os_linux.cpp revision 7051:0420e825bb3c
198186Sgordon/* 298186Sgordon * Copyright (c) 1999, 2014, Oracle and/or its affiliates. All rights reserved. 378344Sobrien * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 498186Sgordon * 578344Sobrien * This code is free software; you can redistribute it and/or modify it 678344Sobrien * under the terms of the GNU General Public License version 2 only, as 778344Sobrien * published by the Free Software Foundation. 878344Sobrien * 978344Sobrien * This code is distributed in the hope that it will be useful, but WITHOUT 1078344Sobrien * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 1178344Sobrien * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 1278344Sobrien * version 2 for more details (a copy is included in the LICENSE file that 1378344Sobrien * accompanied this code). 1478344Sobrien * 1578344Sobrien * You should have received a copy of the GNU General Public License version 1678344Sobrien * 2 along with this work; if not, write to the Free Software Foundation, 1778344Sobrien * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 1878344Sobrien * 1978344Sobrien * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 2078344Sobrien * or visit www.oracle.com if you need additional information or have any 2178344Sobrien * questions. 2278344Sobrien * 2378344Sobrien */ 2478344Sobrien 2578344Sobrien// no precompiled headers 2678344Sobrien#include "classfile/classLoader.hpp" 2778344Sobrien#include "classfile/systemDictionary.hpp" 2878344Sobrien#include "classfile/vmSymbols.hpp" 2978344Sobrien#include "code/icBuffer.hpp" 3078344Sobrien#include "code/vtableStubs.hpp" 3178344Sobrien#include "compiler/compileBroker.hpp" 3278344Sobrien#include "compiler/disassembler.hpp" 3378344Sobrien#include "interpreter/interpreter.hpp" 3478344Sobrien#include "jvm_linux.h" 3578344Sobrien#include "memory/allocation.inline.hpp" 3678344Sobrien#include "memory/filemap.hpp" 3778344Sobrien#include "mutex_linux.inline.hpp" 3878344Sobrien#include "oops/oop.inline.hpp" 3978344Sobrien#include "os_linux.inline.hpp" 4078344Sobrien#include "os_share_linux.hpp" 4178344Sobrien#include "prims/jniFastGetField.hpp" 4278344Sobrien#include "prims/jvm.h" 4398186Sgordon#include "prims/jvm_misc.hpp" 4498186Sgordon#include "runtime/arguments.hpp" 4598186Sgordon#include "runtime/atomic.inline.hpp" 4698186Sgordon#include "runtime/extendedPC.hpp" 4798186Sgordon#include "runtime/globals.hpp" 4898186Sgordon#include "runtime/interfaceSupport.hpp" 49103018Sgordon#include "runtime/init.hpp" 5098186Sgordon#include "runtime/java.hpp" 51103018Sgordon#include "runtime/javaCalls.hpp" 5298186Sgordon#include "runtime/mutexLocker.hpp" 5398186Sgordon#include "runtime/objectMonitor.hpp" 5498186Sgordon#include "runtime/orderAccess.inline.hpp" 5598186Sgordon#include "runtime/osThread.hpp" 5698186Sgordon#include "runtime/perfMemory.hpp" 5798186Sgordon#include "runtime/sharedRuntime.hpp" 5898186Sgordon#include "runtime/statSampler.hpp" 5998186Sgordon#include "runtime/stubRoutines.hpp" 6098186Sgordon#include "runtime/thread.inline.hpp" 6178344Sobrien#include "runtime/threadCritical.hpp" 6278344Sobrien#include "runtime/timer.hpp" 6378344Sobrien#include "services/attachListener.hpp" 6478344Sobrien#include "services/memTracker.hpp" 6598186Sgordon#include "services/runtimeService.hpp" 6698186Sgordon#include "utilities/decoder.hpp" 6798186Sgordon#include "utilities/defaultStream.hpp" 6898186Sgordon#include "utilities/events.hpp" 6998186Sgordon#include "utilities/elfFile.hpp" 7098186Sgordon#include "utilities/growableArray.hpp" 7198186Sgordon#include "utilities/vmError.hpp" 7298186Sgordon 7398186Sgordon// put OS-includes here 7498186Sgordon# include <sys/types.h> 7598186Sgordon# include <sys/mman.h> 7698186Sgordon# include <sys/stat.h> 7798186Sgordon# include <sys/select.h> 7898186Sgordon# include <pthread.h> 7998186Sgordon# include <signal.h> 8098186Sgordon# include <errno.h> 8198186Sgordon# include <dlfcn.h> 82103018Sgordon# include <stdio.h> 8398186Sgordon# include <unistd.h> 8498186Sgordon# include <sys/resource.h> 8598186Sgordon# include <pthread.h> 8698186Sgordon# include <sys/stat.h> 8798186Sgordon# include <sys/time.h> 8898186Sgordon# include <sys/times.h> 8998186Sgordon# include <sys/utsname.h> 9098186Sgordon# include <sys/socket.h> 9198186Sgordon# include <sys/wait.h> 9298186Sgordon# include <pwd.h> 9398186Sgordon# include <poll.h> 9498186Sgordon# include <semaphore.h> 9598186Sgordon# include <fcntl.h> 9698186Sgordon# include <string.h> 9798186Sgordon# include <syscall.h> 9898186Sgordon# include <sys/sysinfo.h> 9998186Sgordon# include <gnu/libc-version.h> 10098186Sgordon# include <sys/ipc.h> 10198186Sgordon# include <sys/shm.h> 10298186Sgordon# include <link.h> 10398186Sgordon# include <stdint.h> 10498186Sgordon# include <inttypes.h> 10598186Sgordon# include <sys/ioctl.h> 10698186Sgordon 10798186SgordonPRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC 10898186Sgordon 10998186Sgordon// if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 11098186Sgordon// getrusage() is prepared to handle the associated failure. 11198186Sgordon#ifndef RUSAGE_THREAD 11298186Sgordon #define RUSAGE_THREAD (1) /* only the calling thread */ 11398186Sgordon#endif 11498186Sgordon 11578344Sobrien#define MAX_PATH (2 * K) 11678344Sobrien 11778344Sobrien#define MAX_SECS 100000000 11878344Sobrien 11978344Sobrien// for timer info max values which include all bits 12078344Sobrien#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 12178344Sobrien 12298186Sgordon#define LARGEPAGES_BIT (1 << 6) 12378344Sobrien//////////////////////////////////////////////////////////////////////////////// 12478344Sobrien// global variables 12578344Sobrienjulong os::Linux::_physical_memory = 0; 12678344Sobrien 12778344Sobrienaddress os::Linux::_initial_thread_stack_bottom = NULL; 12878344Sobrienuintptr_t os::Linux::_initial_thread_stack_size = 0; 12978344Sobrien 13078344Sobrienint (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 13178344Sobrienint (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 13278344SobrienMutex* os::Linux::_createThread_lock = NULL; 13378344Sobrienpthread_t os::Linux::_main_thread; 13478344Sobrienint os::Linux::_page_size = -1; 135106643Sgordonconst int os::Linux::_vm_default_page_size = (8 * K); 