os_linux.cpp revision 4453:17bf4d428955
1/* 2 * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25// no precompiled headers 26#include "classfile/classLoader.hpp" 27#include "classfile/systemDictionary.hpp" 28#include "classfile/vmSymbols.hpp" 29#include "code/icBuffer.hpp" 30#include "code/vtableStubs.hpp" 31#include "compiler/compileBroker.hpp" 32#include "compiler/disassembler.hpp" 33#include "interpreter/interpreter.hpp" 34#include "jvm_linux.h" 35#include "memory/allocation.inline.hpp" 36#include "memory/filemap.hpp" 37#include "mutex_linux.inline.hpp" 38#include "oops/oop.inline.hpp" 39#include "os_share_linux.hpp" 40#include "prims/jniFastGetField.hpp" 41#include "prims/jvm.h" 42#include "prims/jvm_misc.hpp" 43#include "runtime/arguments.hpp" 44#include "runtime/extendedPC.hpp" 45#include "runtime/globals.hpp" 46#include "runtime/interfaceSupport.hpp" 47#include "runtime/init.hpp" 48#include "runtime/java.hpp" 49#include "runtime/javaCalls.hpp" 50#include "runtime/mutexLocker.hpp" 51#include "runtime/objectMonitor.hpp" 52#include "runtime/osThread.hpp" 53#include "runtime/perfMemory.hpp" 54#include "runtime/sharedRuntime.hpp" 55#include "runtime/statSampler.hpp" 56#include "runtime/stubRoutines.hpp" 57#include "runtime/thread.inline.hpp" 58#include "runtime/threadCritical.hpp" 59#include "runtime/timer.hpp" 60#include "services/attachListener.hpp" 61#include "services/memTracker.hpp" 62#include "services/runtimeService.hpp" 63#include "utilities/decoder.hpp" 64#include "utilities/defaultStream.hpp" 65#include "utilities/events.hpp" 66#include "utilities/elfFile.hpp" 67#include "utilities/growableArray.hpp" 68#include "utilities/vmError.hpp" 69 70// put OS-includes here 71# include <sys/types.h> 72# include <sys/mman.h> 73# include <sys/stat.h> 74# include <sys/select.h> 75# include <pthread.h> 76# include <signal.h> 77# include <errno.h> 78# include <dlfcn.h> 79# include <stdio.h> 80# include <unistd.h> 81# include <sys/resource.h> 82# include <pthread.h> 83# include <sys/stat.h> 84# include <sys/time.h> 85# include <sys/times.h> 86# include <sys/utsname.h> 87# include <sys/socket.h> 88# include <sys/wait.h> 89# include <pwd.h> 90# include <poll.h> 91# include <semaphore.h> 92# include <fcntl.h> 93# include <string.h> 94# include <syscall.h> 95# include <sys/sysinfo.h> 96# include <gnu/libc-version.h> 97# include <sys/ipc.h> 98# include <sys/shm.h> 99# include <link.h> 100# include <stdint.h> 101# include <inttypes.h> 102# include <sys/ioctl.h> 103 104#define MAX_PATH (2 * K) 105 106// for timer info max values which include all bits 107#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 108 109#define LARGEPAGES_BIT (1 << 6) 110//////////////////////////////////////////////////////////////////////////////// 111// global variables 112julong os::Linux::_physical_memory = 0; 113 114address os::Linux::_initial_thread_stack_bottom = NULL; 115uintptr_t os::Linux::_initial_thread_stack_size = 0; 116 117int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 118int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 119Mutex* os::Linux::_createThread_lock = NULL; 120pthread_t os::Linux::_main_thread; 121int os::Linux::_page_size = -1; 122bool os::Linux::_is_floating_stack = false; 123bool os::Linux::_is_NPTL = false; 124bool os::Linux::_supports_fast_thread_cpu_time = false; 125const char * os::Linux::_glibc_version = NULL; 126const char * os::Linux::_libpthread_version = NULL; 127 128static jlong initial_time_count=0; 129 130static int clock_tics_per_sec = 100; 131 132// For diagnostics to print a message once. see run_periodic_checks 133static sigset_t check_signal_done; 134static bool check_signals = true;; 135 136static pid_t _initial_pid = 0; 137 138/* Signal number used to suspend/resume a thread */ 139 140/* do not use any signal number less than SIGSEGV, see 4355769 */ 141static int SR_signum = SIGUSR2; 142sigset_t SR_sigset; 143 144/* Used to protect dlsym() calls */ 145static pthread_mutex_t dl_mutex; 146 147#ifdef JAVASE_EMBEDDED 148class MemNotifyThread: public Thread { 149 friend class VMStructs; 150 public: 151 virtual void run(); 152 153 private: 154 static MemNotifyThread* _memnotify_thread; 155 int _fd; 156 157 public: 158 159 // Constructor 160 MemNotifyThread(int fd); 161 162 // Tester 163 bool is_memnotify_thread() const { return true; } 164 165 // Printing 166 char* name() const { return (char*)"Linux MemNotify Thread"; } 167 168 // Returns the single instance of the MemNotifyThread 169 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; } 170 171 // Create and start the single instance of MemNotifyThread 172 static void start(); 173}; 174#endif // JAVASE_EMBEDDED 175 176// utility functions 177 178static int SR_initialize(); 179static int SR_finalize(); 180 181julong os::available_memory() { 182 return Linux::available_memory(); 183} 184 185julong os::Linux::available_memory() { 186 // values in struct sysinfo are "unsigned long" 187 struct sysinfo si; 188 sysinfo(&si); 189 190 return (julong)si.freeram * si.mem_unit; 191} 192 193julong os::physical_memory() { 194 return Linux::physical_memory(); 195} 196 197julong os::allocatable_physical_memory(julong size) { 198#ifdef _LP64 199 return size; 200#else 201 julong result = MIN2(size, (julong)3800*M); 202 if (!is_allocatable(result)) { 203 // See comments under solaris for alignment considerations 204 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); 205 result = MIN2(size, reasonable_size); 206 } 207 return result; 208#endif // _LP64 209} 210 211//////////////////////////////////////////////////////////////////////////////// 212// environment support 213 214bool os::getenv(const char* name, char* buf, int len) { 215 const char* val = ::getenv(name); 216 if (val != NULL && strlen(val) < (size_t)len) { 217 strcpy(buf, val); 218 return true; 219 } 220 if (len > 0) buf[0] = 0; // return a null string 221 return false; 222} 223 224 225// Return true if user is running as root. 226 227bool os::have_special_privileges() { 228 static bool init = false; 229 static bool privileges = false; 230 if (!init) { 231 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 232 init = true; 233 } 234 return privileges; 235} 236 237 238#ifndef SYS_gettid 239// i386: 224, ia64: 1105, amd64: 186, sparc 143 240#ifdef __ia64__ 241#define SYS_gettid 1105 242#elif __i386__ 243#define SYS_gettid 224 244#elif __amd64__ 245#define SYS_gettid 186 246#elif __sparc__ 247#define SYS_gettid 143 248#else 249#error define gettid for the arch 250#endif 251#endif 252 253// Cpu architecture string 254#if defined(ZERO) 255static char cpu_arch[] = ZERO_LIBARCH; 256#elif defined(IA64) 257static char cpu_arch[] = "ia64"; 258#elif defined(IA32) 259static char cpu_arch[] = "i386"; 260#elif defined(AMD64) 261static char cpu_arch[] = "amd64"; 262#elif defined(ARM) 263static char cpu_arch[] = "arm"; 264#elif defined(PPC) 265static char cpu_arch[] = "ppc"; 266#elif defined(SPARC) 267# ifdef _LP64 268static char cpu_arch[] = "sparcv9"; 269# else 270static char cpu_arch[] = "sparc"; 271# endif 272#else 273#error Add appropriate cpu_arch setting 274#endif 275 276 277// pid_t gettid() 278// 279// Returns the kernel thread id of the currently running thread. Kernel 280// thread id is used to access /proc. 281// 282// (Note that getpid() on LinuxThreads returns kernel thread id too; but 283// on NPTL, it returns the same pid for all threads, as required by POSIX.) 284// 285pid_t os::Linux::gettid() { 286 int rslt = syscall(SYS_gettid); 287 if (rslt == -1) { 288 // old kernel, no NPTL support 289 return getpid(); 290 } else { 291 return (pid_t)rslt; 292 } 293} 294 295// Most versions of linux have a bug where the number of processors are 296// determined by looking at the /proc file system. In a chroot environment, 297// the system call returns 1. This causes the VM to act as if it is 298// a single processor and elide locking (see is_MP() call). 299static bool unsafe_chroot_detected = false; 300static const char *unstable_chroot_error = "/proc file system not found.\n" 301 "Java may be unstable running multithreaded in a chroot " 302 "environment on Linux when /proc filesystem is not mounted."; 303 304void os::Linux::initialize_system_info() { 305 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 306 if (processor_count() == 1) { 307 pid_t pid = os::Linux::gettid(); 308 char fname[32]; 309 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 310 FILE *fp = fopen(fname, "r"); 311 if (fp == NULL) { 312 unsafe_chroot_detected = true; 313 } else { 314 fclose(fp); 315 } 316 } 317 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 318 assert(processor_count() > 0, "linux error"); 319} 320 321void os::init_system_properties_values() { 322// char arch[12]; 323// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 324 325 // The next steps are taken in the product version: 326 // 327 // Obtain the JAVA_HOME value from the location of libjvm.so. 328 // This library should be located at: 329 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 330 // 331 // If "/jre/lib/" appears at the right place in the path, then we 332 // assume libjvm.so is installed in a JDK and we use this path. 333 // 334 // Otherwise exit with message: "Could not create the Java virtual machine." 335 // 336 // The following extra steps are taken in the debugging version: 337 // 338 // If "/jre/lib/" does NOT appear at the right place in the path 339 // instead of exit check for $JAVA_HOME environment variable. 340 // 341 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 342 // then we append a fake suffix "hotspot/libjvm.so" to this path so 343 // it looks like libjvm.so is installed there 344 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 345 // 346 // Otherwise exit. 347 // 348 // Important note: if the location of libjvm.so changes this 349 // code needs to be changed accordingly. 350 351 // The next few definitions allow the code to be verbatim: 352#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal) 353#define getenv(n) ::getenv(n) 354 355/* 356 * See ld(1): 357 * The linker uses the following search paths to locate required 358 * shared libraries: 359 * 1: ... 360 * ... 361 * 7: The default directories, normally /lib and /usr/lib. 362 */ 363#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 364#define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 365#else 366#define DEFAULT_LIBPATH "/lib:/usr/lib" 367#endif 368 369#define EXTENSIONS_DIR "/lib/ext" 370#define ENDORSED_DIR "/lib/endorsed" 371#define REG_DIR "/usr/java/packages" 372 373 { 374 /* sysclasspath, java_home, dll_dir */ 375 { 376 char *home_path; 377 char *dll_path; 378 char *pslash; 379 char buf[MAXPATHLEN]; 380 os::jvm_path(buf, sizeof(buf)); 381 382 // Found the full path to libjvm.so. 383 // Now cut the path to <java_home>/jre if we can. 384 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 385 pslash = strrchr(buf, '/'); 386 if (pslash != NULL) 387 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 388 dll_path = malloc(strlen(buf) + 1); 389 if (dll_path == NULL) 390 return; 391 strcpy(dll_path, buf); 392 Arguments::set_dll_dir(dll_path); 393 394 if (pslash != NULL) { 395 pslash = strrchr(buf, '/'); 396 if (pslash != NULL) { 397 *pslash = '\0'; /* get rid of /<arch> */ 398 pslash = strrchr(buf, '/'); 399 if (pslash != NULL) 400 *pslash = '\0'; /* get rid of /lib */ 401 } 402 } 403 404 home_path = malloc(strlen(buf) + 1); 405 if (home_path == NULL) 406 return; 407 strcpy(home_path, buf); 408 Arguments::set_java_home(home_path); 409 410 if (!set_boot_path('/', ':')) 411 return; 412 } 413 414 /* 415 * Where to look for native libraries 416 * 417 * Note: Due to a legacy implementation, most of the library path 418 * is set in the launcher. This was to accomodate linking restrictions 419 * on legacy Linux implementations (which are no longer supported). 420 * Eventually, all the library path setting will be done here. 421 * 422 * However, to prevent the proliferation of improperly built native 423 * libraries, the new path component /usr/java/packages is added here. 424 * Eventually, all the library path setting will be done here. 425 */ 426 { 427 char *ld_library_path; 428 429 /* 430 * Construct the invariant part of ld_library_path. Note that the 431 * space for the colon and the trailing null are provided by the 432 * nulls included by the sizeof operator (so actually we allocate 433 * a byte more than necessary). 434 */ 435 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 436 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 437 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 438 439 /* 440 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 441 * should always exist (until the legacy problem cited above is 442 * addressed). 443 */ 444 char *v = getenv("LD_LIBRARY_PATH"); 445 if (v != NULL) { 446 char *t = ld_library_path; 447 /* That's +1 for the colon and +1 for the trailing '\0' */ 448 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 449 sprintf(ld_library_path, "%s:%s", v, t); 450 } 451 Arguments::set_library_path(ld_library_path); 452 } 453 454 /* 455 * Extensions directories. 456 * 457 * Note that the space for the colon and the trailing null are provided 458 * by the nulls included by the sizeof operator (so actually one byte more 459 * than necessary is allocated). 460 */ 461 { 462 char *buf = malloc(strlen(Arguments::get_java_home()) + 463 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 464 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 465 Arguments::get_java_home()); 466 Arguments::set_ext_dirs(buf); 467 } 468 469 /* Endorsed standards default directory. */ 470 { 471 char * buf; 472 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 473 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 474 Arguments::set_endorsed_dirs(buf); 475 } 476 } 477 478#undef malloc 479#undef getenv 480#undef EXTENSIONS_DIR 481#undef ENDORSED_DIR 482 483 // Done 484 return; 485} 486 487//////////////////////////////////////////////////////////////////////////////// 488// breakpoint support 489 490void os::breakpoint() { 491 BREAKPOINT; 492} 493 494extern "C" void breakpoint() { 495 // use debugger to set breakpoint here 496} 497 498//////////////////////////////////////////////////////////////////////////////// 499// signal support 500 501debug_only(static bool signal_sets_initialized = false); 502static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 503 504bool os::Linux::is_sig_ignored(int sig) { 505 struct sigaction oact; 506 sigaction(sig, (struct sigaction*)NULL, &oact); 507 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 508 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 509 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 510 return true; 511 else 512 return false; 513} 514 515void os::Linux::signal_sets_init() { 516 // Should also have an assertion stating we are still single-threaded. 517 assert(!signal_sets_initialized, "Already initialized"); 518 // Fill in signals that are necessarily unblocked for all threads in 519 // the VM. Currently, we unblock the following signals: 520 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 521 // by -Xrs (=ReduceSignalUsage)); 522 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 523 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 524 // the dispositions or masks wrt these signals. 525 // Programs embedding the VM that want to use the above signals for their 526 // own purposes must, at this time, use the "-Xrs" option to prevent 527 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 528 // (See bug 4345157, and other related bugs). 529 // In reality, though, unblocking these signals is really a nop, since 530 // these signals are not blocked by default. 531 sigemptyset(&unblocked_sigs); 532 sigemptyset(&allowdebug_blocked_sigs); 533 sigaddset(&unblocked_sigs, SIGILL); 534 sigaddset(&unblocked_sigs, SIGSEGV); 535 sigaddset(&unblocked_sigs, SIGBUS); 536 sigaddset(&unblocked_sigs, SIGFPE); 537 sigaddset(&unblocked_sigs, SR_signum); 538 539 if (!ReduceSignalUsage) { 540 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 541 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 542 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 543 } 544 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 545 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 546 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 547 } 548 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 549 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 550 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 551 } 552 } 553 // Fill in signals that are blocked by all but the VM thread. 554 sigemptyset(&vm_sigs); 555 if (!ReduceSignalUsage) 556 sigaddset(&vm_sigs, BREAK_SIGNAL); 557 debug_only(signal_sets_initialized = true); 558 559} 560 561// These are signals that are unblocked while a thread is running Java. 562// (For some reason, they get blocked by default.) 563sigset_t* os::Linux::unblocked_signals() { 564 assert(signal_sets_initialized, "Not initialized"); 565 return &unblocked_sigs; 566} 567 568// These are the signals that are blocked while a (non-VM) thread is 569// running Java. Only the VM thread handles these signals. 570sigset_t* os::Linux::vm_signals() { 571 assert(signal_sets_initialized, "Not initialized"); 572 return &vm_sigs; 573} 574 575// These are signals that are blocked during cond_wait to allow debugger in 576sigset_t* os::Linux::allowdebug_blocked_signals() { 577 assert(signal_sets_initialized, "Not initialized"); 578 return &allowdebug_blocked_sigs; 579} 580 581void os::Linux::hotspot_sigmask(Thread* thread) { 582 583 //Save caller's signal mask before setting VM signal mask 584 sigset_t caller_sigmask; 585 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 586 587 OSThread* osthread = thread->osthread(); 588 osthread->set_caller_sigmask(caller_sigmask); 589 590 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 591 592 if (!ReduceSignalUsage) { 593 if (thread->is_VM_thread()) { 594 // Only the VM thread handles BREAK_SIGNAL ... 595 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 596 } else { 597 // ... all other threads block BREAK_SIGNAL 598 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 599 } 600 } 601} 602 603////////////////////////////////////////////////////////////////////////////// 604// detecting pthread library 605 606void os::Linux::libpthread_init() { 607 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 608 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 609 // generic name for earlier versions. 610 // Define macros here so we can build HotSpot on old systems. 611# ifndef _CS_GNU_LIBC_VERSION 612# define _CS_GNU_LIBC_VERSION 2 613# endif 614# ifndef _CS_GNU_LIBPTHREAD_VERSION 615# define _CS_GNU_LIBPTHREAD_VERSION 3 616# endif 617 618 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 619 if (n > 0) { 620 char *str = (char *)malloc(n, mtInternal); 621 confstr(_CS_GNU_LIBC_VERSION, str, n); 622 os::Linux::set_glibc_version(str); 623 } else { 624 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 625 static char _gnu_libc_version[32]; 626 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 627 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 628 os::Linux::set_glibc_version(_gnu_libc_version); 629 } 630 631 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 632 if (n > 0) { 633 char *str = (char *)malloc(n, mtInternal); 634 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 635 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 636 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 637 // is the case. LinuxThreads has a hard limit on max number of threads. 638 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 639 // On the other hand, NPTL does not have such a limit, sysconf() 640 // will return -1 and errno is not changed. Check if it is really NPTL. 641 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 642 strstr(str, "NPTL") && 643 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 644 free(str); 645 os::Linux::set_libpthread_version("linuxthreads"); 646 } else { 647 os::Linux::set_libpthread_version(str); 648 } 649 } else { 650 // glibc before 2.3.2 only has LinuxThreads. 651 os::Linux::set_libpthread_version("linuxthreads"); 652 } 653 654 if (strstr(libpthread_version(), "NPTL")) { 655 os::Linux::set_is_NPTL(); 656 } else { 657 os::Linux::set_is_LinuxThreads(); 658 } 659 660 // LinuxThreads have two flavors: floating-stack mode, which allows variable 661 // stack size; and fixed-stack mode. NPTL is always floating-stack. 662 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 663 os::Linux::set_is_floating_stack(); 664 } 665} 666 667///////////////////////////////////////////////////////////////////////////// 668// thread stack 669 670// Force Linux kernel to expand current thread stack. If "bottom" is close 671// to the stack guard, caller should block all signals. 672// 673// MAP_GROWSDOWN: 674// A special mmap() flag that is used to implement thread stacks. It tells 675// kernel that the memory region should extend downwards when needed. This 676// allows early versions of LinuxThreads to only mmap the first few pages 677// when creating a new thread. Linux kernel will automatically expand thread 678// stack as needed (on page faults). 679// 680// However, because the memory region of a MAP_GROWSDOWN stack can grow on 681// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 682// region, it's hard to tell if the fault is due to a legitimate stack 683// access or because of reading/writing non-exist memory (e.g. buffer 684// overrun). As a rule, if the fault happens below current stack pointer, 685// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 686// application (see Linux kernel fault.c). 687// 688// This Linux feature can cause SIGSEGV when VM bangs thread stack for 689// stack overflow detection. 690// 691// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 692// not use this flag. However, the stack of initial thread is not created 693// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 694// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 695// and then attach the thread to JVM. 696// 697// To get around the problem and allow stack banging on Linux, we need to 698// manually expand thread stack after receiving the SIGSEGV. 699// 700// There are two ways to expand thread stack to address "bottom", we used 701// both of them in JVM before 1.5: 702// 1. adjust stack pointer first so that it is below "bottom", and then 703// touch "bottom" 704// 2. mmap() the page in question 705// 706// Now alternate signal stack is gone, it's harder to use 2. For instance, 707// if current sp is already near the lower end of page 101, and we need to 708// call mmap() to map page 100, it is possible that part of the mmap() frame 709// will be placed in page 100. When page 100 is mapped, it is zero-filled. 710// That will destroy the mmap() frame and cause VM to crash. 711// 712// The following code works by adjusting sp first, then accessing the "bottom" 713// page to force a page fault. Linux kernel will then automatically expand the 714// stack mapping. 715// 716// _expand_stack_to() assumes its frame size is less than page size, which 717// should always be true if the function is not inlined. 718 719#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 720#define NOINLINE 721#else 722#define NOINLINE __attribute__ ((noinline)) 723#endif 724 725static void _expand_stack_to(address bottom) NOINLINE; 726 727static void _expand_stack_to(address bottom) { 728 address sp; 729 size_t size; 730 volatile char *p; 731 732 // Adjust bottom to point to the largest address within the same page, it 733 // gives us a one-page buffer if alloca() allocates slightly more memory. 