783 2784GrowableArray<int>* os::Linux::_cpu_to_node; 2785os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2786os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2787os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2788os::Linux::numa_available_func_t os::Linux::_numa_available; 2789os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2790os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2791unsigned long* os::Linux::_numa_all_nodes; 2792 2793bool os::pd_uncommit_memory(char* addr, size_t size) { 2794 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2795 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2796 return res != (uintptr_t) MAP_FAILED; 2797} 2798 2799// Linux uses a growable mapping for the stack, and if the mapping for 2800// the stack guard pages is not removed when we detach a thread the 2801// stack cannot grow beyond the pages where the stack guard was 2802// mapped. If at some point later in the process the stack expands to 2803// that point, the Linux kernel cannot expand the stack any further 2804// because the guard pages are in the way, and a segfault occurs. 2805// 2806// However, it's essential not to split the stack region by unmapping 2807// a region (leaving a hole) that's already part of the stack mapping, 2808// so if the stack mapping has already grown beyond the guard pages at 2809// the time we create them, we have to truncate the stack mapping. 2810// So, we need to know the extent of the stack mapping when 2811// create_stack_guard_pages() is called. 2812 2813// Find the bounds of the stack mapping. Return true for success. 2814// 2815// We only need this for stacks that are growable: at the time of 2816// writing thread stacks don't use growable mappings (i.e. those 2817// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 2818// only applies to the main thread. 2819 2820static 2821bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) { 2822 2823 char buf[128]; 2824 int fd, sz; 2825 2826 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) { 2827 return false; 2828 } 2829 2830 const char kw[] = "[stack]"; 2831 const int kwlen = sizeof(kw)-1; 2832 2833 // Address part of /proc/self/maps couldn't be more than 128 bytes 2834 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) { 2835 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) { 2836 // Extract addresses 2837 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) { 2838 uintptr_t sp = (uintptr_t) __builtin_frame_address(0); 2839 if (sp >= *bottom && sp <= *top) { 2840 ::close(fd); 2841 return true; 2842 } 2843 } 2844 } 2845 } 2846 2847 ::close(fd); 2848 return false; 2849} 2850 2851 2852// If the (growable) stack mapping already extends beyond the point 2853// where we're going to put our guard pages, truncate the mapping at 2854// that point by munmap()ping it. This ensures that when we later 2855// munmap() the guard pages we don't leave a hole in the stack 2856// mapping. This only affects the main/initial thread, but guard 2857// against future OS changes 2858bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 2859 uintptr_t stack_extent, stack_base; 2860 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2861 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2862 assert(os::Linux::is_initial_thread(), 2863 "growable stack in non-initial thread"); 2864 if (stack_extent < (uintptr_t)addr) 2865 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent); 2866 } 2867 2868 return os::commit_memory(addr, size); 2869} 2870 2871// If this is a growable mapping, remove the guard pages entirely by 2872// munmap()ping them. If not, just call uncommit_memory(). This only 2873// affects the main/initial thread, but guard against future OS changes 2874bool os::remove_stack_guard_pages(char* addr, size_t size) { 2875 uintptr_t stack_extent, stack_base; 2876 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2877 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2878 assert(os::Linux::is_initial_thread(), 2879 "growable stack in non-initial thread"); 2880 2881 return ::munmap(addr, size) == 0; 2882 } 2883 2884 return os::uncommit_memory(addr, size); 2885} 2886 2887static address _highest_vm_reserved_address = NULL; 2888 2889// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2890// at 'requested_addr'. If there are existing memory mappings at the same 2891// location, however, they will be overwritten. If 'fixed' is false, 2892// 'requested_addr' is only treated as a hint, the return value may or 2893// may not start from the requested address. Unlike Linux mmap(), this 2894// function returns NULL to indicate failure. 2895static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2896 char * addr; 2897 int flags; 2898 2899 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2900 if (fixed) { 2901 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2902 flags |= MAP_FIXED; 2903 } 2904 2905 // Map uncommitted pages PROT_READ and PROT_WRITE, change access 2906 // to PROT_EXEC if executable when we commit the page. 2907 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE, 2908 flags, -1, 0); 2909 2910 if (addr != MAP_FAILED) { 2911 // anon_mmap() should only get called during VM initialization, 2912 // don't need lock (actually we can skip locking even it can be called 2913 // from multiple threads, because _highest_vm_reserved_address is just a 2914 // hint about the upper limit of non-stack memory regions.) 2915 if ((address)addr + bytes > _highest_vm_reserved_address) { 2916 _highest_vm_reserved_address = (address)addr + bytes; 2917 } 2918 } 2919 2920 return addr == MAP_FAILED ? NULL : addr; 2921} 2922 2923// Don't update _highest_vm_reserved_address, because there might be memory 2924// regions above addr + size. If so, releasing a memory region only creates 2925// a hole in the address space, it doesn't help prevent heap-stack collision. 2926// 2927static int anon_munmap(char * addr, size_t size) { 2928 return ::munmap(addr, size) == 0; 2929} 2930 2931char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 2932 size_t alignment_hint) { 2933 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 2934} 2935 2936bool os::pd_release_memory(char* addr, size_t size) { 2937 return anon_munmap(addr, size); 2938} 2939 2940static address highest_vm_reserved_address() { 2941 return _highest_vm_reserved_address; 2942} 2943 2944static bool linux_mprotect(char* addr, size_t size, int prot) { 2945 // Linux wants the mprotect address argument to be page aligned. 2946 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 2947 2948 // According to SUSv3, mprotect() should only be used with mappings 2949 // established by mmap(), and mmap() always maps whole pages. Unaligned 2950 // 'addr' likely indicates problem in the VM (e.g. trying to change 2951 // protection of malloc'ed or statically allocated memory). Check the 2952 // caller if you hit this assert. 2953 assert(addr == bottom, "sanity check"); 2954 2955 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 2956 return ::mprotect(bottom, size, prot) == 0; 2957} 2958 2959// Set protections specified 2960bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 2961 bool is_committed) { 2962 unsigned int p = 0; 2963 switch (prot) { 2964 case MEM_PROT_NONE: p = PROT_NONE; break; 2965 case MEM_PROT_READ: p = PROT_READ; break; 2966 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 2967 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 2968 default: 2969 ShouldNotReachHere(); 2970 } 2971 // is_committed is unused. 2972 return linux_mprotect(addr, bytes, p); 2973} 2974 2975bool os::guard_memory(char* addr, size_t size) { 2976 return linux_mprotect(addr, size, PROT_NONE); 2977} 2978 2979bool os::unguard_memory(char* addr, size_t size) { 2980 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 2981} 2982 2983bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 2984 bool result = false; 2985 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE, 2986 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 2987 -1, 0); 2988 2989 if (p != (void *) -1) { 2990 // We don't know if this really is a huge page or not. 2991 FILE *fp = fopen("/proc/self/maps", "r"); 2992 if (fp) { 2993 while (!feof(fp)) { 2994 char chars[257]; 2995 long x = 0; 2996 if (fgets(chars, sizeof(chars), fp)) { 2997 if (sscanf(chars, "%lx-%*x", &x) == 1 2998 && x == (long)p) { 2999 if (strstr (chars, "hugepage")) { 3000 result = true; 3001 break; 3002 } 3003 } 3004 } 3005 } 3006 fclose(fp); 3007 } 3008 munmap (p, page_size); 3009 if (result) 3010 return true; 3011 } 3012 3013 if (warn) { 3014 warning("HugeTLBFS is not supported by the operating system."); 3015 } 3016 3017 return result; 3018} 3019 3020/* 3021* Set the coredump_filter bits to include largepages in core dump (bit 6) 3022* 3023* From the coredump_filter documentation: 3024* 3025* - (bit 0) anonymous private memory 3026* - (bit 1) anonymous shared memory 3027* - (bit 2) file-backed private memory 3028* - (bit 3) file-backed shared memory 3029* - (bit 4) ELF header pages in file-backed private memory areas (it is 3030* effective only if the bit 2 is cleared) 3031* - (bit 5) hugetlb private memory 3032* - (bit 6) hugetlb shared memory 3033*/ 3034static void set_coredump_filter(void) { 3035 FILE *f; 3036 long cdm; 3037 3038 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3039 return; 3040 } 3041 3042 if (fscanf(f, "%lx", &cdm) != 1) { 3043 fclose(f); 3044 return; 3045 } 3046 3047 rewind(f); 3048 3049 if ((cdm & LARGEPAGES_BIT) == 0) { 3050 cdm |= LARGEPAGES_BIT; 3051 fprintf(f, "%#lx", cdm); 3052 } 3053 3054 fclose(f); 3055} 3056 3057// Large page support 3058 3059static size_t _large_page_size = 0; 3060 3061void os::large_page_init() { 3062 if (!UseLargePages) { 3063 UseHugeTLBFS = false; 3064 UseSHM = false; 3065 return; 3066 } 3067 3068 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) { 3069 // If UseLargePages is specified on the command line try both methods, 3070 // if it's default, then try only HugeTLBFS. 3071 if (FLAG_IS_DEFAULT(UseLargePages)) { 3072 UseHugeTLBFS = true; 3073 } else { 3074 UseHugeTLBFS = UseSHM = true; 3075 } 3076 } 3077 3078 if (LargePageSizeInBytes) { 3079 _large_page_size = LargePageSizeInBytes; 3080 } else { 3081 // large_page_size on Linux is used to round up heap size. x86 uses either 3082 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3083 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3084 // page as large as 256M. 3085 // 3086 // Here we try to figure out page size by parsing /proc/meminfo and looking 3087 // for a line with the following format: 3088 // Hugepagesize: 2048 kB 3089 // 3090 // If we can't determine the value (e.g. /proc is not mounted, or the text 3091 // format has been changed), we'll use the largest page size supported by 3092 // the processor. 3093 3094#ifndef ZERO 3095 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3096 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3097#endif // ZERO 3098 3099 FILE *fp = fopen("/proc/meminfo", "r"); 3100 if (fp) { 3101 while (!feof(fp)) { 3102 int x = 0; 3103 char buf[16]; 3104 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3105 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3106 _large_page_size = x * K; 3107 break; 3108 } 3109 } else { 3110 // skip to next line 3111 for (;;) { 3112 int ch = fgetc(fp); 3113 if (ch == EOF || ch == (int)'\n') break; 3114 } 3115 } 3116 } 3117 fclose(fp); 3118 } 3119 } 3120 3121 // print a warning if any large page related flag is specified on command line 3122 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3123 3124 const size_t default_page_size = (size_t)Linux::page_size(); 3125 if (_large_page_size > default_page_size) { 3126 _page_sizes[0] = _large_page_size; 3127 _page_sizes[1] = default_page_size; 3128 _page_sizes[2] = 0; 3129 } 3130 UseHugeTLBFS = UseHugeTLBFS && 3131 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size); 3132 3133 if (UseHugeTLBFS) 3134 UseSHM = false; 3135 3136 UseLargePages = UseHugeTLBFS || UseSHM; 3137 3138 set_coredump_filter(); 3139} 3140 3141#ifndef SHM_HUGETLB 3142#define SHM_HUGETLB 04000 3143#endif 3144 3145char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) { 3146 // "exec" is passed in but not used. Creating the shared image for 3147 // the code cache doesn't have an SHM_X executable permission to check. 