13678344Sobrienbool os::Linux::_is_floating_stack = false; 13778344Sobrienbool os::Linux::_is_NPTL = false; 13878344Sobrienbool os::Linux::_supports_fast_thread_cpu_time = false; 13978344Sobrienconst char * os::Linux::_glibc_version = NULL; 14078344Sobrienconst char * os::Linux::_libpthread_version = NULL; 14198186Sgordonpthread_condattr_t os::Linux::_condattr[1]; 14298186Sgordon 14378344Sobrienstatic jlong initial_time_count=0; 14498186Sgordon 14598186Sgordonstatic int clock_tics_per_sec = 100; 14698186Sgordon 14798186Sgordon// For diagnostics to print a message once. see run_periodic_checks 14898186Sgordonstatic sigset_t check_signal_done; 14998186Sgordonstatic bool check_signals = true; 15098186Sgordon 15198186Sgordonstatic pid_t _initial_pid = 0; 15298186Sgordon 15378344Sobrien// Signal number used to suspend/resume a thread 15498186Sgordon 15598186Sgordon// do not use any signal number less than SIGSEGV, see 4355769 15698186Sgordonstatic int SR_signum = SIGUSR2; 15798186Sgordonsigset_t SR_sigset; 15898186Sgordon 15978344Sobrien// Used to protect dlsym() calls 16078344Sobrienstatic pthread_mutex_t dl_mutex; 16198186Sgordon 16278344Sobrien// Declarations 16378344Sobrienstatic void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 16478344Sobrien 16578344Sobrien#ifdef JAVASE_EMBEDDED 16678344Sobrienclass MemNotifyThread: public Thread { 16778344Sobrien friend class VMStructs; 16878344Sobrien public: 16978344Sobrien virtual void run(); 17098186Sgordon 17178344Sobrien private: 17278344Sobrien static MemNotifyThread* _memnotify_thread; 17398186Sgordon int _fd; 17478344Sobrien 17578344Sobrien public: 17678344Sobrien 17778344Sobrien // Constructor 17898186Sgordon MemNotifyThread(int fd); 17998186Sgordon 18078344Sobrien // Tester 18198186Sgordon bool is_memnotify_thread() const { return true; } 18298186Sgordon 18378344Sobrien // Printing 18478344Sobrien char* name() const { return (char*)"Linux MemNotify Thread"; } 18578344Sobrien 18678344Sobrien // Returns the single instance of the MemNotifyThread 18778344Sobrien static MemNotifyThread* memnotify_thread() { return _memnotify_thread; } 18898186Sgordon 18978344Sobrien // Create and start the single instance of MemNotifyThread 19098186Sgordon static void start(); 19178344Sobrien}; 19278344Sobrien#endif // JAVASE_EMBEDDED 19398186Sgordon 19478344Sobrien// utility functions 19578344Sobrien 19678344Sobrienstatic int SR_initialize(); 19778344Sobrien 19898186Sgordonjulong os::available_memory() { 19978344Sobrien return Linux::available_memory(); 20078344Sobrien} 20198186Sgordon 20278344Sobrienjulong os::Linux::available_memory() { 20378344Sobrien // values in struct sysinfo are "unsigned long" 20478344Sobrien struct sysinfo si; 20598186Sgordon sysinfo(&si); 20678344Sobrien 20798186Sgordon return (julong)si.freeram * si.mem_unit; 20898186Sgordon} 20978344Sobrien 21078344Sobrienjulong os::physical_memory() { 21178344Sobrien return Linux::physical_memory(); 21278344Sobrien} 21398186Sgordon 21478344Sobrien//////////////////////////////////////////////////////////////////////////////// 21598186Sgordon// environment support 21678344Sobrien 21798186Sgordonbool os::getenv(const char* name, char* buf, int len) { 21898186Sgordon const char* val = ::getenv(name); 21998186Sgordon if (val != NULL && strlen(val) < (size_t)len) { 22098186Sgordon strcpy(buf, val); 22198186Sgordon return true; 22298186Sgordon } 22398186Sgordon if (len > 0) buf[0] = 0; // return a null string 22498186Sgordon return false; 22598186Sgordon} 22698186Sgordon 22798186Sgordon 22898186Sgordon// Return true if user is running as root. 22998186Sgordon 23098186Sgordonbool os::have_special_privileges() { 23198186Sgordon static bool init = false; 23298186Sgordon static bool privileges = false; 23398186Sgordon if (!init) { 23498186Sgordon privileges = (getuid() != geteuid()) || (getgid() != getegid()); 23598186Sgordon init = true; 23698186Sgordon } 23798186Sgordon return privileges; 23898186Sgordon} 23998186Sgordon 24098186Sgordon 24198186Sgordon#ifndef SYS_gettid 24298186Sgordon// i386: 224, ia64: 1105, amd64: 186, sparc 143 24398186Sgordon #ifdef __ia64__ 24498186Sgordon #define SYS_gettid 1105 24598186Sgordon #elif __i386__ 24678344Sobrien #define SYS_gettid 224 24798186Sgordon #elif __amd64__ 24898186Sgordon #define SYS_gettid 186 24998186Sgordon #elif __sparc__ 25098186Sgordon #define SYS_gettid 143 25198186Sgordon #else 25298186Sgordon #error define gettid for the arch 25378344Sobrien #endif 25498186Sgordon#endif 25598186Sgordon 25698186Sgordon// Cpu architecture string 25798186Sgordon#if defined(ZERO) 25898186Sgordonstatic char cpu_arch[] = ZERO_LIBARCH; 25998186Sgordon#elif defined(IA64) 26098186Sgordonstatic char cpu_arch[] = "ia64"; 26198186Sgordon#elif defined(IA32) 26298186Sgordonstatic char cpu_arch[] = "i386"; 26398186Sgordon#elif defined(AMD64) 26498186Sgordonstatic char cpu_arch[] = "amd64"; 26598186Sgordon#elif defined(ARM) 26698186Sgordonstatic char cpu_arch[] = "arm"; 26798186Sgordon#elif defined(PPC32) 26898186Sgordonstatic char cpu_arch[] = "ppc"; 26998186Sgordon#elif defined(PPC64) 27098186Sgordonstatic char cpu_arch[] = "ppc64"; 27198186Sgordon#elif defined(SPARC) 27298186Sgordon #ifdef _LP64 27398186Sgordonstatic char cpu_arch[] = "sparcv9"; 27498186Sgordon #else 27598186Sgordonstatic char cpu_arch[] = "sparc"; 27698186Sgordon #endif 27798186Sgordon#else 27898186Sgordon #error Add appropriate cpu_arch setting 27998186Sgordon#endif 28098186Sgordon 28198186Sgordon 28298186Sgordon// pid_t gettid() 28398186Sgordon// 28498186Sgordon// Returns the kernel thread id of the currently running thread. Kernel 28598186Sgordon// thread id is used to access /proc. 28698186Sgordon// 28798186Sgordon// (Note that getpid() on LinuxThreads returns kernel thread id too; but 28898186Sgordon// on NPTL, it returns the same pid for all threads, as required by POSIX.) 28998186Sgordon// 29098186Sgordonpid_t os::Linux::gettid() { 29198186Sgordon int rslt = syscall(SYS_gettid); 29298186Sgordon if (rslt == -1) { 29398186Sgordon // old kernel, no NPTL support 29498186Sgordon return getpid(); 29598186Sgordon } else { 29698186Sgordon return (pid_t)rslt; 29798186Sgordon } 29898186Sgordon} 29998186Sgordon 30098186Sgordon// Most versions of linux have a bug where the number of processors are 30198186Sgordon// determined by looking at the /proc file system. In a chroot environment, 30278344Sobrien// the system call returns 1. This causes the VM to act as if it is 30398186Sgordon// a single processor and elide locking (see is_MP() call). 