734 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 735 bottom += os::Linux::page_size() - 1; 736 737 // sp might be slightly above current stack pointer; if that's the case, we 738 // will alloca() a little more space than necessary, which is OK. Don't use 739 // os::current_stack_pointer(), as its result can be slightly below current 740 // stack pointer, causing us to not alloca enough to reach "bottom". 741 sp = (address)&sp; 742 743 if (sp > bottom) { 744 size = sp - bottom; 745 p = (volatile char *)alloca(size); 746 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 747 p[0] = '\0'; 748 } 749} 750 751bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 752 assert(t!=NULL, "just checking"); 753 assert(t->osthread()->expanding_stack(), "expand should be set"); 754 assert(t->stack_base() != NULL, "stack_base was not initialized"); 755 756 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 757 sigset_t mask_all, old_sigset; 758 sigfillset(&mask_all); 759 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 760 _expand_stack_to(addr); 761 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 762 return true; 763 } 764 return false; 765} 766 767////////////////////////////////////////////////////////////////////////////// 768// create new thread 769 770static address highest_vm_reserved_address(); 771 772// check if it's safe to start a new thread 773static bool _thread_safety_check(Thread* thread) { 774 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 775 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 776 // Heap is mmap'ed at lower end of memory space. Thread stacks are 777 // allocated (MAP_FIXED) from high address space. Every thread stack 778 // occupies a fixed size slot (usually 2Mbytes, but user can change 779 // it to other values if they rebuild LinuxThreads). 780 // 781 // Problem with MAP_FIXED is that mmap() can still succeed even part of 782 // the memory region has already been mmap'ed. That means if we have too 783 // many threads and/or very large heap, eventually thread stack will 784 // collide with heap. 785 // 786 // Here we try to prevent heap/stack collision by comparing current 787 // stack bottom with the highest address that has been mmap'ed by JVM 788 // plus a safety margin for memory maps created by native code. 789 // 790 // This feature can be disabled by setting ThreadSafetyMargin to 0 791 // 792 if (ThreadSafetyMargin > 0) { 793 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 794 795 // not safe if our stack extends below the safety margin 796 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 797 } else { 798 return true; 799 } 800 } else { 801 // Floating stack LinuxThreads or NPTL: 802 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 803 // there's not enough space left, pthread_create() will fail. If we come 804 // here, that means enough space has been reserved for stack. 805 return true; 806 } 807} 808 809// Thread start routine for all newly created threads 810static void *java_start(Thread *thread) { 811 // Try to randomize the cache line index of hot stack frames. 812 // This helps when threads of the same stack traces evict each other's 813 // cache lines. The threads can be either from the same JVM instance, or 814 // from different JVM instances. The benefit is especially true for 815 // processors with hyperthreading technology. 816 static int counter = 0; 817 int pid = os::current_process_id(); 818 alloca(((pid ^ counter++) & 7) * 128); 819 820 ThreadLocalStorage::set_thread(thread); 821 822 OSThread* osthread = thread->osthread(); 823 Monitor* sync = osthread->startThread_lock(); 824 825 // non floating stack LinuxThreads needs extra check, see above 826 if (!_thread_safety_check(thread)) { 827 // notify parent thread 828 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 829 osthread->set_state(ZOMBIE); 830 sync->notify_all(); 831 return NULL; 832 } 833 834 // thread_id is kernel thread id (similar to Solaris LWP id) 835 osthread->set_thread_id(os::Linux::gettid()); 836 837 if (UseNUMA) { 838 int lgrp_id = os::numa_get_group_id(); 839 if (lgrp_id != -1) { 840 thread->set_lgrp_id(lgrp_id); 841 } 842 } 843 // initialize signal mask for this thread 844 os::Linux::hotspot_sigmask(thread); 845 846 // initialize floating point control register 847 os::Linux::init_thread_fpu_state(); 848 849 // handshaking with parent thread 850 { 851 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 852 853 // notify parent thread 854 osthread->set_state(INITIALIZED); 855 sync->notify_all(); 856 857 // wait until os::start_thread() 858 while (osthread->get_state() == INITIALIZED) { 859 sync->wait(Mutex::_no_safepoint_check_flag); 860 } 861 } 862 863 // call one more level start routine 864 thread->run(); 865 866 return 0; 867} 868 869bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 870 assert(thread->osthread() == NULL, "caller responsible"); 871 872 // Allocate the OSThread object 873 OSThread* osthread = new OSThread(NULL, NULL); 874 if (osthread == NULL) { 875 return false; 876 } 877 878 // set the correct thread state 879 osthread->set_thread_type(thr_type); 880 881 // Initial state is ALLOCATED but not INITIALIZED 882 osthread->set_state(ALLOCATED); 883 884 thread->set_osthread(osthread); 885 886 // init thread attributes 887 pthread_attr_t attr; 888 pthread_attr_init(&attr); 889 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 890 891 // stack size 892 if (os::Linux::supports_variable_stack_size()) { 893 // calculate stack size if it's not specified by caller 894 if (stack_size == 0) { 895 stack_size = os::Linux::default_stack_size(thr_type); 896 897 switch (thr_type) { 898 case os::java_thread: 899 // Java threads use ThreadStackSize which default value can be 900 // changed with the flag -Xss 901 assert (JavaThread::stack_size_at_create() > 0, "this should be set"); 902 stack_size = JavaThread::stack_size_at_create(); 903 break; 904 case os::compiler_thread: 905 if (CompilerThreadStackSize > 0) { 906 stack_size = (size_t)(CompilerThreadStackSize * K); 907 break; 908 } // else fall through: 909 // use VMThreadStackSize if CompilerThreadStackSize is not defined 910 case os::vm_thread: 911 case os::pgc_thread: 912 case os::cgc_thread: 913 case os::watcher_thread: 914 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 915 break; 916 } 917 } 918 919 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 920 pthread_attr_setstacksize(&attr, stack_size); 921 } else { 922 // let pthread_create() pick the default value. 923 } 924 925 // glibc guard page 926 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 927 928 ThreadState state; 929 930 { 931 // Serialize thread creation if we are running with fixed stack LinuxThreads 932 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 933 if (lock) { 934 os::Linux::createThread_lock()->lock_without_safepoint_check(); 935 } 936 937 pthread_t tid; 938 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 939 940 pthread_attr_destroy(&attr); 941 942 if (ret != 0) { 943 if (PrintMiscellaneous && (Verbose || WizardMode)) { 944 perror("pthread_create()"); 945 } 946 // Need to clean up stuff we've allocated so far 947 thread->set_osthread(NULL); 948 delete osthread; 949 if (lock) os::Linux::createThread_lock()->unlock(); 950 return false; 951 } 952 953 // Store pthread info into the OSThread 954 osthread->set_pthread_id(tid); 955 956 // Wait until child thread is either initialized or aborted 957 { 958 Monitor* sync_with_child = osthread->startThread_lock(); 959 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 960 while ((state = osthread->get_state()) == ALLOCATED) { 961 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 962 } 963 } 964 965 if (lock) { 966 os::Linux::createThread_lock()->unlock(); 967 } 968 } 969 970 // Aborted due to thread limit being reached 971 if (state == ZOMBIE) { 972 thread->set_osthread(NULL); 973 delete osthread; 974 return false; 975 } 976 977 // The thread is returned suspended (in state INITIALIZED), 978 // and is started higher up in the call chain 979 assert(state == INITIALIZED, "race condition"); 980 return true; 981} 982 983///////////////////////////////////////////////////////////////////////////// 984// attach existing thread 985 986// bootstrap the main thread 987bool os::create_main_thread(JavaThread* thread) { 988 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 989 return create_attached_thread(thread); 990} 991 992bool os::create_attached_thread(JavaThread* thread) { 993#ifdef ASSERT 994 thread->verify_not_published(); 995#endif 996 997 // Allocate the OSThread object 998 OSThread* osthread = new OSThread(NULL, NULL); 999 1000 if (osthread == NULL) { 1001 return false; 1002 } 1003 1004 // Store pthread info into the OSThread 1005 osthread->set_thread_id(os::Linux::gettid()); 1006 osthread->set_pthread_id(::pthread_self()); 1007 1008 // initialize floating point control register 1009 os::Linux::init_thread_fpu_state(); 1010 1011 // Initial thread state is RUNNABLE 1012 osthread->set_state(RUNNABLE); 1013 1014 thread->set_osthread(osthread); 1015 1016 if (UseNUMA) { 1017 int lgrp_id = os::numa_get_group_id(); 1018 if (lgrp_id != -1) { 1019 thread->set_lgrp_id(lgrp_id); 1020 } 1021 } 1022 1023 if (os::Linux::is_initial_thread()) { 1024 // If current thread is initial thread, its stack is mapped on demand, 1025 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 1026 // the entire stack region to avoid SEGV in stack banging. 1027 // It is also useful to get around the heap-stack-gap problem on SuSE 1028 // kernel (see 4821821 for details). We first expand stack to the top 1029 // of yellow zone, then enable stack yellow zone (order is significant, 1030 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1031 // is no gap between the last two virtual memory regions. 1032 1033 JavaThread *jt = (JavaThread *)thread; 1034 address addr = jt->stack_yellow_zone_base(); 1035 assert(addr != NULL, "initialization problem?"); 1036 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1037 1038 osthread->set_expanding_stack(); 1039 os::Linux::manually_expand_stack(jt, addr); 1040 osthread->clear_expanding_stack(); 1041 } 1042 1043 // initialize signal mask for this thread 1044 // and save the caller's signal mask 1045 os::Linux::hotspot_sigmask(thread); 1046 1047 return true; 1048} 1049 1050void os::pd_start_thread(Thread* thread) { 1051 OSThread * osthread = thread->osthread(); 1052 assert(osthread->get_state() != INITIALIZED, "just checking"); 1053 Monitor* sync_with_child = osthread->startThread_lock(); 1054 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1055 sync_with_child->notify(); 1056} 1057 1058// Free Linux resources related to the OSThread 1059void os::free_thread(OSThread* osthread) { 1060 assert(osthread != NULL, "osthread not set"); 1061 1062 if (Thread::current()->osthread() == osthread) { 1063 // Restore caller's signal mask 1064 sigset_t sigmask = osthread->caller_sigmask(); 1065 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1066 } 1067 1068 delete osthread; 1069} 1070 1071////////////////////////////////////////////////////////////////////////////// 1072// thread local storage 1073 1074int os::allocate_thread_local_storage() { 1075 pthread_key_t key; 1076 int rslt = pthread_key_create(&key, NULL); 1077 assert(rslt == 0, "cannot allocate thread local storage"); 1078 return (int)key; 1079} 1080 1081// Note: This is currently not used by VM, as we don't destroy TLS key 1082// on VM exit. 1083void os::free_thread_local_storage(int index) { 1084 int rslt = pthread_key_delete((pthread_key_t)index); 1085 assert(rslt == 0, "invalid index"); 1086} 1087 1088void os::thread_local_storage_at_put(int index, void* value) { 1089 int rslt = pthread_setspecific((pthread_key_t)index, value); 1090 assert(rslt == 0, "pthread_setspecific failed"); 1091} 1092 1093extern "C" Thread* get_thread() { 1094 return ThreadLocalStorage::thread(); 1095} 1096 1097////////////////////////////////////////////////////////////////////////////// 1098// initial thread 1099 1100// Check if current thread is the initial thread, similar to Solaris thr_main. 1101bool os::Linux::is_initial_thread(void) { 1102 char dummy; 1103 // If called before init complete, thread stack bottom will be null. 1104 // Can be called if fatal error occurs before initialization. 1105 if (initial_thread_stack_bottom() == NULL) return false; 1106 assert(initial_thread_stack_bottom() != NULL && 1107 initial_thread_stack_size() != 0, 1108 "os::init did not locate initial thread's stack region"); 1109 if ((address)&dummy >= initial_thread_stack_bottom() && 1110 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1111 return true; 1112 else return false; 1113} 1114 1115// Find the virtual memory area that contains addr 1116static bool find_vma(address addr, address* vma_low, address* vma_high) { 1117 FILE *fp = fopen("/proc/self/maps", "r"); 1118 if (fp) { 1119 address low, high; 1120 while (!feof(fp)) { 1121 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1122 if (low <= addr && addr < high) { 1123 if (vma_low) *vma_low = low; 1124 if (vma_high) *vma_high = high; 1125 fclose (fp); 1126 return true; 1127 } 1128 } 1129 for (;;) { 1130 int ch = fgetc(fp); 1131 if (ch == EOF || ch == (int)'\n') break; 1132 } 1133 } 1134 fclose(fp); 1135 } 1136 return false; 1137} 1138 1139// Locate initial thread stack. This special handling of initial thread stack 1140// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1141// bogus value for initial thread. 1142void os::Linux::capture_initial_stack(size_t max_size) { 1143 // stack size is the easy part, get it from RLIMIT_STACK 1144 size_t stack_size; 1145 struct rlimit rlim; 1146 getrlimit(RLIMIT_STACK, &rlim); 1147 stack_size = rlim.rlim_cur; 1148 1149 // 6308388: a bug in ld.so will relocate its own .data section to the 1150 // lower end of primordial stack; reduce ulimit -s value a little bit 1151 // so we won't install guard page on ld.so's data section. 1152 stack_size -= 2 * page_size(); 1153 1154 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1155 // 7.1, in both cases we will get 2G in return value. 1156 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1157 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1158 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1159 // in case other parts in glibc still assumes 2M max stack size. 1160 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1161 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1162 if (stack_size > 2 * K * K IA64_ONLY(*2)) 1163 stack_size = 2 * K * K IA64_ONLY(*2); 1164 // Try to figure out where the stack base (top) is. This is harder. 1165 // 1166 // When an application is started, glibc saves the initial stack pointer in 1167 // a global variable "__libc_stack_end", which is then used by system 1168 // libraries. __libc_stack_end should be pretty close to stack top. The 1169 // variable is available since the very early days. However, because it is 1170 // a private interface, it could disappear in the future. 1171 // 1172 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1173 // to __libc_stack_end, it is very close to stack top, but isn't the real 1174 // stack top. Note that /proc may not exist if VM is running as a chroot 1175 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1176 // /proc/<pid>/stat could change in the future (though unlikely). 1177 // 1178 // We try __libc_stack_end first. If that doesn't work, look for 1179 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1180 // as a hint, which should work well in most cases. 1181 1182 uintptr_t stack_start; 1183 1184 // try __libc_stack_end first 1185 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1186 if (p && *p) { 1187 stack_start = *p; 1188 } else { 1189 // see if we can get the start_stack field from /proc/self/stat 1190 FILE *fp; 1191 int pid; 1192 char state; 1193 int ppid; 1194 int pgrp; 1195 int session; 1196 int nr; 1197 int tpgrp; 1198 unsigned long flags; 1199 unsigned long minflt; 1200 unsigned long cminflt; 1201 unsigned long majflt; 1202 unsigned long cmajflt; 1203 unsigned long utime; 1204 unsigned long stime; 1205 long cutime; 1206 long cstime; 1207 long prio; 1208 long nice; 1209 long junk; 1210 long it_real; 1211 uintptr_t start; 1212 uintptr_t vsize; 1213 intptr_t rss; 1214 uintptr_t rsslim; 1215 uintptr_t scodes; 1216 uintptr_t ecode; 1217 int i; 1218 1219 // Figure what the primordial thread stack base is. Code is inspired 1220 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1221 // followed by command name surrounded by parentheses, state, etc. 1222 char stat[2048]; 1223 int statlen; 1224 1225 fp = fopen("/proc/self/stat", "r"); 1226 if (fp) { 1227 statlen = fread(stat, 1, 2047, fp); 1228 stat[statlen] = '\0'; 1229 fclose(fp); 1230 1231 // Skip pid and the command string. Note that we could be dealing with 1232 // weird command names, e.g. user could decide to rename java launcher 1233 // to "java 1.4.2 :)", then the stat file would look like 1234 // 1234 (java 1.4.2 :)) R ... ... 1235 // We don't really need to know the command string, just find the last 1236 // occurrence of ")" and then start parsing from there. See bug 4726580. 1237 char * s = strrchr(stat, ')'); 1238 1239 i = 0; 1240 if (s) { 1241 // Skip blank chars 1242 do s++; while (isspace(*s)); 1243 1244#define _UFM UINTX_FORMAT 1245#define _DFM INTX_FORMAT 1246 1247 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1248 /* 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 */ 1249 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, 1250 &state, /* 3 %c */ 1251 &ppid, /* 4 %d */ 1252 &pgrp, /* 5 %d */ 1253 &session, /* 6 %d */ 1254 &nr, /* 7 %d */ 1255 &tpgrp, /* 8 %d */ 1256 &flags, /* 9 %lu */ 1257 &minflt, /* 10 %lu */ 1258 &cminflt, /* 11 %lu */ 1259 &majflt, /* 12 %lu */ 1260 &cmajflt, /* 13 %lu */ 1261 &utime, /* 14 %lu */ 1262 &stime, /* 15 %lu */ 1263 &cutime, /* 16 %ld */ 1264 &cstime, /* 17 %ld */ 1265 &prio, /* 18 %ld */ 1266 &nice, /* 19 %ld */ 1267 &junk, /* 20 %ld */ 1268 &it_real, /* 21 %ld */ 1269 &start, /* 22 UINTX_FORMAT */ 1270 &vsize, /* 23 UINTX_FORMAT */ 1271 &rss, /* 24 INTX_FORMAT */ 1272 &rsslim, /* 25 UINTX_FORMAT */ 1273 &scodes, /* 26 UINTX_FORMAT */ 1274 &ecode, /* 27 UINTX_FORMAT */ 1275 &stack_start); /* 28 UINTX_FORMAT */ 1276 } 1277 1278#undef _UFM 1279#undef _DFM 1280 1281 if (i != 28 - 2) { 1282 assert(false, "Bad conversion from /proc/self/stat"); 1283 // product mode - assume we are the initial thread, good luck in the 1284 // embedded case. 1285 warning("Can't detect initial thread stack location - bad conversion"); 1286 stack_start = (uintptr_t) &rlim; 1287 } 1288 } else { 1289 // For some reason we can't open /proc/self/stat (for example, running on 1290 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1291 // most cases, so don't abort: 1292 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1293 stack_start = (uintptr_t) &rlim; 1294 } 1295 } 1296 1297 // Now we have a pointer (stack_start) very close to the stack top, the 1298 // next thing to do is to figure out the exact location of stack top. We 1299 // can find out the virtual memory area that contains stack_start by 1300 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1301 // and its upper limit is the real stack top. (again, this would fail if 1302 // running inside chroot, because /proc may not exist.) 1303 1304 uintptr_t stack_top; 1305 address low, high; 1306 if (find_vma((address)stack_start, &low, &high)) { 1307 // success, "high" is the true stack top. (ignore "low", because initial 1308 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1309 stack_top = (uintptr_t)high; 1310 } else { 1311 // failed, likely because /proc/self/maps does not exist 1312 warning("Can't detect initial thread stack location - find_vma failed"); 1313 // best effort: stack_start is normally within a few pages below the real 1314 // stack top, use it as stack top, and reduce stack size so we won't put 1315 // guard page outside stack. 1316 stack_top = stack_start; 1317 stack_size -= 16 * page_size(); 1318 } 1319 1320 // stack_top could be partially down the page so align it 1321 stack_top = align_size_up(stack_top, page_size()); 1322 1323 if (max_size && stack_size > max_size) { 1324 _initial_thread_stack_size = max_size; 1325 } else { 1326 _initial_thread_stack_size = stack_size; 1327 } 1328 1329 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1330 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1331} 1332 1333//////////////////////////////////////////////////////////////////////////////// 1334// time support 1335 1336// Time since start-up in seconds to a fine granularity. 1337// Used by VMSelfDestructTimer and the MemProfiler. 1338double os::elapsedTime() { 1339 1340 return (double)(os::elapsed_counter()) * 0.000001; 1341} 1342 1343jlong os::elapsed_counter() { 1344 timeval time; 1345 int status = gettimeofday(&time, NULL); 1346 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1347} 1348 1349jlong os::elapsed_frequency() { 1350 return (1000 * 1000); 1351} 1352 1353// For now, we say that linux does not support vtime. I have no idea 1354// whether it can actually be made to (DLD, 9/13/05). 1355 1356bool os::supports_vtime() { return false; } 1357bool os::enable_vtime() { return false; } 1358bool os::vtime_enabled() { return false; } 1359double os::elapsedVTime() { 1360 // better than nothing, but not much 1361 return elapsedTime(); 1362} 1363 1364jlong os::javaTimeMillis() { 1365 timeval time; 1366 int status = gettimeofday(&time, NULL); 1367 assert(status != -1, "linux error"); 1368 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1369} 1370 1371#ifndef CLOCK_MONOTONIC 1372#define CLOCK_MONOTONIC (1) 1373#endif 1374 1375void os::Linux::clock_init() { 1376 // we do dlopen's in this particular order due to bug in linux 1377 // dynamical loader (see 6348968) leading to crash on exit 1378 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1379 if (handle == NULL) { 1380 handle = dlopen("librt.so", RTLD_LAZY); 1381 } 1382 1383 if (handle) { 1384 int (*clock_getres_func)(clockid_t, struct timespec*) = 1385 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1386 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1387 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1388 if (clock_getres_func && clock_gettime_func) { 1389 // See if monotonic clock is supported by the kernel. Note that some 1390 // early implementations simply return kernel jiffies (updated every 1391 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1392 // for nano time (though the monotonic property is still nice to have). 1393 // It's fixed in newer kernels, however clock_getres() still returns 1394 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1395 // resolution for now. Hopefully as people move to new kernels, this 1396 // won't be a problem. 