os_linux.cpp revision 5245:4472884d8b37
159191Skris/* 2296341Sdelphij * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved. 3296341Sdelphij * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4160814Ssimon * 559191Skris * This code is free software; you can redistribute it and/or modify it 659191Skris * under the terms of the GNU General Public License version 2 only, as 7160814Ssimon * published by the Free Software Foundation. 859191Skris * 959191Skris * This code is distributed in the hope that it will be useful, but WITHOUT 1059191Skris * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 1159191Skris * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 1259191Skris * version 2 for more details (a copy is included in the LICENSE file that 1359191Skris * accompanied this code). 14296341Sdelphij * 1559191Skris * You should have received a copy of the GNU General Public License version 1659191Skris * 2 along with this work; if not, write to the Free Software Foundation, 1759191Skris * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 1859191Skris * 1959191Skris * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 2059191Skris * or visit www.oracle.com if you need additional information or have any 2159191Skris * questions. 2259191Skris * 2359191Skris */ 2459191Skris 2559191Skris// no precompiled headers 2659191Skris#include "classfile/classLoader.hpp" 2759191Skris#include "classfile/systemDictionary.hpp" 2859191Skris#include "classfile/vmSymbols.hpp" 2959191Skris#include "code/icBuffer.hpp" 3059191Skris#include "code/vtableStubs.hpp" 3159191Skris#include "compiler/compileBroker.hpp" 3259191Skris#include "compiler/disassembler.hpp" 3359191Skris#include "interpreter/interpreter.hpp" 3459191Skris#include "jvm_linux.h" 3559191Skris#include "memory/allocation.inline.hpp" 3659191Skris#include "memory/filemap.hpp" 3759191Skris#include "mutex_linux.inline.hpp" 3859191Skris#include "oops/oop.inline.hpp" 3959191Skris#include "os_share_linux.hpp" 4059191Skris#include "prims/jniFastGetField.hpp" 4159191Skris#include "prims/jvm.h" 4259191Skris#include "prims/jvm_misc.hpp" 4359191Skris#include "runtime/arguments.hpp" 4459191Skris#include "runtime/extendedPC.hpp" 4559191Skris#include "runtime/globals.hpp" 4659191Skris#include "runtime/interfaceSupport.hpp" 4759191Skris#include "runtime/init.hpp" 4859191Skris#include "runtime/java.hpp" 4959191Skris#include "runtime/javaCalls.hpp" 5059191Skris#include "runtime/mutexLocker.hpp" 5159191Skris#include "runtime/objectMonitor.hpp" 5259191Skris#include "runtime/osThread.hpp" 5359191Skris#include "runtime/perfMemory.hpp" 5459191Skris#include "runtime/sharedRuntime.hpp" 5559191Skris#include "runtime/statSampler.hpp" 5659191Skris#include "runtime/stubRoutines.hpp" 5759191Skris#include "runtime/thread.inline.hpp" 5859191Skris#include "runtime/threadCritical.hpp" 5959191Skris#include "runtime/timer.hpp" 6059191Skris#include "services/attachListener.hpp" 6159191Skris#include "services/memTracker.hpp" 6259191Skris#include "services/runtimeService.hpp" 6359191Skris#include "utilities/decoder.hpp" 6459191Skris#include "utilities/defaultStream.hpp" 6559191Skris#include "utilities/events.hpp" 6659191Skris#include "utilities/elfFile.hpp" 6759191Skris#include "utilities/growableArray.hpp" 68160814Ssimon#include "utilities/vmError.hpp" 69160814Ssimon 7059191Skris// put OS-includes here 7159191Skris# include <sys/types.h> 7259191Skris# include <sys/mman.h> 7359191Skris# include <sys/stat.h> 74160814Ssimon# include <sys/select.h> 7559191Skris# include <pthread.h> 76296341Sdelphij# include <signal.h> 77296341Sdelphij# include <errno.h> 78296341Sdelphij# include <dlfcn.h> 79296341Sdelphij# include <stdio.h> 80296341Sdelphij# include <unistd.h> 81296341Sdelphij# include <sys/resource.h> 82296341Sdelphij# include <pthread.h> 83296341Sdelphij# include <sys/stat.h> 84296341Sdelphij# include <sys/time.h> 8559191Skris# include <sys/times.h> 8659191Skris# include <sys/utsname.h> 8759191Skris# include <sys/socket.h> 8859191Skris# include <sys/wait.h> 89296341Sdelphij# include <pwd.h> 90296341Sdelphij# include <poll.h> 91296341Sdelphij# include <semaphore.h> 92296341Sdelphij# include <fcntl.h> 93296341Sdelphij# include <string.h> 94296341Sdelphij# include <syscall.h> 95296341Sdelphij# include <sys/sysinfo.h> 96296341Sdelphij# include <gnu/libc-version.h> 97296341Sdelphij# include <sys/ipc.h> 98296341Sdelphij# include <sys/shm.h> 99296341Sdelphij# include <link.h> 100296341Sdelphij# include <stdint.h> 101296341Sdelphij# include <inttypes.h> 102296341Sdelphij# include <sys/ioctl.h> 103296341Sdelphij 104296341Sdelphij// if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 105296341Sdelphij// getrusage() is prepared to handle the associated failure. 106296341Sdelphij#ifndef RUSAGE_THREAD 107296341Sdelphij#define RUSAGE_THREAD (1) /* only the calling thread */ 108296341Sdelphij#endif 109296341Sdelphij 110296341Sdelphij#define MAX_PATH (2 * K) 111296341Sdelphij 112296341Sdelphij// for timer info max values which include all bits 113296341Sdelphij#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 114296341Sdelphij 115296341Sdelphij#define LARGEPAGES_BIT (1 << 6) 116296341Sdelphij//////////////////////////////////////////////////////////////////////////////// 117111147Snectar// global variables 118296341Sdelphijjulong os::Linux::_physical_memory = 0; 119111147Snectar 120109998Smarkmaddress os::Linux::_initial_thread_stack_bottom = NULL; 121296341Sdelphijuintptr_t os::Linux::_initial_thread_stack_size = 0; 122160814Ssimon 123296341Sdelphijint (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 124296341Sdelphijint (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 12559191SkrisMutex* os::Linux::_createThread_lock = NULL; 126296341Sdelphijpthread_t os::Linux::_main_thread; 127109998Smarkmint os::Linux::_page_size = -1; 128296341Sdelphijconst int os::Linux::_vm_default_page_size = (8 * K); 129296341Sdelphijbool os::Linux::_is_floating_stack = false; 130296341Sdelphijbool os::Linux::_is_NPTL = false; 131296341Sdelphijbool os::Linux::_supports_fast_thread_cpu_time = false; 132109998Smarkmconst char * os::Linux::_glibc_version = NULL; 133296341Sdelphijconst char * os::Linux::_libpthread_version = NULL; 134296341Sdelphijpthread_condattr_t os::Linux::_condattr[1]; 135109998Smarkm 136296341Sdelphijstatic jlong initial_time_count=0; 137296341Sdelphij 138296341Sdelphijstatic int clock_tics_per_sec = 100; 139296341Sdelphij 140296341Sdelphij// For diagnostics to print a message once. see run_periodic_checks 141296341Sdelphijstatic sigset_t check_signal_done; 142296341Sdelphijstatic bool check_signals = true;; 143296341Sdelphij 144296341Sdelphijstatic pid_t _initial_pid = 0; 145296341Sdelphij 146296341Sdelphij/* Signal number used to suspend/resume a thread */ 147296341Sdelphij 148296341Sdelphij/* do not use any signal number less than SIGSEGV, see 4355769 */ 149109998Smarkmstatic int SR_signum = SIGUSR2; 150296341Sdelphijsigset_t SR_sigset; 151296341Sdelphij 152296341Sdelphij/* Used to protect dlsym() calls */ 153296341Sdelphijstatic pthread_mutex_t dl_mutex; 15459191Skris 155194206Ssimon// Declarations 156296341Sdelphijstatic void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 157296341Sdelphij 158194206Ssimon#ifdef JAVASE_EMBEDDED 159109998Smarkmclass MemNotifyThread: public Thread { 160296341Sdelphij friend class VMStructs; 161296341Sdelphij public: 162296341Sdelphij virtual void run(); 163296341Sdelphij 164296341Sdelphij private: 165296341Sdelphij static MemNotifyThread* _memnotify_thread; 16659191Skris int _fd; 167109998Smarkm 168296341Sdelphij public: 169296341Sdelphij 170296341Sdelphij // Constructor 171296341Sdelphij MemNotifyThread(int fd); 172296341Sdelphij 173296341Sdelphij // Tester 174109998Smarkm bool is_memnotify_thread() const { return true; } 175162911Ssimon 176296341Sdelphij // Printing 177296341Sdelphij char* name() const { return (char*)"Linux MemNotify Thread"; } 178296341Sdelphij 179296341Sdelphij // Returns the single instance of the MemNotifyThread 180296341Sdelphij static MemNotifyThread* memnotify_thread() { return _memnotify_thread; } 181296341Sdelphij 182162911Ssimon // Create and start the single instance of MemNotifyThread 183296341Sdelphij static void start(); 184296341Sdelphij}; 185296341Sdelphij#endif // JAVASE_EMBEDDED 186296341Sdelphij 187296341Sdelphij// utility functions 188296341Sdelphij 189296341Sdelphijstatic int SR_initialize(); 190296341Sdelphij 191296341Sdelphijjulong os::available_memory() { 192296341Sdelphij return Linux::available_memory(); 193296341Sdelphij} 194296341Sdelphij 195296341Sdelphijjulong os::Linux::available_memory() { 196296341Sdelphij // values in struct sysinfo are "unsigned long" 197296341Sdelphij struct sysinfo si; 198296341Sdelphij sysinfo(&si); 199296341Sdelphij 200296341Sdelphij return (julong)si.freeram * si.mem_unit; 201296341Sdelphij} 202296341Sdelphij 203296341Sdelphijjulong os::physical_memory() { 204296341Sdelphij return Linux::physical_memory(); 205296341Sdelphij} 206296341Sdelphij 207296341Sdelphij//////////////////////////////////////////////////////////////////////////////// 208296341Sdelphij// environment support 209296341Sdelphij 210296341Sdelphijbool os::getenv(const char* name, char* buf, int len) { 211296341Sdelphij const char* val = ::getenv(name); 212296341Sdelphij if (val != NULL && strlen(val) < (size_t)len) { 213296341Sdelphij strcpy(buf, val); 214296341Sdelphij return true; 215296341Sdelphij } 216296341Sdelphij if (len > 0) buf[0] = 0; // return a null string 217296341Sdelphij return false; 218296341Sdelphij} 219296341Sdelphij 220111147Snectar 221296341Sdelphij// Return true if user is running as root. 222296341Sdelphij 223296341Sdelphijbool os::have_special_privileges() { 224296341Sdelphij static bool init = false; 225296341Sdelphij static bool privileges = false; 226111147Snectar if (!init) { 227296341Sdelphij privileges = (getuid() != geteuid()) || (getgid() != getegid()); 228296341Sdelphij init = true; 229296341Sdelphij } 230296341Sdelphij return privileges; 231296341Sdelphij} 232296341Sdelphij 233296341Sdelphij 234296341Sdelphij#ifndef SYS_gettid 235296341Sdelphij// i386: 224, ia64: 1105, amd64: 186, sparc 143 236296341Sdelphij#ifdef __ia64__ 237296341Sdelphij#define SYS_gettid 1105 238296341Sdelphij#elif __i386__ 239296341Sdelphij#define SYS_gettid 224 240296341Sdelphij#elif __amd64__ 241296341Sdelphij#define SYS_gettid 186 242296341Sdelphij#elif __sparc__ 243296341Sdelphij#define SYS_gettid 143 244296341Sdelphij#else 245296341Sdelphij#error define gettid for the arch 246296341Sdelphij#endif 247238405Sjkim#endif 248296341Sdelphij 249296341Sdelphij// Cpu architecture string 250296341Sdelphij#if defined(ZERO) 251296341Sdelphijstatic char cpu_arch[] = ZERO_LIBARCH; 252296341Sdelphij#elif defined(IA64) 253296341Sdelphijstatic char cpu_arch[] = "ia64"; 254296341Sdelphij#elif defined(IA32) 255296341Sdelphijstatic char cpu_arch[] = "i386"; 256296341Sdelphij#elif defined(AMD64) 257296341Sdelphijstatic char cpu_arch[] = "amd64"; 258296341Sdelphij#elif defined(ARM) 259296341Sdelphijstatic char cpu_arch[] = "arm"; 260296341Sdelphij#elif defined(PPC) 261296341Sdelphijstatic char cpu_arch[] = "ppc"; 262296341Sdelphij#elif defined(SPARC) 263296341Sdelphij# ifdef _LP64 264296341Sdelphijstatic char cpu_arch[] = "sparcv9"; 265296341Sdelphij# else 266296341Sdelphijstatic char cpu_arch[] = "sparc"; 267296341Sdelphij# endif 268296341Sdelphij#else 269296341Sdelphij#error Add appropriate cpu_arch setting 270296341Sdelphij#endif 271296341Sdelphij 272296341Sdelphij 273296341Sdelphij// pid_t gettid() 274296341Sdelphij// 275296341Sdelphij// Returns the kernel thread id of the currently running thread. Kernel 276296341Sdelphij// thread id is used to access /proc. 277296341Sdelphij// 278296341Sdelphij// (Note that getpid() on LinuxThreads returns kernel thread id too; but 279296341Sdelphij// on NPTL, it returns the same pid for all threads, as required by POSIX.) 280296341Sdelphij// 281296341Sdelphijpid_t os::Linux::gettid() { 282296341Sdelphij int rslt = syscall(SYS_gettid); 283296341Sdelphij if (rslt == -1) { 284296341Sdelphij // old kernel, no NPTL support 285296341Sdelphij return getpid(); 286296341Sdelphij } else { 287296341Sdelphij return (pid_t)rslt; 288296341Sdelphij } 289296341Sdelphij} 290296341Sdelphij 291296341Sdelphij// Most versions of linux have a bug where the number of processors are 292296341Sdelphij// determined by looking at the /proc file system. In a chroot environment, 293296341Sdelphij// the system call returns 1. This causes the VM to act as if it is 294296341Sdelphij// a single processor and elide locking (see is_MP() call). 295296341Sdelphijstatic bool unsafe_chroot_detected = false; 296296341Sdelphijstatic const char *unstable_chroot_error = "/proc file system not found.\n" 297296341Sdelphij "Java may be unstable running multithreaded in a chroot " 298296341Sdelphij "environment on Linux when /proc filesystem is not mounted."; 299296341Sdelphij 300296341Sdelphijvoid os::Linux::initialize_system_info() { 301296341Sdelphij set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 302296341Sdelphij if (processor_count() == 1) { 303296341Sdelphij pid_t pid = os::Linux::gettid(); 304296341Sdelphij char fname[32]; 305296341Sdelphij jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 306296341Sdelphij FILE *fp = fopen(fname, "r"); 307296341Sdelphij if (fp == NULL) { 308296341Sdelphij unsafe_chroot_detected = true; 309296341Sdelphij } else { 310296341Sdelphij fclose(fp); 311296341Sdelphij } 312296341Sdelphij } 313296341Sdelphij _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 314296341Sdelphij assert(processor_count() > 0, "linux error"); 315296341Sdelphij} 316296341Sdelphij 317296341Sdelphijvoid os::init_system_properties_values() { 318296341Sdelphij// char arch[12]; 319296341Sdelphij// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 320296341Sdelphij 321296341Sdelphij // The next steps are taken in the product version: 322296341Sdelphij // 323296341Sdelphij // Obtain the JAVA_HOME value from the location of libjvm.so. 324296341Sdelphij // This library should be located at: 325296341Sdelphij // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 326296341Sdelphij // 327296341Sdelphij // If "/jre/lib/" appears at the right place in the path, then we 328296341Sdelphij // assume libjvm.so is installed in a JDK and we use this path. 329296341Sdelphij // 330296341Sdelphij // Otherwise exit with message: "Could not create the Java virtual machine." 331296341Sdelphij // 33259191Skris // The following extra steps are taken in the debugging version: 333296341Sdelphij // 334296341Sdelphij // If "/jre/lib/" does NOT appear at the right place in the path 335296341Sdelphij // instead of exit check for $JAVA_HOME environment variable. 336296341Sdelphij // 337160814Ssimon // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 338296341Sdelphij // then we append a fake suffix "hotspot/libjvm.so" to this path so 339296341Sdelphij // it looks like libjvm.so is installed there 340296341Sdelphij // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 341296341Sdelphij // 342296341Sdelphij // Otherwise exit. 343296341Sdelphij // 344296341Sdelphij // Important note: if the location of libjvm.so changes this 345296341Sdelphij // code needs to be changed accordingly. 346296341Sdelphij 347296341Sdelphij // The next few definitions allow the code to be verbatim: 348296341Sdelphij#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal) 349296341Sdelphij#define getenv(n) ::getenv(n) 350296341Sdelphij 351296341Sdelphij/* 352296341Sdelphij * See ld(1): 353296341Sdelphij * The linker uses the following search paths to locate required 354296341Sdelphij * shared libraries: 355296341Sdelphij * 1: ... 356296341Sdelphij * ... 357296341Sdelphij * 7: The default directories, normally /lib and /usr/lib. 358296341Sdelphij */ 359296341Sdelphij#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 360296341Sdelphij#define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 361296341Sdelphij#else 362296341Sdelphij#define DEFAULT_LIBPATH "/lib:/usr/lib" 363296341Sdelphij#endif 364296341Sdelphij 365296341Sdelphij#define EXTENSIONS_DIR "/lib/ext" 366296341Sdelphij#define ENDORSED_DIR "/lib/endorsed" 367296341Sdelphij#define REG_DIR "/usr/java/packages" 368296341Sdelphij 369296341Sdelphij { 370296341Sdelphij /* sysclasspath, java_home, dll_dir */ 371296341Sdelphij { 372296341Sdelphij char *home_path; 373296341Sdelphij char *dll_path; 374296341Sdelphij char *pslash; 37559191Skris char buf[MAXPATHLEN]; 376296341Sdelphij os::jvm_path(buf, sizeof(buf)); 377296341Sdelphij 378296341Sdelphij // Found the full path to libjvm.so. 379296341Sdelphij // Now cut the path to <java_home>/jre if we can. 380296341Sdelphij *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 381296341Sdelphij pslash = strrchr(buf, '/'); 382296341Sdelphij if (pslash != NULL) 383296341Sdelphij *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 384296341Sdelphij dll_path = malloc(strlen(buf) + 1); 385109998Smarkm if (dll_path == NULL) 386296341Sdelphij return; 387296341Sdelphij strcpy(dll_path, buf); 38859191Skris Arguments::set_dll_dir(dll_path); 389194206Ssimon 390296341Sdelphij if (pslash != NULL) { 391194206Ssimon pslash = strrchr(buf, '/'); 392109998Smarkm if (pslash != NULL) { 393296341Sdelphij *pslash = '\0'; /* get rid of /<arch> */ 394296341Sdelphij pslash = strrchr(buf, '/'); 395296341Sdelphij if (pslash != NULL) 39659191Skris *pslash = '\0'; /* get rid of /lib */ 397109998Smarkm } 398296341Sdelphij } 399296341Sdelphij 400296341Sdelphij home_path = malloc(strlen(buf) + 1); 401109998Smarkm if (home_path == NULL) 402162911Ssimon return; 403296341Sdelphij strcpy(home_path, buf); 404296341Sdelphij Arguments::set_java_home(home_path); 405296341Sdelphij 406162911Ssimon if (!set_boot_path('/', ':')) 407296341Sdelphij return; 408296341Sdelphij } 409296341Sdelphij 410296341Sdelphij /* 411296341Sdelphij * Where to look for native libraries 412296341Sdelphij * 413296341Sdelphij * Note: Due to a legacy implementation, most of the library path 414296341Sdelphij * is set in the launcher. This was to accomodate linking restrictions 415296341Sdelphij * on legacy Linux implementations (which are no longer supported). 416296341Sdelphij * Eventually, all the library path setting will be done here. 417296341Sdelphij * 418296341Sdelphij * However, to prevent the proliferation of improperly built native 419296341Sdelphij * libraries, the new path component /usr/java/packages is added here. 420296341Sdelphij * Eventually, all the library path setting will be done here. 421296341Sdelphij */ 422296341Sdelphij { 423296341Sdelphij char *ld_library_path; 424296341Sdelphij 425296341Sdelphij /* 426296341Sdelphij * Construct the invariant part of ld_library_path. Note that the 427296341Sdelphij * space for the colon and the trailing null are provided by the 428296341Sdelphij * nulls included by the sizeof operator (so actually we allocate 429296341Sdelphij * a byte more than necessary). 430296341Sdelphij */ 431296341Sdelphij ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 432296341Sdelphij strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 433296341Sdelphij sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 434296341Sdelphij 435296341Sdelphij /* 436296341Sdelphij * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 437296341Sdelphij * should always exist (until the legacy problem cited above is 438296341Sdelphij * addressed). 439296341Sdelphij */ 440296341Sdelphij char *v = getenv("LD_LIBRARY_PATH"); 441296341Sdelphij if (v != NULL) { 442296341Sdelphij char *t = ld_library_path; 443296341Sdelphij /* That's +1 for the colon and +1 for the trailing '\0' */ 444296341Sdelphij ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 445296341Sdelphij sprintf(ld_library_path, "%s:%s", v, t); 446296341Sdelphij } 447296341Sdelphij Arguments::set_library_path(ld_library_path); 448296341Sdelphij } 449296341Sdelphij 450111147Snectar /* 451296341Sdelphij * Extensions directories. 452296341Sdelphij * 453111147Snectar * Note that the space for the colon and the trailing null are provided 454296341Sdelphij * by the nulls included by the sizeof operator (so actually one byte more 455296341Sdelphij * than necessary is allocated). 456296341Sdelphij */ 457296341Sdelphij { 458296341Sdelphij char *buf = malloc(strlen(Arguments::get_java_home()) + 459296341Sdelphij sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 460296341Sdelphij sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 461296341Sdelphij Arguments::get_java_home()); 462296341Sdelphij Arguments::set_ext_dirs(buf); 463296341Sdelphij } 464111147Snectar 465296341Sdelphij /* Endorsed standards default directory. */ 466111147Snectar { 467109998Smarkm char * buf; 468296341Sdelphij buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 469296341Sdelphij sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 470296341Sdelphij Arguments::set_endorsed_dirs(buf); 471296341Sdelphij } 47259191Skris } 473296341Sdelphij 474296341Sdelphij#undef malloc 475296341Sdelphij#undef getenv 476296341Sdelphij#undef EXTENSIONS_DIR 477296341Sdelphij#undef ENDORSED_DIR 478296341Sdelphij 47959191Skris // Done 480296341Sdelphij return; 48159191Skris} 482296341Sdelphij 483296341Sdelphij//////////////////////////////////////////////////////////////////////////////// 48459191Skris// breakpoint support 485296341Sdelphij 486296341Sdelphijvoid os::breakpoint() { 487296341Sdelphij BREAKPOINT; 488296341Sdelphij} 489296341Sdelphij 490296341Sdelphijextern "C" void breakpoint() { 491296341Sdelphij // use debugger to set breakpoint here 492238405Sjkim} 493296341Sdelphij 494296341Sdelphij//////////////////////////////////////////////////////////////////////////////// 495296341Sdelphij// signal support 496296341Sdelphij 497296341Sdelphijdebug_only(static bool signal_sets_initialized = false); 498296341Sdelphijstatic sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 499296341Sdelphij 50059191Skrisbool os::Linux::is_sig_ignored(int sig) { 501296341Sdelphij struct sigaction oact; 502296341Sdelphij sigaction(sig, (struct sigaction*)NULL, &oact); 503296341Sdelphij void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 504296341Sdelphij : CAST_FROM_FN_PTR(void*, oact.sa_handler); 50559191Skris if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 506296341Sdelphij return true; 507296341Sdelphij else 50859191Skris return false; 509296341Sdelphij} 510296341Sdelphij 511296341Sdelphijvoid os::Linux::signal_sets_init() { 512296341Sdelphij // Should also have an assertion stating we are still single-threaded. 513296341Sdelphij assert(!signal_sets_initialized, "Already initialized"); 514296341Sdelphij // Fill in signals that are necessarily unblocked for all threads in 515296341Sdelphij // the VM. Currently, we unblock the following signals: 516296341Sdelphij // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 517296341Sdelphij // by -Xrs (=ReduceSignalUsage)); 518109998Smarkm // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 519296341Sdelphij // other threads. The "ReduceSignalUsage" boolean tells us not to alter 520296341Sdelphij // the dispositions or masks wrt these signals. 521296341Sdelphij // Programs embedding the VM that want to use the above signals for their 522296341Sdelphij // own purposes must, at this time, use the "-Xrs" option to prevent 523296341Sdelphij // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 524296341Sdelphij // (See bug 4345157, and other related bugs). 525296341Sdelphij // In reality, though, unblocking these signals is really a nop, since 52659191Skris // these signals are not blocked by default. 527296341Sdelphij sigemptyset(&unblocked_sigs); 528296341Sdelphij sigemptyset(&allowdebug_blocked_sigs); 529296341Sdelphij sigaddset(&unblocked_sigs, SIGILL); 530296341Sdelphij sigaddset(&unblocked_sigs, SIGSEGV); 531296341Sdelphij sigaddset(&unblocked_sigs, SIGBUS); 532296341Sdelphij sigaddset(&unblocked_sigs, SIGFPE); 533296341Sdelphij sigaddset(&unblocked_sigs, SR_signum); 53459191Skris 535296341Sdelphij if (!ReduceSignalUsage) { 536296341Sdelphij if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 537296341Sdelphij sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 538296341Sdelphij sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 539296341Sdelphij } 540296341Sdelphij if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 541296341Sdelphij sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 54259191Skris sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 543296341Sdelphij } 544296341Sdelphij if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 545296341Sdelphij sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 546296341Sdelphij sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 547296341Sdelphij } 548296341Sdelphij } 549296341Sdelphij // Fill in signals that are blocked by all but the VM thread. 550296341Sdelphij sigemptyset(&vm_sigs); 55159191Skris if (!ReduceSignalUsage) 552296341Sdelphij sigaddset(&vm_sigs, BREAK_SIGNAL); 553296341Sdelphij debug_only(signal_sets_initialized = true); 554296341Sdelphij 555296341Sdelphij} 556296341Sdelphij 557296341Sdelphij// These are signals that are unblocked while a thread is running Java. 55859191Skris// (For some reason, they get blocked by default.) 559296341Sdelphijsigset_t* os::Linux::unblocked_signals() { 560296341Sdelphij assert(signal_sets_initialized, "Not initialized"); 561296341Sdelphij return &unblocked_sigs; 562296341Sdelphij} 563296341Sdelphij 564296341Sdelphij// These are the signals that are blocked while a (non-VM) thread is 565296341Sdelphij// running Java. Only the VM thread handles these signals. 56659191Skrissigset_t* os::Linux::vm_signals() { 567296341Sdelphij assert(signal_sets_initialized, "Not initialized"); 568296341Sdelphij return &vm_sigs; 569296341Sdelphij} 570296341Sdelphij 571296341Sdelphij// These are signals that are blocked during cond_wait to allow debugger in 572296341Sdelphijsigset_t* os::Linux::allowdebug_blocked_signals() { 573296341Sdelphij assert(signal_sets_initialized, "Not initialized"); 574296341Sdelphij return &allowdebug_blocked_sigs; 575296341Sdelphij} 576296341Sdelphij 577296341Sdelphijvoid os::Linux::hotspot_sigmask(Thread* thread) { 578238405Sjkim 579296341Sdelphij //Save caller's signal mask before setting VM signal mask 580296341Sdelphij sigset_t caller_sigmask; 581296341Sdelphij pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 582296341Sdelphij 583296341Sdelphij OSThread* osthread = thread->osthread(); 584296341Sdelphij osthread->set_caller_sigmask(caller_sigmask); 585296341Sdelphij 586296341Sdelphij pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 587296341Sdelphij 588296341Sdelphij if (!ReduceSignalUsage) { 589296341Sdelphij if (thread->is_VM_thread()) { 590296341Sdelphij // Only the VM thread handles BREAK_SIGNAL ... 591238405Sjkim pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 592296341Sdelphij } else { 593296341Sdelphij // ... all other threads block BREAK_SIGNAL 594296341Sdelphij pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 595296341Sdelphij } 596296341Sdelphij } 597296341Sdelphij} 598296341Sdelphij 599109998Smarkm////////////////////////////////////////////////////////////////////////////// 600296341Sdelphij// detecting pthread library 601296341Sdelphij 602296341Sdelphijvoid os::Linux::libpthread_init() { 603296341Sdelphij // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 60468651Skris // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 605296341Sdelphij // generic name for earlier versions. 