30498186Sgordonstatic bool unsafe_chroot_detected = false; 30598186Sgordonstatic const char *unstable_chroot_error = "/proc file system not found.\n" 30698186Sgordon "Java may be unstable running multithreaded in a chroot " 30778344Sobrien "environment on Linux when /proc filesystem is not mounted."; 30898186Sgordon 30998186Sgordonvoid os::Linux::initialize_system_info() { 31098186Sgordon set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 31178344Sobrien if (processor_count() == 1) { 31278344Sobrien pid_t pid = os::Linux::gettid(); 31378344Sobrien char fname[32]; 31498186Sgordon jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 31598186Sgordon FILE *fp = fopen(fname, "r"); 31698186Sgordon if (fp == NULL) { 31798186Sgordon unsafe_chroot_detected = true; 31898186Sgordon } else { 31978344Sobrien fclose(fp); 32098186Sgordon } 32198186Sgordon } 32278344Sobrien _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 32398186Sgordon assert(processor_count() > 0, "linux error"); 32498186Sgordon} 32578344Sobrien 32678344Sobrienvoid os::init_system_properties_values() { 32778344Sobrien // The next steps are taken in the product version: 32898186Sgordon // 32998186Sgordon // Obtain the JAVA_HOME value from the location of libjvm.so. 33078344Sobrien // This library should be located at: 33178344Sobrien // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 33278344Sobrien // 33398186Sgordon // If "/jre/lib/" appears at the right place in the path, then we 33478344Sobrien // assume libjvm.so is installed in a JDK and we use this path. 33578344Sobrien // 33678344Sobrien // Otherwise exit with message: "Could not create the Java virtual machine." 33778344Sobrien // 33898186Sgordon // The following extra steps are taken in the debugging version: 33998186Sgordon // 34098186Sgordon // If "/jre/lib/" does NOT appear at the right place in the path 34178344Sobrien // instead of exit check for $JAVA_HOME environment variable. 34278344Sobrien // 34398186Sgordon // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 34498186Sgordon // then we append a fake suffix "hotspot/libjvm.so" to this path so 34598186Sgordon // it looks like libjvm.so is installed there 34678344Sobrien // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 34798186Sgordon // 34898186Sgordon // Otherwise exit. 34978344Sobrien // 35078344Sobrien // Important note: if the location of libjvm.so changes this 35178344Sobrien // code needs to be changed accordingly. 35278344Sobrien 35398186Sgordon // See ld(1): 35478344Sobrien // The linker uses the following search paths to locate required 35578344Sobrien // shared libraries: 35678344Sobrien // 1: ... 35778344Sobrien // ... 35878344Sobrien // 7: The default directories, normally /lib and /usr/lib. 35978344Sobrien#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 36078344Sobrien #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 36178344Sobrien#else 36278344Sobrien #define DEFAULT_LIBPATH "/lib:/usr/lib" 36378344Sobrien#endif 36478344Sobrien 36578344Sobrien// Base path of extensions installed on the system. 36698186Sgordon#define SYS_EXT_DIR "/usr/java/packages" 36778344Sobrien#define EXTENSIONS_DIR "/lib/ext" 36878344Sobrien#define ENDORSED_DIR "/lib/endorsed" 36998186Sgordon 37078344Sobrien // Buffer that fits several sprintfs. 37198186Sgordon // Note that the space for the colon and the trailing null are provided 37298186Sgordon // by the nulls included by the sizeof operator. 37398186Sgordon const size_t bufsize = 37478344Sobrien MAX3((size_t)MAXPATHLEN, // For dll_dir & friends. 37598186Sgordon (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir 37678344Sobrien (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir 37778344Sobrien char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 37898186Sgordon 37998186Sgordon // sysclasspath, java_home, dll_dir 38098186Sgordon { 38198186Sgordon char *pslash; 38278344Sobrien os::jvm_path(buf, bufsize); 38398186Sgordon 38478344Sobrien // Found the full path to libjvm.so. 38598186Sgordon // Now cut the path to <java_home>/jre if we can. 38698186Sgordon *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so. 38798186Sgordon pslash = strrchr(buf, '/'); 38898186Sgordon if (pslash != NULL) { 38978344Sobrien *pslash = '\0'; // Get rid of /{client|server|hotspot}. 39078344Sobrien } 39178344Sobrien Arguments::set_dll_dir(buf); 39278344Sobrien 39378344Sobrien if (pslash != NULL) { 39478344Sobrien pslash = strrchr(buf, '/'); 39578344Sobrien if (pslash != NULL) { 39678344Sobrien *pslash = '\0'; // Get rid of /<arch>. 39778344Sobrien pslash = strrchr(buf, '/'); 39878344Sobrien if (pslash != NULL) { 39978344Sobrien *pslash = '\0'; // Get rid of /lib. 40078344Sobrien } 40198186Sgordon } 40298186Sgordon } 40378344Sobrien Arguments::set_java_home(buf); 40498186Sgordon set_boot_path('/', ':'); 40598186Sgordon } 40678344Sobrien 40778344Sobrien // Where to look for native libraries. 40878344Sobrien // 40978344Sobrien // Note: Due to a legacy implementation, most of the library path 41098186Sgordon // is set in the launcher. This was to accomodate linking restrictions 41178344Sobrien // on legacy Linux implementations (which are no longer supported). 41298186Sgordon // Eventually, all the library path setting will be done here. 41398186Sgordon // 41498186Sgordon // However, to prevent the proliferation of improperly built native 41598186Sgordon // libraries, the new path component /usr/java/packages is added here. 41678344Sobrien // Eventually, all the library path setting will be done here. 41798186Sgordon { 41898186Sgordon // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 41978344Sobrien // should always exist (until the legacy problem cited above is 42078344Sobrien // addressed). 42178344Sobrien const char *v = ::getenv("LD_LIBRARY_PATH"); 42278344Sobrien const char *v_colon = ":"; 42398186Sgordon if (v == NULL) { v = ""; v_colon = ""; } 42478344Sobrien // That's +1 for the colon and +1 for the trailing '\0'. 