1397 struct timespec res; 1398 struct timespec tp; 1399 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1400 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1401 // yes, monotonic clock is supported 1402 _clock_gettime = clock_gettime_func; 1403 } else { 1404 // close librt if there is no monotonic clock 1405 dlclose(handle); 1406 } 1407 } 1408 } 1409} 1410 1411#ifndef SYS_clock_getres 1412 1413#if defined(IA32) || defined(AMD64) 1414#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1415#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1416#else 1417#warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1418#define sys_clock_getres(x,y) -1 1419#endif 1420 1421#else 1422#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1423#endif 1424 1425void os::Linux::fast_thread_clock_init() { 1426 if (!UseLinuxPosixThreadCPUClocks) { 1427 return; 1428 } 1429 clockid_t clockid; 1430 struct timespec tp; 1431 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1432 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1433 1434 // Switch to using fast clocks for thread cpu time if 1435 // the sys_clock_getres() returns 0 error code. 1436 // Note, that some kernels may support the current thread 1437 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1438 // returned by the pthread_getcpuclockid(). 1439 // If the fast Posix clocks are supported then the sys_clock_getres() 1440 // must return at least tp.tv_sec == 0 which means a resolution 1441 // better than 1 sec. This is extra check for reliability. 1442 1443 if(pthread_getcpuclockid_func && 1444 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1445 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1446 1447 _supports_fast_thread_cpu_time = true; 1448 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1449 } 1450} 1451 1452jlong os::javaTimeNanos() { 1453 if (Linux::supports_monotonic_clock()) { 1454 struct timespec tp; 1455 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1456 assert(status == 0, "gettime error"); 1457 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1458 return result; 1459 } else { 1460 timeval time; 1461 int status = gettimeofday(&time, NULL); 1462 assert(status != -1, "linux error"); 1463 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1464 return 1000 * usecs; 1465 } 1466} 1467 1468void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1469 if (Linux::supports_monotonic_clock()) { 1470 info_ptr->max_value = ALL_64_BITS; 1471 1472 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1473 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1474 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1475 } else { 1476 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1477 info_ptr->max_value = ALL_64_BITS; 1478 1479 // gettimeofday is a real time clock so it skips 1480 info_ptr->may_skip_backward = true; 1481 info_ptr->may_skip_forward = true; 1482 } 1483 1484 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1485} 1486 1487// Return the real, user, and system times in seconds from an 1488// arbitrary fixed point in the past. 1489bool os::getTimesSecs(double* process_real_time, 1490 double* process_user_time, 1491 double* process_system_time) { 1492 struct tms ticks; 1493 clock_t real_ticks = times(&ticks); 1494 1495 if (real_ticks == (clock_t) (-1)) { 1496 return false; 1497 } else { 1498 double ticks_per_second = (double) clock_tics_per_sec; 1499 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1500 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1501 *process_real_time = ((double) real_ticks) / ticks_per_second; 1502 1503 return true; 1504 } 1505} 1506 1507 1508char * os::local_time_string(char *buf, size_t buflen) { 1509 struct tm t; 1510 time_t long_time; 1511 time(&long_time); 1512 localtime_r(&long_time, &t); 1513 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1514 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1515 t.tm_hour, t.tm_min, t.tm_sec); 1516 return buf; 1517} 1518 1519struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1520 return localtime_r(clock, res); 1521} 1522 1523//////////////////////////////////////////////////////////////////////////////// 1524// runtime exit support 1525 1526// Note: os::shutdown() might be called very early during initialization, or 1527// called from signal handler. Before adding something to os::shutdown(), make 1528// sure it is async-safe and can handle partially initialized VM. 1529void os::shutdown() { 1530 1531 // allow PerfMemory to attempt cleanup of any persistent resources 1532 perfMemory_exit(); 1533 1534 // needs to remove object in file system 1535 AttachListener::abort(); 1536 1537 // flush buffered output, finish log files 1538 ostream_abort(); 1539 1540 // Check for abort hook 1541 abort_hook_t abort_hook = Arguments::abort_hook(); 1542 if (abort_hook != NULL) { 1543 abort_hook(); 1544 } 1545 1546} 1547 1548// Note: os::abort() might be called very early during initialization, or 1549// called from signal handler. Before adding something to os::abort(), make 1550// sure it is async-safe and can handle partially initialized VM. 1551void os::abort(bool dump_core) { 1552 os::shutdown(); 1553 if (dump_core) { 1554#ifndef PRODUCT 1555 fdStream out(defaultStream::output_fd()); 1556 out.print_raw("Current thread is "); 1557 char buf[16]; 1558 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1559 out.print_raw_cr(buf); 1560 out.print_raw_cr("Dumping core ..."); 1561#endif 1562 ::abort(); // dump core 1563 } 1564 1565 ::exit(1); 1566} 1567 1568// Die immediately, no exit hook, no abort hook, no cleanup. 1569void os::die() { 1570 // _exit() on LinuxThreads only kills current thread 1571 ::abort(); 1572} 1573 1574// unused on linux for now. 1575void os::set_error_file(const char *logfile) {} 1576 1577 1578// This method is a copy of JDK's sysGetLastErrorString 1579// from src/solaris/hpi/src/system_md.c 1580 1581size_t os::lasterror(char *buf, size_t len) { 1582 1583 if (errno == 0) return 0; 1584 1585 const char *s = ::strerror(errno); 1586 size_t n = ::strlen(s); 1587 if (n >= len) { 1588 n = len - 1; 1589 } 1590 ::strncpy(buf, s, n); 1591 buf[n] = '\0'; 1592 return n; 1593} 1594 1595intx os::current_thread_id() { return (intx)pthread_self(); } 1596int os::current_process_id() { 1597 1598 // Under the old linux thread library, linux gives each thread 1599 // its own process id. Because of this each thread will return 1600 // a different pid if this method were to return the result 1601 // of getpid(2). Linux provides no api that returns the pid 1602 // of the launcher thread for the vm. This implementation 1603 // returns a unique pid, the pid of the launcher thread 1604 // that starts the vm 'process'. 1605 1606 // Under the NPTL, getpid() returns the same pid as the 1607 // launcher thread rather than a unique pid per thread. 1608 // Use gettid() if you want the old pre NPTL behaviour. 1609 1610 // if you are looking for the result of a call to getpid() that 1611 // returns a unique pid for the calling thread, then look at the 1612 // OSThread::thread_id() method in osThread_linux.hpp file 1613 1614 return (int)(_initial_pid ? _initial_pid : getpid()); 1615} 1616 1617// DLL functions 1618 1619const char* os::dll_file_extension() { return ".so"; } 1620 1621// This must be hard coded because it's the system's temporary 1622// directory not the java application's temp directory, ala java.io.tmpdir. 1623const char* os::get_temp_directory() { return "/tmp"; } 1624 1625static bool file_exists(const char* filename) { 1626 struct stat statbuf; 1627 if (filename == NULL || strlen(filename) == 0) { 1628 return false; 1629 } 1630 return os::stat(filename, &statbuf) == 0; 1631} 1632 1633bool os::dll_build_name(char* buffer, size_t buflen, 1634 const char* pname, const char* fname) { 1635 bool retval = false; 1636 // Copied from libhpi 1637 const size_t pnamelen = pname ? strlen(pname) : 0; 1638 1639 // Return error on buffer overflow. 1640 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1641 return retval; 1642 } 1643 1644 if (pnamelen == 0) { 1645 snprintf(buffer, buflen, "lib%s.so", fname); 1646 retval = true; 1647 } else if (strchr(pname, *os::path_separator()) != NULL) { 1648 int n; 1649 char** pelements = split_path(pname, &n); 1650 if (pelements == NULL) { 1651 return false; 1652 } 1653 for (int i = 0 ; i < n ; i++) { 1654 // Really shouldn't be NULL, but check can't hurt 1655 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1656 continue; // skip the empty path values 1657 } 1658 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1659 if (file_exists(buffer)) { 1660 retval = true; 1661 break; 1662 } 1663 } 1664 // release the storage 1665 for (int i = 0 ; i < n ; i++) { 1666 if (pelements[i] != NULL) { 1667 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1668 } 1669 } 1670 if (pelements != NULL) { 1671 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1672 } 1673 } else { 1674 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1675 retval = true; 1676 } 1677 return retval; 1678} 1679 1680const char* os::get_current_directory(char *buf, int buflen) { 1681 return getcwd(buf, buflen); 1682} 1683 1684// check if addr is inside libjvm.so 1685bool os::address_is_in_vm(address addr) { 1686 static address libjvm_base_addr; 1687 Dl_info dlinfo; 1688 1689 if (libjvm_base_addr == NULL) { 1690 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); 1691 libjvm_base_addr = (address)dlinfo.dli_fbase; 1692 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1693 } 1694 1695 if (dladdr((void *)addr, &dlinfo)) { 1696 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1697 } 1698 1699 return false; 1700} 1701 1702bool os::dll_address_to_function_name(address addr, char *buf, 1703 int buflen, int *offset) { 1704 Dl_info dlinfo; 1705 1706 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { 1707 if (buf != NULL) { 1708 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1709 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1710 } 1711 } 1712 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1713 return true; 1714 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) { 1715 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1716 buf, buflen, offset, dlinfo.dli_fname)) { 1717 return true; 1718 } 1719 } 1720 1721 if (buf != NULL) buf[0] = '\0'; 1722 if (offset != NULL) *offset = -1; 1723 return false; 1724} 1725 1726struct _address_to_library_name { 1727 address addr; // input : memory address 1728 size_t buflen; // size of fname 1729 char* fname; // output: library name 1730 address base; // library base addr 1731}; 1732 1733static int address_to_library_name_callback(struct dl_phdr_info *info, 1734 size_t size, void *data) { 1735 int i; 1736 bool found = false; 1737 address libbase = NULL; 1738 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1739 1740 // iterate through all loadable segments 1741 for (i = 0; i < info->dlpi_phnum; i++) { 1742 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1743 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1744 // base address of a library is the lowest address of its loaded 1745 // segments. 1746 if (libbase == NULL || libbase > segbase) { 1747 libbase = segbase; 1748 } 1749 // see if 'addr' is within current segment 1750 if (segbase <= d->addr && 1751 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1752 found = true; 1753 } 1754 } 1755 } 1756 1757 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1758 // so dll_address_to_library_name() can fall through to use dladdr() which 1759 // can figure out executable name from argv[0]. 1760 if (found && info->dlpi_name && info->dlpi_name[0]) { 1761 d->base = libbase; 1762 if (d->fname) { 1763 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1764 } 1765 return 1; 1766 } 1767 return 0; 1768} 1769 1770bool os::dll_address_to_library_name(address addr, char* buf, 1771 int buflen, int* offset) { 1772 Dl_info dlinfo; 1773 struct _address_to_library_name data; 1774 1775 // There is a bug in old glibc dladdr() implementation that it could resolve 1776 // to wrong library name if the .so file has a base address != NULL. Here 1777 // we iterate through the program headers of all loaded libraries to find 1778 // out which library 'addr' really belongs to. This workaround can be 1779 // removed once the minimum requirement for glibc is moved to 2.3.x. 1780 data.addr = addr; 1781 data.fname = buf; 1782 data.buflen = buflen; 1783 data.base = NULL; 1784 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1785 1786 if (rslt) { 1787 // buf already contains library name 1788 if (offset) *offset = addr - data.base; 1789 return true; 1790 } else if (dladdr((void*)addr, &dlinfo)){ 1791 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1792 if (offset) *offset = addr - (address)dlinfo.dli_fbase; 1793 return true; 1794 } else { 1795 if (buf) buf[0] = '\0'; 1796 if (offset) *offset = -1; 1797 return false; 1798 } 1799} 1800 1801 // Loads .dll/.so and 1802 // in case of error it checks if .dll/.so was built for the 1803 // same architecture as Hotspot is running on 1804 1805 1806// Remember the stack's state. The Linux dynamic linker will change 1807// the stack to 'executable' at most once, so we must safepoint only once. 1808bool os::Linux::_stack_is_executable = false; 1809 1810// VM operation that loads a library. This is necessary if stack protection 1811// of the Java stacks can be lost during loading the library. If we 1812// do not stop the Java threads, they can stack overflow before the stacks 1813// are protected again. 1814class VM_LinuxDllLoad: public VM_Operation { 1815 private: 1816 const char *_filename; 1817 char *_ebuf; 1818 int _ebuflen; 1819 void *_lib; 1820 public: 1821 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1822 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1823 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1824 void doit() { 1825 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1826 os::Linux::_stack_is_executable = true; 1827 } 1828 void* loaded_library() { return _lib; } 1829}; 1830 1831void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1832{ 1833 void * result = NULL; 1834 bool load_attempted = false; 1835 1836 // Check whether the library to load might change execution rights 1837 // of the stack. If they are changed, the protection of the stack 1838 // guard pages will be lost. We need a safepoint to fix this. 1839 // 1840 // See Linux man page execstack(8) for more info. 1841 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1842 ElfFile ef(filename); 1843 if (!ef.specifies_noexecstack()) { 1844 if (!is_init_completed()) { 1845 os::Linux::_stack_is_executable = true; 1846 // This is OK - No Java threads have been created yet, and hence no 1847 // stack guard pages to fix. 1848 // 1849 // This should happen only when you are building JDK7 using a very 1850 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1851 // 1852 // Dynamic loader will make all stacks executable after 1853 // this function returns, and will not do that again. 1854 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1855 } else { 1856 warning("You have loaded library %s which might have disabled stack guard. " 1857 "The VM will try to fix the stack guard now.\n" 1858 "It's highly recommended that you fix the library with " 1859 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1860 filename); 1861 1862 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1863 JavaThread *jt = JavaThread::current(); 1864 if (jt->thread_state() != _thread_in_native) { 1865 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1866 // that requires ExecStack. Cannot enter safe point. Let's give up. 1867 warning("Unable to fix stack guard. Giving up."); 1868 } else { 1869 if (!LoadExecStackDllInVMThread) { 1870 // This is for the case where the DLL has an static 1871 // constructor function that executes JNI code. We cannot 1872 // load such DLLs in the VMThread. 1873 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1874 } 1875 1876 ThreadInVMfromNative tiv(jt); 1877 debug_only(VMNativeEntryWrapper vew;) 1878 1879 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1880 VMThread::execute(&op); 1881 if (LoadExecStackDllInVMThread) { 1882 result = op.loaded_library(); 1883 } 1884 load_attempted = true; 1885 } 1886 } 1887 } 1888 } 1889 1890 if (!load_attempted) { 1891 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1892 } 1893 1894 if (result != NULL) { 1895 // Successful loading 1896 return result; 1897 } 1898 1899 Elf32_Ehdr elf_head; 1900 int diag_msg_max_length=ebuflen-strlen(ebuf); 1901 char* diag_msg_buf=ebuf+strlen(ebuf); 1902 1903 if (diag_msg_max_length==0) { 1904 // No more space in ebuf for additional diagnostics message 1905 return NULL; 1906 } 1907 1908 1909 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1910 1911 if (file_descriptor < 0) { 1912 // Can't open library, report dlerror() message 1913 return NULL; 1914 } 1915 1916 bool failed_to_read_elf_head= 1917 (sizeof(elf_head)!= 1918 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1919 1920 ::close(file_descriptor); 1921 if (failed_to_read_elf_head) { 1922 // file i/o error - report dlerror() msg 1923 return NULL; 1924 } 1925 1926 typedef struct { 1927 Elf32_Half code; // Actual value as defined in elf.h 1928 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1929 char elf_class; // 32 or 64 bit 1930 char endianess; // MSB or LSB 1931 char* name; // String representation 1932 } arch_t; 1933 1934 #ifndef EM_486 1935 #define EM_486 6 /* Intel 80486 */ 1936 #endif 1937 1938 static const arch_t arch_array[]={ 1939 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1940 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1941 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1942 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1943 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1944 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1945 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1946 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1947 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1948 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1949 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1950 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1951 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1952 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1953 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1954 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1955 }; 1956 1957 #if (defined IA32) 1958 static Elf32_Half running_arch_code=EM_386; 1959 #elif (defined AMD64) 1960 static Elf32_Half running_arch_code=EM_X86_64; 1961 #elif (defined IA64) 1962 static Elf32_Half running_arch_code=EM_IA_64; 1963 #elif (defined __sparc) && (defined _LP64) 1964 static Elf32_Half running_arch_code=EM_SPARCV9; 1965 #elif (defined __sparc) && (!defined _LP64) 1966 static Elf32_Half running_arch_code=EM_SPARC; 1967 #elif (defined __powerpc64__) 1968 static Elf32_Half running_arch_code=EM_PPC64; 1969 #elif (defined __powerpc__) 1970 static Elf32_Half running_arch_code=EM_PPC; 1971 #elif (defined ARM) 1972 static Elf32_Half running_arch_code=EM_ARM; 1973 #elif (defined S390) 1974 static Elf32_Half running_arch_code=EM_S390; 1975 #elif (defined ALPHA) 1976 static Elf32_Half running_arch_code=EM_ALPHA; 1977 #elif (defined MIPSEL) 1978 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1979 #elif (defined PARISC) 1980 static Elf32_Half running_arch_code=EM_PARISC; 1981 #elif (defined MIPS) 1982 static Elf32_Half running_arch_code=EM_MIPS; 1983 #elif (defined M68K) 1984 static Elf32_Half running_arch_code=EM_68K; 1985 #else 1986 #error Method os::dll_load requires that one of following is defined:\ 1987 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 1988 #endif 1989 1990 // Identify compatability class for VM's architecture and library's architecture 1991 // Obtain string descriptions for architectures 1992 1993 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1994 int running_arch_index=-1; 1995 1996 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1997 if (running_arch_code == arch_array[i].code) { 1998 running_arch_index = i; 1999 } 2000 if (lib_arch.code == arch_array[i].code) { 2001 lib_arch.compat_class = arch_array[i].compat_class; 2002 lib_arch.name = arch_array[i].name; 2003 } 2004 } 2005 2006 assert(running_arch_index != -1, 2007 "Didn't find running architecture code (running_arch_code) in arch_array"); 2008 if (running_arch_index == -1) { 2009 // Even though running architecture detection failed 2010 // we may still continue with reporting dlerror() message 2011 return NULL; 2012 } 2013 2014 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 2015 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 2016 return NULL; 2017 } 2018 2019#ifndef S390 2020 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 2021 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 2022 return NULL; 2023 } 2024#endif // !S390 2025 2026 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2027 if ( lib_arch.name!=NULL ) { 2028 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2029 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2030 lib_arch.name, arch_array[running_arch_index].name); 2031 } else { 2032 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2033 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2034 lib_arch.code, 2035 arch_array[running_arch_index].name); 2036 } 2037 } 2038 2039 return NULL; 2040} 2041 2042void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) { 2043 void * result = ::dlopen(filename, RTLD_LAZY); 2044 if (result == NULL) { 2045 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2046 ebuf[ebuflen-1] = '\0'; 2047 } 2048 return result; 2049} 2050 2051void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) { 2052 void * result = NULL; 2053 if (LoadExecStackDllInVMThread) { 2054 result = dlopen_helper(filename, ebuf, ebuflen); 2055 } 2056 2057 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2058 // library that requires an executable stack, or which does not have this 2059 // stack attribute set, dlopen changes the stack attribute to executable. The 2060 // read protection of the guard pages gets lost. 2061 // 2062 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2063 // may have been queued at the same time. 2064 2065 if (!_stack_is_executable) { 2066 JavaThread *jt = Threads::first(); 2067 2068 while (jt) { 2069 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2070 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2071 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2072 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2073 warning("Attempt to reguard stack yellow zone failed."); 2074 } 2075 } 2076 jt = jt->next(); 2077 } 2078 } 2079 2080 return result; 2081} 2082 2083/* 2084 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2085 * chances are you might want to run the generated bits against glibc-2.0 2086 * libdl.so, so always use locking for any version of glibc. 2087 */ 2088void* os::dll_lookup(void* handle, const char* name) { 2089 pthread_mutex_lock(&dl_mutex); 2090 void* res = dlsym(handle, name); 2091 pthread_mutex_unlock(&dl_mutex); 2092 return res; 2093} 2094 2095 2096static bool _print_ascii_file(const char* filename, outputStream* st) { 2097 int fd = ::open(filename, O_RDONLY); 2098 if (fd == -1) { 2099 return false; 2100 } 2101 2102 char buf[32]; 2103 int bytes; 2104 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2105 st->print_raw(buf, bytes); 2106 } 2107 2108 ::close(fd); 2109 2110 return true; 2111} 2112 2113void os::print_dll_info(outputStream *st) { 2114 st->print_cr("Dynamic libraries:"); 2115 2116 char fname[32]; 2117 pid_t pid = os::Linux::gettid(); 2118 2119 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2120 2121 if (!