60659191Skris // Define macros here so we can build HotSpot on old systems. 607296341Sdelphij# ifndef _CS_GNU_LIBC_VERSION 608296341Sdelphij# define _CS_GNU_LIBC_VERSION 2 609296341Sdelphij# endif 610296341Sdelphij# ifndef _CS_GNU_LIBPTHREAD_VERSION 611296341Sdelphij# define _CS_GNU_LIBPTHREAD_VERSION 3 612296341Sdelphij# endif 613296341Sdelphij 61459191Skris size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 615296341Sdelphij if (n > 0) { 616109998Smarkm char *str = (char *)malloc(n, mtInternal); 617296341Sdelphij confstr(_CS_GNU_LIBC_VERSION, str, n); 618296341Sdelphij os::Linux::set_glibc_version(str); 619296341Sdelphij } else { 620296341Sdelphij // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 621296341Sdelphij static char _gnu_libc_version[32]; 622296341Sdelphij jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 623296341Sdelphij "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 624296341Sdelphij os::Linux::set_glibc_version(_gnu_libc_version); 625296341Sdelphij } 626296341Sdelphij 627296341Sdelphij n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 628296341Sdelphij if (n > 0) { 629296341Sdelphij char *str = (char *)malloc(n, mtInternal); 630296341Sdelphij confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 631296341Sdelphij // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 632296341Sdelphij // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 633296341Sdelphij // is the case. LinuxThreads has a hard limit on max number of threads. 634296341Sdelphij // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 635296341Sdelphij // On the other hand, NPTL does not have such a limit, sysconf() 636296341Sdelphij // will return -1 and errno is not changed. Check if it is really NPTL. 637296341Sdelphij if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 638296341Sdelphij strstr(str, "NPTL") && 639296341Sdelphij sysconf(_SC_THREAD_THREADS_MAX) > 0) { 640296341Sdelphij free(str); 641296341Sdelphij os::Linux::set_libpthread_version("linuxthreads"); 642296341Sdelphij } else { 643296341Sdelphij os::Linux::set_libpthread_version(str); 644296341Sdelphij } 645296341Sdelphij } else { 646296341Sdelphij // glibc before 2.3.2 only has LinuxThreads. 647296341Sdelphij os::Linux::set_libpthread_version("linuxthreads"); 648296341Sdelphij } 649296341Sdelphij 650296341Sdelphij if (strstr(libpthread_version(), "NPTL")) { 651296341Sdelphij os::Linux::set_is_NPTL(); 652296341Sdelphij } else { 653296341Sdelphij os::Linux::set_is_LinuxThreads(); 654296341Sdelphij } 655296341Sdelphij 656296341Sdelphij // LinuxThreads have two flavors: floating-stack mode, which allows variable 657296341Sdelphij // stack size; and fixed-stack mode. NPTL is always floating-stack. 658296341Sdelphij if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 659296341Sdelphij os::Linux::set_is_floating_stack(); 660296341Sdelphij } 661296341Sdelphij} 662296341Sdelphij 66359191Skris///////////////////////////////////////////////////////////////////////////// 664296341Sdelphij// thread stack 665296341Sdelphij 666296341Sdelphij// Force Linux kernel to expand current thread stack. If "bottom" is close 667296341Sdelphij// to the stack guard, caller should block all signals. 66859191Skris// 669296341Sdelphij// MAP_GROWSDOWN: 670296341Sdelphij// A special mmap() flag that is used to implement thread stacks. It tells 671296341Sdelphij// kernel that the memory region should extend downwards when needed. This 672296341Sdelphij// allows early versions of LinuxThreads to only mmap the first few pages 673296341Sdelphij// when creating a new thread. Linux kernel will automatically expand thread 674296341Sdelphij// stack as needed (on page faults). 675296341Sdelphij// 676296341Sdelphij// However, because the memory region of a MAP_GROWSDOWN stack can grow on 677296341Sdelphij// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 678296341Sdelphij// region, it's hard to tell if the fault is due to a legitimate stack 679296341Sdelphij// access or because of reading/writing non-exist memory (e.g. buffer 680296341Sdelphij// overrun). As a rule, if the fault happens below current stack pointer, 681296341Sdelphij// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 682296341Sdelphij// application (see Linux kernel fault.c). 683296341Sdelphij// 684296341Sdelphij// This Linux feature can cause SIGSEGV when VM bangs thread stack for 685296341Sdelphij// stack overflow detection. 686296341Sdelphij// 687296341Sdelphij// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 688296341Sdelphij// not use this flag. However, the stack of initial thread is not created 689296341Sdelphij// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 690296341Sdelphij// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 691296341Sdelphij// and then attach the thread to JVM. 692296341Sdelphij// 693296341Sdelphij// To get around the problem and allow stack banging on Linux, we need to 694296341Sdelphij// manually expand thread stack after receiving the SIGSEGV. 695296341Sdelphij// 696296341Sdelphij// There are two ways to expand thread stack to address "bottom", we used 697296341Sdelphij// both of them in JVM before 1.5: 698296341Sdelphij// 1. adjust stack pointer first so that it is below "bottom", and then 699296341Sdelphij// touch "bottom" 700296341Sdelphij// 2. mmap() the page in question 701296341Sdelphij// 702296341Sdelphij// Now alternate signal stack is gone, it's harder to use 2. For instance, 703296341Sdelphij// if current sp is already near the lower end of page 101, and we need to 704296341Sdelphij// call mmap() to map page 100, it is possible that part of the mmap() frame 705296341Sdelphij// will be placed in page 100. When page 100 is mapped, it is zero-filled. 706296341Sdelphij// That will destroy the mmap() frame and cause VM to crash. 707296341Sdelphij// 708296341Sdelphij// The following code works by adjusting sp first, then accessing the "bottom" 709296341Sdelphij// page to force a page fault. Linux kernel will then automatically expand the 710296341Sdelphij// stack mapping. 711296341Sdelphij// 712296341Sdelphij// _expand_stack_to() assumes its frame size is less than page size, which 713296341Sdelphij// should always be true if the function is not inlined. 714296341Sdelphij 715296341Sdelphij#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 716296341Sdelphij#define NOINLINE 717296341Sdelphij#else 718296341Sdelphij#define NOINLINE __attribute__ ((noinline)) 719296341Sdelphij#endif 720296341Sdelphij 721296341Sdelphijstatic void _expand_stack_to(address bottom) NOINLINE; 722296341Sdelphij 723296341Sdelphijstatic void _expand_stack_to(address bottom) { 724296341Sdelphij address sp; 725296341Sdelphij size_t size; 726296341Sdelphij volatile char *p; 727296341Sdelphij 728296341Sdelphij // Adjust bottom to point to the largest address within the same page, it 729296341Sdelphij // gives us a one-page buffer if alloca() allocates slightly more memory. 730296341Sdelphij bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 731296341Sdelphij bottom += os::Linux::page_size() - 1; 732296341Sdelphij 733296341Sdelphij // sp might be slightly above current stack pointer; if that's the case, we 734296341Sdelphij // will alloca() a little more space than necessary, which is OK. Don't use 735296341Sdelphij // os::current_stack_pointer(), as its result can be slightly below current 736296341Sdelphij // stack pointer, causing us to not alloca enough to reach "bottom". 737296341Sdelphij sp = (address)&sp; 738296341Sdelphij 73959191Skris if (sp > bottom) { 74059191Skris size = sp - bottom; 74159191Skris p = (volatile char *)alloca(size); 742296341Sdelphij assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 743296341Sdelphij p[0] = '\0'; 744296341Sdelphij } 745296341Sdelphij} 746296341Sdelphij 747296341Sdelphijbool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 748296341Sdelphij assert(t!=NULL, "just checking"); 749296341Sdelphij assert(t->osthread()->expanding_stack(), "expand should be set"); 750296341Sdelphij assert(t->stack_base() != NULL, "stack_base was not initialized"); 751296341Sdelphij 752296341Sdelphij if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 753296341Sdelphij sigset_t mask_all, old_sigset; 754296341Sdelphij sigfillset(&mask_all); 755160814Ssimon pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 756160814Ssimon _expand_stack_to(addr); 757160814Ssimon pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 758160814Ssimon return true; 759296341Sdelphij } 760296341Sdelphij return false; 761160814Ssimon} 762296341Sdelphij 763160814Ssimon////////////////////////////////////////////////////////////////////////////// 764296341Sdelphij// create new thread 765296341Sdelphij 766296341Sdelphijstatic address highest_vm_reserved_address(); 767160814Ssimon 768296341Sdelphij// check if it's safe to start a new thread 769160814Ssimonstatic bool _thread_safety_check(Thread* thread) { 770296341Sdelphij if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 771160814Ssimon // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 772296341Sdelphij // Heap is mmap'ed at lower end of memory space. Thread stacks are 773 // allocated (MAP_FIXED) from high address space. Every thread stack 774 // occupies a fixed size slot (usually 2Mbytes, but user can change 775 // it to other values if they rebuild LinuxThreads). 776 // 777 // Problem with MAP_FIXED is that mmap() can still succeed even part of 778 // the memory region has already been mmap'ed. That means if we have too 779 // many threads and/or very large heap, eventually thread stack will 780 // collide with heap. 781 // 782 // Here we try to prevent heap/stack collision by comparing current 783 // stack bottom with the highest address that has been mmap'ed by JVM 784 // plus a safety margin for memory maps created by native code. 785 // 786 // This feature can be disabled by setting ThreadSafetyMargin to 0 787 // 788 if (ThreadSafetyMargin > 0) { 789 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 790 791 // not safe if our stack extends below the safety margin 792 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 793 } else { 794 return true; 795 } 796 } else { 797 // Floating stack LinuxThreads or NPTL: 798 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 799 // there's not enough space left, pthread_create() will fail. If we come 800 // here, that means enough space has been reserved for stack. 801 return true; 802 } 803} 804 805// Thread start routine for all newly created threads 806static void *java_start(Thread *thread) { 807 // Try to randomize the cache line index of hot stack frames. 808 // This helps when threads of the same stack traces evict each other's 809 // cache lines. The threads can be either from the same JVM instance, or 810 // from different JVM instances. The benefit is especially true for 811 // processors with hyperthreading technology. 812 static int counter = 0; 813 int pid = os::current_process_id(); 814 alloca(((pid ^ counter++) & 7) * 128); 815 816 ThreadLocalStorage::set_thread(thread); 817 818 OSThread* osthread = thread->osthread(); 819 Monitor* sync = osthread->startThread_lock(); 820 821 // non floating stack LinuxThreads needs extra check, see above 822 if (!_thread_safety_check(thread)) { 823 // notify parent thread 824 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 825 osthread->set_state(ZOMBIE); 826 sync->notify_all(); 827 return NULL; 828 } 829 830 // thread_id is kernel thread id (similar to Solaris LWP id) 831 osthread->set_thread_id(os::Linux::gettid()); 832 833 if (UseNUMA) { 834 int lgrp_id = os::numa_get_group_id(); 835 if (lgrp_id != -1) { 836 thread->set_lgrp_id(lgrp_id); 837 } 838 } 839 // initialize signal mask for this thread 840 os::Linux::hotspot_sigmask(thread); 841 842 // initialize floating point control register 843 os::Linux::init_thread_fpu_state(); 844 845 // handshaking with parent thread 846 { 847 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 848 849 // notify parent thread 850 osthread->set_state(INITIALIZED); 851 sync->notify_all(); 852 853 // wait until os::start_thread() 854 while (osthread->get_state() == INITIALIZED) { 855 sync->wait(Mutex::_no_safepoint_check_flag); 856 } 857 } 858 859 // call one more level start routine 860 thread->run(); 861 862 return 0; 863} 864 865bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 866 assert(thread->osthread() == NULL, "caller responsible"); 867 868 // Allocate the OSThread object 869 OSThread* osthread = new OSThread(NULL, NULL); 870 if (osthread == NULL) { 871 return false; 872 } 873 874 // set the correct thread state 875 osthread->set_thread_type(thr_type); 876 877 // Initial state is ALLOCATED but not INITIALIZED 878 osthread->set_state(ALLOCATED); 879 880 thread->set_osthread(osthread); 881 882 // init thread attributes 883 pthread_attr_t attr; 884 pthread_attr_init(&attr); 885 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 886 887 // stack size 888 if (os::Linux::supports_variable_stack_size()) { 889 // calculate stack size if it's not specified by caller 890 if (stack_size == 0) { 891 stack_size = os::Linux::default_stack_size(thr_type); 892 893 switch (thr_type) { 894 case os::java_thread: 895 // Java threads use ThreadStackSize which default value can be 896 // changed with the flag -Xss 897 assert (JavaThread::stack_size_at_create() > 0, "this should be set"); 898 stack_size = JavaThread::stack_size_at_create(); 899 break; 900 case os::compiler_thread: 901 if (CompilerThreadStackSize > 0) { 902 stack_size = (size_t)(CompilerThreadStackSize * K); 903 break; 904 } // else fall through: 905 // use VMThreadStackSize if CompilerThreadStackSize is not defined 906 case os::vm_thread: 907 case os::pgc_thread: 908 case os::cgc_thread: 909 case os::watcher_thread: 910 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 911 break; 912 } 913 } 914 915 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 916 pthread_attr_setstacksize(&attr, stack_size); 917 } else { 918 // let pthread_create() pick the default value. 919 } 920 921 // glibc guard page 922 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 923 924 ThreadState state; 925 926 { 927 // Serialize thread creation if we are running with fixed stack LinuxThreads 928 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 929 if (lock) { 930 os::Linux::createThread_lock()->lock_without_safepoint_check(); 931 } 932 933 pthread_t tid; 934 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 935 936 pthread_attr_destroy(&attr); 937 938 if (ret != 0) { 939 if (PrintMiscellaneous && (Verbose || WizardMode)) { 940 perror("pthread_create()"); 941 } 942 // Need to clean up stuff we've allocated so far 943 thread->set_osthread(NULL); 944 delete osthread; 945 if (lock) os::Linux::createThread_lock()->unlock(); 946 return false; 947 } 948 949 // Store pthread info into the OSThread 950 osthread->set_pthread_id(tid); 951 952 // Wait until child thread is either initialized or aborted 953 { 954 Monitor* sync_with_child = osthread->startThread_lock(); 955 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 956 while ((state = osthread->get_state()) == ALLOCATED) { 957 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 958 } 959 } 960 961 if (lock) { 962 os::Linux::createThread_lock()->unlock(); 963 } 964 } 965 966 // Aborted due to thread limit being reached 967 if (state == ZOMBIE) { 968 thread->set_osthread(NULL); 969 delete osthread; 970 return false; 971 } 972 973 // The thread is returned suspended (in state INITIALIZED), 974 // and is started higher up in the call chain 975 assert(state == INITIALIZED, "race condition"); 976 return true; 977} 978 979///////////////////////////////////////////////////////////////////////////// 980// attach existing thread 981 982// bootstrap the main thread 983bool os::create_main_thread(JavaThread* thread) { 984 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 985 return create_attached_thread(thread); 986} 987 988bool os::create_attached_thread(JavaThread* thread) { 989#ifdef ASSERT 990 thread->verify_not_published(); 991#endif 992 993 // Allocate the OSThread object 994 OSThread* osthread = new OSThread(NULL, NULL); 995 996 if (osthread == NULL) { 997 return false; 998 } 999 1000 // Store pthread info into the OSThread 1001 osthread->set_thread_id(os::Linux::gettid()); 1002 osthread->set_pthread_id(::pthread_self()); 1003 1004 // initialize floating point control register 1005 os::Linux::init_thread_fpu_state(); 1006 1007 // Initial thread state is RUNNABLE 1008 osthread->set_state(RUNNABLE); 1009 1010 thread->set_osthread(osthread); 1011 1012 if (UseNUMA) { 1013 int lgrp_id = os::numa_get_group_id(); 1014 if (lgrp_id != -1) { 1015 thread->set_lgrp_id(lgrp_id); 1016 } 1017 } 1018 1019 if (os::Linux::is_initial_thread()) { 1020 // If current thread is initial thread, its stack is mapped on demand, 1021 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 1022 // the entire stack region to avoid SEGV in stack banging. 1023 // It is also useful to get around the heap-stack-gap problem on SuSE 1024 // kernel (see 4821821 for details). We first expand stack to the top 1025 // of yellow zone, then enable stack yellow zone (order is significant, 1026 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1027 // is no gap between the last two virtual memory regions. 1028 1029 JavaThread *jt = (JavaThread *)thread; 1030 address addr = jt->stack_yellow_zone_base(); 1031 assert(addr != NULL, "initialization problem?"); 1032 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1033 1034 osthread->set_expanding_stack(); 1035 os::Linux::manually_expand_stack(jt, addr); 1036 osthread->clear_expanding_stack(); 1037 } 1038 1039 // initialize signal mask for this thread 1040 // and save the caller's signal mask 1041 os::Linux::hotspot_sigmask(thread); 1042 1043 return true; 1044} 1045 1046void os::pd_start_thread(Thread* thread) { 1047 OSThread * osthread = thread->osthread(); 1048 assert(osthread->get_state() != INITIALIZED, "just checking"); 1049 Monitor* sync_with_child = osthread->startThread_lock(); 1050 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1051 sync_with_child->notify(); 1052} 1053 1054// Free Linux resources related to the OSThread 1055void os::free_thread(OSThread* osthread) { 1056 assert(osthread != NULL, "osthread not set"); 1057 1058 if (Thread::current()->osthread() == osthread) { 1059 // Restore caller's signal mask 1060 sigset_t sigmask = osthread->caller_sigmask(); 1061 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1062 } 1063 1064 delete osthread; 1065} 1066 1067////////////////////////////////////////////////////////////////////////////// 1068// thread local storage 1069 1070int os::allocate_thread_local_storage() { 1071 pthread_key_t key; 1072 int rslt = pthread_key_create(&key, NULL); 1073 assert(rslt == 0, "cannot allocate thread local storage"); 1074 return (int)key; 1075} 1076 1077// Note: This is currently not used by VM, as we don't destroy TLS key 1078// on VM exit. 1079void os::free_thread_local_storage(int index) { 1080 int rslt = pthread_key_delete((pthread_key_t)index); 1081 assert(rslt == 0, "invalid index"); 1082} 1083 1084void os::thread_local_storage_at_put(int index, void* value) { 1085 int rslt = pthread_setspecific((pthread_key_t)index, value); 1086 assert(rslt == 0, "pthread_setspecific failed"); 1087} 1088 1089extern "C" Thread* get_thread() { 1090 return ThreadLocalStorage::thread(); 1091} 1092 1093////////////////////////////////////////////////////////////////////////////// 1094// initial thread 1095 1096// Check if current thread is the initial thread, similar to Solaris thr_main. 1097bool os::Linux::is_initial_thread(void) { 1098 char dummy; 1099 // If called before init complete, thread stack bottom will be null. 1100 // Can be called if fatal error occurs before initialization. 1101 if (initial_thread_stack_bottom() == NULL) return false; 1102 assert(initial_thread_stack_bottom() != NULL && 1103 initial_thread_stack_size() != 0, 1104 "os::init did not locate initial thread's stack region"); 1105 if ((address)&dummy >= initial_thread_stack_bottom() && 1106 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1107 return true; 1108 else return false; 1109} 1110 1111// Find the virtual memory area that contains addr 1112static bool find_vma(address addr, address* vma_low, address* vma_high) { 1113 FILE *fp = fopen("/proc/self/maps", "r"); 1114 if (fp) { 1115 address low, high; 1116 while (!feof(fp)) { 1117 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1118 if (low <= addr && addr < high) { 1119 if (vma_low) *vma_low = low; 1120 if (vma_high) *vma_high = high; 1121 fclose (fp); 1122 return true; 1123 } 1124 } 1125 for (;;) { 1126 int ch = fgetc(fp); 1127 if (ch == EOF || ch == (int)'\n') break; 1128 } 1129 } 1130 fclose(fp); 1131 } 1132 return false; 1133} 1134 1135// Locate initial thread stack. This special handling of initial thread stack 1136// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1137// bogus value for initial thread. 1138void os::Linux::capture_initial_stack(size_t max_size) { 1139 // stack size is the easy part, get it from RLIMIT_STACK 1140 size_t stack_size; 1141 struct rlimit rlim; 1142 getrlimit(RLIMIT_STACK, &rlim); 1143 stack_size = rlim.rlim_cur; 1144 1145 // 6308388: a bug in ld.so will relocate its own .data section to the 1146 // lower end of primordial stack; reduce ulimit -s value a little bit 1147 // so we won't install guard page on ld.so's data section. 1148 stack_size -= 2 * page_size(); 1149 1150 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1151 // 7.1, in both cases we will get 2G in return value. 1152 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1153 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1154 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1155 // in case other parts in glibc still assumes 2M max stack size. 1156 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1157 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1158 if (stack_size > 2 * K * K IA64_ONLY(*2)) 1159 stack_size = 2 * K * K IA64_ONLY(*2); 1160 // Try to figure out where the stack base (top) is. This is harder. 1161 // 1162 // When an application is started, glibc saves the initial stack pointer in 1163 // a global variable "__libc_stack_end", which is then used by system 1164 // libraries. __libc_stack_end should be pretty close to stack top. The 1165 // variable is available since the very early days. However, because it is 1166 // a private interface, it could disappear in the future. 1167 // 1168 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1169 // to __libc_stack_end, it is very close to stack top, but isn't the real 1170 // stack top. Note that /proc may not exist if VM is running as a chroot 1171 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1172 // /proc/<pid>/stat could change in the future (though unlikely). 1173 // 1174 // We try __libc_stack_end first. If that doesn't work, look for 1175 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1176 // as a hint, which should work well in most cases. 1177 1178 uintptr_t stack_start; 1179 1180 // try __libc_stack_end first 1181 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1182 if (p && *p) { 1183 stack_start = *p; 1184 } else { 1185 // see if we can get the start_stack field from /proc/self/stat 1186 FILE *fp; 1187 int pid; 1188 char state; 1189 int ppid; 1190 int pgrp; 1191 int session; 1192 int nr; 1193 int tpgrp; 1194 unsigned long flags; 1195 unsigned long minflt; 1196 unsigned long cminflt; 1197 unsigned long majflt; 1198 unsigned long cmajflt; 1199 unsigned long utime; 1200 unsigned long stime; 1201 long cutime; 1202 long cstime; 1203 long prio; 1204 long nice; 1205 long junk; 1206 long it_real; 1207 uintptr_t start; 1208 uintptr_t vsize; 1209 intptr_t rss; 1210 uintptr_t rsslim; 1211 uintptr_t scodes; 1212 uintptr_t ecode; 1213 int i; 1214 1215 // Figure what the primordial thread stack base is. Code is inspired 1216 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1217 // followed by command name surrounded by parentheses, state, etc. 1218 char stat[2048]; 1219 int statlen; 1220 1221 fp = fopen("/proc/self/stat", "r"); 1222 if (fp) { 1223 statlen = fread(stat, 1, 2047, fp); 1224 stat[statlen] = '\0'; 1225 fclose(fp); 1226 1227 // Skip pid and the command string. Note that we could be dealing with 1228 // weird command names, e.g. user could decide to rename java launcher 1229 // to "java 1.4.2 :)", then the stat file would look like 1230 // 1234 (java 1.4.2 :)) R ... ... 1231 // We don't really need to know the command string, just find the last 1232 // occurrence of ")" and then start parsing from there. See bug 4726580. 1233 char * s = strrchr(stat, ')'); 1234 1235 i = 0; 1236 if (s) { 1237 // Skip blank chars 1238 do s++; while (isspace(*s)); 1239 1240#define _UFM UINTX_FORMAT 1241#define _DFM INTX_FORMAT 1242 1243 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1244 /* 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 */ 1245 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, 1246 &state, /* 3 %c */ 1247 &ppid, /* 4 %d */ 1248 &pgrp, /* 5 %d */ 1249 &session, /* 6 %d */ 1250 &nr, /* 7 %d */ 1251 &tpgrp, /* 8 %d */ 1252 &flags, /* 9 %lu */ 1253 &minflt, /* 10 %lu */ 1254 &cminflt, /* 11 %lu */ 1255 &majflt, /* 12 %lu */ 1256 &cmajflt, /* 13 %lu */ 1257 &utime, /* 14 %lu */ 1258 &stime, /* 15 %lu */ 1259 &cutime, /* 16 %ld */ 1260 &cstime, /* 17 %ld */ 1261 &prio, /* 18 %ld */ 1262 &nice, /* 19 %ld */ 1263 &junk, /* 20 %ld */ 1264 &it_real, /* 21 %ld */ 1265 &start, /* 22 UINTX_FORMAT */ 1266 &vsize, /* 23 UINTX_FORMAT */ 1267 &rss, /* 24 INTX_FORMAT */ 1268 &rsslim, /* 25 UINTX_FORMAT */ 1269 &scodes, /* 26 UINTX_FORMAT */ 1270 &ecode, /* 27 UINTX_FORMAT */ 1271 &stack_start); /* 28 UINTX_FORMAT */ 1272 } 1273 1274#undef _UFM 1275#undef _DFM 1276 1277 if (i != 28 - 2) { 1278 assert(false, "Bad conversion from /proc/self/stat"); 1279 // product mode - assume we are the initial thread, good luck in the 1280 // embedded case. 1281 warning("Can't detect initial thread stack location - bad conversion"); 1282 stack_start = (uintptr_t) &rlim; 1283 } 1284 } else { 1285 // For some reason we can't open /proc/self/stat (for example, running on 1286 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1287 // most cases, so don't abort: 1288 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1289 stack_start = (uintptr_t) &rlim; 1290 } 1291 } 1292 1293 // Now we have a pointer (stack_start) very close to the stack top, the 1294 // next thing to do is to figure out the exact location of stack top. We 1295 // can find out the virtual memory area that contains stack_start by 1296 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1297 // and its upper limit is the real stack top. (again, this would fail if 1298 // running inside chroot, because /proc may not exist.) 1299 1300 uintptr_t stack_top; 1301 address low, high; 1302 if (find_vma((address)stack_start, &low, &high)) { 1303 // success, "high" is the true stack top. (ignore "low", because initial 1304 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1305 stack_top = (uintptr_t)high; 1306 } else { 1307 // failed, likely because /proc/self/maps does not exist 1308 warning("Can't detect initial thread stack location - find_vma failed"); 1309 // best effort: stack_start is normally within a few pages below the real 1310 // stack top, use it as stack top, and reduce stack size so we won't put 1311 // guard page outside stack. 