42598186Sgordon char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 42698186Sgordon strlen(v) + 1 + 42798186Sgordon sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1, 42898186Sgordon mtInternal); 42998186Sgordon sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch); 43098186Sgordon Arguments::set_library_path(ld_library_path); 43198186Sgordon FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal); 43298186Sgordon } 43398186Sgordon 43498186Sgordon // Extensions directories. 43598186Sgordon sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 43698186Sgordon Arguments::set_ext_dirs(buf); 43798186Sgordon 43898186Sgordon // Endorsed standards default directory. 43998186Sgordon sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 44098186Sgordon Arguments::set_endorsed_dirs(buf); 44198186Sgordon 44298186Sgordon FREE_C_HEAP_ARRAY(char, buf, mtInternal); 44398186Sgordon 44498186Sgordon#undef DEFAULT_LIBPATH 44598186Sgordon#undef SYS_EXT_DIR 44698186Sgordon#undef EXTENSIONS_DIR 44798186Sgordon#undef ENDORSED_DIR 44898186Sgordon} 44978344Sobrien 45078344Sobrien//////////////////////////////////////////////////////////////////////////////// 45198186Sgordon// breakpoint support 45278344Sobrien 45398186Sgordonvoid os::breakpoint() { 45478344Sobrien BREAKPOINT; 45578344Sobrien} 45698186Sgordon 45778344Sobrienextern "C" void breakpoint() { 45898186Sgordon // use debugger to set breakpoint here 45998186Sgordon} 46078344Sobrien 46178344Sobrien//////////////////////////////////////////////////////////////////////////////// 46298186Sgordon// signal support 46398186Sgordon 46478344Sobriendebug_only(static bool signal_sets_initialized = false); 46578344Sobrienstatic sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 46678344Sobrien 46778344Sobrienbool os::Linux::is_sig_ignored(int sig) { 46878344Sobrien struct sigaction oact; 46978344Sobrien sigaction(sig, (struct sigaction*)NULL, &oact); 47098186Sgordon void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 47198186Sgordon : CAST_FROM_FN_PTR(void*, oact.sa_handler); 47298186Sgordon if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) { 47398186Sgordon return true; 47498186Sgordon } else { 47578344Sobrien return false; 47698186Sgordon } 47778344Sobrien} 47898186Sgordon 47998186Sgordonvoid os::Linux::signal_sets_init() { 48078344Sobrien // Should also have an assertion stating we are still single-threaded. 48198186Sgordon assert(!signal_sets_initialized, "Already initialized"); 48278344Sobrien // Fill in signals that are necessarily unblocked for all threads in 48398186Sgordon // the VM. Currently, we unblock the following signals: 48498186Sgordon // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 48598186Sgordon // by -Xrs (=ReduceSignalUsage)); 48678344Sobrien // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 48778344Sobrien // other threads. The "ReduceSignalUsage" boolean tells us not to alter 48898186Sgordon // the dispositions or masks wrt these signals. 48978344Sobrien // Programs embedding the VM that want to use the above signals for their 49078344Sobrien // own purposes must, at this time, use the "-Xrs" option to prevent 49178344Sobrien // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 49298186Sgordon // (See bug 4345157, and other related bugs). 49378344Sobrien // In reality, though, unblocking these signals is really a nop, since 49478344Sobrien // these signals are not blocked by default. 49578344Sobrien sigemptyset(&unblocked_sigs); 49678344Sobrien sigemptyset(&allowdebug_blocked_sigs); 49798186Sgordon sigaddset(&unblocked_sigs, SIGILL); 49878344Sobrien sigaddset(&unblocked_sigs, SIGSEGV); 49998186Sgordon sigaddset(&unblocked_sigs, SIGBUS); 50078344Sobrien sigaddset(&unblocked_sigs, SIGFPE); 50198186Sgordon#if defined(PPC64) 50298186Sgordon sigaddset(&unblocked_sigs, SIGTRAP); 50398186Sgordon#endif 50478344Sobrien sigaddset(&unblocked_sigs, SR_signum); 50598186Sgordon 50698186Sgordon if (!ReduceSignalUsage) { 50798186Sgordon if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 50898186Sgordon sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 50998186Sgordon sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 51098186Sgordon } 51178344Sobrien if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 51298186Sgordon sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 51378344Sobrien sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 51478344Sobrien } 51578344Sobrien if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 51698186Sgordon sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 51778344Sobrien sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 51878344Sobrien } 51978344Sobrien } 52078344Sobrien // Fill in signals that are blocked by all but the VM thread. 52178344Sobrien sigemptyset(&vm_sigs); 52278344Sobrien if (!ReduceSignalUsage) { 52378344Sobrien sigaddset(&vm_sigs, BREAK_SIGNAL); 52478344Sobrien } 52598186Sgordon debug_only(signal_sets_initialized = true); 52678344Sobrien 52778344Sobrien} 52878344Sobrien 52978344Sobrien// These are signals that are unblocked while a thread is running Java. 53078344Sobrien// (For some reason, they get blocked by default.) 53178344Sobriensigset_t* os::Linux::unblocked_signals() { 53298186Sgordon assert(signal_sets_initialized, "Not initialized"); 53398186Sgordon return &unblocked_sigs; 53478344Sobrien} 53598186Sgordon 53678344Sobrien// These are the signals that are blocked while a (non-VM) thread is 53778344Sobrien// running Java. Only the VM thread handles these signals. 53878344Sobriensigset_t* os::Linux::vm_signals() { 53998186Sgordon assert(signal_sets_initialized, "Not initialized"); 54078344Sobrien return &vm_sigs; 54178344Sobrien} 54278344Sobrien 54398186Sgordon// These are signals that are blocked during cond_wait to allow debugger in 54498186Sgordonsigset_t* os::Linux::allowdebug_blocked_signals() { 54598186Sgordon assert(signal_sets_initialized, "Not initialized"); 54698186Sgordon return &allowdebug_blocked_sigs; 54778344Sobrien} 54878344Sobrien 54978344Sobrienvoid os::Linux::hotspot_sigmask(Thread* thread) { 55098186Sgordon 55178344Sobrien //Save caller's signal mask before setting VM signal mask 55278344Sobrien sigset_t caller_sigmask; 55398186Sgordon pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 55498186Sgordon 55578344Sobrien OSThread* osthread = thread->osthread(); 55678344Sobrien osthread->set_caller_sigmask(caller_sigmask); 55778344Sobrien 55878344Sobrien pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 55978344Sobrien 56078344Sobrien if (!ReduceSignalUsage) { 56178344Sobrien if (thread->is_VM_thread()) { 56298186Sgordon // Only the VM thread handles BREAK_SIGNAL ... 