3148 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3149 3150 key_t key = IPC_PRIVATE; 3151 char *addr; 3152 3153 bool warn_on_failure = UseLargePages && 3154 (!FLAG_IS_DEFAULT(UseLargePages) || 3155 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 3156 ); 3157 char msg[128]; 3158 3159 // Create a large shared memory region to attach to based on size. 3160 // Currently, size is the total size of the heap 3161 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3162 if (shmid == -1) { 3163 // Possible reasons for shmget failure: 3164 // 1. shmmax is too small for Java heap. 3165 // > check shmmax value: cat /proc/sys/kernel/shmmax 3166 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3167 // 2. not enough large page memory. 3168 // > check available large pages: cat /proc/meminfo 3169 // > increase amount of large pages: 3170 // echo new_value > /proc/sys/vm/nr_hugepages 3171 // Note 1: different Linux may use different name for this property, 3172 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3173 // Note 2: it's possible there's enough physical memory available but 3174 // they are so fragmented after a long run that they can't 3175 // coalesce into large pages. Try to reserve large pages when 3176 // the system is still "fresh". 3177 if (warn_on_failure) { 3178 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3179 warning(msg); 3180 } 3181 return NULL; 3182 } 3183 3184 // attach to the region 3185 addr = (char*)shmat(shmid, req_addr, 0); 3186 int err = errno; 3187 3188 // Remove shmid. If shmat() is successful, the actual shared memory segment 3189 // will be deleted when it's detached by shmdt() or when the process 3190 // terminates. If shmat() is not successful this will remove the shared 3191 // segment immediately. 3192 shmctl(shmid, IPC_RMID, NULL); 3193 3194 if ((intptr_t)addr == -1) { 3195 if (warn_on_failure) { 3196 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3197 warning(msg); 3198 } 3199 return NULL; 3200 } 3201 3202 if ((addr != NULL) && UseNUMAInterleaving) { 3203 numa_make_global(addr, bytes); 3204 } 3205 3206 // The memory is committed 3207 address pc = CALLER_PC; 3208 MemTracker::record_virtual_memory_reserve((address)addr, bytes, pc); 3209 MemTracker::record_virtual_memory_commit((address)addr, bytes, pc); 3210 3211 return addr; 3212} 3213 3214bool os::release_memory_special(char* base, size_t bytes) { 3215 // detaching the SHM segment will also delete it, see reserve_memory_special() 3216 int rslt = shmdt(base); 3217 if (rslt == 0) { 3218 MemTracker::record_virtual_memory_uncommit((address)base, bytes); 3219 MemTracker::record_virtual_memory_release((address)base, bytes); 3220 return true; 3221 } else { 3222 return false; 3223 } 3224} 3225 3226size_t os::large_page_size() { 3227 return _large_page_size; 3228} 3229 3230// HugeTLBFS allows application to commit large page memory on demand; 3231// with SysV SHM the entire memory region must be allocated as shared 3232// memory. 3233bool os::can_commit_large_page_memory() { 3234 return UseHugeTLBFS; 3235} 3236 3237bool os::can_execute_large_page_memory() { 3238 return UseHugeTLBFS; 3239} 3240 3241// Reserve memory at an arbitrary address, only if that area is 3242// available (and not reserved for something else). 3243 3244char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3245 const int max_tries = 10; 3246 char* base[max_tries]; 3247 size_t size[max_tries]; 3248 const size_t gap = 0x000000; 3249 3250 // Assert only that the size is a multiple of the page size, since 3251 // that's all that mmap requires, and since that's all we really know 3252 // about at this low abstraction level. If we need higher alignment, 3253 // we can either pass an alignment to this method or verify alignment 3254 // in one of the methods further up the call chain. See bug 5044738. 3255 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3256 3257 // Repeatedly allocate blocks until the block is allocated at the 3258 // right spot. Give up after max_tries. Note that reserve_memory() will 3259 // automatically update _highest_vm_reserved_address if the call is 3260 // successful. The variable tracks the highest memory address every reserved 3261 // by JVM. It is used to detect heap-stack collision if running with 3262 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3263 // space than needed, it could confuse the collision detecting code. To 3264 // solve the problem, save current _highest_vm_reserved_address and 3265 // calculate the correct value before return. 3266 address old_highest = _highest_vm_reserved_address; 3267 3268 // Linux mmap allows caller to pass an address as hint; give it a try first, 3269 // if kernel honors the hint then we can return immediately. 3270 char * addr = anon_mmap(requested_addr, bytes, false); 3271 if (addr == requested_addr) { 3272 return requested_addr; 3273 } 3274 3275 if (addr != NULL) { 3276 // mmap() is successful but it fails to reserve at the requested address 3277 anon_munmap(addr, bytes); 3278 } 3279 3280 int i; 3281 for (i = 0; i < max_tries; ++i) { 3282 base[i] = reserve_memory(bytes); 3283 3284 if (base[i] != NULL) { 3285 // Is this the block we wanted? 3286 if (base[i] == requested_addr) { 3287 size[i] = bytes; 3288 break; 3289 } 3290 3291 // Does this overlap the block we wanted? Give back the overlapped 3292 // parts and try again. 3293 3294 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3295 if (top_overlap >= 0 && top_overlap < bytes) { 3296 unmap_memory(base[i], top_overlap); 3297 base[i] += top_overlap; 3298 size[i] = bytes - top_overlap; 3299 } else { 3300 size_t bottom_overlap = base[i] + bytes - requested_addr; 3301 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3302 unmap_memory(requested_addr, bottom_overlap); 3303 size[i] = bytes - bottom_overlap; 3304 } else { 3305 size[i] = bytes; 3306 } 3307 } 3308 } 3309 } 3310 3311 // Give back the unused reserved pieces. 3312 3313 for (int j = 0; j < i; ++j) { 3314 if (base[j] != NULL) { 3315 unmap_memory(base[j], size[j]); 3316 } 3317 } 3318 3319 if (i < max_tries) { 3320 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3321 return requested_addr; 3322 } else { 3323 _highest_vm_reserved_address = old_highest; 3324 return NULL; 3325 } 3326} 3327 3328size_t os::read(int fd, void *buf, unsigned int nBytes) { 3329 return ::read(fd, buf, nBytes); 3330} 3331 3332// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 3333// Solaris uses poll(), linux uses park(). 3334// Poll() is likely a better choice, assuming that Thread.interrupt() 3335// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 3336// SIGSEGV, see 4355769. 3337 3338int os::sleep(Thread* thread, jlong millis, bool interruptible) { 3339 assert(thread == Thread::current(), "thread consistency check"); 3340 3341 ParkEvent * const slp = thread->_SleepEvent ; 3342 slp->reset() ; 3343 OrderAccess::fence() ; 3344 3345 if (interruptible) { 3346 jlong prevtime = javaTimeNanos(); 3347 3348 for (;;) { 3349 if (os::is_interrupted(thread, true)) { 3350 return OS_INTRPT; 3351 } 3352 3353 jlong newtime = javaTimeNanos(); 3354 3355 if (newtime - prevtime < 0) { 3356 // time moving backwards, should only happen if no monotonic clock 3357 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3358 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3359 } else { 3360 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3361 } 3362 3363 if(millis <= 0) { 3364 return OS_OK; 3365 } 3366 3367 prevtime = newtime; 3368 3369 { 3370 assert(thread->is_Java_thread(), "sanity check"); 3371 JavaThread *jt = (JavaThread *) thread; 3372 ThreadBlockInVM tbivm(jt); 3373 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 3374 3375 jt->set_suspend_equivalent(); 3376 // cleared by handle_special_suspend_equivalent_condition() or 3377 // java_suspend_self() via check_and_wait_while_suspended() 3378 3379 slp->park(millis); 3380 3381 // were we externally suspended while we were waiting? 3382 jt->check_and_wait_while_suspended(); 3383 } 3384 } 3385 } else { 3386 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 3387 jlong prevtime = javaTimeNanos(); 3388 3389 for (;;) { 3390 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 3391 // the 1st iteration ... 3392 jlong newtime = javaTimeNanos(); 3393 3394 if (newtime - prevtime < 0) { 3395 // time moving backwards, should only happen if no monotonic clock 3396 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3397 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3398 } else { 3399 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3400 } 3401 3402 if(millis <= 0) break ; 3403 3404 prevtime = newtime; 3405 slp->park(millis); 3406 } 3407 return OS_OK ; 3408 } 3409} 3410 3411int os::naked_sleep() { 3412 // %% make the sleep time an integer flag. for now use 1 millisec. 3413 return os::sleep(Thread::current(), 1, false); 3414} 3415 3416// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3417void os::infinite_sleep() { 3418 while (true) { // sleep forever ... 3419 ::sleep(100); // ... 100 seconds at a time 3420 } 3421} 3422 3423// Used to convert frequent JVM_Yield() to nops 3424bool os::dont_yield() { 3425 return DontYieldALot; 3426} 3427 3428void os::yield() { 3429 sched_yield(); 3430} 3431 3432os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3433 3434void os::yield_all(int attempts) { 3435 // Yields to all threads, including threads with lower priorities 3436 // Threads on Linux are all with same priority. The Solaris style 3437 // os::yield_all() with nanosleep(1ms) is not necessary. 3438 sched_yield(); 3439} 3440 3441// Called from the tight loops to possibly influence time-sharing heuristics 3442void os::loop_breaker(int attempts) { 3443 os::yield_all(attempts); 3444} 3445 3446//////////////////////////////////////////////////////////////////////////////// 3447// thread priority support 3448 3449// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3450// only supports dynamic priority, static priority must be zero. For real-time 3451// applications, Linux supports SCHED_RR which allows static priority (1-99). 3452// However, for large multi-threaded applications, SCHED_RR is not only slower 3453// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3454// of 5 runs - Sep 2005). 3455// 3456// The following code actually changes the niceness of kernel-thread/LWP. It 3457// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3458// not the entire user process, and user level threads are 1:1 mapped to kernel 3459// threads. It has always been the case, but could change in the future. For 3460// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3461// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3462 3463int os::java_to_os_priority[CriticalPriority + 1] = { 3464 19, // 0 Entry should never be used 3465 3466 4, // 1 MinPriority 3467 3, // 2 3468 2, // 3 3469 3470 1, // 4 3471 0, // 5 NormPriority 3472 -1, // 6 3473 3474 -2, // 7 3475 -3, // 8 3476 -4, // 9 NearMaxPriority 3477 3478 -5, // 10 MaxPriority 3479 3480 -5 // 11 CriticalPriority 3481}; 3482 3483static int prio_init() { 3484 if (ThreadPriorityPolicy == 1) { 3485 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3486 // if effective uid is not root. Perhaps, a more elegant way of doing 3487 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3488 if (geteuid() != 0) { 3489 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3490 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3491 } 3492 ThreadPriorityPolicy = 0; 3493 } 3494 } 3495 if (UseCriticalJavaThreadPriority) { 3496 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3497 } 3498 return 0; 3499} 3500 3501OSReturn os::set_native_priority(Thread* thread, int newpri) { 3502 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3503 3504 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3505 return (ret == 0) ? OS_OK : OS_ERR; 3506} 3507 3508OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3509 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3510 *priority_ptr = java_to_os_priority[NormPriority]; 3511 return OS_OK; 3512 } 3513 3514 errno = 0; 3515 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3516 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3517} 3518 3519// Hint to the underlying OS that a task switch would not be good. 3520// Void return because it's a hint and can fail. 3521void os::hint_no_preempt() {} 3522 3523//////////////////////////////////////////////////////////////////////////////// 3524// suspend/resume support 3525 3526// the low-level signal-based suspend/resume support is a remnant from the 3527// old VM-suspension that used to be for java-suspension, safepoints etc, 3528// within hotspot. Now there is a single use-case for this: 3529// - calling get_thread_pc() on the VMThread by the flat-profiler task 3530// that runs in the watcher thread. 