1312 stack_top = stack_start; 1313 stack_size -= 16 * page_size(); 1314 } 1315 1316 // stack_top could be partially down the page so align it 1317 stack_top = align_size_up(stack_top, page_size()); 1318 1319 if (max_size && stack_size > max_size) { 1320 _initial_thread_stack_size = max_size; 1321 } else { 1322 _initial_thread_stack_size = stack_size; 1323 } 1324 1325 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1326 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1327} 1328 1329//////////////////////////////////////////////////////////////////////////////// 1330// time support 1331 1332// Time since start-up in seconds to a fine granularity. 1333// Used by VMSelfDestructTimer and the MemProfiler. 1334double os::elapsedTime() { 1335 1336 return (double)(os::elapsed_counter()) * 0.000001; 1337} 1338 1339jlong os::elapsed_counter() { 1340 timeval time; 1341 int status = gettimeofday(&time, NULL); 1342 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1343} 1344 1345jlong os::elapsed_frequency() { 1346 return (1000 * 1000); 1347} 1348 1349bool os::supports_vtime() { return true; } 1350bool os::enable_vtime() { return false; } 1351bool os::vtime_enabled() { return false; } 1352 1353double os::elapsedVTime() { 1354 struct rusage usage; 1355 int retval = getrusage(RUSAGE_THREAD, &usage); 1356 if (retval == 0) { 1357 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000); 1358 } else { 1359 // better than nothing, but not much 1360 return elapsedTime(); 1361 } 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 return; 1404 } else { 1405 // close librt if there is no monotonic clock 1406 dlclose(handle); 1407 } 1408 } 1409 } 1410 warning("No monotonic clock was available - timed services may " \ 1411 "be adversely affected if the time-of-day clock changes"); 1412} 1413 1414#ifndef SYS_clock_getres 1415 1416#if defined(IA32) || defined(AMD64) 1417#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1418#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1419#else 1420#warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1421#define sys_clock_getres(x,y) -1 1422#endif 1423 1424#else 1425#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1426#endif 1427 1428void os::Linux::fast_thread_clock_init() { 1429 if (!UseLinuxPosixThreadCPUClocks) { 1430 return; 1431 } 1432 clockid_t clockid; 1433 struct timespec tp; 1434 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1435 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1436 1437 // Switch to using fast clocks for thread cpu time if 1438 // the sys_clock_getres() returns 0 error code. 1439 // Note, that some kernels may support the current thread 1440 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1441 // returned by the pthread_getcpuclockid(). 1442 // If the fast Posix clocks are supported then the sys_clock_getres() 1443 // must return at least tp.tv_sec == 0 which means a resolution 1444 // better than 1 sec. This is extra check for reliability. 1445 1446 if(pthread_getcpuclockid_func && 1447 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1448 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1449 1450 _supports_fast_thread_cpu_time = true; 1451 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1452 } 1453} 1454 1455jlong os::javaTimeNanos() { 1456 if (Linux::supports_monotonic_clock()) { 1457 struct timespec tp; 1458 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1459 assert(status == 0, "gettime error"); 1460 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1461 return result; 1462 } else { 1463 timeval time; 1464 int status = gettimeofday(&time, NULL); 1465 assert(status != -1, "linux error"); 1466 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1467 return 1000 * usecs; 1468 } 1469} 1470 1471void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1472 if (Linux::supports_monotonic_clock()) { 1473 info_ptr->max_value = ALL_64_BITS; 1474 1475 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1476 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1477 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1478 } else { 1479 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1480 info_ptr->max_value = ALL_64_BITS; 1481 1482 // gettimeofday is a real time clock so it skips 1483 info_ptr->may_skip_backward = true; 1484 info_ptr->may_skip_forward = true; 1485 } 1486 1487 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1488} 1489 1490// Return the real, user, and system times in seconds from an 1491// arbitrary fixed point in the past. 1492bool os::getTimesSecs(double* process_real_time, 1493 double* process_user_time, 1494 double* process_system_time) { 1495 struct tms ticks; 1496 clock_t real_ticks = times(&ticks); 1497 1498 if (real_ticks == (clock_t) (-1)) { 1499 return false; 1500 } else { 1501 double ticks_per_second = (double) clock_tics_per_sec; 1502 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1503 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1504 *process_real_time = ((double) real_ticks) / ticks_per_second; 1505 1506 return true; 1507 } 1508} 1509 1510 1511char * os::local_time_string(char *buf, size_t buflen) { 1512 struct tm t; 1513 time_t long_time; 1514 time(&long_time); 1515 localtime_r(&long_time, &t); 1516 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1517 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1518 t.tm_hour, t.tm_min, t.tm_sec); 1519 return buf; 1520} 1521 1522struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1523 return localtime_r(clock, res); 1524} 1525 1526//////////////////////////////////////////////////////////////////////////////// 1527// runtime exit support 1528 1529// Note: os::shutdown() might be called very early during initialization, or 1530// called from signal handler. Before adding something to os::shutdown(), make 1531// sure it is async-safe and can handle partially initialized VM. 1532void os::shutdown() { 1533 1534 // allow PerfMemory to attempt cleanup of any persistent resources 1535 perfMemory_exit(); 1536 1537 // needs to remove object in file system 1538 AttachListener::abort(); 1539 1540 // flush buffered output, finish log files 1541 ostream_abort(); 1542 1543 // Check for abort hook 1544 abort_hook_t abort_hook = Arguments::abort_hook(); 1545 if (abort_hook != NULL) { 1546 abort_hook(); 1547 } 1548 1549} 1550 1551// Note: os::abort() might be called very early during initialization, or 1552// called from signal handler. Before adding something to os::abort(), make 1553// sure it is async-safe and can handle partially initialized VM. 1554void os::abort(bool dump_core) { 1555 os::shutdown(); 1556 if (dump_core) { 1557#ifndef PRODUCT 1558 fdStream out(defaultStream::output_fd()); 1559 out.print_raw("Current thread is "); 1560 char buf[16]; 1561 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1562 out.print_raw_cr(buf); 1563 out.print_raw_cr("Dumping core ..."); 1564#endif 1565 ::abort(); // dump core 1566 } 1567 1568 ::exit(1); 1569} 1570 1571// Die immediately, no exit hook, no abort hook, no cleanup. 1572void os::die() { 1573 // _exit() on LinuxThreads only kills current thread 1574 ::abort(); 1575} 1576 1577// unused on linux for now. 1578void os::set_error_file(const char *logfile) {} 1579 1580 1581// This method is a copy of JDK's sysGetLastErrorString 1582// from src/solaris/hpi/src/system_md.c 1583 1584size_t os::lasterror(char *buf, size_t len) { 1585 1586 if (errno == 0) return 0; 1587 1588 const char *s = ::strerror(errno); 1589 size_t n = ::strlen(s); 1590 if (n >= len) { 1591 n = len - 1; 1592 } 1593 ::strncpy(buf, s, n); 1594 buf[n] = '\0'; 1595 return n; 1596} 1597 1598intx os::current_thread_id() { return (intx)pthread_self(); } 1599int os::current_process_id() { 1600 1601 // Under the old linux thread library, linux gives each thread 1602 // its own process id. Because of this each thread will return 1603 // a different pid if this method were to return the result 1604 // of getpid(2). Linux provides no api that returns the pid 1605 // of the launcher thread for the vm. This implementation 1606 // returns a unique pid, the pid of the launcher thread 1607 // that starts the vm 'process'. 1608 1609 // Under the NPTL, getpid() returns the same pid as the 1610 // launcher thread rather than a unique pid per thread. 1611 // Use gettid() if you want the old pre NPTL behaviour. 1612 1613 // if you are looking for the result of a call to getpid() that 1614 // returns a unique pid for the calling thread, then look at the 1615 // OSThread::thread_id() method in osThread_linux.hpp file 1616 1617 return (int)(_initial_pid ? _initial_pid : getpid()); 1618} 1619 1620// DLL functions 1621 1622const char* os::dll_file_extension() { return ".so"; } 1623 1624// This must be hard coded because it's the system's temporary 1625// directory not the java application's temp directory, ala java.io.tmpdir. 1626const char* os::get_temp_directory() { return "/tmp"; } 1627 1628static bool file_exists(const char* filename) { 1629 struct stat statbuf; 1630 if (filename == NULL || strlen(filename) == 0) { 1631 return false; 1632 } 1633 return os::stat(filename, &statbuf) == 0; 1634} 1635 1636bool os::dll_build_name(char* buffer, size_t buflen, 1637 const char* pname, const char* fname) { 1638 bool retval = false; 1639 // Copied from libhpi 1640 const size_t pnamelen = pname ? strlen(pname) : 0; 1641 1642 // Return error on buffer overflow. 1643 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1644 return retval; 1645 } 1646 1647 if (pnamelen == 0) { 1648 snprintf(buffer, buflen, "lib%s.so", fname); 1649 retval = true; 1650 } else if (strchr(pname, *os::path_separator()) != NULL) { 1651 int n; 1652 char** pelements = split_path(pname, &n); 1653 if (pelements == NULL) { 1654 return false; 1655 } 1656 for (int i = 0 ; i < n ; i++) { 1657 // Really shouldn't be NULL, but check can't hurt 1658 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1659 continue; // skip the empty path values 1660 } 1661 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1662 if (file_exists(buffer)) { 1663 retval = true; 1664 break; 1665 } 1666 } 1667 // release the storage 1668 for (int i = 0 ; i < n ; i++) { 1669 if (pelements[i] != NULL) { 1670 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1671 } 1672 } 1673 if (pelements != NULL) { 1674 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1675 } 1676 } else { 1677 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1678 retval = true; 1679 } 1680 return retval; 1681} 1682 1683// check if addr is inside libjvm.so 1684bool os::address_is_in_vm(address addr) { 1685 static address libjvm_base_addr; 1686 Dl_info dlinfo; 1687 1688 if (libjvm_base_addr == NULL) { 1689 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1690 libjvm_base_addr = (address)dlinfo.dli_fbase; 1691 } 1692 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1693 } 1694 1695 if (dladdr((void *)addr, &dlinfo) != 0) { 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 // buf is not optional, but offset is optional 1705 assert(buf != NULL, "sanity check"); 1706 1707 Dl_info dlinfo; 1708 1709 if (dladdr((void*)addr, &dlinfo) != 0) { 1710 // see if we have a matching symbol 1711 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1712 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1713 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1714 } 1715 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1716 return true; 1717 } 1718 // no matching symbol so try for just file info 1719 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1720 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1721 buf, buflen, offset, dlinfo.dli_fname)) { 1722 return true; 1723 } 1724 } 1725 } 1726 1727 buf[0] = '\0'; 1728 if (offset != NULL) *offset = -1; 1729 return false; 1730} 1731 1732struct _address_to_library_name { 1733 address addr; // input : memory address 1734 size_t buflen; // size of fname 1735 char* fname; // output: library name 1736 address base; // library base addr 1737}; 1738 1739static int address_to_library_name_callback(struct dl_phdr_info *info, 1740 size_t size, void *data) { 1741 int i; 1742 bool found = false; 1743 address libbase = NULL; 1744 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1745 1746 // iterate through all loadable segments 1747 for (i = 0; i < info->dlpi_phnum; i++) { 1748 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1749 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1750 // base address of a library is the lowest address of its loaded 1751 // segments. 1752 if (libbase == NULL || libbase > segbase) { 1753 libbase = segbase; 1754 } 1755 // see if 'addr' is within current segment 1756 if (segbase <= d->addr && 1757 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1758 found = true; 1759 } 1760 } 1761 } 1762 1763 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1764 // so dll_address_to_library_name() can fall through to use dladdr() which 1765 // can figure out executable name from argv[0]. 1766 if (found && info->dlpi_name && info->dlpi_name[0]) { 1767 d->base = libbase; 1768 if (d->fname) { 1769 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1770 } 1771 return 1; 1772 } 1773 return 0; 1774} 1775 1776bool os::dll_address_to_library_name(address addr, char* buf, 1777 int buflen, int* offset) { 1778 // buf is not optional, but offset is optional 1779 assert(buf != NULL, "sanity check"); 1780 1781 Dl_info dlinfo; 1782 struct _address_to_library_name data; 1783 1784 // There is a bug in old glibc dladdr() implementation that it could resolve 1785 // to wrong library name if the .so file has a base address != NULL. Here 1786 // we iterate through the program headers of all loaded libraries to find 1787 // out which library 'addr' really belongs to. This workaround can be 1788 // removed once the minimum requirement for glibc is moved to 2.3.x. 1789 data.addr = addr; 1790 data.fname = buf; 1791 data.buflen = buflen; 1792 data.base = NULL; 1793 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1794 1795 if (rslt) { 1796 // buf already contains library name 1797 if (offset) *offset = addr - data.base; 1798 return true; 1799 } 1800 if (dladdr((void*)addr, &dlinfo) != 0) { 1801 if (dlinfo.dli_fname != NULL) { 1802 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1803 } 1804 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1805 *offset = addr - (address)dlinfo.dli_fbase; 1806 } 1807 return true; 1808 } 1809 1810 buf[0] = '\0'; 1811 if (offset) *offset = -1; 1812 return false; 1813} 1814 1815 // Loads .dll/.so and 1816 // in case of error it checks if .dll/.so was built for the 1817 // same architecture as Hotspot is running on 1818 1819 1820// Remember the stack's state. The Linux dynamic linker will change 1821// the stack to 'executable' at most once, so we must safepoint only once. 1822bool os::Linux::_stack_is_executable = false; 1823 1824// VM operation that loads a library. This is necessary if stack protection 1825// of the Java stacks can be lost during loading the library. If we 1826// do not stop the Java threads, they can stack overflow before the stacks 1827// are protected again. 1828class VM_LinuxDllLoad: public VM_Operation { 1829 private: 1830 const char *_filename; 1831 char *_ebuf; 1832 int _ebuflen; 1833 void *_lib; 1834 public: 1835 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1836 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1837 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1838 void doit() { 1839 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1840 os::Linux::_stack_is_executable = true; 1841 } 1842 void* loaded_library() { return _lib; } 1843}; 1844 1845void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1846{ 1847 void * result = NULL; 1848 bool load_attempted = false; 1849 1850 // Check whether the library to load might change execution rights 1851 // of the stack. If they are changed, the protection of the stack 1852 // guard pages will be lost. We need a safepoint to fix this. 1853 // 1854 // See Linux man page execstack(8) for more info. 1855 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1856 ElfFile ef(filename); 1857 if (!ef.specifies_noexecstack()) { 1858 if (!is_init_completed()) { 1859 os::Linux::_stack_is_executable = true; 1860 // This is OK - No Java threads have been created yet, and hence no 1861 // stack guard pages to fix. 1862 // 1863 // This should happen only when you are building JDK7 using a very 1864 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1865 // 1866 // Dynamic loader will make all stacks executable after 1867 // this function returns, and will not do that again. 1868 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1869 } else { 1870 warning("You have loaded library %s which might have disabled stack guard. " 1871 "The VM will try to fix the stack guard now.\n" 1872 "It's highly recommended that you fix the library with " 1873 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1874 filename); 1875 1876 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1877 JavaThread *jt = JavaThread::current(); 1878 if (jt->thread_state() != _thread_in_native) { 1879 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1880 // that requires ExecStack. Cannot enter safe point. Let's give up. 1881 warning("Unable to fix stack guard. Giving up."); 1882 } else { 1883 if (!LoadExecStackDllInVMThread) { 1884 // This is for the case where the DLL has an static 1885 // constructor function that executes JNI code. We cannot 1886 // load such DLLs in the VMThread. 1887 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1888 } 1889 1890 ThreadInVMfromNative tiv(jt); 1891 debug_only(VMNativeEntryWrapper vew;) 1892 1893 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1894 VMThread::execute(&op); 1895 if (LoadExecStackDllInVMThread) { 1896 result = op.loaded_library(); 1897 } 1898 load_attempted = true; 1899 } 1900 } 1901 } 1902 } 1903 1904 if (!load_attempted) { 1905 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1906 } 1907 1908 if (result != NULL) { 1909 // Successful loading 1910 return result; 1911 } 1912 1913 Elf32_Ehdr elf_head; 1914 int diag_msg_max_length=ebuflen-strlen(ebuf); 1915 char* diag_msg_buf=ebuf+strlen(ebuf); 1916 1917 if (diag_msg_max_length==0) { 1918 // No more space in ebuf for additional diagnostics message 1919 return NULL; 1920 } 1921 1922 1923 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1924 1925 if (file_descriptor < 0) { 1926 // Can't open library, report dlerror() message 1927 return NULL; 1928 } 1929 1930 bool failed_to_read_elf_head= 1931 (sizeof(elf_head)!= 1932 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1933 1934 ::close(file_descriptor); 1935 if (failed_to_read_elf_head) { 1936 // file i/o error - report dlerror() msg 1937 return NULL; 1938 } 1939 1940 typedef struct { 1941 Elf32_Half code; // Actual value as defined in elf.h 1942 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1943 char elf_class; // 32 or 64 bit 1944 char endianess; // MSB or LSB 1945 char* name; // String representation 1946 } arch_t; 1947 1948 #ifndef EM_486 1949 #define EM_486 6 /* Intel 80486 */ 1950 #endif 1951 1952 static const arch_t arch_array[]={ 1953 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1954 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1955 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1956 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1957 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1958 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1959 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1960 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1961 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1962 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1963 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1964 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1965 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1966 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1967 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1968 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1969 }; 1970 1971 #if (defined IA32) 1972 static Elf32_Half running_arch_code=EM_386; 1973 #elif (defined AMD64) 1974 static Elf32_Half running_arch_code=EM_X86_64; 1975 #elif (defined IA64) 1976 static Elf32_Half running_arch_code=EM_IA_64; 1977 #elif (defined __sparc) && (defined _LP64) 1978 static Elf32_Half running_arch_code=EM_SPARCV9; 1979 #elif (defined __sparc) && (!defined _LP64) 1980 static Elf32_Half running_arch_code=EM_SPARC; 1981 #elif (defined __powerpc64__) 1982 static Elf32_Half running_arch_code=EM_PPC64; 1983 #elif (defined __powerpc__) 1984 static Elf32_Half running_arch_code=EM_PPC; 1985 #elif (defined ARM) 1986 static Elf32_Half running_arch_code=EM_ARM; 1987 #elif (defined S390) 1988 static Elf32_Half running_arch_code=EM_S390; 1989 #elif (defined ALPHA) 1990 static Elf32_Half running_arch_code=EM_ALPHA; 1991 #elif (defined MIPSEL) 1992 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1993 #elif (defined PARISC) 1994 static Elf32_Half running_arch_code=EM_PARISC; 1995 #elif (defined MIPS) 1996 static Elf32_Half running_arch_code=EM_MIPS; 1997 #elif (defined M68K) 1998 static Elf32_Half running_arch_code=EM_68K; 1999 #else 2000 #error Method os::dll_load requires that one of following is defined:\ 2001 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 2002 #endif 2003 2004 // Identify compatability class for VM's architecture and library's architecture 2005 // Obtain string descriptions for architectures 2006 2007 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 2008 int running_arch_index=-1; 2009 2010 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 2011 if (running_arch_code == arch_array[i].code) { 2012 running_arch_index = i; 2013 } 2014 if (lib_arch.code == arch_array[i].code) { 2015 lib_arch.compat_class = arch_array[i].compat_class; 2016 lib_arch.name = arch_array[i].name; 2017 } 2018 } 2019 2020 assert(running_arch_index != -1, 2021 "Didn't find running architecture code (running_arch_code) in arch_array"); 2022 if (running_arch_index == -1) { 2023 // Even though running architecture detection failed 2024 // we may still continue with reporting dlerror() message 2025 return NULL; 2026 } 2027 2028 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 2029 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 2030 return NULL; 2031 } 2032 2033#ifndef S390 2034 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 2035 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 2036 return NULL; 2037 } 2038#endif // !S390 2039 2040 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2041 if ( lib_arch.name!=NULL ) { 2042 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2043 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2044 lib_arch.name, arch_array[running_arch_index].name); 2045 } else { 2046 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2047 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2048 lib_arch.code, 2049 arch_array[running_arch_index].name); 2050 } 2051 } 2052 2053 return NULL; 2054} 2055 2056void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) { 2057 void * result = ::dlopen(filename, RTLD_LAZY); 2058 if (result == NULL) { 2059 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2060 ebuf[ebuflen-1] = '\0'; 2061 } 2062 return result; 2063} 2064 2065void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) { 2066 void * result = NULL; 2067 if (LoadExecStackDllInVMThread) { 2068 result = dlopen_helper(filename, ebuf, ebuflen); 2069 } 2070 2071 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2072 // library that requires an executable stack, or which does not have this 2073 // stack attribute set, dlopen changes the stack attribute to executable. The 2074 // read protection of the guard pages gets lost. 2075 // 2076 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2077 // may have been queued at the same time. 2078 2079 if (!_stack_is_executable) { 2080 JavaThread *jt = Threads::first(); 2081 2082 while (jt) { 2083 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2084 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2085 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2086 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2087 warning("Attempt to reguard stack yellow zone failed."); 2088 } 2089 } 2090 jt = jt->next(); 2091 } 2092 } 2093 2094 return result; 2095} 2096 2097/* 2098 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2099 * chances are you might want to run the generated bits against glibc-2.0 2100 * libdl.so, so always use locking for any version of glibc. 2101 */ 2102void* os::dll_lookup(void* handle, const char* name) { 2103 pthread_mutex_lock(&dl_mutex); 2104 void* res = dlsym(handle, name); 2105 pthread_mutex_unlock(&dl_mutex); 2106 return res; 2107} 2108 2109 2110static bool _print_ascii_file(const char* filename, outputStream* st) { 2111 int fd = ::open(filename, O_RDONLY); 2112 if (fd == -1) { 2113 return false; 2114 } 2115 2116 char buf[32]; 2117 int bytes; 2118 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2119 st->print_raw(buf, bytes); 2120 } 2121 2122 ::close(fd); 2123 2124 return true; 2125} 2126 2127void os::print_dll_info(outputStream *st) { 2128 st->print_cr("Dynamic libraries:"); 2129 2130 char fname[32]; 2131 pid_t pid = os::Linux::gettid(); 2132 2133 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2134 2135 if (!_print_ascii_file(fname, st)) { 2136 st->print("Can not get library information for pid = %d\n", pid); 2137 } 2138} 2139 2140void os::print_os_info_brief(outputStream* st) { 2141 os::Linux::print_distro_info(st); 2142 2143 os::Posix::print_uname_info(st); 2144 2145 os::Linux::print_libversion_info(st); 2146 2147} 2148 2149void os::print_os_info(outputStream* st) { 2150 st->print("OS:"); 2151 2152 os::Linux::print_distro_info(st); 2153 2154 os::Posix::print_uname_info(st); 2155 2156 // Print warning if unsafe chroot environment detected 2157 if (unsafe_chroot_detected) { 2158 st->print("WARNING!! "); 2159 st->print_cr(unstable_chroot_error); 2160 } 2161 2162 os::Linux::print_libversion_info(st); 2163 2164 os::Posix::print_rlimit_info(st); 2165 2166 os::Posix::print_load_average(st); 2167 2168 os::Linux::print_full_memory_info(st); 2169} 2170 2171// Try to identify popular distros. 2172// Most Linux distributions have a /etc/XXX-release file, which contains 2173// the OS version string. Newer Linux distributions have a /etc/lsb-release 2174// file that also contains the OS version string. Some have more than one 2175// /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2176// /etc/redhat-release.), so the order is important. 2177// Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2178// their own specific XXX-release file as well as a redhat-release file. 2179// Because of this the XXX-release file needs to be searched for before the 2180// redhat-release file. 2181// Since Red Hat has a lsb-release file that is not very descriptive the 2182// search for redhat-release needs to be before lsb-release. 2183// Since the lsb-release file is the new standard it needs to be searched 2184// before the older style release files. 2185// Searching system-release (Red Hat) and os-release (other Linuxes) are a 2186// next to last resort. The os-release file is a new standard that contains 2187// distribution information and the system-release file seems to be an old 2188// standard that has been replaced by the lsb-release and os-release files. 