56398186Sgordon pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 56478344Sobrien } else { 56578344Sobrien // ... all other threads block BREAK_SIGNAL 56678344Sobrien pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 56778344Sobrien } 56898186Sgordon } 56978344Sobrien} 57078344Sobrien 57178344Sobrien////////////////////////////////////////////////////////////////////////////// 57278344Sobrien// detecting pthread library 57378344Sobrien 57478344Sobrienvoid os::Linux::libpthread_init() { 57578344Sobrien // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 57678344Sobrien // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 57778344Sobrien // generic name for earlier versions. 57898186Sgordon // Define macros here so we can build HotSpot on old systems. 57978344Sobrien#ifndef _CS_GNU_LIBC_VERSION 58078344Sobrien #define _CS_GNU_LIBC_VERSION 2 58178344Sobrien#endif 58278344Sobrien#ifndef _CS_GNU_LIBPTHREAD_VERSION 58378344Sobrien #define _CS_GNU_LIBPTHREAD_VERSION 3 58478344Sobrien#endif 58578344Sobrien 58698186Sgordon size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 58778344Sobrien if (n > 0) { 58878344Sobrien char *str = (char *)malloc(n, mtInternal); 58978344Sobrien confstr(_CS_GNU_LIBC_VERSION, str, n); 59078344Sobrien os::Linux::set_glibc_version(str); 59178344Sobrien } else { 59278344Sobrien // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 59378344Sobrien static char _gnu_libc_version[32]; 59498186Sgordon jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 59578344Sobrien "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 59678344Sobrien os::Linux::set_glibc_version(_gnu_libc_version); 59778344Sobrien } 59878344Sobrien 59978344Sobrien n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 60078344Sobrien if (n > 0) { 60178344Sobrien char *str = (char *)malloc(n, mtInternal); 60278344Sobrien confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 60398186Sgordon // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 60478344Sobrien // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 60578344Sobrien // is the case. LinuxThreads has a hard limit on max number of threads. 60678344Sobrien // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 60778344Sobrien // On the other hand, NPTL does not have such a limit, sysconf() 60878344Sobrien // will return -1 and errno is not changed. Check if it is really NPTL. 60978344Sobrien if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 61078344Sobrien strstr(str, "NPTL") && 61178344Sobrien sysconf(_SC_THREAD_THREADS_MAX) > 0) { 61278344Sobrien free(str); 61378344Sobrien os::Linux::set_libpthread_version("linuxthreads"); 61498186Sgordon } else { 61578344Sobrien os::Linux::set_libpthread_version(str); 61678344Sobrien } 61778344Sobrien } else { 61878344Sobrien // glibc before 2.3.2 only has LinuxThreads. 61998186Sgordon os::Linux::set_libpthread_version("linuxthreads"); 62098186Sgordon } 62198186Sgordon 62298186Sgordon if (strstr(libpthread_version(), "NPTL")) { 62378344Sobrien os::Linux::set_is_NPTL(); 62498186Sgordon } else { 62598186Sgordon os::Linux::set_is_LinuxThreads(); 62698186Sgordon } 62798186Sgordon 62898186Sgordon // LinuxThreads have two flavors: floating-stack mode, which allows variable 62998186Sgordon // stack size; and fixed-stack mode. NPTL is always floating-stack. 63098186Sgordon if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 63198186Sgordon os::Linux::set_is_floating_stack(); 63298186Sgordon } 63398186Sgordon} 63498186Sgordon 63598186Sgordon///////////////////////////////////////////////////////////////////////////// 63678344Sobrien// thread stack 63778344Sobrien 63878344Sobrien// Force Linux kernel to expand current thread stack. If "bottom" is close 63998186Sgordon// to the stack guard, caller should block all signals. 64078344Sobrien// 64178344Sobrien// MAP_GROWSDOWN: 64278344Sobrien// A special mmap() flag that is used to implement thread stacks. It tells 64378344Sobrien// kernel that the memory region should extend downwards when needed. This 64478344Sobrien// allows early versions of LinuxThreads to only mmap the first few pages 64578344Sobrien// when creating a new thread. Linux kernel will automatically expand thread 64678344Sobrien// stack as needed (on page faults). 64778344Sobrien// 64878344Sobrien// However, because the memory region of a MAP_GROWSDOWN stack can grow on 64998186Sgordon// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 65098186Sgordon// region, it's hard to tell if the fault is due to a legitimate stack 65198186Sgordon// access or because of reading/writing non-exist memory (e.g. buffer 65298186Sgordon// overrun). As a rule, if the fault happens below current stack pointer, 65378344Sobrien// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 65478344Sobrien// application (see Linux kernel fault.c). 65598186Sgordon// 65698186Sgordon// This Linux feature can cause SIGSEGV when VM bangs thread stack for 65798186Sgordon// stack overflow detection. 65878344Sobrien// 65998186Sgordon// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 66098186Sgordon// not use this flag. However, the stack of initial thread is not created 66198186Sgordon// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 66298186Sgordon// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 66398186Sgordon// and then attach the thread to JVM. 66498186Sgordon// 66598186Sgordon// To get around the problem and allow stack banging on Linux, we need to 66698186Sgordon// manually expand thread stack after receiving the SIGSEGV. 66798186Sgordon// 66898186Sgordon// There are two ways to expand thread stack to address "bottom", we used 66998186Sgordon// both of them in JVM before 1.5: 67098186Sgordon// 1. adjust stack pointer first so that it is below "bottom", and then 67198186Sgordon// touch "bottom" 67298186Sgordon// 2. mmap() the page in question 67398186Sgordon// 67498186Sgordon// Now alternate signal stack is gone, it's harder to use 2. For instance, 67578344Sobrien// if current sp is already near the lower end of page 101, and we need to 67678344Sobrien// call mmap() to map page 100, it is possible that part of the mmap() frame 67778344Sobrien// will be placed in page 100. When page 100 is mapped, it is zero-filled. 