3531// The remaining code is greatly simplified from the more general suspension 3532// code that used to be used. 3533// 3534// The protocol is quite simple: 3535// - suspend: 3536// - sends a signal to the target thread 3537// - polls the suspend state of the osthread using a yield loop 3538// - target thread signal handler (SR_handler) sets suspend state 3539// and blocks in sigsuspend until continued 3540// - resume: 3541// - sets target osthread state to continue 3542// - sends signal to end the sigsuspend loop in the SR_handler 3543// 3544// Note that the SR_lock plays no role in this suspend/resume protocol. 3545// 3546 3547static void resume_clear_context(OSThread *osthread) { 3548 osthread->set_ucontext(NULL); 3549 osthread->set_siginfo(NULL); 3550 3551 // notify the suspend action is completed, we have now resumed 3552 osthread->sr.clear_suspended(); 3553} 3554 3555static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3556 osthread->set_ucontext(context); 3557 osthread->set_siginfo(siginfo); 3558} 3559 3560// 3561// Handler function invoked when a thread's execution is suspended or 3562// resumed. We have to be careful that only async-safe functions are 3563// called here (Note: most pthread functions are not async safe and 3564// should be avoided.) 3565// 3566// Note: sigwait() is a more natural fit than sigsuspend() from an 3567// interface point of view, but sigwait() prevents the signal hander 3568// from being run. libpthread would get very confused by not having 3569// its signal handlers run and prevents sigwait()'s use with the 3570// mutex granting granting signal. 3571// 3572// Currently only ever called on the VMThread 3573// 3574static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3575 // Save and restore errno to avoid confusing native code with EINTR 3576 // after sigsuspend. 3577 int old_errno = errno; 3578 3579 Thread* thread = Thread::current(); 3580 OSThread* osthread = thread->osthread(); 3581 assert(thread->is_VM_thread(), "Must be VMThread"); 3582 // read current suspend action 3583 int action = osthread->sr.suspend_action(); 3584 if (action == os::Linux::SuspendResume::SR_SUSPEND) { 3585 suspend_save_context(osthread, siginfo, context); 3586 3587 // Notify the suspend action is about to be completed. do_suspend() 3588 // waits until SR_SUSPENDED is set and then returns. We will wait 3589 // here for a resume signal and that completes the suspend-other 3590 // action. do_suspend/do_resume is always called as a pair from 3591 // the same thread - so there are no races 3592 3593 // notify the caller 3594 osthread->sr.set_suspended(); 3595 3596 sigset_t suspend_set; // signals for sigsuspend() 3597 3598 // get current set of blocked signals and unblock resume signal 3599 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3600 sigdelset(&suspend_set, SR_signum); 3601 3602 // wait here until we are resumed 3603 do { 3604 sigsuspend(&suspend_set); 3605 // ignore all returns until we get a resume signal 3606 } while (osthread->sr.suspend_action() != os::Linux::SuspendResume::SR_CONTINUE); 3607 3608 resume_clear_context(osthread); 3609 3610 } else { 3611 assert(action == os::Linux::SuspendResume::SR_CONTINUE, "unexpected sr action"); 3612 // nothing special to do - just leave the handler 3613 } 3614 3615 errno = old_errno; 3616} 3617 3618 3619static int SR_initialize() { 3620 struct sigaction act; 3621 char *s; 3622 /* Get signal number to use for suspend/resume */ 3623 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3624 int sig = ::strtol(s, 0, 10); 3625 if (sig > 0 || sig < _NSIG) { 3626 SR_signum = sig; 3627 } 3628 } 3629 3630 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3631 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3632 3633 sigemptyset(&SR_sigset); 3634 sigaddset(&SR_sigset, SR_signum); 3635 3636 /* Set up signal handler for suspend/resume */ 3637 act.sa_flags = SA_RESTART|SA_SIGINFO; 3638 act.sa_handler = (void (*)(int)) SR_handler; 3639 3640 // SR_signum is blocked by default. 3641 // 4528190 - We also need to block pthread restart signal (32 on all 3642 // supported Linux platforms). Note that LinuxThreads need to block 3643 // this signal for all threads to work properly. So we don't have 3644 // to use hard-coded signal number when setting up the mask. 3645 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 3646 3647 if (sigaction(SR_signum, &act, 0) == -1) { 3648 return -1; 3649 } 3650 3651 // Save signal flag 3652 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 3653 return 0; 3654} 3655 3656 3657// returns true on success and false on error - really an error is fatal 3658// but this seems the normal response to library errors 3659static bool do_suspend(OSThread* osthread) { 3660 // mark as suspended and send signal 3661 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_SUSPEND); 3662 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3663 assert_status(status == 0, status, "pthread_kill"); 3664 3665 // check status and wait until notified of suspension 3666 if (status == 0) { 3667 for (int i = 0; !osthread->sr.is_suspended(); i++) { 3668 os::yield_all(i); 3669 } 3670 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE); 3671 return true; 3672 } 3673 else { 3674 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE); 3675 return false; 3676 } 3677} 3678 3679static void do_resume(OSThread* osthread) { 3680 assert(osthread->sr.is_suspended(), "thread should be suspended"); 3681 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_CONTINUE); 3682 3683 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3684 assert_status(status == 0, status, "pthread_kill"); 3685 // check status and wait unit notified of resumption 3686 if (status == 0) { 3687 for (int i = 0; osthread->sr.is_suspended(); i++) { 3688 os::yield_all(i); 3689 } 3690 } 3691 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE); 3692} 3693 3694//////////////////////////////////////////////////////////////////////////////// 3695// interrupt support 3696 3697void os::interrupt(Thread* thread) { 3698 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3699 "possibility of dangling Thread pointer"); 3700 3701 OSThread* osthread = thread->osthread(); 3702 3703 if (!osthread->interrupted()) { 3704 osthread->set_interrupted(true); 3705 // More than one thread can get here with the same value of osthread, 3706 // resulting in multiple notifications. We do, however, want the store 3707 // to interrupted() to be visible to other threads before we execute unpark(). 3708 OrderAccess::fence(); 3709 ParkEvent * const slp = thread->_SleepEvent ; 3710 if (slp != NULL) slp->unpark() ; 3711 } 3712 3713 // For JSR166. Unpark even if interrupt status already was set 3714 if (thread->is_Java_thread()) 3715 ((JavaThread*)thread)->parker()->unpark(); 3716 3717 ParkEvent * ev = thread->_ParkEvent ; 3718 if (ev != NULL) ev->unpark() ; 3719 3720} 3721 3722bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 3723 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3724 "possibility of dangling Thread pointer"); 3725 3726 OSThread* osthread = thread->osthread(); 3727 3728 bool interrupted = osthread->interrupted(); 3729 3730 if (interrupted && clear_interrupted) { 3731 osthread->set_interrupted(false); 3732 // consider thread->_SleepEvent->reset() ... optional optimization 3733 } 3734 3735 return interrupted; 3736} 3737 3738/////////////////////////////////////////////////////////////////////////////////// 3739// signal handling (except suspend/resume) 3740 3741// This routine may be used by user applications as a "hook" to catch signals. 3742// The user-defined signal handler must pass unrecognized signals to this 3743// routine, and if it returns true (non-zero), then the signal handler must 3744// return immediately. If the flag "abort_if_unrecognized" is true, then this 3745// routine will never retun false (zero), but instead will execute a VM panic 3746// routine kill the process. 3747// 3748// If this routine returns false, it is OK to call it again. This allows 3749// the user-defined signal handler to perform checks either before or after 3750// the VM performs its own checks. Naturally, the user code would be making 3751// a serious error if it tried to handle an exception (such as a null check 3752// or breakpoint) that the VM was generating for its own correct operation. 3753// 3754// This routine may recognize any of the following kinds of signals: 3755// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 3756// It should be consulted by handlers for any of those signals. 3757// 3758// The caller of this routine must pass in the three arguments supplied 3759// to the function referred to in the "sa_sigaction" (not the "sa_handler") 3760// field of the structure passed to sigaction(). This routine assumes that 3761// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 3762// 3763// Note that the VM will print warnings if it detects conflicting signal 3764// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 3765// 3766extern "C" JNIEXPORT int 3767JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 3768 void* ucontext, int abort_if_unrecognized); 3769 3770void signalHandler(int sig, siginfo_t* info, void* uc) { 3771 assert(info != NULL && uc != NULL, "it must be old kernel"); 3772 int orig_errno = errno; // Preserve errno value over signal handler. 3773 JVM_handle_linux_signal(sig, info, uc, true); 3774 errno = orig_errno; 3775} 3776 3777 3778// This boolean allows users to forward their own non-matching signals 3779// to JVM_handle_linux_signal, harmlessly. 3780bool os::Linux::signal_handlers_are_installed = false; 3781 3782// For signal-chaining 3783struct sigaction os::Linux::sigact[MAXSIGNUM]; 3784unsigned int os::Linux::sigs = 0; 3785bool os::Linux::libjsig_is_loaded = false; 3786typedef struct sigaction *(*get_signal_t)(int); 3787get_signal_t os::Linux::get_signal_action = NULL; 3788 3789struct sigaction* os::Linux::get_chained_signal_action(int sig) { 3790 struct sigaction *actp = NULL; 3791 3792 if (libjsig_is_loaded) { 3793 // Retrieve the old signal handler from libjsig 3794 actp = (*get_signal_action)(sig); 3795 } 3796 if (actp == NULL) { 3797 // Retrieve the preinstalled signal handler from jvm 3798 actp = get_preinstalled_handler(sig); 3799 } 3800 3801 return actp; 3802} 3803 3804static bool call_chained_handler(struct sigaction *actp, int sig, 3805 siginfo_t *siginfo, void *context) { 3806 // Call the old signal handler 3807 if (actp->sa_handler == SIG_DFL) { 3808 // It's more reasonable to let jvm treat it as an unexpected exception 3809 // instead of taking the default action. 