2189// Searching for the debian_version file is the last resort. It contains 2190// an informative string like "6.0.6" or "wheezy/sid". Because of this 2191// "Debian " is printed before the contents of the debian_version file. 2192void os::Linux::print_distro_info(outputStream* st) { 2193 if (!_print_ascii_file("/etc/oracle-release", st) && 2194 !_print_ascii_file("/etc/mandriva-release", st) && 2195 !_print_ascii_file("/etc/mandrake-release", st) && 2196 !_print_ascii_file("/etc/sun-release", st) && 2197 !_print_ascii_file("/etc/redhat-release", st) && 2198 !_print_ascii_file("/etc/lsb-release", st) && 2199 !_print_ascii_file("/etc/SuSE-release", st) && 2200 !_print_ascii_file("/etc/turbolinux-release", st) && 2201 !_print_ascii_file("/etc/gentoo-release", st) && 2202 !_print_ascii_file("/etc/ltib-release", st) && 2203 !_print_ascii_file("/etc/angstrom-version", st) && 2204 !_print_ascii_file("/etc/system-release", st) && 2205 !_print_ascii_file("/etc/os-release", st)) { 2206 2207 if (file_exists("/etc/debian_version")) { 2208 st->print("Debian "); 2209 _print_ascii_file("/etc/debian_version", st); 2210 } else { 2211 st->print("Linux"); 2212 } 2213 } 2214 st->cr(); 2215} 2216 2217void os::Linux::print_libversion_info(outputStream* st) { 2218 // libc, pthread 2219 st->print("libc:"); 2220 st->print(os::Linux::glibc_version()); st->print(" "); 2221 st->print(os::Linux::libpthread_version()); st->print(" "); 2222 if (os::Linux::is_LinuxThreads()) { 2223 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2224 } 2225 st->cr(); 2226} 2227 2228void os::Linux::print_full_memory_info(outputStream* st) { 2229 st->print("\n/proc/meminfo:\n"); 2230 _print_ascii_file("/proc/meminfo", st); 2231 st->cr(); 2232} 2233 2234void os::print_memory_info(outputStream* st) { 2235 2236 st->print("Memory:"); 2237 st->print(" %dk page", os::vm_page_size()>>10); 2238 2239 // values in struct sysinfo are "unsigned long" 2240 struct sysinfo si; 2241 sysinfo(&si); 2242 2243 st->print(", physical " UINT64_FORMAT "k", 2244 os::physical_memory() >> 10); 2245 st->print("(" UINT64_FORMAT "k free)", 2246 os::available_memory() >> 10); 2247 st->print(", swap " UINT64_FORMAT "k", 2248 ((jlong)si.totalswap * si.mem_unit) >> 10); 2249 st->print("(" UINT64_FORMAT "k free)", 2250 ((jlong)si.freeswap * si.mem_unit) >> 10); 2251 st->cr(); 2252} 2253 2254void os::pd_print_cpu_info(outputStream* st) { 2255 st->print("\n/proc/cpuinfo:\n"); 2256 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2257 st->print(" <Not Available>"); 2258 } 2259 st->cr(); 2260} 2261 2262// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 2263// but they're the same for all the linux arch that we support 2264// and they're the same for solaris but there's no common place to put this. 2265const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 2266 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 2267 "ILL_COPROC", "ILL_BADSTK" }; 2268 2269const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 2270 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 2271 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 2272 2273const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 2274 2275const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 2276 2277void os::print_siginfo(outputStream* st, void* siginfo) { 2278 st->print("siginfo:"); 2279 2280 const int buflen = 100; 2281 char buf[buflen]; 2282 siginfo_t *si = (siginfo_t*)siginfo; 2283 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 2284 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 2285 st->print("si_errno=%s", buf); 2286 } else { 2287 st->print("si_errno=%d", si->si_errno); 2288 } 2289 const int c = si->si_code; 2290 assert(c > 0, "unexpected si_code"); 2291 switch (si->si_signo) { 2292 case SIGILL: 2293 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 2294 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2295 break; 2296 case SIGFPE: 2297 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 2298 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2299 break; 2300 case SIGSEGV: 2301 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 2302 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2303 break; 2304 case SIGBUS: 2305 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 2306 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2307 break; 2308 default: 2309 st->print(", si_code=%d", si->si_code); 2310 // no si_addr 2311 } 2312 2313 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2314 UseSharedSpaces) { 2315 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2316 if (mapinfo->is_in_shared_space(si->si_addr)) { 2317 st->print("\n\nError accessing class data sharing archive." \ 2318 " Mapped file inaccessible during execution, " \ 2319 " possible disk/network problem."); 2320 } 2321 } 2322 st->cr(); 2323} 2324 2325 2326static void print_signal_handler(outputStream* st, int sig, 2327 char* buf, size_t buflen); 2328 2329void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2330 st->print_cr("Signal Handlers:"); 2331 print_signal_handler(st, SIGSEGV, buf, buflen); 2332 print_signal_handler(st, SIGBUS , buf, buflen); 2333 print_signal_handler(st, SIGFPE , buf, buflen); 2334 print_signal_handler(st, SIGPIPE, buf, buflen); 2335 print_signal_handler(st, SIGXFSZ, buf, buflen); 2336 print_signal_handler(st, SIGILL , buf, buflen); 2337 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2338 print_signal_handler(st, SR_signum, buf, buflen); 2339 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2340 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2341 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2342 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2343} 2344 2345static char saved_jvm_path[MAXPATHLEN] = {0}; 2346 2347// Find the full path to the current module, libjvm.so 2348void os::jvm_path(char *buf, jint buflen) { 2349 // Error checking. 2350 if (buflen < MAXPATHLEN) { 2351 assert(false, "must use a large-enough buffer"); 2352 buf[0] = '\0'; 2353 return; 2354 } 2355 // Lazy resolve the path to current module. 2356 if (saved_jvm_path[0] != 0) { 2357 strcpy(buf, saved_jvm_path); 2358 return; 2359 } 2360 2361 char dli_fname[MAXPATHLEN]; 2362 bool ret = dll_address_to_library_name( 2363 CAST_FROM_FN_PTR(address, os::jvm_path), 2364 dli_fname, sizeof(dli_fname), NULL); 2365 assert(ret, "cannot locate libjvm"); 2366 char *rp = NULL; 2367 if (ret && dli_fname[0] != '\0') { 2368 rp = realpath(dli_fname, buf); 2369 } 2370 if (rp == NULL) 2371 return; 2372 2373 if (Arguments::created_by_gamma_launcher()) { 2374 // Support for the gamma launcher. Typical value for buf is 2375 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 2376 // the right place in the string, then assume we are installed in a JDK and 2377 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2378 // up the path so it looks like libjvm.so is installed there (append a 2379 // fake suffix hotspot/libjvm.so). 2380 const char *p = buf + strlen(buf) - 1; 2381 for (int count = 0; p > buf && count < 5; ++count) { 2382 for (--p; p > buf && *p != '/'; --p) 2383 /* empty */ ; 2384 } 2385 2386 if (strncmp(p, "/jre/lib/", 9) != 0) { 2387 // Look for JAVA_HOME in the environment. 2388 char* java_home_var = ::getenv("JAVA_HOME"); 2389 if (java_home_var != NULL && java_home_var[0] != 0) { 2390 char* jrelib_p; 2391 int len; 2392 2393 // Check the current module name "libjvm.so". 2394 p = strrchr(buf, '/'); 2395 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2396 2397 rp = realpath(java_home_var, buf); 2398 if (rp == NULL) 2399 return; 2400 2401 // determine if this is a legacy image or modules image 2402 // modules image doesn't have "jre" subdirectory 2403 len = strlen(buf); 2404 jrelib_p = buf + len; 2405 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2406 if (0 != access(buf, F_OK)) { 2407 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2408 } 2409 2410 if (0 == access(buf, F_OK)) { 2411 // Use current module name "libjvm.so" 2412 len = strlen(buf); 2413 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2414 } else { 2415 // Go back to path of .so 2416 rp = realpath(dli_fname, buf); 2417 if (rp == NULL) 2418 return; 2419 } 2420 } 2421 } 2422 } 2423 2424 strcpy(saved_jvm_path, buf); 2425} 2426 2427void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2428 // no prefix required, not even "_" 2429} 2430 2431void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2432 // no suffix required 2433} 2434 2435//////////////////////////////////////////////////////////////////////////////// 2436// sun.misc.Signal support 2437 2438static volatile jint sigint_count = 0; 2439 2440static void 2441UserHandler(int sig, void *siginfo, void *context) { 2442 // 4511530 - sem_post is serialized and handled by the manager thread. When 2443 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2444 // don't want to flood the manager thread with sem_post requests. 2445 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2446 return; 2447 2448 // Ctrl-C is pressed during error reporting, likely because the error 2449 // handler fails to abort. Let VM die immediately. 2450 if (sig == SIGINT && is_error_reported()) { 2451 os::die(); 2452 } 2453 2454 os::signal_notify(sig); 2455} 2456 2457void* os::user_handler() { 2458 return CAST_FROM_FN_PTR(void*, UserHandler); 2459} 2460 2461class Semaphore : public StackObj { 2462 public: 2463 Semaphore(); 2464 ~Semaphore(); 2465 void signal(); 2466 void wait(); 2467 bool trywait(); 2468 bool timedwait(unsigned int sec, int nsec); 2469 private: 2470 sem_t _semaphore; 2471}; 2472 2473 2474Semaphore::Semaphore() { 2475 sem_init(&_semaphore, 0, 0); 2476} 2477 2478Semaphore::~Semaphore() { 2479 sem_destroy(&_semaphore); 2480} 2481 2482void Semaphore::signal() { 2483 sem_post(&_semaphore); 2484} 2485 2486void Semaphore::wait() { 2487 sem_wait(&_semaphore); 2488} 2489 2490bool Semaphore::trywait() { 2491 return sem_trywait(&_semaphore) == 0; 2492} 2493 2494bool Semaphore::timedwait(unsigned int sec, int nsec) { 2495 struct timespec ts; 2496 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec); 2497 2498 while (1) { 2499 int result = sem_timedwait(&_semaphore, &ts); 2500 if (result == 0) { 2501 return true; 2502 } else if (errno == EINTR) { 2503 continue; 2504 } else if (errno == ETIMEDOUT) { 2505 return false; 2506 } else { 2507 return false; 2508 } 2509 } 2510} 2511 2512extern "C" { 2513 typedef void (*sa_handler_t)(int); 2514 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2515} 2516 2517void* os::signal(int signal_number, void* handler) { 2518 struct sigaction sigAct, oldSigAct; 2519 2520 sigfillset(&(sigAct.sa_mask)); 2521 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2522 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2523 2524 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2525 // -1 means registration failed 2526 return (void *)-1; 2527 } 2528 2529 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2530} 2531 2532void os::signal_raise(int signal_number) { 2533 ::raise(signal_number); 2534} 2535 2536/* 2537 * The following code is moved from os.cpp for making this 2538 * code platform specific, which it is by its very nature. 2539 */ 2540 2541// Will be modified when max signal is changed to be dynamic 2542int os::sigexitnum_pd() { 2543 return NSIG; 2544} 2545 2546// a counter for each possible signal value 2547static volatile jint pending_signals[NSIG+1] = { 0 }; 2548 2549// Linux(POSIX) specific hand shaking semaphore. 2550static sem_t sig_sem; 2551static Semaphore sr_semaphore; 2552 2553void os::signal_init_pd() { 2554 // Initialize signal structures 2555 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2556 2557 // Initialize signal semaphore 2558 ::sem_init(&sig_sem, 0, 0); 2559} 2560 2561void os::signal_notify(int sig) { 2562 Atomic::inc(&pending_signals[sig]); 2563 ::sem_post(&sig_sem); 2564} 2565 2566static int check_pending_signals(bool wait) { 2567 Atomic::store(0, &sigint_count); 2568 for (;;) { 2569 for (int i = 0; i < NSIG + 1; i++) { 2570 jint n = pending_signals[i]; 2571 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2572 return i; 2573 } 2574 } 2575 if (!wait) { 2576 return -1; 2577 } 2578 JavaThread *thread = JavaThread::current(); 2579 ThreadBlockInVM tbivm(thread); 2580 2581 bool threadIsSuspended; 2582 do { 2583 thread->set_suspend_equivalent(); 2584 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2585 ::sem_wait(&sig_sem); 2586 2587 // were we externally suspended while we were waiting? 2588 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2589 if (threadIsSuspended) { 2590 // 2591 // The semaphore has been incremented, but while we were waiting 2592 // another thread suspended us. We don't want to continue running 2593 // while suspended because that would surprise the thread that 2594 // suspended us. 2595 // 2596 ::sem_post(&sig_sem); 2597 2598 thread->java_suspend_self(); 2599 } 2600 } while (threadIsSuspended); 2601 } 2602} 2603 2604int os::signal_lookup() { 2605 return check_pending_signals(false); 2606} 2607 2608int os::signal_wait() { 2609 return check_pending_signals(true); 2610} 2611 2612//////////////////////////////////////////////////////////////////////////////// 2613// Virtual Memory 2614 2615int os::vm_page_size() { 2616 // Seems redundant as all get out 2617 assert(os::Linux::page_size() != -1, "must call os::init"); 2618 return os::Linux::page_size(); 2619} 2620 2621// Solaris allocates memory by pages. 2622int os::vm_allocation_granularity() { 2623 assert(os::Linux::page_size() != -1, "must call os::init"); 2624 return os::Linux::page_size(); 2625} 2626 2627// Rationale behind this function: 2628// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2629// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2630// samples for JITted code. Here we create private executable mapping over the code cache 2631// and then we can use standard (well, almost, as mapping can change) way to provide 2632// info for the reporting script by storing timestamp and location of symbol 2633void linux_wrap_code(char* base, size_t size) { 2634 static volatile jint cnt = 0; 2635 2636 if (!UseOprofile) { 2637 return; 2638 } 2639 2640 char buf[PATH_MAX+1]; 2641 int num = Atomic::add(1, &cnt); 2642 2643 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2644 os::get_temp_directory(), os::current_process_id(), num); 2645 unlink(buf); 2646 2647 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2648 2649 if (fd != -1) { 2650 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2651 if (rv != (off_t)-1) { 2652 if (::write(fd, "", 1) == 1) { 2653 mmap(base, size, 2654 PROT_READ|PROT_WRITE|PROT_EXEC, 2655 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2656 } 2657 } 2658 ::close(fd); 2659 unlink(buf); 2660 } 2661} 2662 2663static bool recoverable_mmap_error(int err) { 2664 // See if the error is one we can let the caller handle. This 2665 // list of errno values comes from JBS-6843484. I can't find a 2666 // Linux man page that documents this specific set of errno 2667 // values so while this list currently matches Solaris, it may 2668 // change as we gain experience with this failure mode. 2669 switch (err) { 2670 case EBADF: 2671 case EINVAL: 2672 case ENOTSUP: 2673 // let the caller deal with these errors 2674 return true; 2675 2676 default: 2677 // Any remaining errors on this OS can cause our reserved mapping 2678 // to be lost. That can cause confusion where different data 2679 // structures think they have the same memory mapped. The worst 2680 // scenario is if both the VM and a library think they have the 2681 // same memory mapped. 2682 return false; 2683 } 2684} 2685 2686static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2687 int err) { 2688 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2689 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2690 strerror(err), err); 2691} 2692 2693static void warn_fail_commit_memory(char* addr, size_t size, 2694 size_t alignment_hint, bool exec, 2695 int err) { 2696 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2697 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2698 alignment_hint, exec, strerror(err), err); 2699} 2700 2701// NOTE: Linux kernel does not really reserve the pages for us. 2702// All it does is to check if there are enough free pages 2703// left at the time of mmap(). This could be a potential 2704// problem. 2705int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2706 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2707 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2708 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2709 if (res != (uintptr_t) MAP_FAILED) { 2710 if (UseNUMAInterleaving) { 2711 numa_make_global(addr, size); 2712 } 2713 return 0; 2714 } 2715 2716 int err = errno; // save errno from mmap() call above 2717 2718 if (!recoverable_mmap_error(err)) { 2719 warn_fail_commit_memory(addr, size, exec, err); 2720 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2721 } 2722 2723 return err; 2724} 2725 2726bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2727 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2728} 2729 2730void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2731 const char* mesg) { 2732 assert(mesg != NULL, "mesg must be specified"); 2733 int err = os::Linux::commit_memory_impl(addr, size, exec); 2734 if (err != 0) { 2735 // the caller wants all commit errors to exit with the specified mesg: 2736 warn_fail_commit_memory(addr, size, exec, err); 2737 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2738 } 2739} 2740 2741// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2742#ifndef MAP_HUGETLB 2743#define MAP_HUGETLB 0x40000 2744#endif 2745 2746// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2747#ifndef MADV_HUGEPAGE 2748#define MADV_HUGEPAGE 14 2749#endif 2750 2751int os::Linux::commit_memory_impl(char* addr, size_t size, 2752 size_t alignment_hint, bool exec) { 2753 int err = os::Linux::commit_memory_impl(addr, size, exec); 2754 if (err == 0) { 2755 realign_memory(addr, size, alignment_hint); 2756 } 2757 return err; 2758} 2759 2760bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2761 bool exec) { 2762 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2763} 2764 2765void os::pd_commit_memory_or_exit(char* addr, size_t size, 2766 size_t alignment_hint, bool exec, 2767 const char* mesg) { 2768 assert(mesg != NULL, "mesg must be specified"); 2769 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2770 if (err != 0) { 2771 // the caller wants all commit errors to exit with the specified mesg: 2772 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2773 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2774 } 2775} 2776 2777void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2778 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2779 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2780 // be supported or the memory may already be backed by huge pages. 2781 ::madvise(addr, bytes, MADV_HUGEPAGE); 2782 } 2783} 2784 2785void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2786 // This method works by doing an mmap over an existing mmaping and effectively discarding 2787 // the existing pages. However it won't work for SHM-based large pages that cannot be 2788 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2789 // small pages on top of the SHM segment. This method always works for small pages, so we 2790 // allow that in any case. 2791 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2792 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2793 } 2794} 2795 2796void os::numa_make_global(char *addr, size_t bytes) { 2797 Linux::numa_interleave_memory(addr, bytes); 2798} 2799 2800// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2801// bind policy to MPOL_PREFERRED for the current thread. 2802#define USE_MPOL_PREFERRED 0 2803 2804void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2805 // To make NUMA and large pages more robust when both enabled, we need to ease 2806 // the requirements on where the memory should be allocated. MPOL_BIND is the 2807 // default policy and it will force memory to be allocated on the specified 2808 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2809 // the specified node, but will not force it. Using this policy will prevent 2810 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2811 // free large pages. 2812 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2813 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2814} 2815 2816bool os::numa_topology_changed() { return false; } 2817 2818size_t os::numa_get_groups_num() { 2819 int max_node = Linux::numa_max_node(); 2820 return max_node > 0 ? max_node + 1 : 1; 2821} 2822 2823int os::numa_get_group_id() { 2824 int cpu_id = Linux::sched_getcpu(); 2825 if (cpu_id != -1) { 2826 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2827 if (lgrp_id != -1) { 2828 return lgrp_id; 2829 } 2830 } 2831 return 0; 2832} 2833 2834size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2835 for (size_t i = 0; i < size; i++) { 2836 ids[i] = i; 2837 } 2838 return size; 2839} 2840 2841bool os::get_page_info(char *start, page_info* info) { 2842 return false; 2843} 2844 2845char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2846 return end; 2847} 2848 2849 2850int os::Linux::sched_getcpu_syscall(void) { 2851 unsigned int cpu; 2852 int retval = -1; 2853 2854#if defined(IA32) 2855# ifndef SYS_getcpu 2856# define SYS_getcpu 318 2857# endif 2858 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2859#elif defined(AMD64) 2860// Unfortunately we have to bring all these macros here from vsyscall.h 2861// to be able to compile on old linuxes. 2862# define __NR_vgetcpu 2 2863# define VSYSCALL_START (-10UL << 20) 2864# define VSYSCALL_SIZE 1024 2865# define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2866 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2867 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2868 retval = vgetcpu(&cpu, NULL, NULL); 2869#endif 2870 2871 return (retval == -1) ? retval : cpu; 2872} 2873 2874// Something to do with the numa-aware allocator needs these symbols 2875extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2876extern "C" JNIEXPORT void numa_error(char *where) { } 2877extern "C" JNIEXPORT int fork1() { return fork(); } 2878 2879 2880// If we are running with libnuma version > 2, then we should 2881// be trying to use symbols with versions 1.1 2882// If we are running with earlier version, which did not have symbol versions, 2883// we should use the base version. 2884void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2885 void *f = dlvsym(handle, name, "libnuma_1.1"); 2886 if (f == NULL) { 2887 f = dlsym(handle, name); 2888 } 2889 return f; 2890} 2891 2892bool os::Linux::libnuma_init() { 2893 // sched_getcpu() should be in libc. 2894 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2895 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2896 2897 // If it's not, try a direct syscall. 2898 if (sched_getcpu() == -1) 2899 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall)); 2900 2901 if (sched_getcpu() != -1) { // Does it work? 2902 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2903 if (handle != NULL) { 2904 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2905 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2906 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2907 libnuma_dlsym(handle, "numa_max_node"))); 2908 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2909 libnuma_dlsym(handle, "numa_available"))); 2910 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2911 libnuma_dlsym(handle, "numa_tonode_memory"))); 2912 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2913 libnuma_dlsym(handle, "numa_interleave_memory"))); 2914 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2915 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2916 2917 2918 if (numa_available() != -1) { 2919 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2920 // Create a cpu -> node mapping 2921 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2922 rebuild_cpu_to_node_map(); 2923 return true; 2924 } 2925 } 2926 } 2927 return false; 2928} 2929 2930// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2931// The table is later used in get_node_by_cpu(). 2932void os::Linux::rebuild_cpu_to_node_map() { 2933 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2934 // in libnuma (possible values are starting from 16, 2935 // and continuing up with every other power of 2, but less 2936 // than the maximum number of CPUs supported by kernel), and 2937 // is a subject to change (in libnuma version 2 the requirements 2938 // are more reasonable) we'll just hardcode the number they use 2939 // in the library. 2940 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2941 2942 size_t cpu_num = os::active_processor_count(); 2943 size_t cpu_map_size = NCPUS / BitsPerCLong; 2944 size_t cpu_map_valid_size = 2945 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2946 2947 cpu_to_node()->clear(); 2948 cpu_to_node()->at_grow(cpu_num - 1); 2949 size_t node_num = numa_get_groups_num(); 2950 2951 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2952 for (size_t i = 0; i < node_num; i++) { 2953 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2954 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2955 if (cpu_map[j] != 0) { 2956 for (size_t k = 0; k < BitsPerCLong; k++) { 2957 if (cpu_map[j] & (1UL << k)) { 2958 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2959 } 2960 } 2961 } 2962 } 2963 } 2964 } 2965 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2966} 2967 2968int os::Linux::get_node_by_cpu(int cpu_id) { 2969 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2970 return cpu_to_node()->at(cpu_id); 2971 } 2972 return -1; 2973} 2974 2975GrowableArray<int>* os::Linux::_cpu_to_node; 2976os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2977os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2978os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2979os::Linux::numa_available_func_t os::Linux::_numa_available; 2980os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2981os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2982os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2983unsigned long* os::Linux::_numa_all_nodes; 2984 2985bool os::pd_uncommit_memory(char* addr, size_t size) { 2986 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2987 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2988 return res != (uintptr_t) MAP_FAILED; 2989} 2990 2991static 2992address get_stack_commited_bottom(address bottom, size_t size) { 2993 address nbot = bottom; 2994 address ntop = bottom + size; 2995 2996 size_t page_sz = os::vm_page_size(); 2997 unsigned pages = size / page_sz; 2998 2999 unsigned char vec[1]; 3000 unsigned imin = 1, imax = pages + 1, imid; 3001 int mincore_return_value; 3002 3003 while (imin < imax) { 3004 imid = (imax + imin) / 2; 3005 nbot = ntop - (imid * page_sz); 3006 3007 // Use a trick with mincore to check whether the page is mapped or not. 3008 // mincore sets vec to 1 if page resides in memory and to 0 if page 3009 // is swapped output but if page we are asking for is unmapped 3010 // it returns -1,ENOMEM 3011 mincore_return_value = mincore(nbot, page_sz, vec); 3012 3013 if (mincore_return_value == -1) { 3014 // Page is not mapped go up 3015 // to find first mapped page 3016 if (errno != EAGAIN) { 3017 assert(errno == ENOMEM, "Unexpected mincore errno"); 3018 imax = imid; 3019 } 3020 } else { 3021 // Page is mapped go down 3022 // to find first not mapped page 3023 imin = imid + 1; 3024 } 3025 } 3026 3027 nbot = nbot + page_sz; 3028 3029 // Adjust stack bottom one page up if last checked page is not mapped 3030 if (mincore_return_value == -1) { 3031 nbot = nbot + page_sz; 3032 } 3033 3034 return nbot; 3035} 3036 3037 3038// Linux uses a growable mapping for the stack, and if the mapping for 3039// the stack guard pages is not removed when we detach a thread the 3040// stack cannot grow beyond the pages where the stack guard was 3041// mapped. If at some point later in the process the stack expands to 3042// that point, the Linux kernel cannot expand the stack any further 3043// because the guard pages are in the way, and a segfault occurs. 