67898186Sgordon// That will destroy the mmap() frame and cause VM to crash. 67978344Sobrien// 68078344Sobrien// The following code works by adjusting sp first, then accessing the "bottom" 68178344Sobrien// page to force a page fault. Linux kernel will then automatically expand the 68278344Sobrien// stack mapping. 68378344Sobrien// 68478344Sobrien// _expand_stack_to() assumes its frame size is less than page size, which 68578344Sobrien// should always be true if the function is not inlined. 68678344Sobrien 68778344Sobrien#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 68898186Sgordon #define NOINLINE 68978344Sobrien#else 69078344Sobrien #define NOINLINE __attribute__ ((noinline)) 69198186Sgordon#endif 69298186Sgordon 69398186Sgordonstatic void _expand_stack_to(address bottom) NOINLINE; 69498186Sgordon 69598186Sgordonstatic void _expand_stack_to(address bottom) { 69698186Sgordon address sp; 69798186Sgordon size_t size; 69898186Sgordon volatile char *p; 69978344Sobrien 70078344Sobrien // Adjust bottom to point to the largest address within the same page, it 70178344Sobrien // gives us a one-page buffer if alloca() allocates slightly more memory. 70298186Sgordon bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 70378344Sobrien bottom += os::Linux::page_size() - 1; 70478344Sobrien 70578344Sobrien // sp might be slightly above current stack pointer; if that's the case, we 70678344Sobrien // will alloca() a little more space than necessary, which is OK. Don't use 70778344Sobrien // os::current_stack_pointer(), as its result can be slightly below current 70878344Sobrien // stack pointer, causing us to not alloca enough to reach "bottom". 70978344Sobrien sp = (address)&sp; 71078344Sobrien 71178344Sobrien if (sp > bottom) { 71278344Sobrien size = sp - bottom; 71398186Sgordon p = (volatile char *)alloca(size); 71498186Sgordon assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 71598186Sgordon p[0] = '\0'; 71698186Sgordon } 71778344Sobrien} 71878344Sobrien 71998186Sgordonbool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 72098186Sgordon assert(t!=NULL, "just checking"); 72198186Sgordon assert(t->osthread()->expanding_stack(), "expand should be set"); 72298186Sgordon assert(t->stack_base() != NULL, "stack_base was not initialized"); 72398186Sgordon 72498186Sgordon if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 72578344Sobrien sigset_t mask_all, old_sigset; 72678344Sobrien sigfillset(&mask_all); 72778344Sobrien pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 72878344Sobrien _expand_stack_to(addr); 72978344Sobrien pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 73078344Sobrien return true; 73178344Sobrien } 73278344Sobrien return false; 73378344Sobrien} 73478344Sobrien 73578344Sobrien////////////////////////////////////////////////////////////////////////////// 73678344Sobrien// create new thread 73778344Sobrien 73878344Sobrienstatic address highest_vm_reserved_address(); 73978344Sobrien 74078344Sobrien// check if it's safe to start a new thread 74178344Sobrienstatic bool _thread_safety_check(Thread* thread) { 74278344Sobrien if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 74378344Sobrien // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 74498186Sgordon // Heap is mmap'ed at lower end of memory space. Thread stacks are 74578344Sobrien // allocated (MAP_FIXED) from high address space. Every thread stack 74678344Sobrien // occupies a fixed size slot (usually 2Mbytes, but user can change 74778344Sobrien // it to other values if they rebuild LinuxThreads). 74878344Sobrien // 74978344Sobrien // Problem with MAP_FIXED is that mmap() can still succeed even part of 75078344Sobrien // the memory region has already been mmap'ed. That means if we have too 75178344Sobrien // many threads and/or very large heap, eventually thread stack will 75278344Sobrien // collide with heap. 75398186Sgordon // 75498186Sgordon // Here we try to prevent heap/stack collision by comparing current 75598186Sgordon // stack bottom with the highest address that has been mmap'ed by JVM 75678344Sobrien // plus a safety margin for memory maps created by native code. 75778344Sobrien // 75878344Sobrien // This feature can be disabled by setting ThreadSafetyMargin to 0 75978344Sobrien // 76078344Sobrien if (ThreadSafetyMargin > 0) { 76178344Sobrien address stack_bottom = os::current_stack_base() - os::current_stack_size(); 76278344Sobrien 76378344Sobrien // not safe if our stack extends below the safety margin 76478344Sobrien return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 76598186Sgordon } else { 76698186Sgordon return true; 76798186Sgordon } 76898186Sgordon } else { 76998186Sgordon // Floating stack LinuxThreads or NPTL: 77098186Sgordon // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 77198186Sgordon // there's not enough space left, pthread_create() will fail. If we come 77278344Sobrien // here, that means enough space has been reserved for stack. 77378344Sobrien return true; 77478344Sobrien } 77578344Sobrien} 77698186Sgordon 77798186Sgordon// Thread start routine for all newly created threads 77898186Sgordonstatic void *java_start(Thread *thread) { 77978344Sobrien // Try to randomize the cache line index of hot stack frames. 78098186Sgordon // This helps when threads of the same stack traces evict each other's 78198186Sgordon // cache lines. The threads can be either from the same JVM instance, or 78298186Sgordon // from different JVM instances. The benefit is especially true for 78398186Sgordon // processors with hyperthreading technology. 78498186Sgordon static int counter = 0; 78598186Sgordon int pid = os::current_process_id(); 78698186Sgordon alloca(((pid ^ counter++) & 7) * 128); 78798186Sgordon 78878344Sobrien ThreadLocalStorage::set_thread(thread); 78978344Sobrien 79078344Sobrien OSThread* osthread = thread->osthread(); 79178344Sobrien Monitor* sync = osthread->startThread_lock(); 79278344Sobrien 79378344Sobrien // non floating stack LinuxThreads needs extra check, see above 79478344Sobrien if (!