3810 return false; 3811 } else if (actp->sa_handler != SIG_IGN) { 3812 if ((actp->sa_flags & SA_NODEFER) == 0) { 3813 // automaticlly block the signal 3814 sigaddset(&(actp->sa_mask), sig); 3815 } 3816 3817 sa_handler_t hand; 3818 sa_sigaction_t sa; 3819 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 3820 // retrieve the chained handler 3821 if (siginfo_flag_set) { 3822 sa = actp->sa_sigaction; 3823 } else { 3824 hand = actp->sa_handler; 3825 } 3826 3827 if ((actp->sa_flags & SA_RESETHAND) != 0) { 3828 actp->sa_handler = SIG_DFL; 3829 } 3830 3831 // try to honor the signal mask 3832 sigset_t oset; 3833 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 3834 3835 // call into the chained handler 3836 if (siginfo_flag_set) { 3837 (*sa)(sig, siginfo, context); 3838 } else { 3839 (*hand)(sig); 3840 } 3841 3842 // restore the signal mask 3843 pthread_sigmask(SIG_SETMASK, &oset, 0); 3844 } 3845 // Tell jvm's signal handler the signal is taken care of. 3846 return true; 3847} 3848 3849bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 3850 bool chained = false; 3851 // signal-chaining 3852 if (UseSignalChaining) { 3853 struct sigaction *actp = get_chained_signal_action(sig); 3854 if (actp != NULL) { 3855 chained = call_chained_handler(actp, sig, siginfo, context); 3856 } 3857 } 3858 return chained; 3859} 3860 3861struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 3862 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 3863 return &sigact[sig]; 3864 } 3865 return NULL; 3866} 3867 3868void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 3869 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3870 sigact[sig] = oldAct; 3871 sigs |= (unsigned int)1 << sig; 3872} 3873 3874// for diagnostic 3875int os::Linux::sigflags[MAXSIGNUM]; 3876 3877int os::Linux::get_our_sigflags(int sig) { 3878 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3879 return sigflags[sig]; 3880} 3881 3882void os::Linux::set_our_sigflags(int sig, int flags) { 3883 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3884 sigflags[sig] = flags; 3885} 3886 3887void os::Linux::set_signal_handler(int sig, bool set_installed) { 3888 // Check for overwrite. 3889 struct sigaction oldAct; 3890 sigaction(sig, (struct sigaction*)NULL, &oldAct); 3891 3892 void* oldhand = oldAct.sa_sigaction 3893 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3894 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3895 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 3896 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 3897 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 3898 if (AllowUserSignalHandlers || !set_installed) { 3899 // Do not overwrite; user takes responsibility to forward to us. 3900 return; 3901 } else if (UseSignalChaining) { 3902 // save the old handler in jvm 3903 save_preinstalled_handler(sig, oldAct); 3904 // libjsig also interposes the sigaction() call below and saves the 3905 // old sigaction on it own. 3906 } else { 3907 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 3908 "%#lx for signal %d.", (long)oldhand, sig)); 3909 } 3910 } 3911 3912 struct sigaction sigAct; 3913 sigfillset(&(sigAct.sa_mask)); 3914 sigAct.sa_handler = SIG_DFL; 3915 if (!set_installed) { 3916 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3917 } else { 3918 sigAct.sa_sigaction = signalHandler; 3919 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3920 } 3921 // Save flags, which are set by ours 3922 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3923 sigflags[sig] = sigAct.sa_flags; 3924 3925 int ret = sigaction(sig, &sigAct, &oldAct); 3926 assert(ret == 0, "check"); 3927 3928 void* oldhand2 = oldAct.sa_sigaction 3929 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3930 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3931 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 3932} 3933 3934// install signal handlers for signals that HotSpot needs to 3935// handle in order to support Java-level exception handling. 3936 3937void os::Linux::install_signal_handlers() { 3938 if (!signal_handlers_are_installed) { 3939 signal_handlers_are_installed = true; 3940 3941 // signal-chaining 3942 typedef void (*signal_setting_t)(); 3943 signal_setting_t begin_signal_setting = NULL; 3944 signal_setting_t end_signal_setting = NULL; 3945 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3946 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 3947 if (begin_signal_setting != NULL) { 3948 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3949 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 3950 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 3951 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 3952 libjsig_is_loaded = true; 3953 assert(UseSignalChaining, "should enable signal-chaining"); 3954 } 3955 if (libjsig_is_loaded) { 3956 // Tell libjsig jvm is setting signal handlers 3957 (*begin_signal_setting)(); 3958 } 3959 3960 set_signal_handler(SIGSEGV, true); 3961 set_signal_handler(SIGPIPE, true); 3962 set_signal_handler(SIGBUS, true); 3963 set_signal_handler(SIGILL, true); 3964 set_signal_handler(SIGFPE, true); 3965 set_signal_handler(SIGXFSZ, true); 3966 3967 if (libjsig_is_loaded) { 3968 // Tell libjsig jvm finishes setting signal handlers 3969 (*end_signal_setting)(); 3970 } 3971 3972 // We don't activate signal checker if libjsig is in place, we trust ourselves 3973 // and if UserSignalHandler is installed all bets are off. 3974 // Log that signal checking is off only if -verbose:jni is specified. 3975 if (CheckJNICalls) { 3976 if (libjsig_is_loaded) { 3977 if (PrintJNIResolving) { 3978 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 3979 } 3980 check_signals = false; 3981 } 3982 if (AllowUserSignalHandlers) { 3983 if (PrintJNIResolving) { 3984 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 3985 } 3986 check_signals = false; 3987 } 3988 } 3989 } 3990} 3991 3992// This is the fastest way to get thread cpu time on Linux. 3993// Returns cpu time (user+sys) for any thread, not only for current. 3994// POSIX compliant clocks are implemented in the kernels 2.6.16+. 3995// It might work on 2.6.10+ with a special kernel/glibc patch. 3996// For reference, please, see IEEE Std 1003.1-2004: 3997// http://www.unix.org/single_unix_specification 3998 3999jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4000 struct timespec tp; 4001 int rc = os::Linux::clock_gettime(clockid, &tp); 4002 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4003 4004 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4005} 4006 4007///// 4008// glibc on Linux platform uses non-documented flag 4009// to indicate, that some special sort of signal 4010// trampoline is used. 4011// We will never set this flag, and we should 4012// ignore this flag in our diagnostic 4013#ifdef SIGNIFICANT_SIGNAL_MASK 4014#undef SIGNIFICANT_SIGNAL_MASK 4015#endif 4016#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4017 4018static const char* get_signal_handler_name(address handler, 4019 char* buf, int buflen) { 4020 int offset; 4021 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4022 if (found) { 4023 // skip directory names 4024 const char *p1, *p2; 4025 p1 = buf; 4026 size_t len = strlen(os::file_separator()); 4027 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4028 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4029 } else { 4030 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4031 } 4032 return buf; 4033} 4034 4035static void print_signal_handler(outputStream* st, int sig, 4036 char* buf, size_t buflen) { 4037 struct sigaction sa; 4038 4039 sigaction(sig, NULL, &sa); 4040 4041 // See comment for SIGNIFICANT_SIGNAL_MASK define 4042 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4043 4044 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4045 4046 address handler = (sa.sa_flags & SA_SIGINFO) 4047 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4048 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4049 4050 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4051 st->print("SIG_DFL"); 4052 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4053 st->print("SIG_IGN"); 4054 } else { 4055 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4056 } 4057 4058 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 4059 4060 address rh = VMError::get_resetted_sighandler(sig); 4061 // May be, handler was resetted by VMError? 4062 if(rh != NULL) { 4063 handler = rh; 4064 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4065 } 4066 4067 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 4068 4069 // Check: is it our handler? 4070 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4071 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4072 // It is our signal handler 4073 // check for flags, reset system-used one! 4074 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4075 st->print( 4076 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4077 os::Linux::get_our_sigflags(sig)); 4078 } 4079 } 4080 st->cr(); 4081} 4082 4083 4084#define DO_SIGNAL_CHECK(sig) \ 4085 if (!sigismember(&check_signal_done, sig)) \ 4086 os::Linux::check_signal_handler(sig) 4087 4088// This method is a periodic task to check for misbehaving JNI applications 4089// under CheckJNI, we can add any periodic checks here 4090 4091void os::run_periodic_checks() { 4092 4093 if (check_signals == false) return; 4094 4095 // SEGV and BUS if overridden could potentially prevent 4096 // generation of hs*.log in the event of a crash, debugging 4097 // such a case can be very challenging, so we absolutely 4098 // check the following for a good measure: 4099 DO_SIGNAL_CHECK(SIGSEGV); 4100 DO_SIGNAL_CHECK(SIGILL); 4101 DO_SIGNAL_CHECK(SIGFPE); 4102 DO_SIGNAL_CHECK(SIGBUS); 4103 DO_SIGNAL_CHECK(SIGPIPE); 4104 DO_SIGNAL_CHECK(SIGXFSZ); 4105 4106 4107 // ReduceSignalUsage allows the user to override these handlers 4108 // see comments at the very top and jvm_solaris.h 4109 if (!ReduceSignalUsage) { 4110 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4111 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4112 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4113 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4114 } 4115 4116 DO_SIGNAL_CHECK(SR_signum); 4117 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4118} 4119 4120typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4121 4122static os_sigaction_t os_sigaction = NULL; 4123 4124void os::Linux::check_signal_handler(int sig) { 4125 char buf[O_BUFLEN]; 4126 address jvmHandler = NULL; 4127 4128 4129 struct sigaction act; 4130 if (os_sigaction == NULL) { 4131 // only trust the default sigaction, in case it has been interposed 4132 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4133 if (os_sigaction == NULL) return; 4134 } 4135 4136 os_sigaction(sig, (struct sigaction*)NULL, &act); 4137 4138 4139 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4140 4141 address thisHandler = (act.sa_flags & SA_SIGINFO) 4142 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4143 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 4144 4145 4146 switch(sig) { 4147 case SIGSEGV: 4148 case SIGBUS: 4149 case SIGFPE: 4150 case SIGPIPE: 4151 case SIGILL: 4152 case SIGXFSZ: 4153 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4154 break; 4155 4156 case SHUTDOWN1_SIGNAL: 4157 case SHUTDOWN2_SIGNAL: 4158 case SHUTDOWN3_SIGNAL: 4159 case BREAK_SIGNAL: 4160 jvmHandler = (address)user_handler(); 4161 break; 4162 4163 case INTERRUPT_SIGNAL: 4164 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4165 break; 4166 4167 default: 4168 if (sig == SR_signum) { 4169 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4170 } else { 4171 return; 4172 } 4173 break; 4174 } 4175 4176 if (thisHandler != jvmHandler) { 4177 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4178 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4179 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4180 // No need to check this sig any longer 4181 sigaddset(&check_signal_done, sig); 4182 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4183 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4184 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4185 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4186 // No need to check this sig any longer 4187 sigaddset(&check_signal_done, sig); 4188 } 4189 4190 // Dump all the signal 4191 if (sigismember(&check_signal_done, sig)) { 4192 print_signal_handlers(tty, buf, O_BUFLEN); 4193 } 4194} 4195 4196extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 4197 4198extern bool signal_name(int signo, char* buf, size_t len); 4199 4200const char* os::exception_name(int exception_code, char* buf, size_t size) { 4201 if (0 < exception_code && exception_code <= SIGRTMAX) { 4202 // signal 4203 if (!signal_name(exception_code, buf, size)) { 4204 jio_snprintf(buf, size, "SIG%d", exception_code); 4205 } 4206 return buf; 4207 } else { 4208 return NULL; 4209 } 4210} 4211 4212// this is called _before_ the most of global arguments have been parsed 4213void os::init(void) { 4214 char dummy; /* used to get a guess on initial stack address */ 4215// first_hrtime = gethrtime(); 4216 4217 // With LinuxThreads the JavaMain thread pid (primordial thread) 4218 // is different than the pid of the java launcher thread. 