3044// 3045// However, it's essential not to split the stack region by unmapping 3046// a region (leaving a hole) that's already part of the stack mapping, 3047// so if the stack mapping has already grown beyond the guard pages at 3048// the time we create them, we have to truncate the stack mapping. 3049// So, we need to know the extent of the stack mapping when 3050// create_stack_guard_pages() is called. 3051 3052// We only need this for stacks that are growable: at the time of 3053// writing thread stacks don't use growable mappings (i.e. those 3054// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3055// only applies to the main thread. 3056 3057// If the (growable) stack mapping already extends beyond the point 3058// where we're going to put our guard pages, truncate the mapping at 3059// that point by munmap()ping it. This ensures that when we later 3060// munmap() the guard pages we don't leave a hole in the stack 3061// mapping. This only affects the main/initial thread 3062 3063bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3064 3065 if (os::Linux::is_initial_thread()) { 3066 // As we manually grow stack up to bottom inside create_attached_thread(), 3067 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3068 // we don't need to do anything special. 3069 // Check it first, before calling heavy function. 3070 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3071 unsigned char vec[1]; 3072 3073 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3074 // Fallback to slow path on all errors, including EAGAIN 3075 stack_extent = (uintptr_t) get_stack_commited_bottom( 3076 os::Linux::initial_thread_stack_bottom(), 3077 (size_t)addr - stack_extent); 3078 } 3079 3080 if (stack_extent < (uintptr_t)addr) { 3081 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3082 } 3083 } 3084 3085 return os::commit_memory(addr, size, !ExecMem); 3086} 3087 3088// If this is a growable mapping, remove the guard pages entirely by 3089// munmap()ping them. If not, just call uncommit_memory(). This only 3090// affects the main/initial thread, but guard against future OS changes 3091// It's safe to always unmap guard pages for initial thread because we 3092// always place it right after end of the mapped region 3093 3094bool os::remove_stack_guard_pages(char* addr, size_t size) { 3095 uintptr_t stack_extent, stack_base; 3096 3097 if (os::Linux::is_initial_thread()) { 3098 return ::munmap(addr, size) == 0; 3099 } 3100 3101 return os::uncommit_memory(addr, size); 3102} 3103 3104static address _highest_vm_reserved_address = NULL; 3105 3106// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3107// at 'requested_addr'. If there are existing memory mappings at the same 3108// location, however, they will be overwritten. If 'fixed' is false, 3109// 'requested_addr' is only treated as a hint, the return value may or 3110// may not start from the requested address. Unlike Linux mmap(), this 3111// function returns NULL to indicate failure. 3112static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3113 char * addr; 3114 int flags; 3115 3116 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3117 if (fixed) { 3118 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3119 flags |= MAP_FIXED; 3120 } 3121 3122 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3123 // touch an uncommitted page. Otherwise, the read/write might 3124 // succeed if we have enough swap space to back the physical page. 3125 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3126 flags, -1, 0); 3127 3128 if (addr != MAP_FAILED) { 3129 // anon_mmap() should only get called during VM initialization, 3130 // don't need lock (actually we can skip locking even it can be called 3131 // from multiple threads, because _highest_vm_reserved_address is just a 3132 // hint about the upper limit of non-stack memory regions.) 3133 if ((address)addr + bytes > _highest_vm_reserved_address) { 3134 _highest_vm_reserved_address = (address)addr + bytes; 3135 } 3136 } 3137 3138 return addr == MAP_FAILED ? NULL : addr; 3139} 3140 3141// Don't update _highest_vm_reserved_address, because there might be memory 3142// regions above addr + size. If so, releasing a memory region only creates 3143// a hole in the address space, it doesn't help prevent heap-stack collision. 3144// 3145static int anon_munmap(char * addr, size_t size) { 3146 return ::munmap(addr, size) == 0; 3147} 3148 3149char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3150 size_t alignment_hint) { 3151 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3152} 3153 3154bool os::pd_release_memory(char* addr, size_t size) { 3155 return anon_munmap(addr, size); 3156} 3157 3158static address highest_vm_reserved_address() { 3159 return _highest_vm_reserved_address; 3160} 3161 3162static bool linux_mprotect(char* addr, size_t size, int prot) { 3163 // Linux wants the mprotect address argument to be page aligned. 3164 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3165 3166 // According to SUSv3, mprotect() should only be used with mappings 3167 // established by mmap(), and mmap() always maps whole pages. Unaligned 3168 // 'addr' likely indicates problem in the VM (e.g. trying to change 3169 // protection of malloc'ed or statically allocated memory). Check the 3170 // caller if you hit this assert. 3171 assert(addr == bottom, "sanity check"); 3172 3173 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3174 return ::mprotect(bottom, size, prot) == 0; 3175} 3176 3177// Set protections specified 3178bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3179 bool is_committed) { 3180 unsigned int p = 0; 3181 switch (prot) { 3182 case MEM_PROT_NONE: p = PROT_NONE; break; 3183 case MEM_PROT_READ: p = PROT_READ; break; 3184 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3185 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3186 default: 3187 ShouldNotReachHere(); 3188 } 3189 // is_committed is unused. 3190 return linux_mprotect(addr, bytes, p); 3191} 3192 3193bool os::guard_memory(char* addr, size_t size) { 3194 return linux_mprotect(addr, size, PROT_NONE); 3195} 3196 3197bool os::unguard_memory(char* addr, size_t size) { 3198 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3199} 3200 3201bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) { 3202 bool result = false; 3203 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3204 MAP_ANONYMOUS|MAP_PRIVATE, 3205 -1, 0); 3206 if (p != MAP_FAILED) { 3207 void *aligned_p = align_ptr_up(p, page_size); 3208 3209 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3210 3211 munmap(p, page_size * 2); 3212 } 3213 3214 if (warn && !result) { 3215 warning("TransparentHugePages is not supported by the operating system."); 3216 } 3217 3218 return result; 3219} 3220 3221bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3222 bool result = false; 3223 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3224 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3225 -1, 0); 3226 3227 if (p != MAP_FAILED) { 3228 // We don't know if this really is a huge page or not. 3229 FILE *fp = fopen("/proc/self/maps", "r"); 3230 if (fp) { 3231 while (!feof(fp)) { 3232 char chars[257]; 3233 long x = 0; 3234 if (fgets(chars, sizeof(chars), fp)) { 3235 if (sscanf(chars, "%lx-%*x", &x) == 1 3236 && x == (long)p) { 3237 if (strstr (chars, "hugepage")) { 3238 result = true; 3239 break; 3240 } 3241 } 3242 } 3243 } 3244 fclose(fp); 3245 } 3246 munmap(p, page_size); 3247 } 3248 3249 if (warn && !result) { 3250 warning("HugeTLBFS is not supported by the operating system."); 3251 } 3252 3253 return result; 3254} 3255 3256/* 3257* Set the coredump_filter bits to include largepages in core dump (bit 6) 3258* 3259* From the coredump_filter documentation: 3260* 3261* - (bit 0) anonymous private memory 3262* - (bit 1) anonymous shared memory 3263* - (bit 2) file-backed private memory 3264* - (bit 3) file-backed shared memory 3265* - (bit 4) ELF header pages in file-backed private memory areas (it is 3266* effective only if the bit 2 is cleared) 3267* - (bit 5) hugetlb private memory 3268* - (bit 6) hugetlb shared memory 3269*/ 3270static void set_coredump_filter(void) { 3271 FILE *f; 3272 long cdm; 3273 3274 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3275 return; 3276 } 3277 3278 if (fscanf(f, "%lx", &cdm) != 1) { 3279 fclose(f); 3280 return; 3281 } 3282 3283 rewind(f); 3284 3285 if ((cdm & LARGEPAGES_BIT) == 0) { 3286 cdm |= LARGEPAGES_BIT; 3287 fprintf(f, "%#lx", cdm); 3288 } 3289 3290 fclose(f); 3291} 3292 3293// Large page support 3294 3295static size_t _large_page_size = 0; 3296 3297size_t os::Linux::find_large_page_size() { 3298 size_t large_page_size = 0; 3299 3300 // large_page_size on Linux is used to round up heap size. x86 uses either 3301 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3302 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3303 // page as large as 256M. 3304 // 3305 // Here we try to figure out page size by parsing /proc/meminfo and looking 3306 // for a line with the following format: 3307 // Hugepagesize: 2048 kB 3308 // 3309 // If we can't determine the value (e.g. /proc is not mounted, or the text 3310 // format has been changed), we'll use the largest page size supported by 3311 // the processor. 3312 3313#ifndef ZERO 3314 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3315 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3316#endif // ZERO 3317 3318 FILE *fp = fopen("/proc/meminfo", "r"); 3319 if (fp) { 3320 while (!feof(fp)) { 3321 int x = 0; 3322 char buf[16]; 3323 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3324 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3325 large_page_size = x * K; 3326 break; 3327 } 3328 } else { 3329 // skip to next line 3330 for (;;) { 3331 int ch = fgetc(fp); 3332 if (ch == EOF || ch == (int)'\n') break; 3333 } 3334 } 3335 } 3336 fclose(fp); 3337 } 3338 3339 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3340 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3341 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3342 proper_unit_for_byte_size(large_page_size)); 3343 } 3344 3345 return large_page_size; 3346} 3347 3348size_t os::Linux::setup_large_page_size() { 3349 _large_page_size = Linux::find_large_page_size(); 3350 const size_t default_page_size = (size_t)Linux::page_size(); 3351 if (_large_page_size > default_page_size) { 3352 _page_sizes[0] = _large_page_size; 3353 _page_sizes[1] = default_page_size; 3354 _page_sizes[2] = 0; 3355 } 3356 3357 return _large_page_size; 3358} 3359 3360bool os::Linux::setup_large_page_type(size_t page_size) { 3361 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3362 FLAG_IS_DEFAULT(UseSHM) && 3363 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3364 // If UseLargePages is specified on the command line try all methods, 3365 // if it's default, then try only UseTransparentHugePages. 3366 if (FLAG_IS_DEFAULT(UseLargePages)) { 3367 UseTransparentHugePages = true; 3368 } else { 3369 UseHugeTLBFS = UseTransparentHugePages = UseSHM = true; 3370 } 3371 } 3372 3373 if (UseTransparentHugePages) { 3374 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3375 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3376 UseHugeTLBFS = false; 3377 UseSHM = false; 3378 return true; 3379 } 3380 UseTransparentHugePages = false; 3381 } 3382 3383 if (UseHugeTLBFS) { 3384 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3385 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3386 UseSHM = false; 3387 return true; 3388 } 3389 UseHugeTLBFS = false; 3390 } 3391 3392 return UseSHM; 3393} 3394 3395void os::large_page_init() { 3396 if (!UseLargePages) { 3397 UseHugeTLBFS = false; 3398 UseTransparentHugePages = false; 3399 UseSHM = false; 3400 return; 3401 } 3402 3403 size_t large_page_size = Linux::setup_large_page_size(); 3404 UseLargePages = Linux::setup_large_page_type(large_page_size); 3405 3406 set_coredump_filter(); 3407} 3408 3409#ifndef SHM_HUGETLB 3410#define SHM_HUGETLB 04000 3411#endif 3412 3413char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3414 // "exec" is passed in but not used. Creating the shared image for 3415 // the code cache doesn't have an SHM_X executable permission to check. 3416 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3417 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3418 3419 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3420 return NULL; // Fallback to small pages. 3421 } 3422 3423 key_t key = IPC_PRIVATE; 3424 char *addr; 3425 3426 bool warn_on_failure = UseLargePages && 3427 (!FLAG_IS_DEFAULT(UseLargePages) || 3428 !FLAG_IS_DEFAULT(UseSHM) || 3429 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 3430 ); 3431 char msg[128]; 3432 3433 // Create a large shared memory region to attach to based on size. 3434 // Currently, size is the total size of the heap 3435 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3436 if (shmid == -1) { 3437 // Possible reasons for shmget failure: 3438 // 1. shmmax is too small for Java heap. 3439 // > check shmmax value: cat /proc/sys/kernel/shmmax 3440 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3441 // 2. not enough large page memory. 3442 // > check available large pages: cat /proc/meminfo 3443 // > increase amount of large pages: 3444 // echo new_value > /proc/sys/vm/nr_hugepages 3445 // Note 1: different Linux may use different name for this property, 3446 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3447 // Note 2: it's possible there's enough physical memory available but 3448 // they are so fragmented after a long run that they can't 3449 // coalesce into large pages. Try to reserve large pages when 3450 // the system is still "fresh". 3451 if (warn_on_failure) { 3452 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3453 warning(msg); 3454 } 3455 return NULL; 3456 } 3457 3458 // attach to the region 3459 addr = (char*)shmat(shmid, req_addr, 0); 3460 int err = errno; 3461 3462 // Remove shmid. If shmat() is successful, the actual shared memory segment 3463 // will be deleted when it's detached by shmdt() or when the process 3464 // terminates. If shmat() is not successful this will remove the shared 3465 // segment immediately. 3466 shmctl(shmid, IPC_RMID, NULL); 3467 3468 if ((intptr_t)addr == -1) { 3469 if (warn_on_failure) { 3470 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3471 warning(msg); 3472 } 3473 return NULL; 3474 } 3475 3476 return addr; 3477} 3478 3479static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) { 3480 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3481 3482 bool warn_on_failure = UseLargePages && 3483 (!FLAG_IS_DEFAULT(UseLargePages) || 3484 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3485 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3486 3487 if (warn_on_failure) { 3488 char msg[128]; 3489 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3490 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3491 warning(msg); 3492 } 3493} 3494 3495char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) { 3496 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3497 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3498 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3499 3500 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3501 char* addr = (char*)::mmap(req_addr, bytes, prot, 3502 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3503 -1, 0); 3504 3505 if (addr == MAP_FAILED) { 3506 warn_on_large_pages_failure(req_addr, bytes, errno); 3507 return NULL; 3508 } 3509 3510 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3511 3512 return addr; 3513} 3514 3515char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3516 size_t large_page_size = os::large_page_size(); 3517 3518 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3519 3520 // Allocate small pages. 3521 3522 char* start; 3523 if (req_addr != NULL) { 3524 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3525 assert(is_size_aligned(bytes, alignment), "Must be"); 3526 start = os::reserve_memory(bytes, req_addr); 3527 assert(start == NULL || start == req_addr, "Must be"); 3528 } else { 3529 start = os::reserve_memory_aligned(bytes, alignment); 3530 } 3531 3532 if (start == NULL) { 3533 return NULL; 3534 } 3535 3536 assert(is_ptr_aligned(start, alignment), "Must be"); 3537 3538 // os::reserve_memory_special will record this memory area. 3539 // Need to release it here to prevent overlapping reservations. 3540 MemTracker::record_virtual_memory_release((address)start, bytes); 3541 3542 char* end = start + bytes; 3543 3544 // Find the regions of the allocated chunk that can be promoted to large pages. 3545 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3546 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3547 3548 size_t lp_bytes = lp_end - lp_start; 3549 3550 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3551 3552 if (lp_bytes == 0) { 3553 // The mapped region doesn't even span the start and the end of a large page. 3554 // Fall back to allocate a non-special area. 3555 ::munmap(start, end - start); 3556 return NULL; 3557 } 3558 3559 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3560 3561 3562 void* result; 3563 3564 if (start != lp_start) { 3565 result = ::mmap(start, lp_start - start, prot, 3566 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3567 -1, 0); 3568 if (result == MAP_FAILED) { 3569 ::munmap(lp_start, end - lp_start); 3570 return NULL; 3571 } 3572 } 3573 3574 result = ::mmap(lp_start, lp_bytes, prot, 3575 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3576 -1, 0); 3577 if (result == MAP_FAILED) { 3578 warn_on_large_pages_failure(req_addr, bytes, errno); 3579 // If the mmap above fails, the large pages region will be unmapped and we 3580 // have regions before and after with small pages. Release these regions. 3581 // 3582 // | mapped | unmapped | mapped | 3583 // ^ ^ ^ ^ 3584 // start lp_start lp_end end 3585 // 3586 ::munmap(start, lp_start - start); 3587 ::munmap(lp_end, end - lp_end); 3588 return NULL; 3589 } 3590 3591 if (lp_end != end) { 3592 result = ::mmap(lp_end, end - lp_end, prot, 3593 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3594 -1, 0); 3595 if (result == MAP_FAILED) { 3596 ::munmap(start, lp_end - start); 3597 return NULL; 3598 } 3599 } 3600 3601 return start; 3602} 3603 3604char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3605 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3606 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3607 assert(is_power_of_2(alignment), "Must be"); 3608 assert(is_power_of_2(os::large_page_size()), "Must be"); 3609 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3610 3611 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3612 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3613 } else { 3614 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3615 } 3616} 3617 3618char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3619 assert(UseLargePages, "only for large pages"); 3620 3621 char* addr; 3622 if (UseSHM) { 3623 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3624 } else { 3625 assert(UseHugeTLBFS, "must be"); 3626 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3627 } 3628 3629 if (addr != NULL) { 3630 if (UseNUMAInterleaving) { 3631 numa_make_global(addr, bytes); 3632 } 3633 3634 // The memory is committed 3635 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC); 3636 } 3637 3638 return addr; 3639} 3640 3641bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3642 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3643 return shmdt(base) == 0; 3644} 3645 3646bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3647 return pd_release_memory(base, bytes); 3648} 3649 3650bool os::release_memory_special(char* base, size_t bytes) { 3651 assert(UseLargePages, "only for large pages"); 3652 3653 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3654 3655 bool res; 3656 if (UseSHM) { 3657 res = os::Linux::release_memory_special_shm(base, bytes); 3658 } else { 3659 assert(UseHugeTLBFS, "must be"); 3660 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3661 } 3662 3663 if (res) { 3664 tkr.record((address)base, bytes); 3665 } else { 3666 tkr.discard(); 3667 } 3668 3669 return res; 3670} 3671 3672size_t os::large_page_size() { 3673 return _large_page_size; 3674} 3675 3676// With SysV SHM the entire memory region must be allocated as shared 3677// memory. 3678// HugeTLBFS allows application to commit large page memory on demand. 3679// However, when committing memory with HugeTLBFS fails, the region 3680// that was supposed to be committed will lose the old reservation 3681// and allow other threads to steal that memory region. Because of this 3682// behavior we can't commit HugeTLBFS memory. 3683bool os::can_commit_large_page_memory() { 3684 return UseTransparentHugePages; 3685} 3686 3687bool os::can_execute_large_page_memory() { 3688 return UseTransparentHugePages || UseHugeTLBFS; 3689} 3690 3691// Reserve memory at an arbitrary address, only if that area is 3692// available (and not reserved for something else). 3693 3694char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3695 const int max_tries = 10; 3696 char* base[max_tries]; 3697 size_t size[max_tries]; 3698 const size_t gap = 0x000000; 3699 3700 // Assert only that the size is a multiple of the page size, since 3701 // that's all that mmap requires, and since that's all we really know 3702 // about at this low abstraction level. If we need higher alignment, 3703 // we can either pass an alignment to this method or verify alignment 3704 // in one of the methods further up the call chain. See bug 5044738. 3705 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3706 3707 // Repeatedly allocate blocks until the block is allocated at the 3708 // right spot. Give up after max_tries. Note that reserve_memory() will 3709 // automatically update _highest_vm_reserved_address if the call is 3710 // successful. The variable tracks the highest memory address every reserved 3711 // by JVM. It is used to detect heap-stack collision if running with 3712 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3713 // space than needed, it could confuse the collision detecting code. To 3714 // solve the problem, save current _highest_vm_reserved_address and 3715 // calculate the correct value before return. 3716 address old_highest = _highest_vm_reserved_address; 3717 3718 // Linux mmap allows caller to pass an address as hint; give it a try first, 3719 // if kernel honors the hint then we can return immediately. 3720 char * addr = anon_mmap(requested_addr, bytes, false); 3721 if (addr == requested_addr) { 3722 return requested_addr; 3723 } 3724 3725 if (addr != NULL) { 3726 // mmap() is successful but it fails to reserve at the requested address 3727 anon_munmap(addr, bytes); 3728 } 3729 3730 int i; 3731 for (i = 0; i < max_tries; ++i) { 3732 base[i] = reserve_memory(bytes); 3733 3734 if (base[i] != NULL) { 3735 // Is this the block we wanted? 3736 if (base[i] == requested_addr) { 3737 size[i] = bytes; 3738 break; 3739 } 3740 3741 // Does this overlap the block we wanted? Give back the overlapped 3742 // parts and try again. 3743 3744 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3745 if (top_overlap >= 0 && top_overlap < bytes) { 3746 unmap_memory(base[i], top_overlap); 3747 base[i] += top_overlap; 3748 size[i] = bytes - top_overlap; 3749 } else { 3750 size_t bottom_overlap = base[i] + bytes - requested_addr; 3751 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3752 unmap_memory(requested_addr, bottom_overlap); 3753 size[i] = bytes - bottom_overlap; 3754 } else { 3755 size[i] = bytes; 3756 } 3757 } 3758 } 3759 } 3760 3761 // Give back the unused reserved pieces. 3762 3763 for (int j = 0; j < i; ++j) { 3764 if (base[j] != NULL) { 3765 unmap_memory(base[j], size[j]); 3766 } 3767 } 3768 3769 if (i < max_tries) { 3770 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3771 return requested_addr; 3772 } else { 3773 _highest_vm_reserved_address = old_highest; 3774 return NULL; 3775 } 3776} 3777 3778size_t os::read(int fd, void *buf, unsigned int nBytes) { 3779 return ::read(fd, buf, nBytes); 3780} 3781 3782// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 3783// Solaris uses poll(), linux uses park(). 3784// Poll() is likely a better choice, assuming that Thread.interrupt() 3785// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 3786// SIGSEGV, see 4355769. 3787 3788int os::sleep(Thread* thread, jlong millis, bool interruptible) { 3789 assert(thread == Thread::current(), "thread consistency check"); 3790 3791 ParkEvent * const slp = thread->_SleepEvent ; 3792 slp->reset() ; 3793 OrderAccess::fence() ; 3794 3795 if (interruptible) { 3796 jlong prevtime = javaTimeNanos(); 3797 3798 for (;;) { 3799 if (os::is_interrupted(thread, true)) { 3800 return OS_INTRPT; 3801 } 3802 3803 jlong newtime = javaTimeNanos(); 3804 3805 if (newtime - prevtime < 0) { 3806 // time moving backwards, should only happen if no monotonic clock 3807 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3808 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3809 } else { 3810 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3811 } 3812 3813 if(millis <= 0) { 3814 return OS_OK; 3815 } 3816 3817 prevtime = newtime; 3818 3819 { 3820 assert(thread->is_Java_thread(), "sanity check"); 3821 JavaThread *jt = (JavaThread *) thread; 3822 ThreadBlockInVM tbivm(jt); 3823 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 3824 3825 jt->set_suspend_equivalent(); 3826 // cleared by handle_special_suspend_equivalent_condition() or 3827 // java_suspend_self() via check_and_wait_while_suspended() 3828 3829 slp->park(millis); 3830 3831 // were we externally suspended while we were waiting? 3832 jt->check_and_wait_while_suspended(); 3833 } 3834 } 3835 } else { 3836 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 3837 jlong prevtime = javaTimeNanos(); 3838 3839 for (;;) { 3840 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 3841 // the 1st iteration ... 3842 jlong newtime = javaTimeNanos(); 3843 3844 if (newtime - prevtime < 0) { 3845 // time moving backwards, should only happen if no monotonic clock 3846 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3847 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3848 } else { 3849 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3850 } 3851 3852 if(millis <= 0) break ; 3853 3854 prevtime = newtime; 3855 slp->park(millis); 3856 } 3857 return OS_OK ; 3858 } 3859} 3860 3861int os::naked_sleep() { 3862 // %% make the sleep time an integer flag. for now use 1 millisec. 3863 return os::sleep(Thread::current(), 1, false); 3864} 3865 3866// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3867void os::infinite_sleep() { 3868 while (true) { // sleep forever ... 