_thread_safety_check(thread)) { 79578344Sobrien // notify parent thread 79678344Sobrien MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 79778344Sobrien osthread->set_state(ZOMBIE); 79878344Sobrien sync->notify_all(); 79978344Sobrien return NULL; 80078344Sobrien } 80178344Sobrien 80278344Sobrien // thread_id is kernel thread id (similar to Solaris LWP id) 80398186Sgordon osthread->set_thread_id(os::Linux::gettid()); 80498186Sgordon 80598186Sgordon if (UseNUMA) { 80698186Sgordon int lgrp_id = os::numa_get_group_id(); 80798186Sgordon if (lgrp_id != -1) { 80898186Sgordon thread->set_lgrp_id(lgrp_id); 80998186Sgordon } 81098186Sgordon } 81198186Sgordon // initialize signal mask for this thread 81298186Sgordon os::Linux::hotspot_sigmask(thread); 81398186Sgordon 81478344Sobrien // initialize floating point control register 81598186Sgordon os::Linux::init_thread_fpu_state(); 81678344Sobrien 81778344Sobrien // handshaking with parent thread 818101850Sgordon { 819101850Sgordon MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 820101850Sgordon 821103018Sgordon // notify parent thread 822101850Sgordon osthread->set_state(INITIALIZED); 823101850Sgordon sync->notify_all(); 824101850Sgordon 825101850Sgordon // wait until os::start_thread() 826101850Sgordon while (osthread->get_state() == INITIALIZED) { 827101850Sgordon sync->wait(Mutex::_no_safepoint_check_flag); 828101850Sgordon } 829101850Sgordon } 830101850Sgordon 831101850Sgordon // call one more level start routine 832101850Sgordon thread->run(); 83378344Sobrien 83478344Sobrien return 0; 83578344Sobrien} 83678344Sobrien 83778344Sobrienbool os::create_thread(Thread* thread, ThreadType thr_type, 83878344Sobrien size_t stack_size) { 83978344Sobrien assert(thread->osthread() == NULL, "caller responsible"); 84078344Sobrien 84178344Sobrien // Allocate the OSThread object 842106643Sgordon OSThread* osthread = new OSThread(NULL, NULL); 84378344Sobrien if (osthread == NULL) { 84478344Sobrien return false; 84578344Sobrien } 84678344Sobrien 84778344Sobrien // set the correct thread state 84878344Sobrien osthread->set_thread_type(thr_type); 84978344Sobrien 85078344Sobrien // Initial state is ALLOCATED but not INITIALIZED 85178344Sobrien osthread->set_state(ALLOCATED); 85278344Sobrien 85378344Sobrien thread->set_osthread(osthread); 85478344Sobrien 85578344Sobrien // init thread attributes 85678344Sobrien pthread_attr_t attr; 85778344Sobrien pthread_attr_init(&attr); 85878344Sobrien pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 85978344Sobrien 86078344Sobrien // stack size 86178344Sobrien if (os::Linux::supports_variable_stack_size()) { 862106643Sgordon // calculate stack size if it's not specified by caller 863106643Sgordon if (stack_size == 0) { 864106643Sgordon stack_size = os::Linux::default_stack_size(thr_type); 865106643Sgordon 86678344Sobrien switch (thr_type) { 86778344Sobrien case os::java_thread: 86878344Sobrien // Java threads use ThreadStackSize which default value can be 86978344Sobrien // changed with the flag -Xss 87078344Sobrien assert(JavaThread::stack_size_at_create() > 0, "this should be set"); 87178344Sobrien stack_size = JavaThread::stack_size_at_create(); 87278344Sobrien break; 87378344Sobrien case os::compiler_thread: 87478344Sobrien if (CompilerThreadStackSize > 0) { 875106643Sgordon stack_size = (size_t)(CompilerThreadStackSize * K); 876106643Sgordon break; 877106643Sgordon } // else fall through: 878106643Sgordon // use VMThreadStackSize if CompilerThreadStackSize is not defined 87978344Sobrien case os::vm_thread: 88098186Sgordon case os::pgc_thread: 88198186Sgordon case os::cgc_thread: 88298186Sgordon case os::watcher_thread: 88398186Sgordon if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 88498186Sgordon break; 88598186Sgordon } 88698186Sgordon } 887106643Sgordon 888106643Sgordon stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 889106643Sgordon pthread_attr_setstacksize(&attr, stack_size); 890106643Sgordon } else { 89198186Sgordon // let pthread_create() pick the default value. 89298186Sgordon } 89398186Sgordon 89498186Sgordon // glibc guard page 895106643Sgordon pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 89698186Sgordon 89798186Sgordon ThreadState state; 89898186Sgordon 89998186Sgordon { 90098186Sgordon // Serialize thread creation if we are running with fixed stack LinuxThreads 90198186Sgordon bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 90298186Sgordon if (lock) { 903106643Sgordon os::Linux::createThread_lock()->lock_without_safepoint_check(); 90498186Sgordon } 90598186Sgordon 90698186Sgordon pthread_t tid; 90798186Sgordon int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 90898186Sgordon 90998186Sgordon pthread_attr_destroy(&attr); 91098186Sgordon 91198186Sgordon if (ret != 0) { 91298186Sgordon if (PrintMiscellaneous && (Verbose || WizardMode)) { 91398186Sgordon perror("pthread_create()"); 91498186Sgordon } 91598186Sgordon // Need to clean up stuff we've allocated so far 91698186Sgordon thread->set_osthread(NULL); 91798186Sgordon delete osthread; 91898186Sgordon if (lock) os::Linux::createThread_lock()->unlock(); 91998186Sgordon return false; 92098186Sgordon } 92198186Sgordon 92298186Sgordon // Store pthread info into the OSThread 92398186Sgordon osthread->set_pthread_id(tid); 92498186Sgordon 92598186Sgordon // Wait until child thread is either initialized or aborted 92698186Sgordon { 92798186Sgordon Monitor* sync_with_child = osthread->startThread_lock(); 92898186Sgordon MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 92998186Sgordon while ((state = osthread->get_state()) == ALLOCATED) { 93098186Sgordon sync_with_child->wait(Mutex::_no_safepoint_check_flag); 93198186Sgordon } 93298186Sgordon } 93398186Sgordon 93498186Sgordon if (lock) { 93598186Sgordon os::Linux::createThread_lock()->unlock(); 93698186Sgordon } 93798186Sgordon } 93898186Sgordon 93998186Sgordon // Aborted due to thread limit being reached 94098186Sgordon if (state == ZOMBIE) { 94198186Sgordon thread->set_osthread(NULL); 94298186Sgordon delete osthread; 94398186Sgordon return false; 94498186Sgordon } 94598186Sgordon 94698186Sgordon // The thread is returned suspended (in state INITIALIZED), 94798186Sgordon // and is started higher up in the call chain 94898186Sgordon assert(state == INITIALIZED, "race condition"); 