4219 // So, on Linux, the launcher thread pid is passed to the VM 4220 // via the sun.java.launcher.pid property. 4221 // Use this property instead of getpid() if it was correctly passed. 4222 // See bug 6351349. 4223 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4224 4225 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4226 4227 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4228 4229 init_random(1234567); 4230 4231 ThreadCritical::initialize(); 4232 4233 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4234 if (Linux::page_size() == -1) { 4235 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4236 strerror(errno))); 4237 } 4238 init_page_sizes((size_t) Linux::page_size()); 4239 4240 Linux::initialize_system_info(); 4241 4242 // main_thread points to the aboriginal thread 4243 Linux::_main_thread = pthread_self(); 4244 4245 Linux::clock_init(); 4246 initial_time_count = os::elapsed_counter(); 4247 pthread_mutex_init(&dl_mutex, NULL); 4248} 4249 4250// To install functions for atexit system call 4251extern "C" { 4252 static void perfMemory_exit_helper() { 4253 perfMemory_exit(); 4254 } 4255} 4256 4257// this is called _after_ the global arguments have been parsed 4258jint os::init_2(void) 4259{ 4260 Linux::fast_thread_clock_init(); 4261 4262 // Allocate a single page and mark it as readable for safepoint polling 4263 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4264 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 4265 4266 os::set_polling_page( polling_page ); 4267 4268#ifndef PRODUCT 4269 if(Verbose && PrintMiscellaneous) 4270 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 4271#endif 4272 4273 if (!UseMembar) { 4274 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4275 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); 4276 os::set_memory_serialize_page( mem_serialize_page ); 4277 4278#ifndef PRODUCT 4279 if(Verbose && PrintMiscellaneous) 4280 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 4281#endif 4282 } 4283 4284 os::large_page_init(); 4285 4286 // initialize suspend/resume support - must do this before signal_sets_init() 4287 if (SR_initialize() != 0) { 4288 perror("SR_initialize failed"); 4289 return JNI_ERR; 4290 } 4291 4292 Linux::signal_sets_init(); 4293 Linux::install_signal_handlers(); 4294 4295 // Check minimum allowable stack size for thread creation and to initialize 4296 // the java system classes, including StackOverflowError - depends on page 4297 // size. Add a page for compiler2 recursion in main thread. 4298 // Add in 2*BytesPerWord times page size to account for VM stack during 4299 // class initialization depending on 32 or 64 bit VM. 4300 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4301 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+ 4302 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size()); 4303 4304 size_t threadStackSizeInBytes = ThreadStackSize * K; 4305 if (threadStackSizeInBytes != 0 && 4306 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4307 tty->print_cr("\nThe stack size specified is too small, " 4308 "Specify at least %dk", 4309 os::Linux::min_stack_allowed/ K); 4310 return JNI_ERR; 4311 } 4312 4313 // Make the stack size a multiple of the page size so that 4314 // the yellow/red zones can be guarded. 4315 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4316 vm_page_size())); 4317 4318 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4319 4320 Linux::libpthread_init(); 4321 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4322 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4323 Linux::glibc_version(), Linux::libpthread_version(), 4324 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4325 } 4326 4327 if (UseNUMA) { 4328 if (!Linux::libnuma_init()) { 4329 UseNUMA = false; 4330 } else { 4331 if ((Linux::numa_max_node() < 1)) { 4332 // There's only one node(they start from 0), disable NUMA. 4333 UseNUMA = false; 4334 } 4335 } 4336 // With SHM large pages we cannot uncommit a page, so there's not way 4337 // we can make the adaptive lgrp chunk resizing work. If the user specified 4338 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and 4339 // disable adaptive resizing. 4340 if (UseNUMA && UseLargePages && UseSHM) { 4341 if (!FLAG_IS_DEFAULT(UseNUMA)) { 4342 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) { 4343 UseLargePages = false; 4344 } else { 4345 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing"); 4346 UseAdaptiveSizePolicy = false; 4347 UseAdaptiveNUMAChunkSizing = false; 4348 } 4349 } else { 4350 UseNUMA = false; 4351 } 4352 } 4353 if (!UseNUMA && ForceNUMA) { 4354 UseNUMA = true; 4355 } 4356 } 4357 4358 if (MaxFDLimit) { 4359 // set the number of file descriptors to max. print out error 4360 // if getrlimit/setrlimit fails but continue regardless. 4361 struct rlimit nbr_files; 4362 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4363 if (status != 0) { 4364 if (PrintMiscellaneous && (Verbose || WizardMode)) 4365 perror("os::init_2 getrlimit failed"); 4366 } else { 4367 nbr_files.rlim_cur = nbr_files.rlim_max; 4368 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4369 if (status != 0) { 4370 if (PrintMiscellaneous && (Verbose || WizardMode)) 4371 perror("os::init_2 setrlimit failed"); 4372 } 4373 } 4374 } 4375 4376 // Initialize lock used to serialize thread creation (see os::create_thread) 4377 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4378 4379 // at-exit methods are called in the reverse order of their registration. 4380 // atexit functions are called on return from main or as a result of a 4381 // call to exit(3C). There can be only 32 of these functions registered 4382 // and atexit() does not set errno. 4383 4384 if (PerfAllowAtExitRegistration) { 4385 // only register atexit functions if PerfAllowAtExitRegistration is set. 4386 // atexit functions can be delayed until process exit time, which 4387 // can be problematic for embedded VM situations. Embedded VMs should 4388 // call DestroyJavaVM() to assure that VM resources are released. 4389 4390 // note: perfMemory_exit_helper atexit function may be removed in 4391 // the future if the appropriate cleanup code can be added to the 4392 // VM_Exit VMOperation's doit method. 4393 if (atexit(perfMemory_exit_helper) != 0) { 4394 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 4395 } 4396 } 4397 4398 // initialize thread priority policy 4399 prio_init(); 4400 4401 return JNI_OK; 4402} 4403 4404// this is called at the end of vm_initialization 4405void os::init_3(void) 4406{ 4407#ifdef JAVASE_EMBEDDED 4408 // Start the MemNotifyThread 4409 if (LowMemoryProtection) { 4410 MemNotifyThread::start(); 4411 } 4412 return; 4413#endif 4414} 4415 4416// Mark the polling page as unreadable 4417void os::make_polling_page_unreadable(void) { 4418 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4419 fatal("Could not disable polling page"); 4420}; 4421 4422// Mark the polling page as readable 4423void os::make_polling_page_readable(void) { 4424 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4425 fatal("Could not enable polling page"); 4426 } 4427}; 4428 4429int os::active_processor_count() { 4430 // Linux doesn't yet have a (official) notion of processor sets, 4431 // so just return the number of online processors. 4432 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4433 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4434 return online_cpus; 4435} 4436 4437void os::set_native_thread_name(const char *name) { 4438 // Not yet implemented. 4439 return; 4440} 4441 4442bool os::distribute_processes(uint length, uint* distribution) { 4443 // Not yet implemented. 4444 return false; 4445} 4446 4447bool os::bind_to_processor(uint processor_id) { 4448 // Not yet implemented. 4449 return false; 4450} 4451 4452/// 4453 4454// Suspends the target using the signal mechanism and then grabs the PC before 4455// resuming the target. Used by the flat-profiler only 4456ExtendedPC os::get_thread_pc(Thread* thread) { 4457 // Make sure that it is called by the watcher for the VMThread 4458 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4459 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4460 4461 ExtendedPC epc; 4462 4463 OSThread* osthread = thread->osthread(); 4464 if (do_suspend(osthread)) { 4465 if (osthread->ucontext() != NULL) { 4466 epc = os::Linux::ucontext_get_pc(osthread->ucontext()); 4467 } else { 4468 // NULL context is unexpected, double-check this is the VMThread 4469 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4470 } 4471 do_resume(osthread); 4472 } 4473 // failure means pthread_kill failed for some reason - arguably this is 4474 // a fatal problem, but such problems are ignored elsewhere 4475 4476 return epc; 4477} 4478 4479int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4480{ 4481 if (is_NPTL()) { 4482 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4483 } else { 4484 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4485 // word back to default 64bit precision if condvar is signaled. Java 4486 // wants 53bit precision. Save and restore current value. 4487 int fpu = get_fpu_control_word(); 4488 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4489 set_fpu_control_word(fpu); 4490 return status; 4491 } 4492} 4493 4494//////////////////////////////////////////////////////////////////////////////// 4495// debug support 4496 4497bool os::find(address addr, outputStream* st) { 4498 Dl_info dlinfo; 4499 memset(&dlinfo, 0, sizeof(dlinfo)); 4500 if (dladdr(addr, &dlinfo)) { 4501 st->print(PTR_FORMAT ": ", addr); 4502 if (dlinfo.dli_sname != NULL) { 4503 st->print("%s+%#x", dlinfo.dli_sname, 4504 addr - (intptr_t)dlinfo.dli_saddr); 4505 } else if (dlinfo.dli_fname) { 4506 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4507 } else { 4508 st->print("<absolute address>"); 4509 } 4510 if (dlinfo.dli_fname) { 4511 st->print(" in %s", dlinfo.dli_fname); 4512 } 4513 if (dlinfo.dli_fbase) { 4514 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4515 } 4516 st->cr(); 4517 4518 if (Verbose) { 4519 // decode some bytes around the PC 4520 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 4521 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 4522 address lowest = (address) dlinfo.dli_sname; 4523 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4524 if (begin < lowest) begin = lowest; 4525 Dl_info dlinfo2; 4526 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr 4527 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 4528 end = (address) dlinfo2.dli_saddr; 4529 Disassembler::decode(begin, end, st); 4530 } 4531 return true; 4532 } 4533 return false; 4534} 4535 4536//////////////////////////////////////////////////////////////////////////////// 4537// misc 4538 4539// This does not do anything on Linux. This is basically a hook for being 4540// able to use structured exception handling (thread-local exception filters) 4541// on, e.g., Win32. 