3869 ::sleep(100); // ... 100 seconds at a time 3870 } 3871} 3872 3873// Used to convert frequent JVM_Yield() to nops 3874bool os::dont_yield() { 3875 return DontYieldALot; 3876} 3877 3878void os::yield() { 3879 sched_yield(); 3880} 3881 3882os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3883 3884void os::yield_all(int attempts) { 3885 // Yields to all threads, including threads with lower priorities 3886 // Threads on Linux are all with same priority. The Solaris style 3887 // os::yield_all() with nanosleep(1ms) is not necessary. 3888 sched_yield(); 3889} 3890 3891// Called from the tight loops to possibly influence time-sharing heuristics 3892void os::loop_breaker(int attempts) { 3893 os::yield_all(attempts); 3894} 3895 3896//////////////////////////////////////////////////////////////////////////////// 3897// thread priority support 3898 3899// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3900// only supports dynamic priority, static priority must be zero. For real-time 3901// applications, Linux supports SCHED_RR which allows static priority (1-99). 3902// However, for large multi-threaded applications, SCHED_RR is not only slower 3903// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3904// of 5 runs - Sep 2005). 3905// 3906// The following code actually changes the niceness of kernel-thread/LWP. It 3907// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3908// not the entire user process, and user level threads are 1:1 mapped to kernel 3909// threads. It has always been the case, but could change in the future. For 3910// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3911// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3912 3913int os::java_to_os_priority[CriticalPriority + 1] = { 3914 19, // 0 Entry should never be used 3915 3916 4, // 1 MinPriority 3917 3, // 2 3918 2, // 3 3919 3920 1, // 4 3921 0, // 5 NormPriority 3922 -1, // 6 3923 3924 -2, // 7 3925 -3, // 8 3926 -4, // 9 NearMaxPriority 3927 3928 -5, // 10 MaxPriority 3929 3930 -5 // 11 CriticalPriority 3931}; 3932 3933static int prio_init() { 3934 if (ThreadPriorityPolicy == 1) { 3935 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3936 // if effective uid is not root. Perhaps, a more elegant way of doing 3937 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3938 if (geteuid() != 0) { 3939 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3940 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3941 } 3942 ThreadPriorityPolicy = 0; 3943 } 3944 } 3945 if (UseCriticalJavaThreadPriority) { 3946 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3947 } 3948 return 0; 3949} 3950 3951OSReturn os::set_native_priority(Thread* thread, int newpri) { 3952 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3953 3954 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3955 return (ret == 0) ? OS_OK : OS_ERR; 3956} 3957 3958OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3959 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3960 *priority_ptr = java_to_os_priority[NormPriority]; 3961 return OS_OK; 3962 } 3963 3964 errno = 0; 3965 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3966 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3967} 3968 3969// Hint to the underlying OS that a task switch would not be good. 3970// Void return because it's a hint and can fail. 3971void os::hint_no_preempt() {} 3972 3973//////////////////////////////////////////////////////////////////////////////// 3974// suspend/resume support 3975 3976// the low-level signal-based suspend/resume support is a remnant from the 3977// old VM-suspension that used to be for java-suspension, safepoints etc, 3978// within hotspot. Now there is a single use-case for this: 3979// - calling get_thread_pc() on the VMThread by the flat-profiler task 3980// that runs in the watcher thread. 3981// The remaining code is greatly simplified from the more general suspension 3982// code that used to be used. 3983// 3984// The protocol is quite simple: 3985// - suspend: 3986// - sends a signal to the target thread 3987// - polls the suspend state of the osthread using a yield loop 3988// - target thread signal handler (SR_handler) sets suspend state 3989// and blocks in sigsuspend until continued 3990// - resume: 3991// - sets target osthread state to continue 3992// - sends signal to end the sigsuspend loop in the SR_handler 3993// 3994// Note that the SR_lock plays no role in this suspend/resume protocol. 3995// 3996 3997static void resume_clear_context(OSThread *osthread) { 3998 osthread->set_ucontext(NULL); 3999 osthread->set_siginfo(NULL); 4000} 4001 4002static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 4003 osthread->set_ucontext(context); 4004 osthread->set_siginfo(siginfo); 4005} 4006 4007// 4008// Handler function invoked when a thread's execution is suspended or 4009// resumed. We have to be careful that only async-safe functions are 4010// called here (Note: most pthread functions are not async safe and 4011// should be avoided.) 4012// 4013// Note: sigwait() is a more natural fit than sigsuspend() from an 4014// interface point of view, but sigwait() prevents the signal hander 4015// from being run. libpthread would get very confused by not having 4016// its signal handlers run and prevents sigwait()'s use with the 4017// mutex granting granting signal. 4018// 4019// Currently only ever called on the VMThread and JavaThreads (PC sampling) 4020// 4021static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 4022 // Save and restore errno to avoid confusing native code with EINTR 4023 // after sigsuspend. 4024 int old_errno = errno; 4025 4026 Thread* thread = Thread::current(); 4027 OSThread* osthread = thread->osthread(); 4028 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 4029 4030 os::SuspendResume::State current = osthread->sr.state(); 4031 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 4032 suspend_save_context(osthread, siginfo, context); 4033 4034 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4035 os::SuspendResume::State state = osthread->sr.suspended(); 4036 if (state == os::SuspendResume::SR_SUSPENDED) { 4037 sigset_t suspend_set; // signals for sigsuspend() 4038 4039 // get current set of blocked signals and unblock resume signal 4040 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4041 sigdelset(&suspend_set, SR_signum); 4042 4043 sr_semaphore.signal(); 4044 // wait here until we are resumed 4045 while (1) { 4046 sigsuspend(&suspend_set); 4047 4048 os::SuspendResume::State result = osthread->sr.running(); 4049 if (result == os::SuspendResume::SR_RUNNING) { 4050 sr_semaphore.signal(); 4051 break; 4052 } 4053 } 4054 4055 } else if (state == os::SuspendResume::SR_RUNNING) { 4056 // request was cancelled, continue 4057 } else { 4058 ShouldNotReachHere(); 4059 } 4060 4061 resume_clear_context(osthread); 4062 } else if (current == os::SuspendResume::SR_RUNNING) { 4063 // request was cancelled, continue 4064 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4065 // ignore 4066 } else { 4067 // ignore 4068 } 4069 4070 errno = old_errno; 4071} 4072 4073 4074static int SR_initialize() { 4075 struct sigaction act; 4076 char *s; 4077 /* Get signal number to use for suspend/resume */ 4078 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4079 int sig = ::strtol(s, 0, 10); 4080 if (sig > 0 || sig < _NSIG) { 4081 SR_signum = sig; 4082 } 4083 } 4084 4085 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4086 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4087 4088 sigemptyset(&SR_sigset); 4089 sigaddset(&SR_sigset, SR_signum); 4090 4091 /* Set up signal handler for suspend/resume */ 4092 act.sa_flags = SA_RESTART|SA_SIGINFO; 4093 act.sa_handler = (void (*)(int)) SR_handler; 4094 4095 // SR_signum is blocked by default. 4096 // 4528190 - We also need to block pthread restart signal (32 on all 4097 // supported Linux platforms). Note that LinuxThreads need to block 4098 // this signal for all threads to work properly. So we don't have 4099 // to use hard-coded signal number when setting up the mask. 4100 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4101 4102 if (sigaction(SR_signum, &act, 0) == -1) { 4103 return -1; 4104 } 4105 4106 // Save signal flag 4107 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4108 return 0; 4109} 4110 4111static int sr_notify(OSThread* osthread) { 4112 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4113 assert_status(status == 0, status, "pthread_kill"); 4114 return status; 4115} 4116 4117// "Randomly" selected value for how long we want to spin 4118// before bailing out on suspending a thread, also how often 4119// we send a signal to a thread we want to resume 4120static const int RANDOMLY_LARGE_INTEGER = 1000000; 4121static const int RANDOMLY_LARGE_INTEGER2 = 100; 4122 4123// returns true on success and false on error - really an error is fatal 4124// but this seems the normal response to library errors 4125static bool do_suspend(OSThread* osthread) { 4126 assert(osthread->sr.is_running(), "thread should be running"); 4127 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4128 4129 // mark as suspended and send signal 4130 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4131 // failed to switch, state wasn't running? 4132 ShouldNotReachHere(); 4133 return false; 4134 } 4135 4136 if (sr_notify(osthread) != 0) { 4137 ShouldNotReachHere(); 4138 } 4139 4140 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4141 while (true) { 4142 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4143 break; 4144 } else { 4145 // timeout 4146 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4147 if (cancelled == os::SuspendResume::SR_RUNNING) { 4148 return false; 4149 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4150 // make sure that we consume the signal on the semaphore as well 4151 sr_semaphore.wait(); 4152 break; 4153 } else { 4154 ShouldNotReachHere(); 4155 return false; 4156 } 4157 } 4158 } 4159 4160 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4161 return true; 4162} 4163 4164static void do_resume(OSThread* osthread) { 4165 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4166 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4167 4168 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4169 // failed to switch to WAKEUP_REQUEST 4170 ShouldNotReachHere(); 4171 return; 4172 } 4173 4174 while (true) { 4175 if (sr_notify(osthread) == 0) { 4176 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4177 if (osthread->sr.is_running()) { 4178 return; 4179 } 4180 } 4181 } else { 4182 ShouldNotReachHere(); 4183 } 4184 } 4185 4186 guarantee(osthread->sr.is_running(), "Must be running!"); 4187} 4188 4189//////////////////////////////////////////////////////////////////////////////// 4190// interrupt support 4191 4192void os::interrupt(Thread* thread) { 4193 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 4194 "possibility of dangling Thread pointer"); 4195 4196 OSThread* osthread = thread->osthread(); 4197 4198 if (!osthread->interrupted()) { 4199 osthread->set_interrupted(true); 4200 // More than one thread can get here with the same value of osthread, 4201 // resulting in multiple notifications. We do, however, want the store 4202 // to interrupted() to be visible to other threads before we execute unpark(). 4203 OrderAccess::fence(); 4204 ParkEvent * const slp = thread->_SleepEvent ; 4205 if (slp != NULL) slp->unpark() ; 4206 } 4207 4208 // For JSR166. Unpark even if interrupt status already was set 4209 if (thread->is_Java_thread()) 4210 ((JavaThread*)thread)->parker()->unpark(); 4211 4212 ParkEvent * ev = thread->_ParkEvent ; 4213 if (ev != NULL) ev->unpark() ; 4214 4215} 4216 4217bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 4218 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 4219 "possibility of dangling Thread pointer"); 4220 4221 OSThread* osthread = thread->osthread(); 4222 4223 bool interrupted = osthread->interrupted(); 4224 4225 if (interrupted && clear_interrupted) { 4226 osthread->set_interrupted(false); 4227 // consider thread->_SleepEvent->reset() ... optional optimization 4228 } 4229 4230 return interrupted; 4231} 4232 4233/////////////////////////////////////////////////////////////////////////////////// 4234// signal handling (except suspend/resume) 4235 4236// This routine may be used by user applications as a "hook" to catch signals. 4237// The user-defined signal handler must pass unrecognized signals to this 4238// routine, and if it returns true (non-zero), then the signal handler must 4239// return immediately. If the flag "abort_if_unrecognized" is true, then this 4240// routine will never retun false (zero), but instead will execute a VM panic 4241// routine kill the process. 4242// 4243// If this routine returns false, it is OK to call it again. This allows 4244// the user-defined signal handler to perform checks either before or after 4245// the VM performs its own checks. Naturally, the user code would be making 4246// a serious error if it tried to handle an exception (such as a null check 4247// or breakpoint) that the VM was generating for its own correct operation. 4248// 4249// This routine may recognize any of the following kinds of signals: 4250// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4251// It should be consulted by handlers for any of those signals. 4252// 4253// The caller of this routine must pass in the three arguments supplied 4254// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4255// field of the structure passed to sigaction(). This routine assumes that 4256// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4257// 4258// Note that the VM will print warnings if it detects conflicting signal 4259// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4260// 4261extern "C" JNIEXPORT int 4262JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 4263 void* ucontext, int abort_if_unrecognized); 4264 4265void signalHandler(int sig, siginfo_t* info, void* uc) { 4266 assert(info != NULL && uc != NULL, "it must be old kernel"); 4267 int orig_errno = errno; // Preserve errno value over signal handler. 4268 JVM_handle_linux_signal(sig, info, uc, true); 4269 errno = orig_errno; 4270} 4271 4272 4273// This boolean allows users to forward their own non-matching signals 4274// to JVM_handle_linux_signal, harmlessly. 4275bool os::Linux::signal_handlers_are_installed = false; 4276 4277// For signal-chaining 4278struct sigaction os::Linux::sigact[MAXSIGNUM]; 4279unsigned int os::Linux::sigs = 0; 4280bool os::Linux::libjsig_is_loaded = false; 4281typedef struct sigaction *(*get_signal_t)(int); 4282get_signal_t os::Linux::get_signal_action = NULL; 4283 4284struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4285 struct sigaction *actp = NULL; 4286 4287 if (libjsig_is_loaded) { 4288 // Retrieve the old signal handler from libjsig 4289 actp = (*get_signal_action)(sig); 4290 } 4291 if (actp == NULL) { 4292 // Retrieve the preinstalled signal handler from jvm 4293 actp = get_preinstalled_handler(sig); 4294 } 4295 4296 return actp; 4297} 4298 4299static bool call_chained_handler(struct sigaction *actp, int sig, 4300 siginfo_t *siginfo, void *context) { 4301 // Call the old signal handler 4302 if (actp->sa_handler == SIG_DFL) { 4303 // It's more reasonable to let jvm treat it as an unexpected exception 4304 // instead of taking the default action. 4305 return false; 4306 } else if (actp->sa_handler != SIG_IGN) { 4307 if ((actp->sa_flags & SA_NODEFER) == 0) { 4308 // automaticlly block the signal 4309 sigaddset(&(actp->sa_mask), sig); 4310 } 4311 4312 sa_handler_t hand; 4313 sa_sigaction_t sa; 4314 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4315 // retrieve the chained handler 4316 if (siginfo_flag_set) { 4317 sa = actp->sa_sigaction; 4318 } else { 4319 hand = actp->sa_handler; 4320 } 4321 4322 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4323 actp->sa_handler = SIG_DFL; 4324 } 4325 4326 // try to honor the signal mask 4327 sigset_t oset; 4328 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4329 4330 // call into the chained handler 4331 if (siginfo_flag_set) { 4332 (*sa)(sig, siginfo, context); 4333 } else { 4334 (*hand)(sig); 4335 } 4336 4337 // restore the signal mask 4338 pthread_sigmask(SIG_SETMASK, &oset, 0); 4339 } 4340 // Tell jvm's signal handler the signal is taken care of. 4341 return true; 4342} 4343 4344bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4345 bool chained = false; 4346 // signal-chaining 4347 if (UseSignalChaining) { 4348 struct sigaction *actp = get_chained_signal_action(sig); 4349 if (actp != NULL) { 4350 chained = call_chained_handler(actp, sig, siginfo, context); 4351 } 4352 } 4353 return chained; 4354} 4355 4356struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4357 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 4358 return &sigact[sig]; 4359 } 4360 return NULL; 4361} 4362 4363void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4364 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4365 sigact[sig] = oldAct; 4366 sigs |= (unsigned int)1 << sig; 4367} 4368 4369// for diagnostic 4370int os::Linux::sigflags[MAXSIGNUM]; 4371 4372int os::Linux::get_our_sigflags(int sig) { 4373 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4374 return sigflags[sig]; 4375} 4376 4377void os::Linux::set_our_sigflags(int sig, int flags) { 4378 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4379 sigflags[sig] = flags; 4380} 4381 4382void os::Linux::set_signal_handler(int sig, bool set_installed) { 4383 // Check for overwrite. 4384 struct sigaction oldAct; 4385 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4386 4387 void* oldhand = oldAct.sa_sigaction 4388 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4389 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4390 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4391 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4392 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4393 if (AllowUserSignalHandlers || !set_installed) { 4394 // Do not overwrite; user takes responsibility to forward to us. 4395 return; 4396 } else if (UseSignalChaining) { 4397 // save the old handler in jvm 4398 save_preinstalled_handler(sig, oldAct); 4399 // libjsig also interposes the sigaction() call below and saves the 4400 // old sigaction on it own. 4401 } else { 4402 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4403 "%#lx for signal %d.", (long)oldhand, sig)); 4404 } 4405 } 4406 4407 struct sigaction sigAct; 4408 sigfillset(&(sigAct.sa_mask)); 4409 sigAct.sa_handler = SIG_DFL; 4410 if (!set_installed) { 4411 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4412 } else { 4413 sigAct.sa_sigaction = signalHandler; 4414 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4415 } 4416 // Save flags, which are set by ours 4417 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4418 sigflags[sig] = sigAct.sa_flags; 4419 4420 int ret = sigaction(sig, &sigAct, &oldAct); 4421 assert(ret == 0, "check"); 4422 4423 void* oldhand2 = oldAct.sa_sigaction 4424 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4425 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4426 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4427} 4428 4429// install signal handlers for signals that HotSpot needs to 4430// handle in order to support Java-level exception handling. 4431 4432void os::Linux::install_signal_handlers() { 4433 if (!signal_handlers_are_installed) { 4434 signal_handlers_are_installed = true; 4435 4436 // signal-chaining 4437 typedef void (*signal_setting_t)(); 4438 signal_setting_t begin_signal_setting = NULL; 4439 signal_setting_t end_signal_setting = NULL; 4440 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4441 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4442 if (begin_signal_setting != NULL) { 4443 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4444 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4445 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4446 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4447 libjsig_is_loaded = true; 4448 assert(UseSignalChaining, "should enable signal-chaining"); 4449 } 4450 if (libjsig_is_loaded) { 4451 // Tell libjsig jvm is setting signal handlers 4452 (*begin_signal_setting)(); 4453 } 4454 4455 set_signal_handler(SIGSEGV, true); 4456 set_signal_handler(SIGPIPE, true); 4457 set_signal_handler(SIGBUS, true); 4458 set_signal_handler(SIGILL, true); 4459 set_signal_handler(SIGFPE, true); 4460 set_signal_handler(SIGXFSZ, true); 4461 4462 if (libjsig_is_loaded) { 4463 // Tell libjsig jvm finishes setting signal handlers 4464 (*end_signal_setting)(); 4465 } 4466 4467 // We don't activate signal checker if libjsig is in place, we trust ourselves 4468 // and if UserSignalHandler is installed all bets are off. 4469 // Log that signal checking is off only if -verbose:jni is specified. 4470 if (CheckJNICalls) { 4471 if (libjsig_is_loaded) { 4472 if (PrintJNIResolving) { 4473 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4474 } 4475 check_signals = false; 4476 } 4477 if (AllowUserSignalHandlers) { 4478 if (PrintJNIResolving) { 4479 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4480 } 4481 check_signals = false; 4482 } 4483 } 4484 } 4485} 4486 4487// This is the fastest way to get thread cpu time on Linux. 4488// Returns cpu time (user+sys) for any thread, not only for current. 4489// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4490// It might work on 2.6.10+ with a special kernel/glibc patch. 4491// For reference, please, see IEEE Std 1003.1-2004: 4492// http://www.unix.org/single_unix_specification 4493 4494jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4495 struct timespec tp; 4496 int rc = os::Linux::clock_gettime(clockid, &tp); 4497 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4498 4499 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4500} 4501 4502///// 4503// glibc on Linux platform uses non-documented flag 4504// to indicate, that some special sort of signal 4505// trampoline is used. 4506// We will never set this flag, and we should 4507// ignore this flag in our diagnostic 4508#ifdef SIGNIFICANT_SIGNAL_MASK 4509#undef SIGNIFICANT_SIGNAL_MASK 4510#endif 4511#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4512 4513static const char* get_signal_handler_name(address handler, 4514 char* buf, int buflen) { 4515 int offset; 4516 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4517 if (found) { 4518 // skip directory names 4519 const char *p1, *p2; 4520 p1 = buf; 4521 size_t len = strlen(os::file_separator()); 4522 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4523 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4524 } else { 4525 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4526 } 4527 return buf; 4528} 4529 4530static void print_signal_handler(outputStream* st, int sig, 4531 char* buf, size_t buflen) { 4532 struct sigaction sa; 4533 4534 sigaction(sig, NULL, &sa); 4535 4536 // See comment for SIGNIFICANT_SIGNAL_MASK define 4537 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4538 4539 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4540 4541 address handler = (sa.sa_flags & SA_SIGINFO) 4542 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4543 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4544 4545 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4546 st->print("SIG_DFL"); 4547 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4548 st->print("SIG_IGN"); 4549 } else { 4550 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4551 } 4552 4553 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 4554 4555 address rh = VMError::get_resetted_sighandler(sig); 4556 // May be, handler was resetted by VMError? 4557 if(rh != NULL) { 4558 handler = rh; 4559 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4560 } 4561 4562 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 4563 4564 // Check: is it our handler? 4565 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4566 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4567 // It is our signal handler 4568 // check for flags, reset system-used one! 4569 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4570 st->print( 4571 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4572 os::Linux::get_our_sigflags(sig)); 4573 } 4574 } 4575 st->cr(); 4576} 4577 4578 4579#define DO_SIGNAL_CHECK(sig) \ 4580 if (!sigismember(&check_signal_done, sig)) \ 4581 os::Linux::check_signal_handler(sig) 4582 4583// This method is a periodic task to check for misbehaving JNI applications 4584// under CheckJNI, we can add any periodic checks here 4585 4586void os::run_periodic_checks() { 4587 4588 if (check_signals == false) return; 4589 4590 // SEGV and BUS if overridden could potentially prevent 4591 // generation of hs*.log in the event of a crash, debugging 4592 // such a case can be very challenging, so we absolutely 4593 // check the following for a good measure: 4594 DO_SIGNAL_CHECK(SIGSEGV); 4595 DO_SIGNAL_CHECK(SIGILL); 4596 DO_SIGNAL_CHECK(SIGFPE); 4597 DO_SIGNAL_CHECK(SIGBUS); 4598 DO_SIGNAL_CHECK(SIGPIPE); 4599 DO_SIGNAL_CHECK(SIGXFSZ); 4600 4601 4602 // ReduceSignalUsage allows the user to override these handlers 4603 // see comments at the very top and jvm_solaris.h 4604 if (!ReduceSignalUsage) { 4605 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4606 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4607 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4608 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4609 } 4610 4611 DO_SIGNAL_CHECK(SR_signum); 4612 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4613} 4614 4615typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4616 4617static os_sigaction_t os_sigaction = NULL; 4618 4619void os::Linux::check_signal_handler(int sig) { 4620 char buf[O_BUFLEN]; 4621 address jvmHandler = NULL; 4622 4623 4624 struct sigaction act; 4625 if (os_sigaction == NULL) { 4626 // only trust the default sigaction, in case it has been interposed 4627 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4628 if (os_sigaction == NULL) return; 4629 } 4630 4631 os_sigaction(sig, (struct sigaction*)NULL, &act); 4632 4633 4634 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4635 4636 address thisHandler = (act.sa_flags & SA_SIGINFO) 4637 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4638 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 4639 4640 4641 switch(sig) { 4642 case SIGSEGV: 4643 case SIGBUS: 4644 case SIGFPE: 4645 case SIGPIPE: 4646 case SIGILL: 4647 case SIGXFSZ: 4648 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4649 break; 4650 4651 case SHUTDOWN1_SIGNAL: 4652 case SHUTDOWN2_SIGNAL: 4653 case SHUTDOWN3_SIGNAL: 4654 case BREAK_SIGNAL: 4655 jvmHandler = (address)user_handler(); 4656 break; 4657 4658 case INTERRUPT_SIGNAL: 4659 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4660 break; 4661 4662 default: 4663 if (sig == SR_signum) { 4664 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4665 } else { 4666 return; 4667 } 4668 break; 4669 } 4670 4671 if (thisHandler != jvmHandler) { 4672 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4673 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4674 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4675 // No need to check this sig any longer 4676 sigaddset(&check_signal_done, sig); 4677 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4678 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4679 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4680 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4681 // No need to check this sig any longer 4682 sigaddset(&check_signal_done, sig); 4683 } 4684 4685 // Dump all the signal 4686 if (sigismember(&check_signal_done, sig)) { 4687 print_signal_handlers(tty, buf, O_BUFLEN); 4688 } 4689} 4690 4691extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 4692 4693extern bool signal_name(int signo, char* buf, size_t len); 4694 4695const char* os::exception_name(int exception_code, char* buf, size_t size) { 4696 if (0 < exception_code && exception_code <= SIGRTMAX) { 4697 // signal 4698 if (!signal_name(exception_code, buf, size)) { 4699 jio_snprintf(buf, size, "SIG%d", exception_code); 4700 } 4701 return buf; 4702 } else { 4703 return NULL; 4704 } 4705} 4706 4707// this is called _before_ the most of global arguments have been parsed 4708void os::init(void) { 4709 char dummy; /* used to get a guess on initial stack address */ 4710// first_hrtime = gethrtime(); 4711 4712 // With LinuxThreads the JavaMain thread pid (primordial thread) 4713 // is different than the pid of the java launcher thread. 