_print_ascii_file(fname, st)) { 2122 st->print("Can not get library information for pid = %d\n", pid); 2123 } 2124} 2125 2126void os::print_os_info_brief(outputStream* st) { 2127 os::Linux::print_distro_info(st); 2128 2129 os::Posix::print_uname_info(st); 2130 2131 os::Linux::print_libversion_info(st); 2132 2133} 2134 2135void os::print_os_info(outputStream* st) { 2136 st->print("OS:"); 2137 2138 os::Linux::print_distro_info(st); 2139 2140 os::Posix::print_uname_info(st); 2141 2142 // Print warning if unsafe chroot environment detected 2143 if (unsafe_chroot_detected) { 2144 st->print("WARNING!! "); 2145 st->print_cr(unstable_chroot_error); 2146 } 2147 2148 os::Linux::print_libversion_info(st); 2149 2150 os::Posix::print_rlimit_info(st); 2151 2152 os::Posix::print_load_average(st); 2153 2154 os::Linux::print_full_memory_info(st); 2155} 2156 2157// Try to identify popular distros. 2158// Most Linux distributions have /etc/XXX-release file, which contains 2159// the OS version string. Some have more than one /etc/XXX-release file 2160// (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 2161// so the order is important. 2162void os::Linux::print_distro_info(outputStream* st) { 2163 if (!_print_ascii_file("/etc/mandrake-release", st) && 2164 !_print_ascii_file("/etc/sun-release", st) && 2165 !_print_ascii_file("/etc/redhat-release", st) && 2166 !_print_ascii_file("/etc/SuSE-release", st) && 2167 !_print_ascii_file("/etc/turbolinux-release", st) && 2168 !_print_ascii_file("/etc/gentoo-release", st) && 2169 !_print_ascii_file("/etc/debian_version", st) && 2170 !_print_ascii_file("/etc/ltib-release", st) && 2171 !_print_ascii_file("/etc/angstrom-version", st)) { 2172 st->print("Linux"); 2173 } 2174 st->cr(); 2175} 2176 2177void os::Linux::print_libversion_info(outputStream* st) { 2178 // libc, pthread 2179 st->print("libc:"); 2180 st->print(os::Linux::glibc_version()); st->print(" "); 2181 st->print(os::Linux::libpthread_version()); st->print(" "); 2182 if (os::Linux::is_LinuxThreads()) { 2183 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2184 } 2185 st->cr(); 2186} 2187 2188void os::Linux::print_full_memory_info(outputStream* st) { 2189 st->print("\n/proc/meminfo:\n"); 2190 _print_ascii_file("/proc/meminfo", st); 2191 st->cr(); 2192} 2193 2194void os::print_memory_info(outputStream* st) { 2195 2196 st->print("Memory:"); 2197 st->print(" %dk page", os::vm_page_size()>>10); 2198 2199 // values in struct sysinfo are "unsigned long" 2200 struct sysinfo si; 2201 sysinfo(&si); 2202 2203 st->print(", physical " UINT64_FORMAT "k", 2204 os::physical_memory() >> 10); 2205 st->print("(" UINT64_FORMAT "k free)", 2206 os::available_memory() >> 10); 2207 st->print(", swap " UINT64_FORMAT "k", 2208 ((jlong)si.totalswap * si.mem_unit) >> 10); 2209 st->print("(" UINT64_FORMAT "k free)", 2210 ((jlong)si.freeswap * si.mem_unit) >> 10); 2211 st->cr(); 2212} 2213 2214void os::pd_print_cpu_info(outputStream* st) { 2215 st->print("\n/proc/cpuinfo:\n"); 2216 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2217 st->print(" <Not Available>"); 2218 } 2219 st->cr(); 2220} 2221 2222// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 2223// but they're the same for all the linux arch that we support 2224// and they're the same for solaris but there's no common place to put this. 2225const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 2226 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 2227 "ILL_COPROC", "ILL_BADSTK" }; 2228 2229const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 2230 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 2231 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 2232 2233const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 2234 2235const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 2236 2237void os::print_siginfo(outputStream* st, void* siginfo) { 2238 st->print("siginfo:"); 2239 2240 const int buflen = 100; 2241 char buf[buflen]; 2242 siginfo_t *si = (siginfo_t*)siginfo; 2243 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 2244 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 2245 st->print("si_errno=%s", buf); 2246 } else { 2247 st->print("si_errno=%d", si->si_errno); 2248 } 2249 const int c = si->si_code; 2250 assert(c > 0, "unexpected si_code"); 2251 switch (si->si_signo) { 2252 case SIGILL: 2253 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 2254 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2255 break; 2256 case SIGFPE: 2257 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 2258 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2259 break; 2260 case SIGSEGV: 2261 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 2262 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2263 break; 2264 case SIGBUS: 2265 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 2266 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2267 break; 2268 default: 2269 st->print(", si_code=%d", si->si_code); 2270 // no si_addr 2271 } 2272 2273 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2274 UseSharedSpaces) { 2275 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2276 if (mapinfo->is_in_shared_space(si->si_addr)) { 2277 st->print("\n\nError accessing class data sharing archive." \ 2278 " Mapped file inaccessible during execution, " \ 2279 " possible disk/network problem."); 2280 } 2281 } 2282 st->cr(); 2283} 2284 2285 2286static void print_signal_handler(outputStream* st, int sig, 2287 char* buf, size_t buflen); 2288 2289void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2290 st->print_cr("Signal Handlers:"); 2291 print_signal_handler(st, SIGSEGV, buf, buflen); 2292 print_signal_handler(st, SIGBUS , buf, buflen); 2293 print_signal_handler(st, SIGFPE , buf, buflen); 2294 print_signal_handler(st, SIGPIPE, buf, buflen); 2295 print_signal_handler(st, SIGXFSZ, buf, buflen); 2296 print_signal_handler(st, SIGILL , buf, buflen); 2297 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2298 print_signal_handler(st, SR_signum, buf, buflen); 2299 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2300 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2301 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2302 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2303} 2304 2305static char saved_jvm_path[MAXPATHLEN] = {0}; 2306 2307// Find the full path to the current module, libjvm.so 2308void os::jvm_path(char *buf, jint buflen) { 2309 // Error checking. 2310 if (buflen < MAXPATHLEN) { 2311 assert(false, "must use a large-enough buffer"); 2312 buf[0] = '\0'; 2313 return; 2314 } 2315 // Lazy resolve the path to current module. 2316 if (saved_jvm_path[0] != 0) { 2317 strcpy(buf, saved_jvm_path); 2318 return; 2319 } 2320 2321 char dli_fname[MAXPATHLEN]; 2322 bool ret = dll_address_to_library_name( 2323 CAST_FROM_FN_PTR(address, os::jvm_path), 2324 dli_fname, sizeof(dli_fname), NULL); 2325 assert(ret != 0, "cannot locate libjvm"); 2326 char *rp = realpath(dli_fname, buf); 2327 if (rp == NULL) 2328 return; 2329 2330 if (Arguments::created_by_gamma_launcher()) { 2331 // Support for the gamma launcher. Typical value for buf is 2332 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 2333 // the right place in the string, then assume we are installed in a JDK and 2334 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2335 // up the path so it looks like libjvm.so is installed there (append a 2336 // fake suffix hotspot/libjvm.so). 2337 const char *p = buf + strlen(buf) - 1; 2338 for (int count = 0; p > buf && count < 5; ++count) { 2339 for (--p; p > buf && *p != '/'; --p) 2340 /* empty */ ; 2341 } 2342 2343 if (strncmp(p, "/jre/lib/", 9) != 0) { 2344 // Look for JAVA_HOME in the environment. 2345 char* java_home_var = ::getenv("JAVA_HOME"); 2346 if (java_home_var != NULL && java_home_var[0] != 0) { 2347 char* jrelib_p; 2348 int len; 2349 2350 // Check the current module name "libjvm.so". 2351 p = strrchr(buf, '/'); 2352 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2353 2354 rp = realpath(java_home_var, buf); 2355 if (rp == NULL) 2356 return; 2357 2358 // determine if this is a legacy image or modules image 2359 // modules image doesn't have "jre" subdirectory 2360 len = strlen(buf); 2361 jrelib_p = buf + len; 2362 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2363 if (0 != access(buf, F_OK)) { 2364 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2365 } 2366 2367 if (0 == access(buf, F_OK)) { 2368 // Use current module name "libjvm.so" 2369 len = strlen(buf); 2370 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2371 } else { 2372 // Go back to path of .so 2373 rp = realpath(dli_fname, buf); 2374 if (rp == NULL) 2375 return; 2376 } 2377 } 2378 } 2379 } 2380 2381 strcpy(saved_jvm_path, buf); 2382} 2383 2384void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2385 // no prefix required, not even "_" 2386} 2387 2388void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2389 // no suffix required 2390} 2391 2392//////////////////////////////////////////////////////////////////////////////// 2393// sun.misc.Signal support 2394 2395static volatile jint sigint_count = 0; 2396 2397static void 2398UserHandler(int sig, void *siginfo, void *context) { 2399 // 4511530 - sem_post is serialized and handled by the manager thread. When 2400 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2401 // don't want to flood the manager thread with sem_post requests. 2402 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2403 return; 2404 2405 // Ctrl-C is pressed during error reporting, likely because the error 2406 // handler fails to abort. Let VM die immediately. 2407 if (sig == SIGINT && is_error_reported()) { 2408 os::die(); 2409 } 2410 2411 os::signal_notify(sig); 2412} 2413 2414void* os::user_handler() { 2415 return CAST_FROM_FN_PTR(void*, UserHandler); 2416} 2417 2418extern "C" { 2419 typedef void (*sa_handler_t)(int); 2420 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2421} 2422 2423void* os::signal(int signal_number, void* handler) { 2424 struct sigaction sigAct, oldSigAct; 2425 2426 sigfillset(&(sigAct.sa_mask)); 2427 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2428 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2429 2430 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2431 // -1 means registration failed 2432 return (void *)-1; 2433 } 2434 2435 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2436} 2437 2438void os::signal_raise(int signal_number) { 2439 ::raise(signal_number); 2440} 2441 2442/* 2443 * The following code is moved from os.cpp for making this 2444 * code platform specific, which it is by its very nature. 2445 */ 2446 2447// Will be modified when max signal is changed to be dynamic 2448int os::sigexitnum_pd() { 2449 return NSIG; 2450} 2451 2452// a counter for each possible signal value 2453static volatile jint pending_signals[NSIG+1] = { 0 }; 2454 2455// Linux(POSIX) specific hand shaking semaphore. 2456static sem_t sig_sem; 2457 2458void os::signal_init_pd() { 2459 // Initialize signal structures 2460 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2461 2462 // Initialize signal semaphore 2463 ::sem_init(&sig_sem, 0, 0); 2464} 2465 2466void os::signal_notify(int sig) { 2467 Atomic::inc(&pending_signals[sig]); 2468 ::sem_post(&sig_sem); 2469} 2470 2471static int check_pending_signals(bool wait) { 2472 Atomic::store(0, &sigint_count); 2473 for (;;) { 2474 for (int i = 0; i < NSIG + 1; i++) { 2475 jint n = pending_signals[i]; 2476 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2477 return i; 2478 } 2479 } 2480 if (!wait) { 2481 return -1; 2482 } 2483 JavaThread *thread = JavaThread::current(); 2484 ThreadBlockInVM tbivm(thread); 2485 2486 bool threadIsSuspended; 2487 do { 2488 thread->set_suspend_equivalent(); 2489 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2490 ::sem_wait(&sig_sem); 2491 2492 // were we externally suspended while we were waiting? 2493 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2494 if (threadIsSuspended) { 2495 // 2496 // The semaphore has been incremented, but while we were waiting 2497 // another thread suspended us. We don't want to continue running 2498 // while suspended because that would surprise the thread that 2499 // suspended us. 2500 // 2501 ::sem_post(&sig_sem); 2502 2503 thread->java_suspend_self(); 2504 } 2505 } while (threadIsSuspended); 2506 } 2507} 2508 2509int os::signal_lookup() { 2510 return check_pending_signals(false); 2511} 2512 2513int os::signal_wait() { 2514 return check_pending_signals(true); 2515} 2516 2517//////////////////////////////////////////////////////////////////////////////// 2518// Virtual Memory 2519 2520int os::vm_page_size() { 2521 // Seems redundant as all get out 2522 assert(os::Linux::page_size() != -1, "must call os::init"); 2523 return os::Linux::page_size(); 2524} 2525 2526// Solaris allocates memory by pages. 2527int os::vm_allocation_granularity() { 2528 assert(os::Linux::page_size() != -1, "must call os::init"); 2529 return os::Linux::page_size(); 2530} 2531 2532// Rationale behind this function: 2533// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2534// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2535// samples for JITted code. Here we create private executable mapping over the code cache 2536// and then we can use standard (well, almost, as mapping can change) way to provide 2537// info for the reporting script by storing timestamp and location of symbol 2538void linux_wrap_code(char* base, size_t size) { 2539 static volatile jint cnt = 0; 2540 2541 if (!UseOprofile) { 2542 return; 2543 } 2544 2545 char buf[PATH_MAX+1]; 2546 int num = Atomic::add(1, &cnt); 2547 2548 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2549 os::get_temp_directory(), os::current_process_id(), num); 2550 unlink(buf); 2551 2552 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2553 2554 if (fd != -1) { 2555 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2556 if (rv != (off_t)-1) { 2557 if (::write(fd, "", 1) == 1) { 2558 mmap(base, size, 2559 PROT_READ|PROT_WRITE|PROT_EXEC, 2560 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2561 } 2562 } 2563 ::close(fd); 2564 unlink(buf); 2565 } 2566} 2567 2568// NOTE: Linux kernel does not really reserve the pages for us. 2569// All it does is to check if there are enough free pages 2570// left at the time of mmap(). This could be a potential 2571// problem. 2572bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2573 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2574 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2575 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2576 if (res != (uintptr_t) MAP_FAILED) { 2577 if (UseNUMAInterleaving) { 2578 numa_make_global(addr, size); 2579 } 2580 return true; 2581 } 2582 return false; 2583} 2584 2585// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2586#ifndef MAP_HUGETLB 2587#define MAP_HUGETLB 0x40000 2588#endif 2589 2590// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2591#ifndef MADV_HUGEPAGE 2592#define MADV_HUGEPAGE 14 2593#endif 2594 2595bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2596 bool exec) { 2597 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) { 2598 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2599 uintptr_t res = 2600 (uintptr_t) ::mmap(addr, size, prot, 2601 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB, 2602 -1, 0); 2603 if (res != (uintptr_t) MAP_FAILED) { 2604 if (UseNUMAInterleaving) { 2605 numa_make_global(addr, size); 2606 } 2607 return true; 2608 } 2609 // Fall through and try to use small pages 2610 } 2611 2612 if (commit_memory(addr, size, exec)) { 2613 realign_memory(addr, size, alignment_hint); 2614 return true; 2615 } 2616 return false; 2617} 2618 2619void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2620 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) { 2621 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2622 // be supported or the memory may already be backed by huge pages. 2623 ::madvise(addr, bytes, MADV_HUGEPAGE); 2624 } 2625} 2626 2627void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2628 // This method works by doing an mmap over an existing mmaping and effectively discarding 2629 // the existing pages. However it won't work for SHM-based large pages that cannot be 2630 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2631 // small pages on top of the SHM segment. This method always works for small pages, so we 2632 // allow that in any case. 2633 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) { 2634 commit_memory(addr, bytes, alignment_hint, false); 2635 } 2636} 2637 2638void os::numa_make_global(char *addr, size_t bytes) { 2639 Linux::numa_interleave_memory(addr, bytes); 2640} 2641 2642void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2643 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2644} 2645 2646bool os::numa_topology_changed() { return false; } 2647 2648size_t os::numa_get_groups_num() { 2649 int max_node = Linux::numa_max_node(); 2650 return max_node > 0 ? max_node + 1 : 1; 2651} 2652 2653int os::numa_get_group_id() { 2654 int cpu_id = Linux::sched_getcpu(); 2655 if (cpu_id != -1) { 2656 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2657 if (lgrp_id != -1) { 2658 return lgrp_id; 2659 } 2660 } 2661 return 0; 2662} 2663 2664size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2665 for (size_t i = 0; i < size; i++) { 2666 ids[i] = i; 2667 } 2668 return size; 2669} 2670 2671bool os::get_page_info(char *start, page_info* info) { 2672 return false; 2673} 2674 2675char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2676 return end; 2677} 2678 2679 2680int os::Linux::sched_getcpu_syscall(void) { 2681 unsigned int cpu; 2682 int retval = -1; 2683 2684#if defined(IA32) 2685# ifndef SYS_getcpu 2686# define SYS_getcpu 318 2687# endif 2688 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2689#elif defined(AMD64) 2690// Unfortunately we have to bring all these macros here from vsyscall.h 2691// to be able to compile on old linuxes. 2692# define __NR_vgetcpu 2 2693# define VSYSCALL_START (-10UL << 20) 2694# define VSYSCALL_SIZE 1024 2695# define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2696 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2697 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2698 retval = vgetcpu(&cpu, NULL, NULL); 2699#endif 2700 2701 return (retval == -1) ? retval : cpu; 2702} 2703 2704// Something to do with the numa-aware allocator needs these symbols 2705extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2706extern "C" JNIEXPORT void numa_error(char *where) { } 2707extern "C" JNIEXPORT int fork1() { return fork(); } 2708 2709 2710// If we are running with libnuma version > 2, then we should 2711// be trying to use symbols with versions 1.1 2712// If we are running with earlier version, which did not have symbol versions, 2713// we should use the base version. 2714void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2715 void *f = dlvsym(handle, name, "libnuma_1.1"); 2716 if (f == NULL) { 2717 f = dlsym(handle, name); 2718 } 2719 return f; 2720} 2721 2722bool os::Linux::libnuma_init() { 2723 // sched_getcpu() should be in libc. 2724 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2725 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2726 2727 // If it's not, try a direct syscall. 2728 if (sched_getcpu() == -1) 2729 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall)); 2730 2731 if (sched_getcpu() != -1) { // Does it work? 2732 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2733 if (handle != NULL) { 2734 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2735 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2736 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2737 libnuma_dlsym(handle, "numa_max_node"))); 2738 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2739 libnuma_dlsym(handle, "numa_available"))); 2740 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2741 libnuma_dlsym(handle, "numa_tonode_memory"))); 2742 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2743 libnuma_dlsym(handle, "numa_interleave_memory"))); 2744 2745 2746 if (numa_available() != -1) { 2747 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2748 // Create a cpu -> node mapping 2749 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2750 rebuild_cpu_to_node_map(); 2751 return true; 2752 } 2753 } 2754 } 2755 return false; 2756} 2757 2758// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2759// The table is later used in get_node_by_cpu(). 2760void os::Linux::rebuild_cpu_to_node_map() { 2761 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2762 // in libnuma (possible values are starting from 16, 2763 // and continuing up with every other power of 2, but less 2764 // than the maximum number of CPUs supported by kernel), and 2765 // is a subject to change (in libnuma version 2 the requirements 2766 // are more reasonable) we'll just hardcode the number they use 2767 // in the library. 2768 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2769 2770 size_t cpu_num = os::active_processor_count(); 2771 size_t cpu_map_size = NCPUS / BitsPerCLong; 2772 size_t cpu_map_valid_size = 2773 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2774 2775 cpu_to_node()->clear(); 2776 cpu_to_node()->at_grow(cpu_num - 1); 2777 size_t node_num = numa_get_groups_num(); 2778 2779 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2780 for (size_t i = 0; i < node_num; i++) { 2781 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2782 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2783 if (cpu_map[j] != 0) { 2784 for (size_t k = 0; k < BitsPerCLong; k++) { 2785 if (cpu_map[j] & (1UL << k)) { 2786 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2787 } 2788 } 2789 } 2790 } 2791 } 2792 } 2793 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2794} 2795 2796int os::Linux::get_node_by_cpu(int cpu_id) { 2797 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2798 return cpu_to_node()->at(cpu_id); 2799 } 2800 return -1; 2801} 2802 2803GrowableArray<int>* os::Linux::_cpu_to_node; 2804os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2805os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2806os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2807os::Linux::numa_available_func_t os::Linux::_numa_available; 2808os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2809os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2810unsigned long* os::Linux::_numa_all_nodes; 2811 2812bool os::pd_uncommit_memory(char* addr, size_t size) { 2813 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2814 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2815 return res != (uintptr_t) MAP_FAILED; 2816} 2817 2818// Linux uses a growable mapping for the stack, and if the mapping for 2819// the stack guard pages is not removed when we detach a thread the 2820// stack cannot grow beyond the pages where the stack guard was 2821// mapped. If at some point later in the process the stack expands to 2822// that point, the Linux kernel cannot expand the stack any further 2823// because the guard pages are in the way, and a segfault occurs. 2824// 2825// However, it's essential not to split the stack region by unmapping 2826// a region (leaving a hole) that's already part of the stack mapping, 2827// so if the stack mapping has already grown beyond the guard pages at 2828// the time we create them, we have to truncate the stack mapping. 2829// So, we need to know the extent of the stack mapping when 2830// create_stack_guard_pages() is called. 2831 2832// Find the bounds of the stack mapping. Return true for success. 