4542void 4543os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4544 JavaCallArguments* args, Thread* thread) { 4545 f(value, method, args, thread); 4546} 4547 4548void os::print_statistics() { 4549} 4550 4551int os::message_box(const char* title, const char* message) { 4552 int i; 4553 fdStream err(defaultStream::error_fd()); 4554 for (i = 0; i < 78; i++) err.print_raw("="); 4555 err.cr(); 4556 err.print_raw_cr(title); 4557 for (i = 0; i < 78; i++) err.print_raw("-"); 4558 err.cr(); 4559 err.print_raw_cr(message); 4560 for (i = 0; i < 78; i++) err.print_raw("="); 4561 err.cr(); 4562 4563 char buf[16]; 4564 // Prevent process from exiting upon "read error" without consuming all CPU 4565 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 4566 4567 return buf[0] == 'y' || buf[0] == 'Y'; 4568} 4569 4570int os::stat(const char *path, struct stat *sbuf) { 4571 char pathbuf[MAX_PATH]; 4572 if (strlen(path) > MAX_PATH - 1) { 4573 errno = ENAMETOOLONG; 4574 return -1; 4575 } 4576 os::native_path(strcpy(pathbuf, path)); 4577 return ::stat(pathbuf, sbuf); 4578} 4579 4580bool os::check_heap(bool force) { 4581 return true; 4582} 4583 4584int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 4585 return ::vsnprintf(buf, count, format, args); 4586} 4587 4588// Is a (classpath) directory empty? 4589bool os::dir_is_empty(const char* path) { 4590 DIR *dir = NULL; 4591 struct dirent *ptr; 4592 4593 dir = opendir(path); 4594 if (dir == NULL) return true; 4595 4596 /* Scan the directory */ 4597 bool result = true; 4598 char buf[sizeof(struct dirent) + MAX_PATH]; 4599 while (result && (ptr = ::readdir(dir)) != NULL) { 4600 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 4601 result = false; 4602 } 4603 } 4604 closedir(dir); 4605 return result; 4606} 4607 4608// This code originates from JDK's sysOpen and open64_w 4609// from src/solaris/hpi/src/system_md.c 4610 4611#ifndef O_DELETE 4612#define O_DELETE 0x10000 4613#endif 4614 4615// Open a file. Unlink the file immediately after open returns 4616// if the specified oflag has the O_DELETE flag set. 4617// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 4618 4619int os::open(const char *path, int oflag, int mode) { 4620 4621 if (strlen(path) > MAX_PATH - 1) { 4622 errno = ENAMETOOLONG; 4623 return -1; 4624 } 4625 int fd; 4626 int o_delete = (oflag & O_DELETE); 4627 oflag = oflag & ~O_DELETE; 4628 4629 fd = ::open64(path, oflag, mode); 4630 if (fd == -1) return -1; 4631 4632 //If the open succeeded, the file might still be a directory 4633 { 4634 struct stat64 buf64; 4635 int ret = ::fstat64(fd, &buf64); 4636 int st_mode = buf64.st_mode; 4637 4638 if (ret != -1) { 4639 if ((st_mode & S_IFMT) == S_IFDIR) { 4640 errno = EISDIR; 4641 ::close(fd); 4642 return -1; 4643 } 4644 } else { 4645 ::close(fd); 4646 return -1; 4647 } 4648 } 4649 4650 /* 4651 * All file descriptors that are opened in the JVM and not 4652 * specifically destined for a subprocess should have the 4653 * close-on-exec flag set. If we don't set it, then careless 3rd 4654 * party native code might fork and exec without closing all 4655 * appropriate file descriptors (e.g. as we do in closeDescriptors in 4656 * UNIXProcess.c), and this in turn might: 4657 * 4658 * - cause end-of-file to fail to be detected on some file 4659 * descriptors, resulting in mysterious hangs, or 4660 * 4661 * - might cause an fopen in the subprocess to fail on a system 4662 * suffering from bug 1085341. 4663 * 4664 * (Yes, the default setting of the close-on-exec flag is a Unix 4665 * design flaw) 4666 * 4667 * See: 4668 * 1085341: 32-bit stdio routines should support file descriptors >255 4669 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 4670 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 4671 */ 4672#ifdef FD_CLOEXEC 4673 { 4674 int flags = ::fcntl(fd, F_GETFD); 4675 if (flags != -1) 4676 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 4677 } 4678#endif 4679 4680 if (o_delete != 0) { 4681 ::unlink(path); 4682 } 4683 return fd; 4684} 4685 4686 4687// create binary file, rewriting existing file if required 4688int os::create_binary_file(const char* path, bool rewrite_existing) { 4689 int oflags = O_WRONLY | O_CREAT; 4690 if (!rewrite_existing) { 4691 oflags |= O_EXCL; 4692 } 4693 return ::open64(path, oflags, S_IREAD | S_IWRITE); 4694} 4695 4696// return current position of file pointer 4697jlong os::current_file_offset(int fd) { 4698 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 4699} 4700 4701// move file pointer to the specified offset 4702jlong os::seek_to_file_offset(int fd, jlong offset) { 4703 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 4704} 4705 4706// This code originates from JDK's sysAvailable 4707// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 4708 4709int os::available(int fd, jlong *bytes) { 4710 jlong cur, end; 4711 int mode; 4712 struct stat64 buf64; 4713 4714 if (::fstat64(fd, &buf64) >= 0) { 4715 mode = buf64.st_mode; 4716 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 4717 /* 4718 * XXX: is the following call interruptible? If so, this might 4719 * need to go through the INTERRUPT_IO() wrapper as for other 4720 * blocking, interruptible calls in this file. 4721 */ 4722 int n; 4723 if (::ioctl(fd, FIONREAD, &n) >= 0) { 4724 *bytes = n; 4725 return 1; 4726 } 4727 } 4728 } 4729 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 4730 return 0; 4731 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 4732 return 0; 4733 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 4734 return 0; 4735 } 4736 *bytes = end - cur; 4737 return 1; 4738} 4739 4740int os::socket_available(int fd, jint *pbytes) { 4741 // Linux doc says EINTR not returned, unlike Solaris 4742 int ret = ::ioctl(fd, FIONREAD, pbytes); 4743 4744 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 4745 // is expected to return 0 on failure and 1 on success to the jdk. 4746 return (ret < 0) ? 0 : 1; 4747} 4748 4749// Map a block of memory. 4750char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 4751 char *addr, size_t bytes, bool read_only, 4752 bool allow_exec) { 4753 int prot; 4754 int flags = MAP_PRIVATE; 4755 4756 if (read_only) { 4757 prot = PROT_READ; 4758 } else { 4759 prot = PROT_READ | PROT_WRITE; 4760 } 4761 4762 if (allow_exec) { 4763 prot |= PROT_EXEC; 4764 } 4765 4766 if (addr != NULL) { 4767 flags |= MAP_FIXED; 4768 } 4769 4770 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 4771 fd, file_offset); 4772 if (mapped_address == MAP_FAILED) { 4773 return NULL; 4774 } 4775 return mapped_address; 4776} 4777 4778 4779// Remap a block of memory. 4780char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 4781 char *addr, size_t bytes, bool read_only, 4782 bool allow_exec) { 4783 // same as map_memory() on this OS 4784 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 4785 allow_exec); 4786} 4787 4788 4789// Unmap a block of memory. 4790bool os::pd_unmap_memory(char* addr, size_t bytes) { 4791 return munmap(addr, bytes) == 0; 4792} 4793 4794static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 4795 4796static clockid_t thread_cpu_clockid(Thread* thread) { 4797 pthread_t tid = thread->osthread()->pthread_id(); 4798 clockid_t clockid; 4799 4800 // Get thread clockid 4801 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 4802 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 4803 return clockid; 4804} 4805 4806// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 4807// are used by JVM M&M and JVMTI to get user+sys or user CPU time 4808// of a thread. 4809// 4810// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 4811// the fast estimate available on the platform. 4812 4813jlong os::current_thread_cpu_time() { 4814 if (os::Linux::supports_fast_thread_cpu_time()) { 4815 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4816 } else { 4817 // return user + sys since the cost is the same 4818 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 4819 } 4820} 4821 4822jlong os::thread_cpu_time(Thread* thread) { 4823 // consistent with what current_thread_cpu_time() returns 4824 if (os::Linux::supports_fast_thread_cpu_time()) { 4825 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4826 } else { 4827 return slow_thread_cpu_time(thread, true /* user + sys */); 4828 } 4829} 4830 4831jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 4832 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4833 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4834 } else { 4835 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 4836 } 4837} 4838 4839jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4840 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4841 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4842 } else { 4843 return slow_thread_cpu_time(thread, user_sys_cpu_time); 4844 } 4845} 4846 4847// 4848// -1 on error. 4849// 4850 4851static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4852 static bool proc_task_unchecked = true; 4853 static const char *proc_stat_path = "/proc/%d/stat"; 4854 pid_t tid = thread->osthread()->thread_id(); 4855 char *s; 4856 char stat[2048]; 4857 int statlen; 4858 char proc_name[64]; 4859 int count; 4860 long sys_time, user_time; 4861 char cdummy; 4862 int idummy; 4863 long ldummy; 4864 FILE *fp; 4865 4866 // The /proc/<tid>/stat aggregates per-process usage on 4867 // new Linux kernels 2.6+ where NPTL is supported. 4868 // The /proc/self/task/<tid>/stat still has the per-thread usage. 4869 // See bug 6328462. 4870 // There possibly can be cases where there is no directory 4871 // /proc/self/task, so we check its availability. 4872 if (proc_task_unchecked && os::Linux::is_NPTL()) { 4873 // This is executed only once 4874 proc_task_unchecked = false; 4875 fp = fopen("/proc/self/task", "r"); 4876 if (fp != NULL) { 4877 proc_stat_path = "/proc/self/task/%d/stat"; 4878 fclose(fp); 4879 } 4880 } 4881 4882 sprintf(proc_name, proc_stat_path, tid); 4883 fp = fopen(proc_name, "r"); 4884 if ( fp == NULL ) return -1; 4885 statlen = fread(stat, 1, 2047, fp); 4886 stat[statlen] = '\0'; 4887 fclose(fp); 4888 4889 // Skip pid and the command string. Note that we could be dealing with 4890 // weird command names, e.g. user could decide to rename java launcher 4891 // to "java 1.4.2 :)", then the stat file would look like 4892 // 1234 (java 1.4.2 :)) R ... ... 4893 // We don't really need to know the command string, just find the last 4894 // occurrence of ")" and then start parsing from there. See bug 4726580. 4895 s = strrchr(stat, ')'); 4896 if (s == NULL ) return -1; 4897 4898 // Skip blank chars 4899 do s++; while (isspace(*s)); 4900 4901 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 4902 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 4903 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 4904 &user_time, &sys_time); 4905 if ( count != 13 ) return -1; 4906 if (user_sys_cpu_time) { 4907 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4908 } else { 4909 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4910 } 4911} 4912 4913void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4914 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4915 info_ptr->may_skip_backward = false; // elapsed time not wall time 4916 info_ptr->may_skip_forward = false; // elapsed time not wall time 4917 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4918} 4919 4920void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4921 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4922 info_ptr->may_skip_backward = false; // elapsed time not wall time 4923 info_ptr->may_skip_forward = false; // elapsed time not wall time 4924 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4925} 4926 4927bool os::is_thread_cpu_time_supported() { 4928 return true; 4929} 4930 4931// System loadavg support. Returns -1 if load average cannot be obtained. 