4714 // So, on Linux, the launcher thread pid is passed to the VM 4715 // via the sun.java.launcher.pid property. 4716 // Use this property instead of getpid() if it was correctly passed. 4717 // See bug 6351349. 4718 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4719 4720 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4721 4722 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4723 4724 init_random(1234567); 4725 4726 ThreadCritical::initialize(); 4727 4728 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4729 if (Linux::page_size() == -1) { 4730 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4731 strerror(errno))); 4732 } 4733 init_page_sizes((size_t) Linux::page_size()); 4734 4735 Linux::initialize_system_info(); 4736 4737 // main_thread points to the aboriginal thread 4738 Linux::_main_thread = pthread_self(); 4739 4740 Linux::clock_init(); 4741 initial_time_count = os::elapsed_counter(); 4742 4743 // pthread_condattr initialization for monotonic clock 4744 int status; 4745 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4746 if ((status = pthread_condattr_init(_condattr)) != 0) { 4747 fatal(err_msg("pthread_condattr_init: %s", strerror(status))); 4748 } 4749 // Only set the clock if CLOCK_MONOTONIC is available 4750 if (Linux::supports_monotonic_clock()) { 4751 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4752 if (status == EINVAL) { 4753 warning("Unable to use monotonic clock with relative timed-waits" \ 4754 " - changes to the time-of-day clock may have adverse affects"); 4755 } else { 4756 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status))); 4757 } 4758 } 4759 } 4760 // else it defaults to CLOCK_REALTIME 4761 4762 pthread_mutex_init(&dl_mutex, NULL); 4763 4764 // If the pagesize of the VM is greater than 8K determine the appropriate 4765 // number of initial guard pages. The user can change this with the 4766 // command line arguments, if needed. 4767 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4768 StackYellowPages = 1; 4769 StackRedPages = 1; 4770 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4771 } 4772} 4773 4774// To install functions for atexit system call 4775extern "C" { 4776 static void perfMemory_exit_helper() { 4777 perfMemory_exit(); 4778 } 4779} 4780 4781// this is called _after_ the global arguments have been parsed 4782jint os::init_2(void) 4783{ 4784 Linux::fast_thread_clock_init(); 4785 4786 // Allocate a single page and mark it as readable for safepoint polling 4787 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4788 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 4789 4790 os::set_polling_page( polling_page ); 4791 4792#ifndef PRODUCT 4793 if(Verbose && PrintMiscellaneous) 4794 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 4795#endif 4796 4797 if (!UseMembar) { 4798 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4799 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4800 os::set_memory_serialize_page( mem_serialize_page ); 4801 4802#ifndef PRODUCT 4803 if(Verbose && PrintMiscellaneous) 4804 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 4805#endif 4806 } 4807 4808 os::large_page_init(); 4809 4810 // initialize suspend/resume support - must do this before signal_sets_init() 4811 if (SR_initialize() != 0) { 4812 perror("SR_initialize failed"); 4813 return JNI_ERR; 4814 } 4815 4816 Linux::signal_sets_init(); 4817 Linux::install_signal_handlers(); 4818 4819 // Check minimum allowable stack size for thread creation and to initialize 4820 // the java system classes, including StackOverflowError - depends on page 4821 // size. Add a page for compiler2 recursion in main thread. 4822 // Add in 2*BytesPerWord times page size to account for VM stack during 4823 // class initialization depending on 32 or 64 bit VM. 4824 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4825 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4826 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4827 4828 size_t threadStackSizeInBytes = ThreadStackSize * K; 4829 if (threadStackSizeInBytes != 0 && 4830 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4831 tty->print_cr("\nThe stack size specified is too small, " 4832 "Specify at least %dk", 4833 os::Linux::min_stack_allowed/ K); 4834 return JNI_ERR; 4835 } 4836 4837 // Make the stack size a multiple of the page size so that 4838 // the yellow/red zones can be guarded. 4839 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4840 vm_page_size())); 4841 4842 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4843 4844 Linux::libpthread_init(); 4845 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4846 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4847 Linux::glibc_version(), Linux::libpthread_version(), 4848 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4849 } 4850 4851 if (UseNUMA) { 4852 if (!Linux::libnuma_init()) { 4853 UseNUMA = false; 4854 } else { 4855 if ((Linux::numa_max_node() < 1)) { 4856 // There's only one node(they start from 0), disable NUMA. 4857 UseNUMA = false; 4858 } 4859 } 4860 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4861 // we can make the adaptive lgrp chunk resizing work. If the user specified 4862 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4863 // disable adaptive resizing. 4864 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4865 if (FLAG_IS_DEFAULT(UseNUMA)) { 4866 UseNUMA = false; 4867 } else { 4868 if (FLAG_IS_DEFAULT(UseLargePages) && 4869 FLAG_IS_DEFAULT(UseSHM) && 4870 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4871 UseLargePages = false; 4872 } else { 4873 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing"); 4874 UseAdaptiveSizePolicy = false; 4875 UseAdaptiveNUMAChunkSizing = false; 4876 } 4877 } 4878 } 4879 if (!UseNUMA && ForceNUMA) { 4880 UseNUMA = true; 4881 } 4882 } 4883 4884 if (MaxFDLimit) { 4885 // set the number of file descriptors to max. print out error 4886 // if getrlimit/setrlimit fails but continue regardless. 4887 struct rlimit nbr_files; 4888 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4889 if (status != 0) { 4890 if (PrintMiscellaneous && (Verbose || WizardMode)) 4891 perror("os::init_2 getrlimit failed"); 4892 } else { 4893 nbr_files.rlim_cur = nbr_files.rlim_max; 4894 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4895 if (status != 0) { 4896 if (PrintMiscellaneous && (Verbose || WizardMode)) 4897 perror("os::init_2 setrlimit failed"); 4898 } 4899 } 4900 } 4901 4902 // Initialize lock used to serialize thread creation (see os::create_thread) 4903 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4904 4905 // at-exit methods are called in the reverse order of their registration. 4906 // atexit functions are called on return from main or as a result of a 4907 // call to exit(3C). There can be only 32 of these functions registered 4908 // and atexit() does not set errno. 4909 4910 if (PerfAllowAtExitRegistration) { 4911 // only register atexit functions if PerfAllowAtExitRegistration is set. 4912 // atexit functions can be delayed until process exit time, which 4913 // can be problematic for embedded VM situations. Embedded VMs should 4914 // call DestroyJavaVM() to assure that VM resources are released. 4915 4916 // note: perfMemory_exit_helper atexit function may be removed in 4917 // the future if the appropriate cleanup code can be added to the 4918 // VM_Exit VMOperation's doit method. 4919 if (atexit(perfMemory_exit_helper) != 0) { 4920 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 4921 } 4922 } 4923 4924 // initialize thread priority policy 4925 prio_init(); 4926 4927 return JNI_OK; 4928} 4929 4930// this is called at the end of vm_initialization 4931void os::init_3(void) 4932{ 4933#ifdef JAVASE_EMBEDDED 4934 // Start the MemNotifyThread 4935 if (LowMemoryProtection) { 4936 MemNotifyThread::start(); 4937 } 4938 return; 4939#endif 4940} 4941 4942// Mark the polling page as unreadable 4943void os::make_polling_page_unreadable(void) { 4944 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4945 fatal("Could not disable polling page"); 4946}; 4947 4948// Mark the polling page as readable 4949void os::make_polling_page_readable(void) { 4950 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4951 fatal("Could not enable polling page"); 4952 } 4953}; 4954 4955int os::active_processor_count() { 4956 // Linux doesn't yet have a (official) notion of processor sets, 4957 // so just return the number of online processors. 4958 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4959 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4960 return online_cpus; 4961} 4962 4963void os::set_native_thread_name(const char *name) { 4964 // Not yet implemented. 4965 return; 4966} 4967 4968bool os::distribute_processes(uint length, uint* distribution) { 4969 // Not yet implemented. 4970 return false; 4971} 4972 4973bool os::bind_to_processor(uint processor_id) { 4974 // Not yet implemented. 4975 return false; 4976} 4977 4978/// 4979 4980void os::SuspendedThreadTask::internal_do_task() { 4981 if (do_suspend(_thread->osthread())) { 4982 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4983 do_task(context); 4984 do_resume(_thread->osthread()); 4985 } 4986} 4987 4988class PcFetcher : public os::SuspendedThreadTask { 4989public: 4990 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4991 ExtendedPC result(); 4992protected: 4993 void do_task(const os::SuspendedThreadTaskContext& context); 4994private: 4995 ExtendedPC _epc; 4996}; 4997 4998ExtendedPC PcFetcher::result() { 4999 guarantee(is_done(), "task is not done yet."); 5000 return _epc; 5001} 5002 5003void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 5004 Thread* thread = context.thread(); 5005 OSThread* osthread = thread->osthread(); 5006 if (osthread->ucontext() != NULL) { 5007 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 5008 } else { 5009 // NULL context is unexpected, double-check this is the VMThread 5010 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 5011 } 5012} 5013 5014// Suspends the target using the signal mechanism and then grabs the PC before 5015// resuming the target. Used by the flat-profiler only 5016ExtendedPC os::get_thread_pc(Thread* thread) { 5017 // Make sure that it is called by the watcher for the VMThread 5018 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 5019 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 5020 5021 PcFetcher fetcher(thread); 5022 fetcher.run(); 5023 return fetcher.result(); 5024} 5025 5026int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 5027{ 5028 if (is_NPTL()) { 5029 return pthread_cond_timedwait(_cond, _mutex, _abstime); 5030 } else { 5031 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 5032 // word back to default 64bit precision if condvar is signaled. Java 5033 // wants 53bit precision. Save and restore current value. 5034 int fpu = get_fpu_control_word(); 5035 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 5036 set_fpu_control_word(fpu); 5037 return status; 5038 } 5039} 5040 5041//////////////////////////////////////////////////////////////////////////////// 5042// debug support 5043 5044bool os::find(address addr, outputStream* st) { 5045 Dl_info dlinfo; 5046 memset(&dlinfo, 0, sizeof(dlinfo)); 5047 if (dladdr(addr, &dlinfo) != 0) { 5048 st->print(PTR_FORMAT ": ", addr); 5049 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5050 st->print("%s+%#x", dlinfo.dli_sname, 5051 addr - (intptr_t)dlinfo.dli_saddr); 5052 } else if (dlinfo.dli_fbase != NULL) { 5053 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 5054 } else { 5055 st->print("<absolute address>"); 5056 } 5057 if (dlinfo.dli_fname != NULL) { 5058 st->print(" in %s", dlinfo.dli_fname); 5059 } 5060 if (dlinfo.dli_fbase != NULL) { 5061 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 5062 } 5063 st->cr(); 5064 5065 if (Verbose) { 5066 // decode some bytes around the PC 5067 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5068 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5069 address lowest = (address) dlinfo.dli_sname; 5070 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5071 if (begin < lowest) begin = lowest; 5072 Dl_info dlinfo2; 5073 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5074 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 5075 end = (address) dlinfo2.dli_saddr; 5076 Disassembler::decode(begin, end, st); 5077 } 5078 return true; 5079 } 5080 return false; 5081} 5082 5083//////////////////////////////////////////////////////////////////////////////// 5084// misc 5085 5086// This does not do anything on Linux. This is basically a hook for being 5087// able to use structured exception handling (thread-local exception filters) 5088// on, e.g., Win32. 5089void 5090os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 5091 JavaCallArguments* args, Thread* thread) { 5092 f(value, method, args, thread); 5093} 5094 5095void os::print_statistics() { 5096} 5097 5098int os::message_box(const char* title, const char* message) { 5099 int i; 5100 fdStream err(defaultStream::error_fd()); 5101 for (i = 0; i < 78; i++) err.print_raw("="); 5102 err.cr(); 5103 err.print_raw_cr(title); 5104 for (i = 0; i < 78; i++) err.print_raw("-"); 5105 err.cr(); 5106 err.print_raw_cr(message); 5107 for (i = 0; i < 78; i++) err.print_raw("="); 5108 err.cr(); 5109 5110 char buf[16]; 5111 // Prevent process from exiting upon "read error" without consuming all CPU 5112 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5113 5114 return buf[0] == 'y' || buf[0] == 'Y'; 5115} 5116 5117int os::stat(const char *path, struct stat *sbuf) { 5118 char pathbuf[MAX_PATH]; 5119 if (strlen(path) > MAX_PATH - 1) { 5120 errno = ENAMETOOLONG; 5121 return -1; 5122 } 5123 os::native_path(strcpy(pathbuf, path)); 5124 return ::stat(pathbuf, sbuf); 5125} 5126 5127bool os::check_heap(bool force) { 5128 return true; 5129} 5130 5131int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 5132 return ::vsnprintf(buf, count, format, args); 5133} 5134 5135// Is a (classpath) directory empty? 5136bool os::dir_is_empty(const char* path) { 5137 DIR *dir = NULL; 5138 struct dirent *ptr; 5139 5140 dir = opendir(path); 5141 if (dir == NULL) return true; 5142 5143 /* Scan the directory */ 5144 bool result = true; 5145 char buf[sizeof(struct dirent) + MAX_PATH]; 5146 while (result && (ptr = ::readdir(dir)) != NULL) { 5147 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5148 result = false; 5149 } 5150 } 5151 closedir(dir); 5152 return result; 5153} 5154 5155// This code originates from JDK's sysOpen and open64_w 5156// from src/solaris/hpi/src/system_md.c 5157 5158#ifndef O_DELETE 5159#define O_DELETE 0x10000 5160#endif 5161 5162// Open a file. Unlink the file immediately after open returns 5163// if the specified oflag has the O_DELETE flag set. 5164// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 5165 5166int os::open(const char *path, int oflag, int mode) { 5167 5168 if (strlen(path) > MAX_PATH - 1) { 5169 errno = ENAMETOOLONG; 5170 return -1; 5171 } 5172 int fd; 5173 int o_delete = (oflag & O_DELETE); 5174 oflag = oflag & ~O_DELETE; 5175 5176 fd = ::open64(path, oflag, mode); 5177 if (fd == -1) return -1; 5178 5179 //If the open succeeded, the file might still be a directory 5180 { 5181 struct stat64 buf64; 5182 int ret = ::fstat64(fd, &buf64); 5183 int st_mode = buf64.st_mode; 5184 5185 if (ret != -1) { 5186 if ((st_mode & S_IFMT) == S_IFDIR) { 5187 errno = EISDIR; 5188 ::close(fd); 5189 return -1; 5190 } 5191 } else { 5192 ::close(fd); 5193 return -1; 5194 } 5195 } 5196 5197 /* 5198 * All file descriptors that are opened in the JVM and not 5199 * specifically destined for a subprocess should have the 5200 * close-on-exec flag set. If we don't set it, then careless 3rd 5201 * party native code might fork and exec without closing all 5202 * appropriate file descriptors (e.g. as we do in closeDescriptors in 5203 * UNIXProcess.c), and this in turn might: 5204 * 5205 * - cause end-of-file to fail to be detected on some file 5206 * descriptors, resulting in mysterious hangs, or 5207 * 5208 * - might cause an fopen in the subprocess to fail on a system 5209 * suffering from bug 1085341. 5210 * 5211 * (Yes, the default setting of the close-on-exec flag is a Unix 5212 * design flaw) 5213 * 5214 * See: 5215 * 1085341: 32-bit stdio routines should support file descriptors >255 5216 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5217 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5218 */ 5219#ifdef FD_CLOEXEC 5220 { 5221 int flags = ::fcntl(fd, F_GETFD); 5222 if (flags != -1) 5223 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5224 } 5225#endif 5226 5227 if (o_delete != 0) { 5228 ::unlink(path); 5229 } 5230 return fd; 5231} 5232 5233 5234// create binary file, rewriting existing file if required 5235int os::create_binary_file(const char* path, bool rewrite_existing) { 5236 int oflags = O_WRONLY | O_CREAT; 5237 if (!rewrite_existing) { 5238 oflags |= O_EXCL; 5239 } 5240 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5241} 5242 5243// return current position of file pointer 5244jlong os::current_file_offset(int fd) { 5245 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5246} 5247 5248// move file pointer to the specified offset 5249jlong os::seek_to_file_offset(int fd, jlong offset) { 5250 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5251} 5252 5253// This code originates from JDK's sysAvailable 5254// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5255 5256int os::available(int fd, jlong *bytes) { 5257 jlong cur, end; 5258 int mode; 5259 struct stat64 buf64; 5260 5261 if (::fstat64(fd, &buf64) >= 0) { 5262 mode = buf64.st_mode; 5263 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5264 /* 5265 * XXX: is the following call interruptible? If so, this might 5266 * need to go through the INTERRUPT_IO() wrapper as for other 5267 * blocking, interruptible calls in this file. 5268 */ 5269 int n; 5270 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5271 *bytes = n; 5272 return 1; 5273 } 5274 } 5275 } 5276 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5277 return 0; 5278 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5279 return 0; 5280 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5281 return 0; 5282 } 5283 *bytes = end - cur; 5284 return 1; 5285} 5286 5287int os::socket_available(int fd, jint *pbytes) { 5288 // Linux doc says EINTR not returned, unlike Solaris 5289 int ret = ::ioctl(fd, FIONREAD, pbytes); 5290 5291 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 5292 // is expected to return 0 on failure and 1 on success to the jdk. 5293 return (ret < 0) ? 0 : 1; 5294} 5295 5296// Map a block of memory. 5297char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5298 char *addr, size_t bytes, bool read_only, 5299 bool allow_exec) { 5300 int prot; 5301 int flags = MAP_PRIVATE; 5302 5303 if (read_only) { 5304 prot = PROT_READ; 5305 } else { 5306 prot = PROT_READ | PROT_WRITE; 5307 } 5308 5309 if (allow_exec) { 5310 prot |= PROT_EXEC; 5311 } 5312 5313 if (addr != NULL) { 5314 flags |= MAP_FIXED; 5315 } 5316 5317 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5318 fd, file_offset); 5319 if (mapped_address == MAP_FAILED) { 5320 return NULL; 5321 } 5322 return mapped_address; 5323} 5324 5325 5326// Remap a block of memory. 5327char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5328 char *addr, size_t bytes, bool read_only, 5329 bool allow_exec) { 5330 // same as map_memory() on this OS 5331 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5332 allow_exec); 5333} 5334 5335 5336// Unmap a block of memory. 5337bool os::pd_unmap_memory(char* addr, size_t bytes) { 5338 return munmap(addr, bytes) == 0; 5339} 5340 5341static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5342 5343static clockid_t thread_cpu_clockid(Thread* thread) { 5344 pthread_t tid = thread->osthread()->pthread_id(); 5345 clockid_t clockid; 5346 5347 // Get thread clockid 5348 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5349 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5350 return clockid; 5351} 5352 5353// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5354// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5355// of a thread. 5356// 5357// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5358// the fast estimate available on the platform. 5359 5360jlong os::current_thread_cpu_time() { 5361 if (os::Linux::supports_fast_thread_cpu_time()) { 5362 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5363 } else { 5364 // return user + sys since the cost is the same 5365 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5366 } 5367} 5368 5369jlong os::thread_cpu_time(Thread* thread) { 5370 // consistent with what current_thread_cpu_time() returns 5371 if (os::Linux::supports_fast_thread_cpu_time()) { 5372 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5373 } else { 5374 return slow_thread_cpu_time(thread, true /* user + sys */); 5375 } 5376} 5377 5378jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5379 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5380 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5381 } else { 5382 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5383 } 5384} 5385 5386jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5387 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5388 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5389 } else { 5390 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5391 } 5392} 5393 5394// 5395// -1 on error. 5396// 5397 5398static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5399 static bool proc_task_unchecked = true; 5400 static const char *proc_stat_path = "/proc/%d/stat"; 5401 pid_t tid = thread->osthread()->thread_id(); 5402 char *s; 5403 char stat[2048]; 5404 int statlen; 5405 char proc_name[64]; 5406 int count; 5407 long sys_time, user_time; 5408 char cdummy; 5409 int idummy; 5410 long ldummy; 5411 FILE *fp; 5412 5413 // The /proc/<tid>/stat aggregates per-process usage on 5414 // new Linux kernels 2.6+ where NPTL is supported. 5415 // The /proc/self/task/<tid>/stat still has the per-thread usage. 5416 // See bug 6328462. 5417 // There possibly can be cases where there is no directory 5418 // /proc/self/task, so we check its availability. 5419 if (proc_task_unchecked && os::Linux::is_NPTL()) { 5420 // This is executed only once 5421 proc_task_unchecked = false; 5422 fp = fopen("/proc/self/task", "r"); 5423 if (fp != NULL) { 5424 proc_stat_path = "/proc/self/task/%d/stat"; 5425 fclose(fp); 5426 } 5427 } 5428 5429 sprintf(proc_name, proc_stat_path, tid); 5430 fp = fopen(proc_name, "r"); 5431 if ( fp == NULL ) return -1; 5432 statlen = fread(stat, 1, 2047, fp); 5433 stat[statlen] = '\0'; 5434 fclose(fp); 5435 5436 // Skip pid and the command string. Note that we could be dealing with 5437 // weird command names, e.g. user could decide to rename java launcher 5438 // to "java 1.4.2 :)", then the stat file would look like 5439 // 1234 (java 1.4.2 :)) R ... ... 5440 // We don't really need to know the command string, just find the last 5441 // occurrence of ")" and then start parsing from there. See bug 4726580. 5442 s = strrchr(stat, ')'); 5443 if (s == NULL ) return -1; 5444 5445 // Skip blank chars 5446 do s++; while (isspace(*s)); 5447 5448 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5449 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5450 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5451 &user_time, &sys_time); 5452 if ( count != 13 ) return -1; 5453 if (user_sys_cpu_time) { 5454 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5455 } else { 5456 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5457 } 5458} 5459 5460void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5461 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5462 info_ptr->may_skip_backward = false; // elapsed time not wall time 5463 info_ptr->may_skip_forward = false; // elapsed time not wall time 5464 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5465} 5466 5467void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5468 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5469 info_ptr->may_skip_backward = false; // elapsed time not wall time 5470 info_ptr->may_skip_forward = false; // elapsed time not wall time 5471 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5472} 5473 5474bool os::is_thread_cpu_time_supported() { 5475 return true; 5476} 5477 5478// System loadavg support. Returns -1 if load average cannot be obtained. 5479// Linux doesn't yet have a (official) notion of processor sets, 5480// so just return the system wide load average. 5481int os::loadavg(double loadavg[], int nelem) { 5482 return ::getloadavg(loadavg, nelem); 5483} 5484 5485void os::pause() { 5486 char filename[MAX_PATH]; 5487 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5488 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5489 } else { 5490 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5491 } 5492 5493 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5494 if (fd != -1) { 5495 struct stat buf; 5496 ::close(fd); 5497 while (::stat(filename, &buf) == 0) { 5498 (void)::poll(NULL, 0, 100); 5499 } 5500 } else { 5501 jio_fprintf(stderr, 5502 "Could not open pause file '%s', continuing immediately.\n", filename); 5503 } 5504} 5505 5506 5507// Refer to the comments in os_solaris.cpp park-unpark. 5508// 5509// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5510// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5511// For specifics regarding the bug see GLIBC BUGID 261237 : 5512// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5513// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5514// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5515// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5516// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5517// and monitorenter when we're using 1-0 locking. All those operations may result in 5518// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5519// of libpthread avoids the problem, but isn't practical. 5520// 5521// Possible remedies: 5522// 5523// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5524// This is palliative and probabilistic, however. If the thread is preempted 5525// between the call to compute_abstime() and pthread_cond_timedwait(), more 5526// than the minimum period may have passed, and the abstime may be stale (in the 5527// past) resultin in a hang. Using this technique reduces the odds of a hang 5528// but the JVM is still vulnerable, particularly on heavily loaded systems. 