2833// 2834// We only need this for stacks that are growable: at the time of 2835// writing thread stacks don't use growable mappings (i.e. those 2836// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 2837// only applies to the main thread. 2838 2839static 2840bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) { 2841 2842 char buf[128]; 2843 int fd, sz; 2844 2845 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) { 2846 return false; 2847 } 2848 2849 const char kw[] = "[stack]"; 2850 const int kwlen = sizeof(kw)-1; 2851 2852 // Address part of /proc/self/maps couldn't be more than 128 bytes 2853 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) { 2854 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) { 2855 // Extract addresses 2856 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) { 2857 uintptr_t sp = (uintptr_t) __builtin_frame_address(0); 2858 if (sp >= *bottom && sp <= *top) { 2859 ::close(fd); 2860 return true; 2861 } 2862 } 2863 } 2864 } 2865 2866 ::close(fd); 2867 return false; 2868} 2869 2870 2871// If the (growable) stack mapping already extends beyond the point 2872// where we're going to put our guard pages, truncate the mapping at 2873// that point by munmap()ping it. This ensures that when we later 2874// munmap() the guard pages we don't leave a hole in the stack 2875// mapping. This only affects the main/initial thread, but guard 2876// against future OS changes 2877bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 2878 uintptr_t stack_extent, stack_base; 2879 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2880 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2881 assert(os::Linux::is_initial_thread(), 2882 "growable stack in non-initial thread"); 2883 if (stack_extent < (uintptr_t)addr) 2884 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent); 2885 } 2886 2887 return os::commit_memory(addr, size); 2888} 2889 2890// If this is a growable mapping, remove the guard pages entirely by 2891// munmap()ping them. If not, just call uncommit_memory(). This only 2892// affects the main/initial thread, but guard against future OS changes 2893bool os::remove_stack_guard_pages(char* addr, size_t size) { 2894 uintptr_t stack_extent, stack_base; 2895 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2896 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2897 assert(os::Linux::is_initial_thread(), 2898 "growable stack in non-initial thread"); 2899 2900 return ::munmap(addr, size) == 0; 2901 } 2902 2903 return os::uncommit_memory(addr, size); 2904} 2905 2906static address _highest_vm_reserved_address = NULL; 2907 2908// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2909// at 'requested_addr'. If there are existing memory mappings at the same 2910// location, however, they will be overwritten. If 'fixed' is false, 2911// 'requested_addr' is only treated as a hint, the return value may or 2912// may not start from the requested address. Unlike Linux mmap(), this 2913// function returns NULL to indicate failure. 2914static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2915 char * addr; 2916 int flags; 2917 2918 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2919 if (fixed) { 2920 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2921 flags |= MAP_FIXED; 2922 } 2923 2924 // Map uncommitted pages PROT_READ and PROT_WRITE, change access 2925 // to PROT_EXEC if executable when we commit the page. 2926 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE, 2927 flags, -1, 0); 2928 2929 if (addr != MAP_FAILED) { 2930 // anon_mmap() should only get called during VM initialization, 2931 // don't need lock (actually we can skip locking even it can be called 2932 // from multiple threads, because _highest_vm_reserved_address is just a 2933 // hint about the upper limit of non-stack memory regions.) 2934 if ((address)addr + bytes > _highest_vm_reserved_address) { 2935 _highest_vm_reserved_address = (address)addr + bytes; 2936 } 2937 } 2938 2939 return addr == MAP_FAILED ? NULL : addr; 2940} 2941 2942// Don't update _highest_vm_reserved_address, because there might be memory 2943// regions above addr + size. If so, releasing a memory region only creates 2944// a hole in the address space, it doesn't help prevent heap-stack collision. 2945// 2946static int anon_munmap(char * addr, size_t size) { 2947 return ::munmap(addr, size) == 0; 2948} 2949 2950char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 2951 size_t alignment_hint) { 2952 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 2953} 2954 2955bool os::pd_release_memory(char* addr, size_t size) { 2956 return anon_munmap(addr, size); 2957} 2958 2959static address highest_vm_reserved_address() { 2960 return _highest_vm_reserved_address; 2961} 2962 2963static bool linux_mprotect(char* addr, size_t size, int prot) { 2964 // Linux wants the mprotect address argument to be page aligned. 2965 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 2966 2967 // According to SUSv3, mprotect() should only be used with mappings 2968 // established by mmap(), and mmap() always maps whole pages. Unaligned 2969 // 'addr' likely indicates problem in the VM (e.g. trying to change 2970 // protection of malloc'ed or statically allocated memory). Check the 2971 // caller if you hit this assert. 2972 assert(addr == bottom, "sanity check"); 2973 2974 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 2975 return ::mprotect(bottom, size, prot) == 0; 2976} 2977 2978// Set protections specified 2979bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 2980 bool is_committed) { 2981 unsigned int p = 0; 2982 switch (prot) { 2983 case MEM_PROT_NONE: p = PROT_NONE; break; 2984 case MEM_PROT_READ: p = PROT_READ; break; 2985 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 2986 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 2987 default: 2988 ShouldNotReachHere(); 2989 } 2990 // is_committed is unused. 2991 return linux_mprotect(addr, bytes, p); 2992} 2993 2994bool os::guard_memory(char* addr, size_t size) { 2995 return linux_mprotect(addr, size, PROT_NONE); 2996} 2997 2998bool os::unguard_memory(char* addr, size_t size) { 2999 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3000} 3001 3002bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3003 bool result = false; 3004 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE, 3005 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3006 -1, 0); 3007 3008 if (p != (void *) -1) { 3009 // We don't know if this really is a huge page or not. 3010 FILE *fp = fopen("/proc/self/maps", "r"); 3011 if (fp) { 3012 while (!feof(fp)) { 3013 char chars[257]; 3014 long x = 0; 3015 if (fgets(chars, sizeof(chars), fp)) { 3016 if (sscanf(chars, "%lx-%*x", &x) == 1 3017 && x == (long)p) { 3018 if (strstr (chars, "hugepage")) { 3019 result = true; 3020 break; 3021 } 3022 } 3023 } 3024 } 3025 fclose(fp); 3026 } 3027 munmap (p, page_size); 3028 if (result) 3029 return true; 3030 } 3031 3032 if (warn) { 3033 warning("HugeTLBFS is not supported by the operating system."); 3034 } 3035 3036 return result; 3037} 3038 3039/* 3040* Set the coredump_filter bits to include largepages in core dump (bit 6) 3041* 3042* From the coredump_filter documentation: 3043* 3044* - (bit 0) anonymous private memory 3045* - (bit 1) anonymous shared memory 3046* - (bit 2) file-backed private memory 3047* - (bit 3) file-backed shared memory 3048* - (bit 4) ELF header pages in file-backed private memory areas (it is 3049* effective only if the bit 2 is cleared) 3050* - (bit 5) hugetlb private memory 3051* - (bit 6) hugetlb shared memory 3052*/ 3053static void set_coredump_filter(void) { 3054 FILE *f; 3055 long cdm; 3056 3057 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3058 return; 3059 } 3060 3061 if (fscanf(f, "%lx", &cdm) != 1) { 3062 fclose(f); 3063 return; 3064 } 3065 3066 rewind(f); 3067 3068 if ((cdm & LARGEPAGES_BIT) == 0) { 3069 cdm |= LARGEPAGES_BIT; 3070 fprintf(f, "%#lx", cdm); 3071 } 3072 3073 fclose(f); 3074} 3075 3076// Large page support 3077 3078static size_t _large_page_size = 0; 3079 3080void os::large_page_init() { 3081 if (!UseLargePages) { 3082 UseHugeTLBFS = false; 3083 UseSHM = false; 3084 return; 3085 } 3086 3087 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) { 3088 // If UseLargePages is specified on the command line try both methods, 3089 // if it's default, then try only HugeTLBFS. 3090 if (FLAG_IS_DEFAULT(UseLargePages)) { 3091 UseHugeTLBFS = true; 3092 } else { 3093 UseHugeTLBFS = UseSHM = true; 3094 } 3095 } 3096 3097 if (LargePageSizeInBytes) { 3098 _large_page_size = LargePageSizeInBytes; 3099 } else { 3100 // large_page_size on Linux is used to round up heap size. x86 uses either 3101 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3102 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3103 // page as large as 256M. 3104 // 3105 // Here we try to figure out page size by parsing /proc/meminfo and looking 3106 // for a line with the following format: 3107 // Hugepagesize: 2048 kB 3108 // 3109 // If we can't determine the value (e.g. /proc is not mounted, or the text 3110 // format has been changed), we'll use the largest page size supported by 3111 // the processor. 3112 3113#ifndef ZERO 3114 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3115 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3116#endif // ZERO 3117 3118 FILE *fp = fopen("/proc/meminfo", "r"); 3119 if (fp) { 3120 while (!feof(fp)) { 3121 int x = 0; 3122 char buf[16]; 3123 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3124 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3125 _large_page_size = x * K; 3126 break; 3127 } 3128 } else { 3129 // skip to next line 3130 for (;;) { 3131 int ch = fgetc(fp); 3132 if (ch == EOF || ch == (int)'\n') break; 3133 } 3134 } 3135 } 3136 fclose(fp); 3137 } 3138 } 3139 3140 // print a warning if any large page related flag is specified on command line 3141 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3142 3143 const size_t default_page_size = (size_t)Linux::page_size(); 3144 if (_large_page_size > default_page_size) { 3145 _page_sizes[0] = _large_page_size; 3146 _page_sizes[1] = default_page_size; 3147 _page_sizes[2] = 0; 3148 } 3149 UseHugeTLBFS = UseHugeTLBFS && 3150 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size); 3151 3152 if (UseHugeTLBFS) 3153 UseSHM = false; 3154 3155 UseLargePages = UseHugeTLBFS || UseSHM; 3156 3157 set_coredump_filter(); 3158} 3159 3160#ifndef SHM_HUGETLB 3161#define SHM_HUGETLB 04000 3162#endif 3163 3164char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) { 3165 // "exec" is passed in but not used. Creating the shared image for 3166 // the code cache doesn't have an SHM_X executable permission to check. 3167 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3168 3169 key_t key = IPC_PRIVATE; 3170 char *addr; 3171 3172 bool warn_on_failure = UseLargePages && 3173 (!FLAG_IS_DEFAULT(UseLargePages) || 3174 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 3175 ); 3176 char msg[128]; 3177 3178 // Create a large shared memory region to attach to based on size. 3179 // Currently, size is the total size of the heap 3180 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3181 if (shmid == -1) { 3182 // Possible reasons for shmget failure: 3183 // 1. shmmax is too small for Java heap. 3184 // > check shmmax value: cat /proc/sys/kernel/shmmax 3185 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3186 // 2. not enough large page memory. 3187 // > check available large pages: cat /proc/meminfo 3188 // > increase amount of large pages: 3189 // echo new_value > /proc/sys/vm/nr_hugepages 3190 // Note 1: different Linux may use different name for this property, 3191 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3192 // Note 2: it's possible there's enough physical memory available but 3193 // they are so fragmented after a long run that they can't 3194 // coalesce into large pages. Try to reserve large pages when 3195 // the system is still "fresh". 3196 if (warn_on_failure) { 3197 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3198 warning(msg); 3199 } 3200 return NULL; 3201 } 3202 3203 // attach to the region 3204 addr = (char*)shmat(shmid, req_addr, 0); 3205 int err = errno; 3206 3207 // Remove shmid. If shmat() is successful, the actual shared memory segment 3208 // will be deleted when it's detached by shmdt() or when the process 3209 // terminates. If shmat() is not successful this will remove the shared 3210 // segment immediately. 3211 shmctl(shmid, IPC_RMID, NULL); 3212 3213 if ((intptr_t)addr == -1) { 3214 if (warn_on_failure) { 3215 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3216 warning(msg); 3217 } 3218 return NULL; 3219 } 3220 3221 if ((addr != NULL) && UseNUMAInterleaving) { 3222 numa_make_global(addr, bytes); 3223 } 3224 3225 // The memory is committed 3226 address pc = CALLER_PC; 3227 MemTracker::record_virtual_memory_reserve((address)addr, bytes, pc); 3228 MemTracker::record_virtual_memory_commit((address)addr, bytes, pc); 3229 3230 return addr; 3231} 3232 3233bool os::release_memory_special(char* base, size_t bytes) { 3234 // detaching the SHM segment will also delete it, see reserve_memory_special() 3235 int rslt = shmdt(base); 3236 if (rslt == 0) { 3237 MemTracker::record_virtual_memory_uncommit((address)base, bytes); 3238 MemTracker::record_virtual_memory_release((address)base, bytes); 3239 return true; 3240 } else { 3241 return false; 3242 } 3243} 3244 3245size_t os::large_page_size() { 3246 return _large_page_size; 3247} 3248 3249// HugeTLBFS allows application to commit large page memory on demand; 3250// with SysV SHM the entire memory region must be allocated as shared 3251// memory. 3252bool os::can_commit_large_page_memory() { 3253 return UseHugeTLBFS; 3254} 3255 3256bool os::can_execute_large_page_memory() { 3257 return UseHugeTLBFS; 3258} 3259 3260// Reserve memory at an arbitrary address, only if that area is 3261// available (and not reserved for something else). 3262 3263char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3264 const int max_tries = 10; 3265 char* base[max_tries]; 3266 size_t size[max_tries]; 3267 const size_t gap = 0x000000; 3268 3269 // Assert only that the size is a multiple of the page size, since 3270 // that's all that mmap requires, and since that's all we really know 3271 // about at this low abstraction level. If we need higher alignment, 3272 // we can either pass an alignment to this method or verify alignment 3273 // in one of the methods further up the call chain. See bug 5044738. 3274 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3275 3276 // Repeatedly allocate blocks until the block is allocated at the 3277 // right spot. Give up after max_tries. Note that reserve_memory() will 3278 // automatically update _highest_vm_reserved_address if the call is 3279 // successful. The variable tracks the highest memory address every reserved 3280 // by JVM. It is used to detect heap-stack collision if running with 3281 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3282 // space than needed, it could confuse the collision detecting code. To 3283 // solve the problem, save current _highest_vm_reserved_address and 3284 // calculate the correct value before return. 3285 address old_highest = _highest_vm_reserved_address; 3286 3287 // Linux mmap allows caller to pass an address as hint; give it a try first, 3288 // if kernel honors the hint then we can return immediately. 3289 char * addr = anon_mmap(requested_addr, bytes, false); 3290 if (addr == requested_addr) { 3291 return requested_addr; 3292 } 3293 3294 if (addr != NULL) { 3295 // mmap() is successful but it fails to reserve at the requested address 3296 anon_munmap(addr, bytes); 3297 } 3298 3299 int i; 3300 for (i = 0; i < max_tries; ++i) { 3301 base[i] = reserve_memory(bytes); 3302 3303 if (base[i] != NULL) { 3304 // Is this the block we wanted? 3305 if (base[i] == requested_addr) { 3306 size[i] = bytes; 3307 break; 3308 } 3309 3310 // Does this overlap the block we wanted? Give back the overlapped 3311 // parts and try again. 3312 3313 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3314 if (top_overlap >= 0 && top_overlap < bytes) { 3315 unmap_memory(base[i], top_overlap); 3316 base[i] += top_overlap; 3317 size[i] = bytes - top_overlap; 3318 } else { 3319 size_t bottom_overlap = base[i] + bytes - requested_addr; 3320 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3321 unmap_memory(requested_addr, bottom_overlap); 3322 size[i] = bytes - bottom_overlap; 3323 } else { 3324 size[i] = bytes; 3325 } 3326 } 3327 } 3328 } 3329 3330 // Give back the unused reserved pieces. 3331 3332 for (int j = 0; j < i; ++j) { 3333 if (base[j] != NULL) { 3334 unmap_memory(base[j], size[j]); 3335 } 3336 } 3337 3338 if (i < max_tries) { 3339 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3340 return requested_addr; 3341 } else { 3342 _highest_vm_reserved_address = old_highest; 3343 return NULL; 3344 } 3345} 3346 3347size_t os::read(int fd, void *buf, unsigned int nBytes) { 3348 return ::read(fd, buf, nBytes); 3349} 3350 3351// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 3352// Solaris uses poll(), linux uses park(). 3353// Poll() is likely a better choice, assuming that Thread.interrupt() 3354// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 3355// SIGSEGV, see 4355769. 3356 3357int os::sleep(Thread* thread, jlong millis, bool interruptible) { 3358 assert(thread == Thread::current(), "thread consistency check"); 3359 3360 ParkEvent * const slp = thread->_SleepEvent ; 3361 slp->reset() ; 3362 OrderAccess::fence() ; 3363 3364 if (interruptible) { 3365 jlong prevtime = javaTimeNanos(); 3366 3367 for (;;) { 3368 if (os::is_interrupted(thread, true)) { 3369 return OS_INTRPT; 3370 } 3371 3372 jlong newtime = javaTimeNanos(); 3373 3374 if (newtime - prevtime < 0) { 3375 // time moving backwards, should only happen if no monotonic clock 3376 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3377 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3378 } else { 3379 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3380 } 3381 3382 if(millis <= 0) { 3383 return OS_OK; 3384 } 3385 3386 prevtime = newtime; 3387 3388 { 3389 assert(thread->is_Java_thread(), "sanity check"); 3390 JavaThread *jt = (JavaThread *) thread; 3391 ThreadBlockInVM tbivm(jt); 3392 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 3393 3394 jt->set_suspend_equivalent(); 3395 // cleared by handle_special_suspend_equivalent_condition() or 3396 // java_suspend_self() via check_and_wait_while_suspended() 3397 3398 slp->park(millis); 3399 3400 // were we externally suspended while we were waiting? 3401 jt->check_and_wait_while_suspended(); 3402 } 3403 } 3404 } else { 3405 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 3406 jlong prevtime = javaTimeNanos(); 3407 3408 for (;;) { 3409 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 3410 // the 1st iteration ... 3411 jlong newtime = javaTimeNanos(); 3412 3413 if (newtime - prevtime < 0) { 3414 // time moving backwards, should only happen if no monotonic clock 3415 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3416 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3417 } else { 3418 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3419 } 3420 3421 if(millis <= 0) break ; 3422 3423 prevtime = newtime; 3424 slp->park(millis); 3425 } 3426 return OS_OK ; 3427 } 3428} 3429 3430int os::naked_sleep() { 3431 // %% make the sleep time an integer flag. for now use 1 millisec. 3432 return os::sleep(Thread::current(), 1, false); 3433} 3434 3435// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3436void os::infinite_sleep() { 3437 while (true) { // sleep forever ... 3438 ::sleep(100); // ... 100 seconds at a time 3439 } 3440} 3441 3442// Used to convert frequent JVM_Yield() to nops 3443bool os::dont_yield() { 3444 return DontYieldALot; 3445} 3446 3447void os::yield() { 3448 sched_yield(); 3449} 3450 3451os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3452 3453void os::yield_all(int attempts) { 3454 // Yields to all threads, including threads with lower priorities 3455 // Threads on Linux are all with same priority. The Solaris style 3456 // os::yield_all() with nanosleep(1ms) is not necessary. 3457 sched_yield(); 3458} 3459 3460// Called from the tight loops to possibly influence time-sharing heuristics 3461void os::loop_breaker(int attempts) { 3462 os::yield_all(attempts); 3463} 3464 3465//////////////////////////////////////////////////////////////////////////////// 3466// thread priority support 3467 3468// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3469// only supports dynamic priority, static priority must be zero. For real-time 3470// applications, Linux supports SCHED_RR which allows static priority (1-99). 3471// However, for large multi-threaded applications, SCHED_RR is not only slower 3472// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3473// of 5 runs - Sep 2005). 3474// 3475// The following code actually changes the niceness of kernel-thread/LWP. It 3476// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3477// not the entire user process, and user level threads are 1:1 mapped to kernel 3478// threads. It has always been the case, but could change in the future. For 3479// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3480// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3481 3482int os::java_to_os_priority[CriticalPriority + 1] = { 3483 19, // 0 Entry should never be used 3484 3485 4, // 1 MinPriority 3486 3, // 2 3487 2, // 3 3488 3489 1, // 4 3490 0, // 5 NormPriority 3491 -1, // 6 3492 3493 -2, // 7 3494 -3, // 8 3495 -4, // 9 NearMaxPriority 3496 3497 -5, // 10 MaxPriority 3498 3499 -5 // 11 CriticalPriority 3500}; 3501 3502static int prio_init() { 3503 if (ThreadPriorityPolicy == 1) { 3504 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3505 // if effective uid is not root. Perhaps, a more elegant way of doing 3506 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3507 if (geteuid() != 0) { 3508 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3509 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3510 } 3511 ThreadPriorityPolicy = 0; 3512 } 3513 } 3514 if (UseCriticalJavaThreadPriority) { 3515 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3516 } 3517 return 0; 3518} 3519 3520OSReturn os::set_native_priority(Thread* thread, int newpri) { 3521 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3522 3523 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3524 return (ret == 0) ? OS_OK : OS_ERR; 3525} 3526 3527OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3528 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3529 *priority_ptr = java_to_os_priority[NormPriority]; 3530 return OS_OK; 3531 } 3532 3533 errno = 0; 3534 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3535 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3536} 3537 3538// Hint to the underlying OS that a task switch would not be good. 3539// Void return because it's a hint and can fail. 3540void os::hint_no_preempt() {} 3541 3542//////////////////////////////////////////////////////////////////////////////// 3543// suspend/resume support 3544 3545// the low-level signal-based suspend/resume support is a remnant from the 3546// old VM-suspension that used to be for java-suspension, safepoints etc, 3547// within hotspot. Now there is a single use-case for this: 3548// - calling get_thread_pc() on the VMThread by the flat-profiler task 3549// that runs in the watcher thread. 