4932// Linux doesn't yet have a (official) notion of processor sets, 4933// so just return the system wide load average. 4934int os::loadavg(double loadavg[], int nelem) { 4935 return ::getloadavg(loadavg, nelem); 4936} 4937 4938void os::pause() { 4939 char filename[MAX_PATH]; 4940 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 4941 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 4942 } else { 4943 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 4944 } 4945 4946 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 4947 if (fd != -1) { 4948 struct stat buf; 4949 ::close(fd); 4950 while (::stat(filename, &buf) == 0) { 4951 (void)::poll(NULL, 0, 100); 4952 } 4953 } else { 4954 jio_fprintf(stderr, 4955 "Could not open pause file '%s', continuing immediately.\n", filename); 4956 } 4957} 4958 4959 4960// Refer to the comments in os_solaris.cpp park-unpark. 4961// 4962// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 4963// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 4964// For specifics regarding the bug see GLIBC BUGID 261237 : 4965// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 4966// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 4967// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 4968// is used. (The simple C test-case provided in the GLIBC bug report manifests the 4969// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 4970// and monitorenter when we're using 1-0 locking. All those operations may result in 4971// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 4972// of libpthread avoids the problem, but isn't practical. 4973// 4974// Possible remedies: 4975// 4976// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 4977// This is palliative and probabilistic, however. If the thread is preempted 4978// between the call to compute_abstime() and pthread_cond_timedwait(), more 4979// than the minimum period may have passed, and the abstime may be stale (in the 4980// past) resultin in a hang. Using this technique reduces the odds of a hang 4981// but the JVM is still vulnerable, particularly on heavily loaded systems. 4982// 4983// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 4984// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 4985// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 4986// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 4987// thread. 4988// 4989// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 4990// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 4991// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 4992// This also works well. In fact it avoids kernel-level scalability impediments 4993// on certain platforms that don't handle lots of active pthread_cond_timedwait() 4994// timers in a graceful fashion. 4995// 4996// 4. When the abstime value is in the past it appears that control returns 4997// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 4998// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 4999// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5000// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5001// It may be possible to avoid reinitialization by checking the return 5002// value from pthread_cond_timedwait(). In addition to reinitializing the 5003// condvar we must establish the invariant that cond_signal() is only called 5004// within critical sections protected by the adjunct mutex. This prevents 5005// cond_signal() from "seeing" a condvar that's in the midst of being 5006// reinitialized or that is corrupt. Sadly, this invariant obviates the 5007// desirable signal-after-unlock optimization that avoids futile context switching. 5008// 5009// I'm also concerned that some versions of NTPL might allocate an auxilliary 5010// structure when a condvar is used or initialized. cond_destroy() would 5011// release the helper structure. Our reinitialize-after-timedwait fix 5012// put excessive stress on malloc/free and locks protecting the c-heap. 5013// 5014// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5015// It may be possible to refine (4) by checking the kernel and NTPL verisons 5016// and only enabling the work-around for vulnerable environments. 5017 5018// utility to compute the abstime argument to timedwait: 5019// millis is the relative timeout time 5020// abstime will be the absolute timeout time 5021// TODO: replace compute_abstime() with unpackTime() 5022 5023static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5024 if (millis < 0) millis = 0; 5025 struct timeval now; 5026 int status = gettimeofday(&now, NULL); 5027 assert(status == 0, "gettimeofday"); 5028 jlong seconds = millis / 1000; 5029 millis %= 1000; 5030 if (seconds > 50000000) { // see man cond_timedwait(3T) 5031 seconds = 50000000; 5032 } 5033 abstime->tv_sec = now.tv_sec + seconds; 5034 long usec = now.tv_usec + millis * 1000; 5035 if (usec >= 1000000) { 5036 abstime->tv_sec += 1; 5037 usec -= 1000000; 5038 } 5039 abstime->tv_nsec = usec * 1000; 5040 return abstime; 5041} 5042 5043 5044// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 5045// Conceptually TryPark() should be equivalent to park(0). 5046 5047int os::PlatformEvent::TryPark() { 5048 for (;;) { 5049 const int v = _Event ; 5050 guarantee ((v == 0) || (v == 1), "invariant") ; 5051 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 5052 } 5053} 5054 5055void os::PlatformEvent::park() { // AKA "down()" 5056 // Invariant: Only the thread associated with the Event/PlatformEvent 5057 // may call park(). 5058 // TODO: assert that _Assoc != NULL or _Assoc == Self 5059 int v ; 5060 for (;;) { 5061 v = _Event ; 5062 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5063 } 5064 guarantee (v >= 0, "invariant") ; 5065 if (v == 0) { 5066 // Do this the hard way by blocking ... 5067 int status = pthread_mutex_lock(_mutex); 5068 assert_status(status == 0, status, "mutex_lock"); 5069 guarantee (_nParked == 0, "invariant") ; 5070 ++ _nParked ; 5071 while (_Event < 0) { 5072 status = pthread_cond_wait(_cond, _mutex); 5073 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5074 // Treat this the same as if the wait was interrupted 5075 if (status == ETIME) { status = EINTR; } 5076 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5077 } 5078 -- _nParked ; 5079 5080 _Event = 0 ; 5081 status = pthread_mutex_unlock(_mutex); 5082 assert_status(status == 0, status, "mutex_unlock"); 5083 // Paranoia to ensure our locked and lock-free paths interact 5084 // correctly with each other. 5085 OrderAccess::fence(); 5086 } 5087 guarantee (_Event >= 0, "invariant") ; 5088} 5089 5090int os::PlatformEvent::park(jlong millis) { 5091 guarantee (_nParked == 0, "invariant") ; 5092 5093 int v ; 5094 for (;;) { 5095 v = _Event ; 5096 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5097 } 5098 guarantee (v >= 0, "invariant") ; 5099 if (v != 0) return OS_OK ; 5100 5101 // We do this the hard way, by blocking the thread. 5102 // Consider enforcing a minimum timeout value. 5103 struct timespec abst; 5104 compute_abstime(&abst, millis); 5105 5106 int ret = OS_TIMEOUT; 5107 int status = pthread_mutex_lock(_mutex); 5108 assert_status(status == 0, status, "mutex_lock"); 5109 guarantee (_nParked == 0, "invariant") ; 5110 ++_nParked ; 5111 5112 // Object.wait(timo) will return because of 5113 // (a) notification 5114 // (b) timeout 5115 // (c) thread.interrupt 5116 // 5117 // Thread.interrupt and object.notify{All} both call Event::set. 5118 // That is, we treat thread.interrupt as a special case of notification. 5119 // The underlying Solaris implementation, cond_timedwait, admits 5120 // spurious/premature wakeups, but the JLS/JVM spec prevents the 5121 // JVM from making those visible to Java code. As such, we must 5122 // filter out spurious wakeups. We assume all ETIME returns are valid. 5123 // 5124 // TODO: properly differentiate simultaneous notify+interrupt. 5125 // In that case, we should propagate the notify to another waiter. 5126 5127 while (_Event < 0) { 5128 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5129 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5130 pthread_cond_destroy (_cond); 5131 pthread_cond_init (_cond, NULL) ; 5132 } 5133 assert_status(status == 0 || status == EINTR || 5134 status == ETIME || status == ETIMEDOUT, 5135 status, "cond_timedwait"); 5136 if (!FilterSpuriousWakeups) break ; // previous semantics 5137 if (status == ETIME || status == ETIMEDOUT) break ; 5138 // We consume and ignore EINTR and spurious wakeups. 5139 } 5140 --_nParked ; 5141 if (_Event >= 0) { 5142 ret = OS_OK; 5143 } 5144 _Event = 0 ; 5145 status = pthread_mutex_unlock(_mutex); 5146 assert_status(status == 0, status, "mutex_unlock"); 5147 assert (_nParked == 0, "invariant") ; 5148 // Paranoia to ensure our locked and lock-free paths interact 5149 // correctly with each other. 5150 OrderAccess::fence(); 5151 return ret; 5152} 5153 5154void os::PlatformEvent::unpark() { 5155 // Transitions for _Event: 5156 // 0 :=> 1 5157 // 1 :=> 1 5158 // -1 :=> either 0 or 1; must signal target thread 5159 // That is, we can safely transition _Event from -1 to either 5160 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back 5161 // unpark() calls. 5162 // See also: "Semaphores in Plan 9" by Mullender & Cox 5163 // 5164 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5165 // that it will take two back-to-back park() calls for the owning 5166 // thread to block. This has the benefit of forcing a spurious return 5167 // from the first park() call after an unpark() call which will help 5168 // shake out uses of park() and unpark() without condition variables. 5169 5170 if (Atomic::xchg(1, &_Event) >= 0) return; 5171 5172 // Wait for the thread associated with the event to vacate 5173 int status = pthread_mutex_lock(_mutex); 5174 assert_status(status == 0, status, "mutex_lock"); 5175 int AnyWaiters = _nParked; 5176 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5177 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5178 AnyWaiters = 0; 5179 pthread_cond_signal(_cond); 5180 } 5181 status = pthread_mutex_unlock(_mutex); 5182 assert_status(status == 0, status, "mutex_unlock"); 5183 if (AnyWaiters != 0) { 5184 status = pthread_cond_signal(_cond); 5185 assert_status(status == 0, status, "cond_signal"); 5186 } 5187 5188 // Note that we signal() _after dropping the lock for "immortal" Events. 5189 // This is safe and avoids a common class of futile wakeups. In rare 5190 // circumstances this can cause a thread to return prematurely from 5191 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5192 // simply re-test the condition and re-park itself. 5193} 5194 5195 5196// JSR166 5197// ------------------------------------------------------- 5198 5199/* 5200 * The solaris and linux implementations of park/unpark are fairly 5201 * conservative for now, but can be improved. They currently use a 5202 * mutex/condvar pair, plus a a count. 5203 * Park decrements count if > 0, else does a condvar wait. Unpark 5204 * sets count to 1 and signals condvar. Only one thread ever waits 5205 * on the condvar. Contention seen when trying to park implies that someone 5206 * is unparking you, so don't wait. And spurious returns are fine, so there 5207 * is no need to track notifications. 5208 */ 5209 5210#define MAX_SECS 100000000 5211/* 5212 * This code is common to linux and solaris and will be moved to a 5213 * common place in dolphin. 5214 * 5215 * The passed in time value is either a relative time in nanoseconds 5216 * or an absolute time in milliseconds. Either way it has to be unpacked 5217 * into suitable seconds and nanoseconds components and stored in the 5218 * given timespec structure. 5219 * Given time is a 64-bit value and the time_t used in the timespec is only 5220 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 5221 * overflow if times way in the future are given. Further on Solaris versions 5222 * prior to 10 there is a restriction (see cond_timedwait) that the specified 5223 * number of seconds, in abstime, is less than current_time + 100,000,000. 