5529// 5530// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5531// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5532// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5533// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5534// thread. 5535// 5536// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5537// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5538// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5539// This also works well. In fact it avoids kernel-level scalability impediments 5540// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5541// timers in a graceful fashion. 5542// 5543// 4. When the abstime value is in the past it appears that control returns 5544// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5545// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5546// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5547// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5548// It may be possible to avoid reinitialization by checking the return 5549// value from pthread_cond_timedwait(). In addition to reinitializing the 5550// condvar we must establish the invariant that cond_signal() is only called 5551// within critical sections protected by the adjunct mutex. This prevents 5552// cond_signal() from "seeing" a condvar that's in the midst of being 5553// reinitialized or that is corrupt. Sadly, this invariant obviates the 5554// desirable signal-after-unlock optimization that avoids futile context switching. 5555// 5556// I'm also concerned that some versions of NTPL might allocate an auxilliary 5557// structure when a condvar is used or initialized. cond_destroy() would 5558// release the helper structure. Our reinitialize-after-timedwait fix 5559// put excessive stress on malloc/free and locks protecting the c-heap. 5560// 5561// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5562// It may be possible to refine (4) by checking the kernel and NTPL verisons 5563// and only enabling the work-around for vulnerable environments. 5564 5565// utility to compute the abstime argument to timedwait: 5566// millis is the relative timeout time 5567// abstime will be the absolute timeout time 5568// TODO: replace compute_abstime() with unpackTime() 5569 5570static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5571 if (millis < 0) millis = 0; 5572 5573 jlong seconds = millis / 1000; 5574 millis %= 1000; 5575 if (seconds > 50000000) { // see man cond_timedwait(3T) 5576 seconds = 50000000; 5577 } 5578 5579 if (os::Linux::supports_monotonic_clock()) { 5580 struct timespec now; 5581 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5582 assert_status(status == 0, status, "clock_gettime"); 5583 abstime->tv_sec = now.tv_sec + seconds; 5584 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5585 if (nanos >= NANOSECS_PER_SEC) { 5586 abstime->tv_sec += 1; 5587 nanos -= NANOSECS_PER_SEC; 5588 } 5589 abstime->tv_nsec = nanos; 5590 } else { 5591 struct timeval now; 5592 int status = gettimeofday(&now, NULL); 5593 assert(status == 0, "gettimeofday"); 5594 abstime->tv_sec = now.tv_sec + seconds; 5595 long usec = now.tv_usec + millis * 1000; 5596 if (usec >= 1000000) { 5597 abstime->tv_sec += 1; 5598 usec -= 1000000; 5599 } 5600 abstime->tv_nsec = usec * 1000; 5601 } 5602 return abstime; 5603} 5604 5605 5606// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 5607// Conceptually TryPark() should be equivalent to park(0). 5608 5609int os::PlatformEvent::TryPark() { 5610 for (;;) { 5611 const int v = _Event ; 5612 guarantee ((v == 0) || (v == 1), "invariant") ; 5613 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 5614 } 5615} 5616 5617void os::PlatformEvent::park() { // AKA "down()" 5618 // Invariant: Only the thread associated with the Event/PlatformEvent 5619 // may call park(). 5620 // TODO: assert that _Assoc != NULL or _Assoc == Self 5621 int v ; 5622 for (;;) { 5623 v = _Event ; 5624 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5625 } 5626 guarantee (v >= 0, "invariant") ; 5627 if (v == 0) { 5628 // Do this the hard way by blocking ... 5629 int status = pthread_mutex_lock(_mutex); 5630 assert_status(status == 0, status, "mutex_lock"); 5631 guarantee (_nParked == 0, "invariant") ; 5632 ++ _nParked ; 5633 while (_Event < 0) { 5634 status = pthread_cond_wait(_cond, _mutex); 5635 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5636 // Treat this the same as if the wait was interrupted 5637 if (status == ETIME) { status = EINTR; } 5638 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5639 } 5640 -- _nParked ; 5641 5642 _Event = 0 ; 5643 status = pthread_mutex_unlock(_mutex); 5644 assert_status(status == 0, status, "mutex_unlock"); 5645 // Paranoia to ensure our locked and lock-free paths interact 5646 // correctly with each other. 5647 OrderAccess::fence(); 5648 } 5649 guarantee (_Event >= 0, "invariant") ; 5650} 5651 5652int os::PlatformEvent::park(jlong millis) { 5653 guarantee (_nParked == 0, "invariant") ; 5654 5655 int v ; 5656 for (;;) { 5657 v = _Event ; 5658 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5659 } 5660 guarantee (v >= 0, "invariant") ; 5661 if (v != 0) return OS_OK ; 5662 5663 // We do this the hard way, by blocking the thread. 5664 // Consider enforcing a minimum timeout value. 5665 struct timespec abst; 5666 compute_abstime(&abst, millis); 5667 5668 int ret = OS_TIMEOUT; 5669 int status = pthread_mutex_lock(_mutex); 5670 assert_status(status == 0, status, "mutex_lock"); 5671 guarantee (_nParked == 0, "invariant") ; 5672 ++_nParked ; 5673 5674 // Object.wait(timo) will return because of 5675 // (a) notification 5676 // (b) timeout 5677 // (c) thread.interrupt 5678 // 5679 // Thread.interrupt and object.notify{All} both call Event::set. 5680 // That is, we treat thread.interrupt as a special case of notification. 5681 // The underlying Solaris implementation, cond_timedwait, admits 5682 // spurious/premature wakeups, but the JLS/JVM spec prevents the 5683 // JVM from making those visible to Java code. As such, we must 5684 // filter out spurious wakeups. We assume all ETIME returns are valid. 5685 // 5686 // TODO: properly differentiate simultaneous notify+interrupt. 5687 // In that case, we should propagate the notify to another waiter. 5688 5689 while (_Event < 0) { 5690 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5691 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5692 pthread_cond_destroy (_cond); 5693 pthread_cond_init (_cond, os::Linux::condAttr()) ; 5694 } 5695 assert_status(status == 0 || status == EINTR || 5696 status == ETIME || status == ETIMEDOUT, 5697 status, "cond_timedwait"); 5698 if (!FilterSpuriousWakeups) break ; // previous semantics 5699 if (status == ETIME || status == ETIMEDOUT) break ; 5700 // We consume and ignore EINTR and spurious wakeups. 5701 } 5702 --_nParked ; 5703 if (_Event >= 0) { 5704 ret = OS_OK; 5705 } 5706 _Event = 0 ; 5707 status = pthread_mutex_unlock(_mutex); 5708 assert_status(status == 0, status, "mutex_unlock"); 5709 assert (_nParked == 0, "invariant") ; 5710 // Paranoia to ensure our locked and lock-free paths interact 5711 // correctly with each other. 5712 OrderAccess::fence(); 5713 return ret; 5714} 5715 5716void os::PlatformEvent::unpark() { 5717 // Transitions for _Event: 5718 // 0 :=> 1 5719 // 1 :=> 1 5720 // -1 :=> either 0 or 1; must signal target thread 5721 // That is, we can safely transition _Event from -1 to either 5722 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back 5723 // unpark() calls. 5724 // See also: "Semaphores in Plan 9" by Mullender & Cox 5725 // 5726 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5727 // that it will take two back-to-back park() calls for the owning 5728 // thread to block. This has the benefit of forcing a spurious return 5729 // from the first park() call after an unpark() call which will help 5730 // shake out uses of park() and unpark() without condition variables. 5731 5732 if (Atomic::xchg(1, &_Event) >= 0) return; 5733 5734 // Wait for the thread associated with the event to vacate 5735 int status = pthread_mutex_lock(_mutex); 5736 assert_status(status == 0, status, "mutex_lock"); 5737 int AnyWaiters = _nParked; 5738 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5739 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5740 AnyWaiters = 0; 5741 pthread_cond_signal(_cond); 5742 } 5743 status = pthread_mutex_unlock(_mutex); 5744 assert_status(status == 0, status, "mutex_unlock"); 5745 if (AnyWaiters != 0) { 5746 status = pthread_cond_signal(_cond); 5747 assert_status(status == 0, status, "cond_signal"); 5748 } 5749 5750 // Note that we signal() _after dropping the lock for "immortal" Events. 5751 // This is safe and avoids a common class of futile wakeups. In rare 5752 // circumstances this can cause a thread to return prematurely from 5753 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5754 // simply re-test the condition and re-park itself. 5755} 5756 5757 5758// JSR166 5759// ------------------------------------------------------- 5760 5761/* 5762 * The solaris and linux implementations of park/unpark are fairly 5763 * conservative for now, but can be improved. They currently use a 5764 * mutex/condvar pair, plus a a count. 5765 * Park decrements count if > 0, else does a condvar wait. Unpark 5766 * sets count to 1 and signals condvar. Only one thread ever waits 5767 * on the condvar. Contention seen when trying to park implies that someone 5768 * is unparking you, so don't wait. And spurious returns are fine, so there 5769 * is no need to track notifications. 5770 */ 5771 5772#define MAX_SECS 100000000 5773/* 5774 * This code is common to linux and solaris and will be moved to a 5775 * common place in dolphin. 5776 * 5777 * The passed in time value is either a relative time in nanoseconds 5778 * or an absolute time in milliseconds. Either way it has to be unpacked 5779 * into suitable seconds and nanoseconds components and stored in the 5780 * given timespec structure. 5781 * Given time is a 64-bit value and the time_t used in the timespec is only 5782 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 5783 * overflow if times way in the future are given. Further on Solaris versions 5784 * prior to 10 there is a restriction (see cond_timedwait) that the specified 5785 * number of seconds, in abstime, is less than current_time + 100,000,000. 5786 * As it will be 28 years before "now + 100000000" will overflow we can 5787 * ignore overflow and just impose a hard-limit on seconds using the value 5788 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 5789 * years from "now". 5790 */ 5791 5792static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5793 assert (time > 0, "convertTime"); 5794 time_t max_secs = 0; 5795 5796 if (!os::Linux::supports_monotonic_clock() || isAbsolute) { 5797 struct timeval now; 5798 int status = gettimeofday(&now, NULL); 5799 assert(status == 0, "gettimeofday"); 5800 5801 max_secs = now.tv_sec + MAX_SECS; 5802 5803 if (isAbsolute) { 5804 jlong secs = time / 1000; 5805 if (secs > max_secs) { 5806 absTime->tv_sec = max_secs; 5807 } else { 5808 absTime->tv_sec = secs; 5809 } 5810 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5811 } else { 5812 jlong secs = time / NANOSECS_PER_SEC; 5813 if (secs >= MAX_SECS) { 5814 absTime->tv_sec = max_secs; 5815 absTime->tv_nsec = 0; 5816 } else { 5817 absTime->tv_sec = now.tv_sec + secs; 5818 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5819 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5820 absTime->tv_nsec -= NANOSECS_PER_SEC; 5821 ++absTime->tv_sec; // note: this must be <= max_secs 5822 } 5823 } 5824 } 5825 } else { 5826 // must be relative using monotonic clock 5827 struct timespec now; 5828 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5829 assert_status(status == 0, status, "clock_gettime"); 5830 max_secs = now.tv_sec + MAX_SECS; 5831 jlong secs = time / NANOSECS_PER_SEC; 5832 if (secs >= MAX_SECS) { 5833 absTime->tv_sec = max_secs; 5834 absTime->tv_nsec = 0; 5835 } else { 5836 absTime->tv_sec = now.tv_sec + secs; 5837 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5838 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5839 absTime->tv_nsec -= NANOSECS_PER_SEC; 5840 ++absTime->tv_sec; // note: this must be <= max_secs 5841 } 5842 } 5843 } 5844 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5845 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5846 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5847 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5848} 5849 5850void Parker::park(bool isAbsolute, jlong time) { 5851 // Ideally we'd do something useful while spinning, such 5852 // as calling unpackTime(). 5853 5854 // Optional fast-path check: 5855 // Return immediately if a permit is available. 5856 // We depend on Atomic::xchg() having full barrier semantics 5857 // since we are doing a lock-free update to _counter. 5858 if (Atomic::xchg(0, &_counter) > 0) return; 5859 5860 Thread* thread = Thread::current(); 5861 assert(thread->is_Java_thread(), "Must be JavaThread"); 5862 JavaThread *jt = (JavaThread *)thread; 5863 5864 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5865 // Check interrupt before trying to wait 5866 if (Thread::is_interrupted(thread, false)) { 5867 return; 5868 } 5869 5870 // Next, demultiplex/decode time arguments 5871 timespec absTime; 5872 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5873 return; 5874 } 5875 if (time > 0) { 5876 unpackTime(&absTime, isAbsolute, time); 5877 } 5878 5879 5880 // Enter safepoint region 5881 // Beware of deadlocks such as 6317397. 5882 // The per-thread Parker:: mutex is a classic leaf-lock. 5883 // In particular a thread must never block on the Threads_lock while 5884 // holding the Parker:: mutex. If safepoints are pending both the 5885 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5886 ThreadBlockInVM tbivm(jt); 5887 5888 // Don't wait if cannot get lock since interference arises from 5889 // unblocking. Also. check interrupt before trying wait 5890 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5891 return; 5892 } 5893 5894 int status ; 5895 if (_counter > 0) { // no wait needed 5896 _counter = 0; 5897 status = pthread_mutex_unlock(_mutex); 5898 assert (status == 0, "invariant") ; 5899 // Paranoia to ensure our locked and lock-free paths interact 5900 // correctly with each other and Java-level accesses. 5901 OrderAccess::fence(); 5902 return; 5903 } 5904 5905#ifdef ASSERT 5906 // Don't catch signals while blocked; let the running threads have the signals. 5907 // (This allows a debugger to break into the running thread.) 5908 sigset_t oldsigs; 5909 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5910 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5911#endif 5912 5913 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5914 jt->set_suspend_equivalent(); 5915 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5916 5917 assert(_cur_index == -1, "invariant"); 5918 if (time == 0) { 5919 _cur_index = REL_INDEX; // arbitrary choice when not timed 5920 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ; 5921 } else { 5922 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5923 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ; 5924 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5925 pthread_cond_destroy (&_cond[_cur_index]) ; 5926 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr()); 5927 } 5928 } 5929 _cur_index = -1; 5930 assert_status(status == 0 || status == EINTR || 5931 status == ETIME || status == ETIMEDOUT, 5932 status, "cond_timedwait"); 5933 5934#ifdef ASSERT 5935 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5936#endif 5937 5938 _counter = 0 ; 5939 status = pthread_mutex_unlock(_mutex) ; 5940 assert_status(status == 0, status, "invariant") ; 5941 // Paranoia to ensure our locked and lock-free paths interact 5942 // correctly with each other and Java-level accesses. 5943 OrderAccess::fence(); 5944 5945 // If externally suspended while waiting, re-suspend 5946 if (jt->handle_special_suspend_equivalent_condition()) { 5947 jt->java_suspend_self(); 5948 } 5949} 5950 5951void Parker::unpark() { 5952 int s, status ; 5953 status = pthread_mutex_lock(_mutex); 5954 assert (status == 0, "invariant") ; 5955 s = _counter; 5956 _counter = 1; 5957 if (s < 1) { 5958 // thread might be parked 5959 if (_cur_index != -1) { 5960 // thread is definitely parked 5961 if (WorkAroundNPTLTimedWaitHang) { 5962 status = pthread_cond_signal (&_cond[_cur_index]); 5963 assert (status == 0, "invariant"); 5964 status = pthread_mutex_unlock(_mutex); 5965 assert (status == 0, "invariant"); 5966 } else { 5967 status = pthread_mutex_unlock(_mutex); 5968 assert (status == 0, "invariant"); 5969 status = pthread_cond_signal (&_cond[_cur_index]); 5970 assert (status == 0, "invariant"); 5971 } 5972 } else { 5973 pthread_mutex_unlock(_mutex); 5974 assert (status == 0, "invariant") ; 5975 } 5976 } else { 5977 pthread_mutex_unlock(_mutex); 5978 assert (status == 0, "invariant") ; 5979 } 5980} 5981 5982 5983extern char** environ; 5984 5985#ifndef __NR_fork 5986#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5987#endif 5988 5989#ifndef __NR_execve 5990#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5991#endif 5992 5993// Run the specified command in a separate process. Return its exit value, 5994// or -1 on failure (e.g. can't fork a new process). 5995// Unlike system(), this function can be called from signal handler. It 5996// doesn't block SIGINT et al. 5997int os::fork_and_exec(char* cmd) { 5998 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5999 6000 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 6001 // pthread_atfork handlers and reset pthread library. All we need is a 6002 // separate process to execve. Make a direct syscall to fork process. 6003 // On IA64 there's no fork syscall, we have to use fork() and hope for 6004 // the best... 6005 pid_t pid = NOT_IA64(syscall(__NR_fork);) 6006 IA64_ONLY(fork();) 6007 6008 if (pid < 0) { 6009 // fork failed 6010 return -1; 6011 6012 } else if (pid == 0) { 6013 // child process 6014 6015 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 6016 // first to kill every thread on the thread list. Because this list is 6017 // not reset by fork() (see notes above), execve() will instead kill 6018 // every thread in the parent process. We know this is the only thread 6019 // in the new process, so make a system call directly. 6020 // IA64 should use normal execve() from glibc to match the glibc fork() 6021 // above. 6022 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 6023 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 6024 6025 // execve failed 6026 _exit(-1); 6027 6028 } else { 6029 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 6030 // care about the actual exit code, for now. 6031 6032 int status; 6033 6034 // Wait for the child process to exit. This returns immediately if 6035 // the child has already exited. */ 6036 while (waitpid(pid, &status, 0) < 0) { 6037 switch (errno) { 6038 case ECHILD: return 0; 6039 case EINTR: break; 6040 default: return -1; 6041 } 6042 } 6043 6044 if (WIFEXITED(status)) { 6045 // The child exited normally; get its exit code. 6046 return WEXITSTATUS(status); 6047 } else if (WIFSIGNALED(status)) { 6048 // The child exited because of a signal 6049 // The best value to return is 0x80 + signal number, 6050 // because that is what all Unix shells do, and because 6051 // it allows callers to distinguish between process exit and 6052 // process death by signal. 6053 return 0x80 + WTERMSIG(status); 6054 } else { 6055 // Unknown exit code; pass it through 6056 return status; 6057 } 6058 } 6059} 6060 6061// is_headless_jre() 6062// 6063// Test for the existence of xawt/libmawt.so or libawt_xawt.so 6064// in order to report if we are running in a headless jre 6065// 6066// Since JDK8 xawt/libmawt.so was moved into the same directory 6067// as libawt.so, and renamed libawt_xawt.so 6068// 6069bool os::is_headless_jre() { 6070 struct stat statbuf; 6071 char buf[MAXPATHLEN]; 6072 char libmawtpath[MAXPATHLEN]; 6073 const char *xawtstr = "/xawt/libmawt.so"; 6074 const char *new_xawtstr = "/libawt_xawt.so"; 6075 char *p; 6076 6077 // Get path to libjvm.so 6078 os::jvm_path(buf, sizeof(buf)); 6079 6080 // Get rid of libjvm.so 6081 p = strrchr(buf, '/'); 6082 if (p == NULL) return false; 6083 else *p = '\0'; 6084 6085 // Get rid of client or server 6086 p = strrchr(buf, '/'); 6087 if (p == NULL) return false; 6088 else *p = '\0'; 6089 6090 // check xawt/libmawt.so 6091 strcpy(libmawtpath, buf); 6092 strcat(libmawtpath, xawtstr); 6093 if (::stat(libmawtpath, &statbuf) == 0) return false; 6094 6095 // check libawt_xawt.so 6096 strcpy(libmawtpath, buf); 6097 strcat(libmawtpath, new_xawtstr); 6098 if (::stat(libmawtpath, &statbuf) == 0) return false; 6099 6100 return true; 6101} 6102 6103// Get the default path to the core file 6104// Returns the length of the string 6105int os::get_core_path(char* buffer, size_t bufferSize) { 6106 const char* p = get_current_directory(buffer, bufferSize); 6107 6108 if (p == NULL) { 6109 assert(p != NULL, "failed to get current directory"); 6110 return 0; 6111 } 6112 6113 return strlen(buffer); 6114} 6115 6116#ifdef JAVASE_EMBEDDED 6117// 6118// A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 6119// 6120MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 6121 6122// ctor 6123// 6124MemNotifyThread::MemNotifyThread(int fd): Thread() { 6125 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 6126 _fd = fd; 6127 6128 if (os::create_thread(this, os::os_thread)) { 6129 _memnotify_thread = this; 6130 os::set_priority(this, NearMaxPriority); 6131 os::start_thread(this); 6132 } 6133} 6134 6135// Where all the work gets done 6136// 6137void MemNotifyThread::run() { 6138 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 6139 6140 // Set up the select arguments 6141 fd_set rfds; 6142 if (_fd != -1) { 6143 FD_ZERO(&rfds); 6144 FD_SET(_fd, &rfds); 6145 } 6146 6147 // Now wait for the mem_notify device to wake up 6148 while (1) { 6149 // Wait for the mem_notify device to signal us.. 6150 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 6151 if (rc == -1) { 6152 perror("select!\n"); 6153 break; 6154 } else if (rc) { 6155 //ssize_t free_before = os::available_memory(); 6156 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 6157 6158 // The kernel is telling us there is not much memory left... 6159 // try to do something about that 6160 6161 // If we are not already in a GC, try one. 6162 if (!Universe::heap()->is_gc_active()) { 6163 Universe::heap()->collect(GCCause::_allocation_failure); 6164 6165 //ssize_t free_after = os::available_memory(); 6166 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 6167 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 6168 } 6169 // We might want to do something like the following if we find the GC's are not helping... 6170 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 6171 } 6172 } 6173} 6174 6175// 6176// See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 6177// 6178void MemNotifyThread::start() { 6179 int fd; 6180 fd = open ("/dev/mem_notify", O_RDONLY, 0); 6181 if (fd < 0) { 6182 return; 6183 } 6184 6185 if (memnotify_thread() == NULL) { 6186 new MemNotifyThread(fd); 6187 } 6188} 6189 6190#endif // JAVASE_EMBEDDED 6191 6192 6193/////////////// Unit tests /////////////// 6194 6195#ifndef PRODUCT 6196 6197#define test_log(...) \ 6198 do {\ 6199 if (VerboseInternalVMTests) { \ 6200 tty->print_cr(__VA_ARGS__); \ 6201 tty->flush(); \ 6202 }\ 6203 } while (false) 6204 6205class TestReserveMemorySpecial : AllStatic { 6206 public: 6207 static void small_page_write(void* addr, size_t size) { 6208 size_t page_size = os::vm_page_size(); 6209 6210 char* end = (char*)addr + size; 6211 for (char* p = (char*)addr; p < end; p += page_size) { 6212 *p = 1; 6213 } 6214 } 6215 6216 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6217 if (!UseHugeTLBFS) { 6218 return; 6219 } 6220 6221 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6222 6223 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6224 6225 if (addr != NULL) { 6226 small_page_write(addr, size); 6227 6228 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6229 } 6230 } 6231 6232 static void test_reserve_memory_special_huge_tlbfs_only() { 6233 if (!UseHugeTLBFS) { 6234 return; 6235 } 6236 6237 size_t lp = os::large_page_size(); 6238 6239 for (size_t size = lp; size <= lp * 10; size += lp) { 6240 test_reserve_memory_special_huge_tlbfs_only(size); 6241 } 6242 } 6243 6244 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) { 6245 if (!UseHugeTLBFS) { 6246 return; 6247 } 6248 6249 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")", 6250 size, alignment); 6251 6252 assert(size >= os::large_page_size(), "Incorrect input to test"); 6253 6254 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6255 6256 if (addr != NULL) { 6257 small_page_write(addr, size); 6258 6259 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6260 } 6261 } 6262 6263 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) { 6264 size_t lp = os::large_page_size(); 6265 size_t ag = os::vm_allocation_granularity(); 6266 6267 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6268 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment); 6269 } 6270 } 6271 6272 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6273 size_t lp = os::large_page_size(); 6274 size_t ag = os::vm_allocation_granularity(); 6275 6276 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp); 6277 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag); 6278 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2); 6279 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2); 6280 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag); 6281 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag); 6282 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2); 6283 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10); 6284 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2); 6285 } 6286 6287 static void test_reserve_memory_special_huge_tlbfs() { 6288 if (!UseHugeTLBFS) { 6289 return; 6290 } 6291 6292 test_reserve_memory_special_huge_tlbfs_only(); 6293 test_reserve_memory_special_huge_tlbfs_mixed(); 6294 } 6295 6296 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6297 if (!UseSHM) { 6298 return; 6299 } 6300 6301 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6302 6303 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6304 6305 if (addr != NULL) { 6306 assert(is_ptr_aligned(addr, alignment), "Check"); 6307 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6308 6309 small_page_write(addr, size); 6310 6311 os::Linux::release_memory_special_shm(addr, size); 6312 } 6313 } 6314 6315 static void test_reserve_memory_special_shm() { 6316 size_t lp = os::large_page_size(); 6317 size_t ag = os::vm_allocation_granularity(); 6318 6319 for (size_t size = ag; size < lp * 3; size += ag) { 6320 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6321 test_reserve_memory_special_shm(size, alignment); 6322 } 6323 } 6324 } 6325 6326 static void test() { 6327 test_reserve_memory_special_huge_tlbfs(); 6328 test_reserve_memory_special_shm(); 6329 } 6330}; 6331 6332void TestReserveMemorySpecial_test() { 6333 TestReserveMemorySpecial::test(); 6334} 6335 6336#endif 6337