3550// The remaining code is greatly simplified from the more general suspension 3551// code that used to be used. 3552// 3553// The protocol is quite simple: 3554// - suspend: 3555// - sends a signal to the target thread 3556// - polls the suspend state of the osthread using a yield loop 3557// - target thread signal handler (SR_handler) sets suspend state 3558// and blocks in sigsuspend until continued 3559// - resume: 3560// - sets target osthread state to continue 3561// - sends signal to end the sigsuspend loop in the SR_handler 3562// 3563// Note that the SR_lock plays no role in this suspend/resume protocol. 3564// 3565 3566static void resume_clear_context(OSThread *osthread) { 3567 osthread->set_ucontext(NULL); 3568 osthread->set_siginfo(NULL); 3569 3570 // notify the suspend action is completed, we have now resumed 3571 osthread->sr.clear_suspended(); 3572} 3573 3574static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3575 osthread->set_ucontext(context); 3576 osthread->set_siginfo(siginfo); 3577} 3578 3579// 3580// Handler function invoked when a thread's execution is suspended or 3581// resumed. We have to be careful that only async-safe functions are 3582// called here (Note: most pthread functions are not async safe and 3583// should be avoided.) 3584// 3585// Note: sigwait() is a more natural fit than sigsuspend() from an 3586// interface point of view, but sigwait() prevents the signal hander 3587// from being run. libpthread would get very confused by not having 3588// its signal handlers run and prevents sigwait()'s use with the 3589// mutex granting granting signal. 3590// 3591// Currently only ever called on the VMThread 3592// 3593static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3594 // Save and restore errno to avoid confusing native code with EINTR 3595 // after sigsuspend. 3596 int old_errno = errno; 3597 3598 Thread* thread = Thread::current(); 3599 OSThread* osthread = thread->osthread(); 3600 assert(thread->is_VM_thread(), "Must be VMThread"); 3601 // read current suspend action 3602 int action = osthread->sr.suspend_action(); 3603 if (action == os::Linux::SuspendResume::SR_SUSPEND) { 3604 suspend_save_context(osthread, siginfo, context); 3605 3606 // Notify the suspend action is about to be completed. do_suspend() 3607 // waits until SR_SUSPENDED is set and then returns. We will wait 3608 // here for a resume signal and that completes the suspend-other 3609 // action. do_suspend/do_resume is always called as a pair from 3610 // the same thread - so there are no races 3611 3612 // notify the caller 3613 osthread->sr.set_suspended(); 3614 3615 sigset_t suspend_set; // signals for sigsuspend() 3616 3617 // get current set of blocked signals and unblock resume signal 3618 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3619 sigdelset(&suspend_set, SR_signum); 3620 3621 // wait here until we are resumed 3622 do { 3623 sigsuspend(&suspend_set); 3624 // ignore all returns until we get a resume signal 3625 } while (osthread->sr.suspend_action() != os::Linux::SuspendResume::SR_CONTINUE); 3626 3627 resume_clear_context(osthread); 3628 3629 } else { 3630 assert(action == os::Linux::SuspendResume::SR_CONTINUE, "unexpected sr action"); 3631 // nothing special to do - just leave the handler 3632 } 3633 3634 errno = old_errno; 3635} 3636 3637 3638static int SR_initialize() { 3639 struct sigaction act; 3640 char *s; 3641 /* Get signal number to use for suspend/resume */ 3642 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3643 int sig = ::strtol(s, 0, 10); 3644 if (sig > 0 || sig < _NSIG) { 3645 SR_signum = sig; 3646 } 3647 } 3648 3649 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3650 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3651 3652 sigemptyset(&SR_sigset); 3653 sigaddset(&SR_sigset, SR_signum); 3654 3655 /* Set up signal handler for suspend/resume */ 3656 act.sa_flags = SA_RESTART|SA_SIGINFO; 3657 act.sa_handler = (void (*)(int)) SR_handler; 3658 3659 // SR_signum is blocked by default. 3660 // 4528190 - We also need to block pthread restart signal (32 on all 3661 // supported Linux platforms). Note that LinuxThreads need to block 3662 // this signal for all threads to work properly. So we don't have 3663 // to use hard-coded signal number when setting up the mask. 3664 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 3665 3666 if (sigaction(SR_signum, &act, 0) == -1) { 3667 return -1; 3668 } 3669 3670 // Save signal flag 3671 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 3672 return 0; 3673} 3674 3675static int SR_finalize() { 3676 return 0; 3677} 3678 3679 3680// returns true on success and false on error - really an error is fatal 3681// but this seems the normal response to library errors 3682static bool do_suspend(OSThread* osthread) { 3683 // mark as suspended and send signal 3684 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_SUSPEND); 3685 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3686 assert_status(status == 0, status, "pthread_kill"); 3687 3688 // check status and wait until notified of suspension 3689 if (status == 0) { 3690 for (int i = 0; !osthread->sr.is_suspended(); i++) { 3691 os::yield_all(i); 3692 } 3693 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE); 3694 return true; 3695 } 3696 else { 3697 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE); 3698 return false; 3699 } 3700} 3701 3702static void do_resume(OSThread* osthread) { 3703 assert(osthread->sr.is_suspended(), "thread should be suspended"); 3704 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_CONTINUE); 3705 3706 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3707 assert_status(status == 0, status, "pthread_kill"); 3708 // check status and wait unit notified of resumption 3709 if (status == 0) { 3710 for (int i = 0; osthread->sr.is_suspended(); i++) { 3711 os::yield_all(i); 3712 } 3713 } 3714 osthread->sr.set_suspend_action(os::Linux::SuspendResume::SR_NONE); 3715} 3716 3717//////////////////////////////////////////////////////////////////////////////// 3718// interrupt support 3719 3720void os::interrupt(Thread* thread) { 3721 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3722 "possibility of dangling Thread pointer"); 3723 3724 OSThread* osthread = thread->osthread(); 3725 3726 if (!osthread->interrupted()) { 3727 osthread->set_interrupted(true); 3728 // More than one thread can get here with the same value of osthread, 3729 // resulting in multiple notifications. We do, however, want the store 3730 // to interrupted() to be visible to other threads before we execute unpark(). 3731 OrderAccess::fence(); 3732 ParkEvent * const slp = thread->_SleepEvent ; 3733 if (slp != NULL) slp->unpark() ; 3734 } 3735 3736 // For JSR166. Unpark even if interrupt status already was set 3737 if (thread->is_Java_thread()) 3738 ((JavaThread*)thread)->parker()->unpark(); 3739 3740 ParkEvent * ev = thread->_ParkEvent ; 3741 if (ev != NULL) ev->unpark() ; 3742 3743} 3744 3745bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 3746 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3747 "possibility of dangling Thread pointer"); 3748 3749 OSThread* osthread = thread->osthread(); 3750 3751 bool interrupted = osthread->interrupted(); 3752 3753 if (interrupted && clear_interrupted) { 3754 osthread->set_interrupted(false); 3755 // consider thread->_SleepEvent->reset() ... optional optimization 3756 } 3757 3758 return interrupted; 3759} 3760 3761/////////////////////////////////////////////////////////////////////////////////// 3762// signal handling (except suspend/resume) 3763 3764// This routine may be used by user applications as a "hook" to catch signals. 3765// The user-defined signal handler must pass unrecognized signals to this 3766// routine, and if it returns true (non-zero), then the signal handler must 3767// return immediately. If the flag "abort_if_unrecognized" is true, then this 3768// routine will never retun false (zero), but instead will execute a VM panic 3769// routine kill the process. 3770// 3771// If this routine returns false, it is OK to call it again. This allows 3772// the user-defined signal handler to perform checks either before or after 3773// the VM performs its own checks. Naturally, the user code would be making 3774// a serious error if it tried to handle an exception (such as a null check 3775// or breakpoint) that the VM was generating for its own correct operation. 3776// 3777// This routine may recognize any of the following kinds of signals: 3778// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 3779// It should be consulted by handlers for any of those signals. 3780// 3781// The caller of this routine must pass in the three arguments supplied 3782// to the function referred to in the "sa_sigaction" (not the "sa_handler") 3783// field of the structure passed to sigaction(). This routine assumes that 3784// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 3785// 3786// Note that the VM will print warnings if it detects conflicting signal 3787// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 3788// 3789extern "C" JNIEXPORT int 3790JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 3791 void* ucontext, int abort_if_unrecognized); 3792 3793void signalHandler(int sig, siginfo_t* info, void* uc) { 3794 assert(info != NULL && uc != NULL, "it must be old kernel"); 3795 int orig_errno = errno; // Preserve errno value over signal handler. 3796 JVM_handle_linux_signal(sig, info, uc, true); 3797 errno = orig_errno; 3798} 3799 3800 3801// This boolean allows users to forward their own non-matching signals 3802// to JVM_handle_linux_signal, harmlessly. 3803bool os::Linux::signal_handlers_are_installed = false; 3804 3805// For signal-chaining 3806struct sigaction os::Linux::sigact[MAXSIGNUM]; 3807unsigned int os::Linux::sigs = 0; 3808bool os::Linux::libjsig_is_loaded = false; 3809typedef struct sigaction *(*get_signal_t)(int); 3810get_signal_t os::Linux::get_signal_action = NULL; 3811 3812struct sigaction* os::Linux::get_chained_signal_action(int sig) { 3813 struct sigaction *actp = NULL; 3814 3815 if (libjsig_is_loaded) { 3816 // Retrieve the old signal handler from libjsig 3817 actp = (*get_signal_action)(sig); 3818 } 3819 if (actp == NULL) { 3820 // Retrieve the preinstalled signal handler from jvm 3821 actp = get_preinstalled_handler(sig); 3822 } 3823 3824 return actp; 3825} 3826 3827static bool call_chained_handler(struct sigaction *actp, int sig, 3828 siginfo_t *siginfo, void *context) { 3829 // Call the old signal handler 3830 if (actp->sa_handler == SIG_DFL) { 3831 // It's more reasonable to let jvm treat it as an unexpected exception 3832 // instead of taking the default action. 3833 return false; 3834 } else if (actp->sa_handler != SIG_IGN) { 3835 if ((actp->sa_flags & SA_NODEFER) == 0) { 3836 // automaticlly block the signal 3837 sigaddset(&(actp->sa_mask), sig); 3838 } 3839 3840 sa_handler_t hand; 3841 sa_sigaction_t sa; 3842 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 3843 // retrieve the chained handler 3844 if (siginfo_flag_set) { 3845 sa = actp->sa_sigaction; 3846 } else { 3847 hand = actp->sa_handler; 3848 } 3849 3850 if ((actp->sa_flags & SA_RESETHAND) != 0) { 3851 actp->sa_handler = SIG_DFL; 3852 } 3853 3854 // try to honor the signal mask 3855 sigset_t oset; 3856 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 3857 3858 // call into the chained handler 3859 if (siginfo_flag_set) { 3860 (*sa)(sig, siginfo, context); 3861 } else { 3862 (*hand)(sig); 3863 } 3864 3865 // restore the signal mask 3866 pthread_sigmask(SIG_SETMASK, &oset, 0); 3867 } 3868 // Tell jvm's signal handler the signal is taken care of. 3869 return true; 3870} 3871 3872bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 3873 bool chained = false; 3874 // signal-chaining 3875 if (UseSignalChaining) { 3876 struct sigaction *actp = get_chained_signal_action(sig); 3877 if (actp != NULL) { 3878 chained = call_chained_handler(actp, sig, siginfo, context); 3879 } 3880 } 3881 return chained; 3882} 3883 3884struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 3885 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 3886 return &sigact[sig]; 3887 } 3888 return NULL; 3889} 3890 3891void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 3892 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3893 sigact[sig] = oldAct; 3894 sigs |= (unsigned int)1 << sig; 3895} 3896 3897// for diagnostic 3898int os::Linux::sigflags[MAXSIGNUM]; 3899 3900int os::Linux::get_our_sigflags(int sig) { 3901 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3902 return sigflags[sig]; 3903} 3904 3905void os::Linux::set_our_sigflags(int sig, int flags) { 3906 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3907 sigflags[sig] = flags; 3908} 3909 3910void os::Linux::set_signal_handler(int sig, bool set_installed) { 3911 // Check for overwrite. 3912 struct sigaction oldAct; 3913 sigaction(sig, (struct sigaction*)NULL, &oldAct); 3914 3915 void* oldhand = oldAct.sa_sigaction 3916 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3917 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3918 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 3919 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 3920 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 3921 if (AllowUserSignalHandlers || !set_installed) { 3922 // Do not overwrite; user takes responsibility to forward to us. 3923 return; 3924 } else if (UseSignalChaining) { 3925 // save the old handler in jvm 3926 save_preinstalled_handler(sig, oldAct); 3927 // libjsig also interposes the sigaction() call below and saves the 3928 // old sigaction on it own. 3929 } else { 3930 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 3931 "%#lx for signal %d.", (long)oldhand, sig)); 3932 } 3933 } 3934 3935 struct sigaction sigAct; 3936 sigfillset(&(sigAct.sa_mask)); 3937 sigAct.sa_handler = SIG_DFL; 3938 if (!set_installed) { 3939 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3940 } else { 3941 sigAct.sa_sigaction = signalHandler; 3942 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3943 } 3944 // Save flags, which are set by ours 3945 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3946 sigflags[sig] = sigAct.sa_flags; 3947 3948 int ret = sigaction(sig, &sigAct, &oldAct); 3949 assert(ret == 0, "check"); 3950 3951 void* oldhand2 = oldAct.sa_sigaction 3952 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3953 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3954 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 3955} 3956 3957// install signal handlers for signals that HotSpot needs to 3958// handle in order to support Java-level exception handling. 3959 3960void os::Linux::install_signal_handlers() { 3961 if (!signal_handlers_are_installed) { 3962 signal_handlers_are_installed = true; 3963 3964 // signal-chaining 3965 typedef void (*signal_setting_t)(); 3966 signal_setting_t begin_signal_setting = NULL; 3967 signal_setting_t end_signal_setting = NULL; 3968 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3969 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 3970 if (begin_signal_setting != NULL) { 3971 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3972 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 3973 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 3974 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 3975 libjsig_is_loaded = true; 3976 assert(UseSignalChaining, "should enable signal-chaining"); 3977 } 3978 if (libjsig_is_loaded) { 3979 // Tell libjsig jvm is setting signal handlers 3980 (*begin_signal_setting)(); 3981 } 3982 3983 set_signal_handler(SIGSEGV, true); 3984 set_signal_handler(SIGPIPE, true); 3985 set_signal_handler(SIGBUS, true); 3986 set_signal_handler(SIGILL, true); 3987 set_signal_handler(SIGFPE, true); 3988 set_signal_handler(SIGXFSZ, true); 3989 3990 if (libjsig_is_loaded) { 3991 // Tell libjsig jvm finishes setting signal handlers 3992 (*end_signal_setting)(); 3993 } 3994 3995 // We don't activate signal checker if libjsig is in place, we trust ourselves 3996 // and if UserSignalHandler is installed all bets are off. 3997 // Log that signal checking is off only if -verbose:jni is specified. 3998 if (CheckJNICalls) { 3999 if (libjsig_is_loaded) { 4000 if (PrintJNIResolving) { 4001 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4002 } 4003 check_signals = false; 4004 } 4005 if (AllowUserSignalHandlers) { 4006 if (PrintJNIResolving) { 4007 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4008 } 4009 check_signals = false; 4010 } 4011 } 4012 } 4013} 4014 4015// This is the fastest way to get thread cpu time on Linux. 4016// Returns cpu time (user+sys) for any thread, not only for current. 4017// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4018// It might work on 2.6.10+ with a special kernel/glibc patch. 4019// For reference, please, see IEEE Std 1003.1-2004: 4020// http://www.unix.org/single_unix_specification 4021 4022jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4023 struct timespec tp; 4024 int rc = os::Linux::clock_gettime(clockid, &tp); 4025 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4026 4027 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4028} 4029 4030///// 4031// glibc on Linux platform uses non-documented flag 4032// to indicate, that some special sort of signal 4033// trampoline is used. 4034// We will never set this flag, and we should 4035// ignore this flag in our diagnostic 4036#ifdef SIGNIFICANT_SIGNAL_MASK 4037#undef SIGNIFICANT_SIGNAL_MASK 4038#endif 4039#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4040 4041static const char* get_signal_handler_name(address handler, 4042 char* buf, int buflen) { 4043 int offset; 4044 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4045 if (found) { 4046 // skip directory names 4047 const char *p1, *p2; 4048 p1 = buf; 4049 size_t len = strlen(os::file_separator()); 4050 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4051 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4052 } else { 4053 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4054 } 4055 return buf; 4056} 4057 4058static void print_signal_handler(outputStream* st, int sig, 4059 char* buf, size_t buflen) { 4060 struct sigaction sa; 4061 4062 sigaction(sig, NULL, &sa); 4063 4064 // See comment for SIGNIFICANT_SIGNAL_MASK define 4065 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4066 4067 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4068 4069 address handler = (sa.sa_flags & SA_SIGINFO) 4070 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4071 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4072 4073 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4074 st->print("SIG_DFL"); 4075 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4076 st->print("SIG_IGN"); 4077 } else { 4078 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4079 } 4080 4081 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 4082 4083 address rh = VMError::get_resetted_sighandler(sig); 4084 // May be, handler was resetted by VMError? 4085 if(rh != NULL) { 4086 handler = rh; 4087 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4088 } 4089 4090 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 4091 4092 // Check: is it our handler? 4093 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4094 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4095 // It is our signal handler 4096 // check for flags, reset system-used one! 4097 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4098 st->print( 4099 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4100 os::Linux::get_our_sigflags(sig)); 4101 } 4102 } 4103 st->cr(); 4104} 4105 4106 4107#define DO_SIGNAL_CHECK(sig) \ 4108 if (!sigismember(&check_signal_done, sig)) \ 4109 os::Linux::check_signal_handler(sig) 4110 4111// This method is a periodic task to check for misbehaving JNI applications 4112// under CheckJNI, we can add any periodic checks here 4113 4114void os::run_periodic_checks() { 4115 4116 if (check_signals == false) return; 4117 4118 // SEGV and BUS if overridden could potentially prevent 4119 // generation of hs*.log in the event of a crash, debugging 4120 // such a case can be very challenging, so we absolutely 4121 // check the following for a good measure: 4122 DO_SIGNAL_CHECK(SIGSEGV); 4123 DO_SIGNAL_CHECK(SIGILL); 4124 DO_SIGNAL_CHECK(SIGFPE); 4125 DO_SIGNAL_CHECK(SIGBUS); 4126 DO_SIGNAL_CHECK(SIGPIPE); 4127 DO_SIGNAL_CHECK(SIGXFSZ); 4128 4129 4130 // ReduceSignalUsage allows the user to override these handlers 4131 // see comments at the very top and jvm_solaris.h 4132 if (!ReduceSignalUsage) { 4133 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4134 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4135 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4136 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4137 } 4138 4139 DO_SIGNAL_CHECK(SR_signum); 4140 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4141} 4142 4143typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4144 4145static os_sigaction_t os_sigaction = NULL; 4146 4147void os::Linux::check_signal_handler(int sig) { 4148 char buf[O_BUFLEN]; 4149 address jvmHandler = NULL; 4150 4151 4152 struct sigaction act; 4153 if (os_sigaction == NULL) { 4154 // only trust the default sigaction, in case it has been interposed 4155 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4156 if (os_sigaction == NULL) return; 4157 } 4158 4159 os_sigaction(sig, (struct sigaction*)NULL, &act); 4160 4161 4162 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4163 4164 address thisHandler = (act.sa_flags & SA_SIGINFO) 4165 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4166 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 4167 4168 4169 switch(sig) { 4170 case SIGSEGV: 4171 case SIGBUS: 4172 case SIGFPE: 4173 case SIGPIPE: 4174 case SIGILL: 4175 case SIGXFSZ: 4176 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4177 break; 4178 4179 case SHUTDOWN1_SIGNAL: 4180 case SHUTDOWN2_SIGNAL: 4181 case SHUTDOWN3_SIGNAL: 4182 case BREAK_SIGNAL: 4183 jvmHandler = (address)user_handler(); 4184 break; 4185 4186 case INTERRUPT_SIGNAL: 4187 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4188 break; 4189 4190 default: 4191 if (sig == SR_signum) { 4192 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4193 } else { 4194 return; 4195 } 4196 break; 4197 } 4198 4199 if (thisHandler != jvmHandler) { 4200 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4201 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4202 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4203 // No need to check this sig any longer 4204 sigaddset(&check_signal_done, sig); 4205 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4206 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4207 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4208 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4209 // No need to check this sig any longer 4210 sigaddset(&check_signal_done, sig); 4211 } 4212 4213 // Dump all the signal 4214 if (sigismember(&check_signal_done, sig)) { 4215 print_signal_handlers(tty, buf, O_BUFLEN); 4216 } 4217} 4218 4219extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 4220 4221extern bool signal_name(int signo, char* buf, size_t len); 4222 4223const char* os::exception_name(int exception_code, char* buf, size_t size) { 4224 if (0 < exception_code && exception_code <= SIGRTMAX) { 4225 // signal 4226 if (!signal_name(exception_code, buf, size)) { 4227 jio_snprintf(buf, size, "SIG%d", exception_code); 4228 } 4229 return buf; 4230 } else { 4231 return NULL; 4232 } 4233} 4234 4235// this is called _before_ the most of global arguments have been parsed 4236void os::init(void) { 4237 char dummy; /* used to get a guess on initial stack address */ 4238// first_hrtime = gethrtime(); 4239 4240 // With LinuxThreads the JavaMain thread pid (primordial thread) 4241 // is different than the pid of the java launcher thread. 