5224 * As it will be 28 years before "now + 100000000" will overflow we can 5225 * ignore overflow and just impose a hard-limit on seconds using the value 5226 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 5227 * years from "now". 5228 */ 5229 5230static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5231 assert (time > 0, "convertTime"); 5232 5233 struct timeval now; 5234 int status = gettimeofday(&now, NULL); 5235 assert(status == 0, "gettimeofday"); 5236 5237 time_t max_secs = now.tv_sec + MAX_SECS; 5238 5239 if (isAbsolute) { 5240 jlong secs = time / 1000; 5241 if (secs > max_secs) { 5242 absTime->tv_sec = max_secs; 5243 } 5244 else { 5245 absTime->tv_sec = secs; 5246 } 5247 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5248 } 5249 else { 5250 jlong secs = time / NANOSECS_PER_SEC; 5251 if (secs >= MAX_SECS) { 5252 absTime->tv_sec = max_secs; 5253 absTime->tv_nsec = 0; 5254 } 5255 else { 5256 absTime->tv_sec = now.tv_sec + secs; 5257 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5258 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5259 absTime->tv_nsec -= NANOSECS_PER_SEC; 5260 ++absTime->tv_sec; // note: this must be <= max_secs 5261 } 5262 } 5263 } 5264 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5265 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5266 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5267 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5268} 5269 5270void Parker::park(bool isAbsolute, jlong time) { 5271 // Ideally we'd do something useful while spinning, such 5272 // as calling unpackTime(). 5273 5274 // Optional fast-path check: 5275 // Return immediately if a permit is available. 5276 // We depend on Atomic::xchg() having full barrier semantics 5277 // since we are doing a lock-free update to _counter. 5278 if (Atomic::xchg(0, &_counter) > 0) return; 5279 5280 Thread* thread = Thread::current(); 5281 assert(thread->is_Java_thread(), "Must be JavaThread"); 5282 JavaThread *jt = (JavaThread *)thread; 5283 5284 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5285 // Check interrupt before trying to wait 5286 if (Thread::is_interrupted(thread, false)) { 5287 return; 5288 } 5289 5290 // Next, demultiplex/decode time arguments 5291 timespec absTime; 5292 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5293 return; 5294 } 5295 if (time > 0) { 5296 unpackTime(&absTime, isAbsolute, time); 5297 } 5298 5299 5300 // Enter safepoint region 5301 // Beware of deadlocks such as 6317397. 5302 // The per-thread Parker:: mutex is a classic leaf-lock. 5303 // In particular a thread must never block on the Threads_lock while 5304 // holding the Parker:: mutex. If safepoints are pending both the 5305 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5306 ThreadBlockInVM tbivm(jt); 5307 5308 // Don't wait if cannot get lock since interference arises from 5309 // unblocking. Also. check interrupt before trying wait 5310 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5311 return; 5312 } 5313 5314 int status ; 5315 if (_counter > 0) { // no wait needed 5316 _counter = 0; 5317 status = pthread_mutex_unlock(_mutex); 5318 assert (status == 0, "invariant") ; 5319 // Paranoia to ensure our locked and lock-free paths interact 5320 // correctly with each other and Java-level accesses. 5321 OrderAccess::fence(); 5322 return; 5323 } 5324 5325#ifdef ASSERT 5326 // Don't catch signals while blocked; let the running threads have the signals. 5327 // (This allows a debugger to break into the running thread.) 5328 sigset_t oldsigs; 5329 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5330 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5331#endif 5332 5333 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5334 jt->set_suspend_equivalent(); 5335 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5336 5337 if (time == 0) { 5338 status = pthread_cond_wait (_cond, _mutex) ; 5339 } else { 5340 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 5341 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5342 pthread_cond_destroy (_cond) ; 5343 pthread_cond_init (_cond, NULL); 5344 } 5345 } 5346 assert_status(status == 0 || status == EINTR || 5347 status == ETIME || status == ETIMEDOUT, 5348 status, "cond_timedwait"); 5349 5350#ifdef ASSERT 5351 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5352#endif 5353 5354 _counter = 0 ; 5355 status = pthread_mutex_unlock(_mutex) ; 5356 assert_status(status == 0, status, "invariant") ; 5357 // Paranoia to ensure our locked and lock-free paths interact 5358 // correctly with each other and Java-level accesses. 5359 OrderAccess::fence(); 5360 5361 // If externally suspended while waiting, re-suspend 5362 if (jt->handle_special_suspend_equivalent_condition()) { 5363 jt->java_suspend_self(); 5364 } 5365} 5366 5367void Parker::unpark() { 5368 int s, status ; 5369 status = pthread_mutex_lock(_mutex); 5370 assert (status == 0, "invariant") ; 5371 s = _counter; 5372 _counter = 1; 5373 if (s < 1) { 5374 if (WorkAroundNPTLTimedWaitHang) { 5375 status = pthread_cond_signal (_cond) ; 5376 assert (status == 0, "invariant") ; 5377 status = pthread_mutex_unlock(_mutex); 5378 assert (status == 0, "invariant") ; 5379 } else { 5380 status = pthread_mutex_unlock(_mutex); 5381 assert (status == 0, "invariant") ; 5382 status = pthread_cond_signal (_cond) ; 5383 assert (status == 0, "invariant") ; 5384 } 5385 } else { 5386 pthread_mutex_unlock(_mutex); 5387 assert (status == 0, "invariant") ; 5388 } 5389} 5390 5391 5392extern char** environ; 5393 5394#ifndef __NR_fork 5395#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5396#endif 5397 5398#ifndef __NR_execve 5399#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5400#endif 5401 5402// Run the specified command in a separate process. Return its exit value, 5403// or -1 on failure (e.g. can't fork a new process). 5404// Unlike system(), this function can be called from signal handler. It 5405// doesn't block SIGINT et al. 5406int os::fork_and_exec(char* cmd) { 5407 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5408 5409 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5410 // pthread_atfork handlers and reset pthread library. All we need is a 5411 // separate process to execve. Make a direct syscall to fork process. 5412 // On IA64 there's no fork syscall, we have to use fork() and hope for 5413 // the best... 5414 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5415 IA64_ONLY(fork();) 5416 5417 if (pid < 0) { 5418 // fork failed 5419 return -1; 5420 5421 } else if (pid == 0) { 5422 // child process 5423 5424 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5425 // first to kill every thread on the thread list. Because this list is 5426 // not reset by fork() (see notes above), execve() will instead kill 5427 // every thread in the parent process. We know this is the only thread 5428 // in the new process, so make a system call directly. 5429 // IA64 should use normal execve() from glibc to match the glibc fork() 5430 // above. 5431 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5432 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5433 5434 // execve failed 5435 _exit(-1); 5436 5437 } else { 5438 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5439 // care about the actual exit code, for now. 5440 5441 int status; 5442 5443 // Wait for the child process to exit. This returns immediately if 5444 // the child has already exited. */ 5445 while (waitpid(pid, &status, 0) < 0) { 5446 switch (errno) { 5447 case ECHILD: return 0; 5448 case EINTR: break; 5449 default: return -1; 5450 } 5451 } 5452 5453 if (WIFEXITED(status)) { 5454 // The child exited normally; get its exit code. 5455 return WEXITSTATUS(status); 5456 } else if (WIFSIGNALED(status)) { 5457 // The child exited because of a signal 5458 // The best value to return is 0x80 + signal number, 5459 // because that is what all Unix shells do, and because 5460 // it allows callers to distinguish between process exit and 5461 // process death by signal. 5462 return 0x80 + WTERMSIG(status); 5463 } else { 5464 // Unknown exit code; pass it through 5465 return status; 5466 } 5467 } 5468} 5469 5470// is_headless_jre() 5471// 5472// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5473// in order to report if we are running in a headless jre 5474// 5475// Since JDK8 xawt/libmawt.so was moved into the same directory 5476// as libawt.so, and renamed libawt_xawt.so 5477// 5478bool os::is_headless_jre() { 5479 struct stat statbuf; 5480 char buf[MAXPATHLEN]; 5481 char libmawtpath[MAXPATHLEN]; 5482 const char *xawtstr = "/xawt/libmawt.so"; 5483 const char *new_xawtstr = "/libawt_xawt.so"; 5484 char *p; 5485 5486 // Get path to libjvm.so 5487 os::jvm_path(buf, sizeof(buf)); 5488 5489 // Get rid of libjvm.so 5490 p = strrchr(buf, '/'); 5491 if (p == NULL) return false; 5492 else *p = '\0'; 5493 5494 // Get rid of client or server 5495 p = strrchr(buf, '/'); 5496 if (p == NULL) return false; 5497 else *p = '\0'; 5498 5499 // check xawt/libmawt.so 5500 strcpy(libmawtpath, buf); 5501 strcat(libmawtpath, xawtstr); 5502 if (::stat(libmawtpath, &statbuf) == 0) return false; 5503 5504 // check libawt_xawt.so 5505 strcpy(libmawtpath, buf); 5506 strcat(libmawtpath, new_xawtstr); 5507 if (::stat(libmawtpath, &statbuf) == 0) return false; 5508 5509 return true; 5510} 5511 5512// Get the default path to the core file 5513// Returns the length of the string 5514int os::get_core_path(char* buffer, size_t bufferSize) { 5515 const char* p = get_current_directory(buffer, bufferSize); 5516 5517 if (p == NULL) { 5518 assert(p != NULL, "failed to get current directory"); 5519 return 0; 5520 } 5521 5522 return strlen(buffer); 5523} 5524 5525#ifdef JAVASE_EMBEDDED 5526// 5527// A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 5528// 5529MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 5530 5531// ctor 5532// 5533MemNotifyThread::MemNotifyThread(int fd): Thread() { 5534 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 5535 _fd = fd; 5536 5537 if (os::create_thread(this, os::os_thread)) { 5538 _memnotify_thread = this; 5539 os::set_priority(this, NearMaxPriority); 5540 os::start_thread(this); 5541 } 5542} 5543 5544// Where all the work gets done 5545// 5546void MemNotifyThread::run() { 5547 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 5548 5549 // Set up the select arguments 5550 fd_set rfds; 5551 if (_fd != -1) { 5552 FD_ZERO(&rfds); 5553 FD_SET(_fd, &rfds); 5554 } 5555 5556 // Now wait for the mem_notify device to wake up 5557 while (1) { 5558 // Wait for the mem_notify device to signal us.. 5559 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 5560 if (rc == -1) { 5561 perror("select!\n"); 5562 break; 5563 } else if (rc) { 5564 //ssize_t free_before = os::available_memory(); 5565 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 5566 5567 // The kernel is telling us there is not much memory left... 5568 // try to do something about that 5569 5570 // If we are not already in a GC, try one. 5571 if (!Universe::heap()->is_gc_active()) { 5572 Universe::heap()->collect(GCCause::_allocation_failure); 5573 5574 //ssize_t free_after = os::available_memory(); 5575 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 5576 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 5577 } 5578 // We might want to do something like the following if we find the GC's are not helping... 5579 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 5580 } 5581 } 5582} 5583 5584// 5585// See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 5586// 5587void MemNotifyThread::start() { 5588 int fd; 5589 fd = open ("/dev/mem_notify", O_RDONLY, 0); 5590 if (fd < 0) { 5591 return; 5592 } 5593 5594 if (memnotify_thread() == NULL) { 5595 new MemNotifyThread(fd); 5596 } 5597} 5598#endif // JAVASE_EMBEDDED 5599