4242 // So, on Linux, the launcher thread pid is passed to the VM 4243 // via the sun.java.launcher.pid property. 4244 // Use this property instead of getpid() if it was correctly passed. 4245 // See bug 6351349. 4246 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4247 4248 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4249 4250 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4251 4252 init_random(1234567); 4253 4254 ThreadCritical::initialize(); 4255 4256 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4257 if (Linux::page_size() == -1) { 4258 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4259 strerror(errno))); 4260 } 4261 init_page_sizes((size_t) Linux::page_size()); 4262 4263 Linux::initialize_system_info(); 4264 4265 // main_thread points to the aboriginal thread 4266 Linux::_main_thread = pthread_self(); 4267 4268 Linux::clock_init(); 4269 initial_time_count = os::elapsed_counter(); 4270 pthread_mutex_init(&dl_mutex, NULL); 4271} 4272 4273// To install functions for atexit system call 4274extern "C" { 4275 static void perfMemory_exit_helper() { 4276 perfMemory_exit(); 4277 } 4278} 4279 4280// this is called _after_ the global arguments have been parsed 4281jint os::init_2(void) 4282{ 4283 Linux::fast_thread_clock_init(); 4284 4285 // Allocate a single page and mark it as readable for safepoint polling 4286 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4287 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 4288 4289 os::set_polling_page( polling_page ); 4290 4291#ifndef PRODUCT 4292 if(Verbose && PrintMiscellaneous) 4293 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 4294#endif 4295 4296 if (!UseMembar) { 4297 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4298 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); 4299 os::set_memory_serialize_page( mem_serialize_page ); 4300 4301#ifndef PRODUCT 4302 if(Verbose && PrintMiscellaneous) 4303 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 4304#endif 4305 } 4306 4307 os::large_page_init(); 4308 4309 // initialize suspend/resume support - must do this before signal_sets_init() 4310 if (SR_initialize() != 0) { 4311 perror("SR_initialize failed"); 4312 return JNI_ERR; 4313 } 4314 4315 Linux::signal_sets_init(); 4316 Linux::install_signal_handlers(); 4317 4318 // Check minimum allowable stack size for thread creation and to initialize 4319 // the java system classes, including StackOverflowError - depends on page 4320 // size. Add a page for compiler2 recursion in main thread. 4321 // Add in 2*BytesPerWord times page size to account for VM stack during 4322 // class initialization depending on 32 or 64 bit VM. 4323 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4324 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+ 4325 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size()); 4326 4327 size_t threadStackSizeInBytes = ThreadStackSize * K; 4328 if (threadStackSizeInBytes != 0 && 4329 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4330 tty->print_cr("\nThe stack size specified is too small, " 4331 "Specify at least %dk", 4332 os::Linux::min_stack_allowed/ K); 4333 return JNI_ERR; 4334 } 4335 4336 // Make the stack size a multiple of the page size so that 4337 // the yellow/red zones can be guarded. 4338 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4339 vm_page_size())); 4340 4341 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4342 4343 Linux::libpthread_init(); 4344 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4345 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4346 Linux::glibc_version(), Linux::libpthread_version(), 4347 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4348 } 4349 4350 if (UseNUMA) { 4351 if (!Linux::libnuma_init()) { 4352 UseNUMA = false; 4353 } else { 4354 if ((Linux::numa_max_node() < 1)) { 4355 // There's only one node(they start from 0), disable NUMA. 4356 UseNUMA = false; 4357 } 4358 } 4359 // With SHM large pages we cannot uncommit a page, so there's not way 4360 // we can make the adaptive lgrp chunk resizing work. If the user specified 4361 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and 4362 // disable adaptive resizing. 4363 if (UseNUMA && UseLargePages && UseSHM) { 4364 if (!FLAG_IS_DEFAULT(UseNUMA)) { 4365 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) { 4366 UseLargePages = false; 4367 } else { 4368 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing"); 4369 UseAdaptiveSizePolicy = false; 4370 UseAdaptiveNUMAChunkSizing = false; 4371 } 4372 } else { 4373 UseNUMA = false; 4374 } 4375 } 4376 if (!UseNUMA && ForceNUMA) { 4377 UseNUMA = true; 4378 } 4379 } 4380 4381 if (MaxFDLimit) { 4382 // set the number of file descriptors to max. print out error 4383 // if getrlimit/setrlimit fails but continue regardless. 4384 struct rlimit nbr_files; 4385 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4386 if (status != 0) { 4387 if (PrintMiscellaneous && (Verbose || WizardMode)) 4388 perror("os::init_2 getrlimit failed"); 4389 } else { 4390 nbr_files.rlim_cur = nbr_files.rlim_max; 4391 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4392 if (status != 0) { 4393 if (PrintMiscellaneous && (Verbose || WizardMode)) 4394 perror("os::init_2 setrlimit failed"); 4395 } 4396 } 4397 } 4398 4399 // Initialize lock used to serialize thread creation (see os::create_thread) 4400 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4401 4402 // at-exit methods are called in the reverse order of their registration. 4403 // atexit functions are called on return from main or as a result of a 4404 // call to exit(3C). There can be only 32 of these functions registered 4405 // and atexit() does not set errno. 4406 4407 if (PerfAllowAtExitRegistration) { 4408 // only register atexit functions if PerfAllowAtExitRegistration is set. 4409 // atexit functions can be delayed until process exit time, which 4410 // can be problematic for embedded VM situations. Embedded VMs should 4411 // call DestroyJavaVM() to assure that VM resources are released. 4412 4413 // note: perfMemory_exit_helper atexit function may be removed in 4414 // the future if the appropriate cleanup code can be added to the 4415 // VM_Exit VMOperation's doit method. 4416 if (atexit(perfMemory_exit_helper) != 0) { 4417 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 4418 } 4419 } 4420 4421 // initialize thread priority policy 4422 prio_init(); 4423 4424 return JNI_OK; 4425} 4426 4427// this is called at the end of vm_initialization 4428void os::init_3(void) 4429{ 4430#ifdef JAVASE_EMBEDDED 4431 // Start the MemNotifyThread 4432 if (LowMemoryProtection) { 4433 MemNotifyThread::start(); 4434 } 4435 return; 4436#endif 4437} 4438 4439// Mark the polling page as unreadable 4440void os::make_polling_page_unreadable(void) { 4441 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4442 fatal("Could not disable polling page"); 4443}; 4444 4445// Mark the polling page as readable 4446void os::make_polling_page_readable(void) { 4447 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4448 fatal("Could not enable polling page"); 4449 } 4450}; 4451 4452int os::active_processor_count() { 4453 // Linux doesn't yet have a (official) notion of processor sets, 4454 // so just return the number of online processors. 4455 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4456 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4457 return online_cpus; 4458} 4459 4460void os::set_native_thread_name(const char *name) { 4461 // Not yet implemented. 4462 return; 4463} 4464 4465bool os::distribute_processes(uint length, uint* distribution) { 4466 // Not yet implemented. 4467 return false; 4468} 4469 4470bool os::bind_to_processor(uint processor_id) { 4471 // Not yet implemented. 4472 return false; 4473} 4474 4475/// 4476 4477// Suspends the target using the signal mechanism and then grabs the PC before 4478// resuming the target. Used by the flat-profiler only 4479ExtendedPC os::get_thread_pc(Thread* thread) { 4480 // Make sure that it is called by the watcher for the VMThread 4481 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4482 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4483 4484 ExtendedPC epc; 4485 4486 OSThread* osthread = thread->osthread(); 4487 if (do_suspend(osthread)) { 4488 if (osthread->ucontext() != NULL) { 4489 epc = os::Linux::ucontext_get_pc(osthread->ucontext()); 4490 } else { 4491 // NULL context is unexpected, double-check this is the VMThread 4492 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4493 } 4494 do_resume(osthread); 4495 } 4496 // failure means pthread_kill failed for some reason - arguably this is 4497 // a fatal problem, but such problems are ignored elsewhere 4498 4499 return epc; 4500} 4501 4502int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4503{ 4504 if (is_NPTL()) { 4505 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4506 } else { 4507 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4508 // word back to default 64bit precision if condvar is signaled. Java 4509 // wants 53bit precision. Save and restore current value. 4510 int fpu = get_fpu_control_word(); 4511 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4512 set_fpu_control_word(fpu); 4513 return status; 4514 } 4515} 4516 4517//////////////////////////////////////////////////////////////////////////////// 4518// debug support 4519 4520static address same_page(address x, address y) { 4521 int page_bits = -os::vm_page_size(); 4522 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) 4523 return x; 4524 else if (x > y) 4525 return (address)(intptr_t(y) | ~page_bits) + 1; 4526 else 4527 return (address)(intptr_t(y) & page_bits); 4528} 4529 4530bool os::find(address addr, outputStream* st) { 4531 Dl_info dlinfo; 4532 memset(&dlinfo, 0, sizeof(dlinfo)); 4533 if (dladdr(addr, &dlinfo)) { 4534 st->print(PTR_FORMAT ": ", addr); 4535 if (dlinfo.dli_sname != NULL) { 4536 st->print("%s+%#x", dlinfo.dli_sname, 4537 addr - (intptr_t)dlinfo.dli_saddr); 4538 } else if (dlinfo.dli_fname) { 4539 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4540 } else { 4541 st->print("<absolute address>"); 4542 } 4543 if (dlinfo.dli_fname) { 4544 st->print(" in %s", dlinfo.dli_fname); 4545 } 4546 if (dlinfo.dli_fbase) { 4547 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4548 } 4549 st->cr(); 4550 4551 if (Verbose) { 4552 // decode some bytes around the PC 4553 address begin = same_page(addr-40, addr); 4554 address end = same_page(addr+40, addr); 4555 address lowest = (address) dlinfo.dli_sname; 4556 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4557 if (begin < lowest) begin = lowest; 4558 Dl_info dlinfo2; 4559 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr 4560 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 4561 end = (address) dlinfo2.dli_saddr; 4562 Disassembler::decode(begin, end, st); 4563 } 4564 return true; 4565 } 4566 return false; 4567} 4568 4569//////////////////////////////////////////////////////////////////////////////// 4570// misc 4571 4572// This does not do anything on Linux. This is basically a hook for being 4573// able to use structured exception handling (thread-local exception filters) 4574// on, e.g., Win32. 4575void 4576os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4577 JavaCallArguments* args, Thread* thread) { 4578 f(value, method, args, thread); 4579} 4580 4581void os::print_statistics() { 4582} 4583 4584int os::message_box(const char* title, const char* message) { 4585 int i; 4586 fdStream err(defaultStream::error_fd()); 4587 for (i = 0; i < 78; i++) err.print_raw("="); 4588 err.cr(); 4589 err.print_raw_cr(title); 4590 for (i = 0; i < 78; i++) err.print_raw("-"); 4591 err.cr(); 4592 err.print_raw_cr(message); 4593 for (i = 0; i < 78; i++) err.print_raw("="); 4594 err.cr(); 4595 4596 char buf[16]; 4597 // Prevent process from exiting upon "read error" without consuming all CPU 4598 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 4599 4600 return buf[0] == 'y' || buf[0] == 'Y'; 4601} 4602 4603int os::stat(const char *path, struct stat *sbuf) { 4604 char pathbuf[MAX_PATH]; 4605 if (strlen(path) > MAX_PATH - 1) { 4606 errno = ENAMETOOLONG; 4607 return -1; 4608 } 4609 os::native_path(strcpy(pathbuf, path)); 4610 return ::stat(pathbuf, sbuf); 4611} 4612 4613bool os::check_heap(bool force) { 4614 return true; 4615} 4616 4617int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 4618 return ::vsnprintf(buf, count, format, args); 4619} 4620 4621// Is a (classpath) directory empty? 4622bool os::dir_is_empty(const char* path) { 4623 DIR *dir = NULL; 4624 struct dirent *ptr; 4625 4626 dir = opendir(path); 4627 if (dir == NULL) return true; 4628 4629 /* Scan the directory */ 4630 bool result = true; 4631 char buf[sizeof(struct dirent) + MAX_PATH]; 4632 while (result && (ptr = ::readdir(dir)) != NULL) { 4633 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 4634 result = false; 4635 } 4636 } 4637 closedir(dir); 4638 return result; 4639} 4640 4641// This code originates from JDK's sysOpen and open64_w 4642// from src/solaris/hpi/src/system_md.c 4643 4644#ifndef O_DELETE 4645#define O_DELETE 0x10000 4646#endif 4647 4648// Open a file. Unlink the file immediately after open returns 4649// if the specified oflag has the O_DELETE flag set. 4650// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 4651 4652int os::open(const char *path, int oflag, int mode) { 4653 4654 if (strlen(path) > MAX_PATH - 1) { 4655 errno = ENAMETOOLONG; 4656 return -1; 4657 } 4658 int fd; 4659 int o_delete = (oflag & O_DELETE); 4660 oflag = oflag & ~O_DELETE; 4661 4662 fd = ::open64(path, oflag, mode); 4663 if (fd == -1) return -1; 4664 4665 //If the open succeeded, the file might still be a directory 4666 { 4667 struct stat64 buf64; 4668 int ret = ::fstat64(fd, &buf64); 4669 int st_mode = buf64.st_mode; 4670 4671 if (ret != -1) { 4672 if ((st_mode & S_IFMT) == S_IFDIR) { 4673 errno = EISDIR; 4674 ::close(fd); 4675 return -1; 4676 } 4677 } else { 4678 ::close(fd); 4679 return -1; 4680 } 4681 } 4682 4683 /* 4684 * All file descriptors that are opened in the JVM and not 4685 * specifically destined for a subprocess should have the 4686 * close-on-exec flag set. If we don't set it, then careless 3rd 4687 * party native code might fork and exec without closing all 4688 * appropriate file descriptors (e.g. as we do in closeDescriptors in 4689 * UNIXProcess.c), and this in turn might: 4690 * 4691 * - cause end-of-file to fail to be detected on some file 4692 * descriptors, resulting in mysterious hangs, or 4693 * 4694 * - might cause an fopen in the subprocess to fail on a system 4695 * suffering from bug 1085341. 4696 * 4697 * (Yes, the default setting of the close-on-exec flag is a Unix 4698 * design flaw) 4699 * 4700 * See: 4701 * 1085341: 32-bit stdio routines should support file descriptors >255 4702 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 4703 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 4704 */ 4705#ifdef FD_CLOEXEC 4706 { 4707 int flags = ::fcntl(fd, F_GETFD); 4708 if (flags != -1) 4709 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 4710 } 4711#endif 4712 4713 if (o_delete != 0) { 4714 ::unlink(path); 4715 } 4716 return fd; 4717} 4718 4719 4720// create binary file, rewriting existing file if required 4721int os::create_binary_file(const char* path, bool rewrite_existing) { 4722 int oflags = O_WRONLY | O_CREAT; 4723 if (!rewrite_existing) { 4724 oflags |= O_EXCL; 4725 } 4726 return ::open64(path, oflags, S_IREAD | S_IWRITE); 4727} 4728 4729// return current position of file pointer 4730jlong os::current_file_offset(int fd) { 4731 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 4732} 4733 4734// move file pointer to the specified offset 4735jlong os::seek_to_file_offset(int fd, jlong offset) { 4736 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 4737} 4738 4739// This code originates from JDK's sysAvailable 4740// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 4741 4742int os::available(int fd, jlong *bytes) { 4743 jlong cur, end; 4744 int mode; 4745 struct stat64 buf64; 4746 4747 if (::fstat64(fd, &buf64) >= 0) { 4748 mode = buf64.st_mode; 4749 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 4750 /* 4751 * XXX: is the following call interruptible? If so, this might 4752 * need to go through the INTERRUPT_IO() wrapper as for other 4753 * blocking, interruptible calls in this file. 4754 */ 4755 int n; 4756 if (::ioctl(fd, FIONREAD, &n) >= 0) { 4757 *bytes = n; 4758 return 1; 4759 } 4760 } 4761 } 4762 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 4763 return 0; 4764 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 4765 return 0; 4766 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 4767 return 0; 4768 } 4769 *bytes = end - cur; 4770 return 1; 4771} 4772 4773int os::socket_available(int fd, jint *pbytes) { 4774 // Linux doc says EINTR not returned, unlike Solaris 4775 int ret = ::ioctl(fd, FIONREAD, pbytes); 4776 4777 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 4778 // is expected to return 0 on failure and 1 on success to the jdk. 4779 return (ret < 0) ? 0 : 1; 4780} 4781 4782// Map a block of memory. 4783char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 4784 char *addr, size_t bytes, bool read_only, 4785 bool allow_exec) { 4786 int prot; 4787 int flags = MAP_PRIVATE; 4788 4789 if (read_only) { 4790 prot = PROT_READ; 4791 } else { 4792 prot = PROT_READ | PROT_WRITE; 4793 } 4794 4795 if (allow_exec) { 4796 prot |= PROT_EXEC; 4797 } 4798 4799 if (addr != NULL) { 4800 flags |= MAP_FIXED; 4801 } 4802 4803 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 4804 fd, file_offset); 4805 if (mapped_address == MAP_FAILED) { 4806 return NULL; 4807 } 4808 return mapped_address; 4809} 4810 4811 4812// Remap a block of memory. 4813char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 4814 char *addr, size_t bytes, bool read_only, 4815 bool allow_exec) { 4816 // same as map_memory() on this OS 4817 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 4818 allow_exec); 4819} 4820 4821 4822// Unmap a block of memory. 4823bool os::pd_unmap_memory(char* addr, size_t bytes) { 4824 return munmap(addr, bytes) == 0; 4825} 4826 4827static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 4828 4829static clockid_t thread_cpu_clockid(Thread* thread) { 4830 pthread_t tid = thread->osthread()->pthread_id(); 4831 clockid_t clockid; 4832 4833 // Get thread clockid 4834 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 4835 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 4836 return clockid; 4837} 4838 4839// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 4840// are used by JVM M&M and JVMTI to get user+sys or user CPU time 4841// of a thread. 4842// 4843// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 4844// the fast estimate available on the platform. 4845 4846jlong os::current_thread_cpu_time() { 4847 if (os::Linux::supports_fast_thread_cpu_time()) { 4848 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4849 } else { 4850 // return user + sys since the cost is the same 4851 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 4852 } 4853} 4854 4855jlong os::thread_cpu_time(Thread* thread) { 4856 // consistent with what current_thread_cpu_time() returns 4857 if (os::Linux::supports_fast_thread_cpu_time()) { 4858 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4859 } else { 4860 return slow_thread_cpu_time(thread, true /* user + sys */); 4861 } 4862} 4863 4864jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 4865 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4866 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4867 } else { 4868 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 4869 } 4870} 4871 4872jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4873 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4874 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4875 } else { 4876 return slow_thread_cpu_time(thread, user_sys_cpu_time); 4877 } 4878} 4879 4880// 4881// -1 on error. 4882// 4883 4884static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4885 static bool proc_task_unchecked = true; 4886 static const char *proc_stat_path = "/proc/%d/stat"; 4887 pid_t tid = thread->osthread()->thread_id(); 4888 char *s; 4889 char stat[2048]; 4890 int statlen; 4891 char proc_name[64]; 4892 int count; 4893 long sys_time, user_time; 4894 char cdummy; 4895 int idummy; 4896 long ldummy; 4897 FILE *fp; 4898 4899 // The /proc/<tid>/stat aggregates per-process usage on 4900 // new Linux kernels 2.6+ where NPTL is supported. 4901 // The /proc/self/task/<tid>/stat still has the per-thread usage. 4902 // See bug 6328462. 4903 // There possibly can be cases where there is no directory 4904 // /proc/self/task, so we check its availability. 4905 if (proc_task_unchecked && os::Linux::is_NPTL()) { 4906 // This is executed only once 4907 proc_task_unchecked = false; 4908 fp = fopen("/proc/self/task", "r"); 4909 if (fp != NULL) { 4910 proc_stat_path = "/proc/self/task/%d/stat"; 4911 fclose(fp); 4912 } 4913 } 4914 4915 sprintf(proc_name, proc_stat_path, tid); 4916 fp = fopen(proc_name, "r"); 4917 if ( fp == NULL ) return -1; 4918 statlen = fread(stat, 1, 2047, fp); 4919 stat[statlen] = '\0'; 4920 fclose(fp); 4921 4922 // Skip pid and the command string. Note that we could be dealing with 4923 // weird command names, e.g. user could decide to rename java launcher 4924 // to "java 1.4.2 :)", then the stat file would look like 4925 // 1234 (java 1.4.2 :)) R ... ... 4926 // We don't really need to know the command string, just find the last 4927 // occurrence of ")" and then start parsing from there. See bug 4726580. 4928 s = strrchr(stat, ')'); 4929 if (s == NULL ) return -1; 4930 4931 // Skip blank chars 4932 do s++; while (isspace(*s)); 4933 4934 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 4935 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 4936 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 4937 &user_time, &sys_time); 4938 if ( count != 13 ) return -1; 4939 if (user_sys_cpu_time) { 4940 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4941 } else { 4942 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4943 } 4944} 4945 4946void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4947 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4948 info_ptr->may_skip_backward = false; // elapsed time not wall time 4949 info_ptr->may_skip_forward = false; // elapsed time not wall time 4950 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4951} 4952 4953void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4954 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4955 info_ptr->may_skip_backward = false; // elapsed time not wall time 4956 info_ptr->may_skip_forward = false; // elapsed time not wall time 4957 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4958} 4959 4960bool os::is_thread_cpu_time_supported() { 4961 return true; 4962} 4963 4964// System loadavg support. Returns -1 if load average cannot be obtained. 4965// Linux doesn't yet have a (official) notion of processor sets, 4966// so just return the system wide load average. 4967int os::loadavg(double loadavg[], int nelem) { 4968 return ::getloadavg(loadavg, nelem); 4969} 4970 4971void os::pause() { 4972 char filename[MAX_PATH]; 4973 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 4974 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 4975 } else { 4976 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 4977 } 4978 4979 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 4980 if (fd != -1) { 4981 struct stat buf; 4982 ::close(fd); 4983 while (::stat(filename, &buf) == 0) { 4984 (void)::poll(NULL, 0, 100); 4985 } 4986 } else { 4987 jio_fprintf(stderr, 4988 "Could not open pause file '%s', continuing immediately.\n", filename); 4989 } 4990} 4991 4992 4993// Refer to the comments in os_solaris.cpp park-unpark. 4994// 4995// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 4996// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 4997// For specifics regarding the bug see GLIBC BUGID 261237 : 4998// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 4999// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5000// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5001// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5002// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5003// and monitorenter when we're using 1-0 locking. All those operations may result in 5004// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5005// of libpthread avoids the problem, but isn't practical. 5006// 5007// Possible remedies: 5008// 5009// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5010// This is palliative and probabilistic, however. If the thread is preempted 5011// between the call to compute_abstime() and pthread_cond_timedwait(), more 5012// than the minimum period may have passed, and the abstime may be stale (in the 5013// past) resultin in a hang. Using this technique reduces the odds of a hang 5014// but the JVM is still vulnerable, particularly on heavily loaded systems. 5015// 5016// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5017// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5018// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5019// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5020// thread. 5021// 5022// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5023// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5024// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5025// This also works well. In fact it avoids kernel-level scalability impediments 5026// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5027// timers in a graceful fashion. 5028// 5029// 4. When the abstime value is in the past it appears that control returns 5030// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5031// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5032// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5033// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5034// It may be possible to avoid reinitialization by checking the return 5035// value from pthread_cond_timedwait(). In addition to reinitializing the 5036// condvar we must establish the invariant that cond_signal() is only called 5037// within critical sections protected by the adjunct mutex. This prevents 5038// cond_signal() from "seeing" a condvar that's in the midst of being 5039// reinitialized or that is corrupt. Sadly, this invariant obviates the 5040// desirable signal-after-unlock optimization that avoids futile context switching. 5041// 5042// I'm also concerned that some versions of NTPL might allocate an auxilliary 5043// structure when a condvar is used or initialized. cond_destroy() would 5044// release the helper structure. Our reinitialize-after-timedwait fix 5045// put excessive stress on malloc/free and locks protecting the c-heap. 5046// 5047// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5048// It may be possible to refine (4) by checking the kernel and NTPL verisons 5049// and only enabling the work-around for vulnerable environments. 5050 5051// utility to compute the abstime argument to timedwait: 5052// millis is the relative timeout time 5053// abstime will be the absolute timeout time 5054// TODO: replace compute_abstime() with unpackTime() 5055 5056static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5057 if (millis < 0) millis = 0; 5058 struct timeval now; 5059 int status = gettimeofday(&now, NULL); 5060 assert(status == 0, "gettimeofday"); 5061 jlong seconds = millis / 1000; 5062 millis %= 1000; 5063 if (seconds > 50000000) { // see man cond_timedwait(3T) 5064 seconds = 50000000; 5065 } 5066 abstime->tv_sec = now.tv_sec + seconds; 5067 long usec = now.tv_usec + millis * 1000; 5068 if (usec >= 1000000) { 5069 abstime->tv_sec += 1; 5070 usec -= 1000000; 5071 } 5072 abstime->tv_nsec = usec * 1000; 5073 return abstime; 5074} 5075 5076 5077// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 5078// Conceptually TryPark() should be equivalent to park(0). 5079 5080int os::PlatformEvent::TryPark() { 5081 for (;;) { 5082 const int v = _Event ; 5083 guarantee ((v == 0) || (v == 1), "invariant") ; 5084 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 5085 } 5086} 5087 5088void os::PlatformEvent::park() { // AKA "down()" 5089 // Invariant: Only the thread associated with the Event/PlatformEvent 5090 // may call park(). 5091 // TODO: assert that _Assoc != NULL or _Assoc == Self 5092 int v ; 5093 for (;;) { 5094 v = _Event ; 5095 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5096 } 5097 guarantee (v >= 0, "invariant") ; 5098 if (v == 0) { 5099 // Do this the hard way by blocking ... 5100 int status = pthread_mutex_lock(_mutex); 5101 assert_status(status == 0, status, "mutex_lock"); 5102 guarantee (_nParked == 0, "invariant") ; 5103 ++ _nParked ; 5104 while (_Event < 0) { 5105 status = pthread_cond_wait(_cond, _mutex); 5106 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5107 // Treat this the same as if the wait was interrupted 5108 if (status == ETIME) { status = EINTR; } 5109 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5110 } 5111 -- _nParked ; 5112 5113 _Event = 0 ; 5114 status = pthread_mutex_unlock(_mutex); 5115 assert_status(status == 0, status, "mutex_unlock"); 5116 // Paranoia to ensure our locked and lock-free paths interact 5117 // correctly with each other. 5118 OrderAccess::fence(); 5119 } 5120 guarantee (_Event >= 0, "invariant") ; 5121} 5122 5123int os::PlatformEvent::park(jlong millis) { 5124 guarantee (_nParked == 0, "invariant") ; 5125 5126 int v ; 5127 for (;;) { 5128 v = _Event ; 5129 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5130 } 5131 guarantee (v >= 0, "invariant") ; 5132 if (v != 0) return OS_OK ; 5133 5134 // We do this the hard way, by blocking the thread. 5135 // Consider enforcing a minimum timeout value. 5136 struct timespec abst; 5137 compute_abstime(&abst, millis); 5138 5139 int ret = OS_TIMEOUT; 5140 int status = pthread_mutex_lock(_mutex); 5141 assert_status(status == 0, status, "mutex_lock"); 5142 guarantee (_nParked == 0, "invariant") ; 5143 ++_nParked ; 5144 5145 // Object.wait(timo) will return because of 5146 // (a) notification 5147 // (b) timeout 5148 // (c) thread.interrupt 5149 // 5150 // Thread.interrupt and object.notify{All} both call Event::set. 5151 // That is, we treat thread.interrupt as a special case of notification. 5152 // The underlying Solaris implementation, cond_timedwait, admits 5153 // spurious/premature wakeups, but the JLS/JVM spec prevents the 5154 // JVM from making those visible to Java code. As such, we must 5155 // filter out spurious wakeups. We assume all ETIME returns are valid. 5156 // 5157 // TODO: properly differentiate simultaneous notify+interrupt. 5158 // In that case, we should propagate the notify to another waiter. 5159 5160 while (_Event < 0) { 5161 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5162 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5163 pthread_cond_destroy (_cond); 5164 pthread_cond_init (_cond, NULL) ; 5165 } 5166 assert_status(status == 0 || status == EINTR || 5167 status == ETIME || status == ETIMEDOUT, 5168 status, "cond_timedwait"); 5169 if (!FilterSpuriousWakeups) break ; // previous semantics 5170 if (status == ETIME || status == ETIMEDOUT) break ; 5171 // We consume and ignore EINTR and spurious wakeups. 5172 } 5173 --_nParked ; 5174 if (_Event >= 0) { 5175 ret = OS_OK; 5176 } 5177 _Event = 0 ; 5178 status = pthread_mutex_unlock(_mutex); 5179 assert_status(status == 0, status, "mutex_unlock"); 5180 assert (_nParked == 0, "invariant") ; 5181 // Paranoia to ensure our locked and lock-free paths interact 5182 // correctly with each other. 5183 OrderAccess::fence(); 5184 return ret; 5185} 5186 5187void os::PlatformEvent::unpark() { 5188 // Transitions for _Event: 5189 // 0 :=> 1 5190 // 1 :=> 1 5191 // -1 :=> either 0 or 1; must signal target thread 5192 // That is, we can safely transition _Event from -1 to either 5193 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back 5194 // unpark() calls. 5195 // See also: "Semaphores in Plan 9" by Mullender & Cox 5196 // 5197 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5198 // that it will take two back-to-back park() calls for the owning 5199 // thread to block. This has the benefit of forcing a spurious return 5200 // from the first park() call after an unpark() call which will help 5201 // shake out uses of park() and unpark() without condition variables. 5202 5203 if (Atomic::xchg(1, &_Event) >= 0) return; 5204 5205 // Wait for the thread associated with the event to vacate 5206 int status = pthread_mutex_lock(_mutex); 5207 assert_status(status == 0, status, "mutex_lock"); 5208 int AnyWaiters = _nParked; 5209 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5210 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5211 AnyWaiters = 0; 5212 pthread_cond_signal(_cond); 5213 } 5214 status = pthread_mutex_unlock(_mutex); 5215 assert_status(status == 0, status, "mutex_unlock"); 5216 if (AnyWaiters != 0) { 5217 status = pthread_cond_signal(_cond); 5218 assert_status(status == 0, status, "cond_signal"); 5219 } 5220 5221 // Note that we signal() _after dropping the lock for "immortal" Events. 5222 // This is safe and avoids a common class of futile wakeups. In rare 5223 // circumstances this can cause a thread to return prematurely from 5224 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5225 // simply re-test the condition and re-park itself. 5226} 5227 5228 5229// JSR166 5230// ------------------------------------------------------- 5231 5232/* 5233 * The solaris and linux implementations of park/unpark are fairly 5234 * conservative for now, but can be improved. They currently use a 5235 * mutex/condvar pair, plus a a count. 5236 * Park decrements count if > 0, else does a condvar wait. Unpark 5237 * sets count to 1 and signals condvar. Only one thread ever waits 5238 * on the condvar. Contention seen when trying to park implies that someone 5239 * is unparking you, so don't wait. And spurious returns are fine, so there 5240 * is no need to track notifications. 5241 */ 5242 5243#define MAX_SECS 100000000 5244/* 5245 * This code is common to linux and solaris and will be moved to a 5246 * common place in dolphin. 5247 * 5248 * The passed in time value is either a relative time in nanoseconds 5249 * or an absolute time in milliseconds. Either way it has to be unpacked 5250 * into suitable seconds and nanoseconds components and stored in the 5251 * given timespec structure. 5252 * Given time is a 64-bit value and the time_t used in the timespec is only 5253 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 5254 * overflow if times way in the future are given. Further on Solaris versions 5255 * prior to 10 there is a restriction (see cond_timedwait) that the specified 5256 * number of seconds, in abstime, is less than current_time + 100,000,000. 5257 * As it will be 28 years before "now + 100000000" will overflow we can 5258 * ignore overflow and just impose a hard-limit on seconds using the value 5259 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 5260 * years from "now". 5261 */ 5262 5263static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5264 assert (time > 0, "convertTime"); 5265 5266 struct timeval now; 5267 int status = gettimeofday(&now, NULL); 5268 assert(status == 0, "gettimeofday"); 5269 5270 time_t max_secs = now.tv_sec + MAX_SECS; 5271 5272 if (isAbsolute) { 5273 jlong secs = time / 1000; 5274 if (secs > max_secs) { 5275 absTime->tv_sec = max_secs; 5276 } 5277 else { 5278 absTime->tv_sec = secs; 5279 } 5280 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5281 } 5282 else { 5283 jlong secs = time / NANOSECS_PER_SEC; 5284 if (secs >= MAX_SECS) { 5285 absTime->tv_sec = max_secs; 5286 absTime->tv_nsec = 0; 5287 } 5288 else { 5289 absTime->tv_sec = now.tv_sec + secs; 5290 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5291 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5292 absTime->tv_nsec -= NANOSECS_PER_SEC; 5293 ++absTime->tv_sec; // note: this must be <= max_secs 5294 } 5295 } 5296 } 5297 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5298 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5299 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5300 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5301} 5302 5303void Parker::park(bool isAbsolute, jlong time) { 5304 // Ideally we'd do something useful while spinning, such 5305 // as calling unpackTime(). 5306 5307 // Optional fast-path check: 5308 // Return immediately if a permit is available. 5309 // We depend on Atomic::xchg() having full barrier semantics 5310 // since we are doing a lock-free update to _counter. 5311 if (Atomic::xchg(0, &_counter) > 0) return; 5312 5313 Thread* thread = Thread::current(); 5314 assert(thread->is_Java_thread(), "Must be JavaThread"); 5315 JavaThread *jt = (JavaThread *)thread; 5316 5317 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5318 // Check interrupt before trying to wait 5319 if (Thread::is_interrupted(thread, false)) { 5320 return; 5321 } 5322 5323 // Next, demultiplex/decode time arguments 5324 timespec absTime; 5325 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5326 return; 5327 } 5328 if (time > 0) { 5329 unpackTime(&absTime, isAbsolute, time); 5330 } 5331 5332 5333 // Enter safepoint region 5334 // Beware of deadlocks such as 6317397. 5335 // The per-thread Parker:: mutex is a classic leaf-lock. 5336 // In particular a thread must never block on the Threads_lock while 5337 // holding the Parker:: mutex. If safepoints are pending both the 5338 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5339 ThreadBlockInVM tbivm(jt); 5340 5341 // Don't wait if cannot get lock since interference arises from 5342 // unblocking. Also. check interrupt before trying wait 5343 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5344 return; 5345 } 5346 5347 int status ; 5348 if (_counter > 0) { // no wait needed 5349 _counter = 0; 5350 status = pthread_mutex_unlock(_mutex); 5351 assert (status == 0, "invariant") ; 5352 // Paranoia to ensure our locked and lock-free paths interact 5353 // correctly with each other and Java-level accesses. 5354 OrderAccess::fence(); 5355 return; 5356 } 5357 5358#ifdef ASSERT 5359 // Don't catch signals while blocked; let the running threads have the signals. 5360 // (This allows a debugger to break into the running thread.) 5361 sigset_t oldsigs; 5362 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5363 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5364#endif 5365 5366 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5367 jt->set_suspend_equivalent(); 5368 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5369 5370 if (time == 0) { 5371 status = pthread_cond_wait (_cond, _mutex) ; 5372 } else { 5373 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 5374 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5375 pthread_cond_destroy (_cond) ; 5376 pthread_cond_init (_cond, NULL); 5377 } 5378 } 5379 assert_status(status == 0 || status == EINTR || 5380 status == ETIME || status == ETIMEDOUT, 5381 status, "cond_timedwait"); 5382 5383#ifdef ASSERT 5384 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5385#endif 5386 5387 _counter = 0 ; 5388 status = pthread_mutex_unlock(_mutex) ; 5389 assert_status(status == 0, status, "invariant") ; 5390 // Paranoia to ensure our locked and lock-free paths interact 5391 // correctly with each other and Java-level accesses. 5392 OrderAccess::fence(); 5393 5394 // If externally suspended while waiting, re-suspend 5395 if (jt->handle_special_suspend_equivalent_condition()) { 5396 jt->java_suspend_self(); 5397 } 5398} 5399 5400void Parker::unpark() { 5401 int s, status ; 5402 status = pthread_mutex_lock(_mutex); 5403 assert (status == 0, "invariant") ; 5404 s = _counter; 5405 _counter = 1; 5406 if (s < 1) { 5407 if (WorkAroundNPTLTimedWaitHang) { 5408 status = pthread_cond_signal (_cond) ; 5409 assert (status == 0, "invariant") ; 5410 status = pthread_mutex_unlock(_mutex); 5411 assert (status == 0, "invariant") ; 5412 } else { 5413 status = pthread_mutex_unlock(_mutex); 5414 assert (status == 0, "invariant") ; 5415 status = pthread_cond_signal (_cond) ; 5416 assert (status == 0, "invariant") ; 5417 } 5418 } else { 5419 pthread_mutex_unlock(_mutex); 5420 assert (status == 0, "invariant") ; 5421 } 5422} 5423 5424 5425extern char** environ; 5426 5427#ifndef __NR_fork 5428#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5429#endif 5430 5431#ifndef __NR_execve 5432#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5433#endif 5434 5435// Run the specified command in a separate process. Return its exit value, 5436// or -1 on failure (e.g. can't fork a new process). 5437// Unlike system(), this function can be called from signal handler. It 5438// doesn't block SIGINT et al. 5439int os::fork_and_exec(char* cmd) { 5440 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5441 5442 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5443 // pthread_atfork handlers and reset pthread library. All we need is a 5444 // separate process to execve. Make a direct syscall to fork process. 5445 // On IA64 there's no fork syscall, we have to use fork() and hope for 5446 // the best... 5447 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5448 IA64_ONLY(fork();) 5449 5450 if (pid < 0) { 5451 // fork failed 5452 return -1; 5453 5454 } else if (pid == 0) { 5455 // child process 5456 5457 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5458 // first to kill every thread on the thread list. Because this list is 5459 // not reset by fork() (see notes above), execve() will instead kill 5460 // every thread in the parent process. We know this is the only thread 5461 // in the new process, so make a system call directly. 5462 // IA64 should use normal execve() from glibc to match the glibc fork() 5463 // above. 5464 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5465 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5466 5467 // execve failed 5468 _exit(-1); 5469 5470 } else { 5471 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5472 // care about the actual exit code, for now. 5473 5474 int status; 5475 5476 // Wait for the child process to exit. This returns immediately if 5477 // the child has already exited. */ 5478 while (waitpid(pid, &status, 0) < 0) { 5479 switch (errno) { 5480 case ECHILD: return 0; 5481 case EINTR: break; 5482 default: return -1; 5483 } 5484 } 5485 5486 if (WIFEXITED(status)) { 5487 // The child exited normally; get its exit code. 5488 return WEXITSTATUS(status); 5489 } else if (WIFSIGNALED(status)) { 5490 // The child exited because of a signal 5491 // The best value to return is 0x80 + signal number, 5492 // because that is what all Unix shells do, and because 5493 // it allows callers to distinguish between process exit and 5494 // process death by signal. 5495 return 0x80 + WTERMSIG(status); 5496 } else { 5497 // Unknown exit code; pass it through 5498 return status; 5499 } 5500 } 5501} 5502 5503// is_headless_jre() 5504// 5505// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5506// in order to report if we are running in a headless jre 5507// 5508// Since JDK8 xawt/libmawt.so was moved into the same directory 5509// as libawt.so, and renamed libawt_xawt.so 5510// 5511bool os::is_headless_jre() { 5512 struct stat statbuf; 5513 char buf[MAXPATHLEN]; 5514 char libmawtpath[MAXPATHLEN]; 5515 const char *xawtstr = "/xawt/libmawt.so"; 5516 const char *new_xawtstr = "/libawt_xawt.so"; 5517 char *p; 5518 5519 // Get path to libjvm.so 5520 os::jvm_path(buf, sizeof(buf)); 5521 5522 // Get rid of libjvm.so 5523 p = strrchr(buf, '/'); 5524 if (p == NULL) return false; 5525 else *p = '\0'; 5526 5527 // Get rid of client or server 5528 p = strrchr(buf, '/'); 5529 if (p == NULL) return false; 5530 else *p = '\0'; 5531 5532 // check xawt/libmawt.so 5533 strcpy(libmawtpath, buf); 5534 strcat(libmawtpath, xawtstr); 5535 if (::stat(libmawtpath, &statbuf) == 0) return false; 5536 5537 // check libawt_xawt.so 5538 strcpy(libmawtpath, buf); 5539 strcat(libmawtpath, new_xawtstr); 5540 if (::stat(libmawtpath, &statbuf) == 0) return false; 5541 5542 return true; 5543} 5544 5545// Get the default path to the core file 5546// Returns the length of the string 5547int os::get_core_path(char* buffer, size_t bufferSize) { 5548 const char* p = get_current_directory(buffer, bufferSize); 5549 5550 if (p == NULL) { 5551 assert(p != NULL, "failed to get current directory"); 5552 return 0; 5553 } 5554 5555 return strlen(buffer); 5556} 5557 5558#ifdef JAVASE_EMBEDDED 5559// 5560// A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 5561// 5562MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 5563 5564// ctor 5565// 5566MemNotifyThread::MemNotifyThread(int fd): Thread() { 5567 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 5568 _fd = fd; 5569 5570 if (os::create_thread(this, os::os_thread)) { 5571 _memnotify_thread = this; 5572 os::set_priority(this, NearMaxPriority); 5573 os::start_thread(this); 5574 } 5575} 5576 5577// Where all the work gets done 5578// 5579void MemNotifyThread::run() { 5580 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 5581 5582 // Set up the select arguments 5583 fd_set rfds; 5584 if (_fd != -1) { 5585 FD_ZERO(&rfds); 5586 FD_SET(_fd, &rfds); 5587 } 5588 5589 // Now wait for the mem_notify device to wake up 5590 while (1) { 5591 // Wait for the mem_notify device to signal us.. 5592 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 5593 if (rc == -1) { 5594 perror("select!\n"); 5595 break; 5596 } else if (rc) { 5597 //ssize_t free_before = os::available_memory(); 5598 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 5599 5600 // The kernel is telling us there is not much memory left... 5601 // try to do something about that 5602 5603 // If we are not already in a GC, try one. 5604 if (!Universe::heap()->is_gc_active()) { 5605 Universe::heap()->collect(GCCause::_allocation_failure); 5606 5607 //ssize_t free_after = os::available_memory(); 5608 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 5609 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 5610 } 5611 // We might want to do something like the following if we find the GC's are not helping... 5612 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 5613 } 5614 } 5615} 5616 5617// 5618// See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 5619// 5620void MemNotifyThread::start() { 5621 int fd; 5622 fd = open ("/dev/mem_notify", O_RDONLY, 0); 5623 if (fd < 0) { 5624 return; 5625 } 5626 5627 if (memnotify_thread() == NULL) { 5628 new MemNotifyThread(fd); 5629 } 5630} 5631#endif // JAVASE_EMBEDDED 5632