94998186Sgordon return true; 95098186Sgordon} 95198186Sgordon 95298186Sgordon///////////////////////////////////////////////////////////////////////////// 95398186Sgordon// attach existing thread 95498186Sgordon 95598186Sgordon// bootstrap the main thread 95698186Sgordonbool os::create_main_thread(JavaThread* thread) { 95798186Sgordon assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 95898186Sgordon return create_attached_thread(thread); 95998186Sgordon} 96098186Sgordon 96198186Sgordonbool os::create_attached_thread(JavaThread* thread) { 96298186Sgordon#ifdef ASSERT 96398186Sgordon thread->verify_not_published(); 96498186Sgordon#endif 96598186Sgordon 96698186Sgordon // Allocate the OSThread object 96798186Sgordon OSThread* osthread = new OSThread(NULL, NULL); 96898186Sgordon 96998186Sgordon if (osthread == NULL) { 97098186Sgordon return false; 97198186Sgordon } 97298186Sgordon 97398186Sgordon // Store pthread info into the OSThread 97498186Sgordon osthread->set_thread_id(os::Linux::gettid()); 97598186Sgordon osthread->set_pthread_id(::pthread_self()); 97698186Sgordon 97798186Sgordon // initialize floating point control register 97898186Sgordon os::Linux::init_thread_fpu_state(); 97998186Sgordon 98098186Sgordon // Initial thread state is RUNNABLE 98198186Sgordon osthread->set_state(RUNNABLE); 98298186Sgordon 98398186Sgordon thread->set_osthread(osthread); 98498186Sgordon 98598186Sgordon if (UseNUMA) { 98698186Sgordon int lgrp_id = os::numa_get_group_id(); 98798186Sgordon if (lgrp_id != -1) { 98898186Sgordon thread->set_lgrp_id(lgrp_id); 98998186Sgordon } 99098186Sgordon } 99198186Sgordon 99298186Sgordon if (os::Linux::is_initial_thread()) { 99398186Sgordon // If current thread is initial thread, its stack is mapped on demand, 994 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 995 // the entire stack region to avoid SEGV in stack banging. 996 // It is also useful to get around the heap-stack-gap problem on SuSE 997 // kernel (see 4821821 for details). We first expand stack to the top 998 // of yellow zone, then enable stack yellow zone (order is significant, 999 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1000 // is no gap between the last two virtual memory regions. 1001 1002 JavaThread *jt = (JavaThread *)thread; 1003 address addr = jt->stack_yellow_zone_base(); 1004 assert(addr != NULL, "initialization problem?"); 1005 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1006 1007 osthread->set_expanding_stack(); 1008 os::Linux::manually_expand_stack(jt, addr); 1009 osthread->clear_expanding_stack(); 1010 } 1011 1012 // initialize signal mask for this thread 1013 // and save the caller's signal mask 1014 os::Linux::hotspot_sigmask(thread); 1015 1016 return true; 1017} 1018 1019void os::pd_start_thread(Thread* thread) { 1020 OSThread * osthread = thread->osthread(); 1021 assert(osthread->get_state() != INITIALIZED, "just checking"); 1022 Monitor* sync_with_child = osthread->startThread_lock(); 1023 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1024 sync_with_child->notify(); 1025} 1026 1027// Free Linux resources related to the OSThread 1028void os::free_thread(OSThread* osthread) { 1029 assert(osthread != NULL, "osthread not set"); 1030 1031 if (Thread::current()->osthread() == osthread) { 1032 // Restore caller's signal mask 1033 sigset_t sigmask = osthread->caller_sigmask(); 1034 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1035 } 1036 1037 delete osthread; 1038} 1039 1040////////////////////////////////////////////////////////////////////////////// 1041// thread local storage 1042 1043// Restore the thread pointer if the destructor is called. This is in case 1044// someone from JNI code sets up a destructor with pthread_key_create to run 1045// detachCurrentThread on thread death. Unless we restore the thread pointer we 1046// will hang or crash. When detachCurrentThread is called the key will be set 1047// to null and we will not be called again. If detachCurrentThread is never 1048// called we could loop forever depending on the pthread implementation. 1049static void restore_thread_pointer(void* p) { 1050 Thread* thread = (Thread*) p; 1051 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread); 1052} 1053 1054int os::allocate_thread_local_storage() { 1055 pthread_key_t key; 1056 int rslt = pthread_key_create(&key, restore_thread_pointer); 1057 assert(rslt == 0, "cannot allocate thread local storage"); 1058 return (int)key; 1059} 1060 1061// Note: This is currently not used by VM, as we don't destroy TLS key 1062// on VM exit. 1063void os::free_thread_local_storage(int index) { 1064 int rslt = pthread_key_delete((pthread_key_t)index); 1065 assert(rslt == 0, "invalid index"); 1066} 1067 1068void os::thread_local_storage_at_put(int index, void* value) { 1069 int rslt = pthread_setspecific((pthread_key_t)index, value); 1070 assert(rslt == 0, "pthread_setspecific failed"); 1071} 1072 1073extern "C" Thread* get_thread() { 1074 return ThreadLocalStorage::thread(); 1075} 1076 1077////////////////////////////////////////////////////////////////////////////// 1078// initial thread 1079 1080// Check if current thread is the initial thread, similar to Solaris thr_main. 1081bool os::Linux::is_initial_thread(void) { 1082 char dummy; 1083 // If called before init complete, thread stack bottom will be null. 1084 // Can be called if fatal error occurs before initialization. 1085 if (initial_thread_stack_bottom() == NULL) return false; 1086 assert(initial_thread_stack_bottom() != NULL && 1087 initial_thread_stack_size() != 0, 1088 "os::init did not locate initial thread's stack region"); 1089 if ((address)&dummy >= initial_thread_stack_bottom() && 1090 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) { 1091 return true; 1092 } else { 1093 return false; 1094 } 1095} 1096 1097// Find the virtual memory area that contains addr 1098static bool find_vma(address addr, address* vma_low, address* vma_high) { 1099 FILE *fp = fopen("/proc/self/maps", "r"); 1100 if (fp) { 1101 address low, high; 1102 while (!feof(fp)) { 1103 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1104 if (low <= addr && addr < high) { 1105 if (vma_low) *vma_low = low; 1106 if (vma_high) *vma_high = high; 1107 fclose(fp); 1108 return true; 1109 } 1110 } 1111 for (;;) { 1112 int ch = fgetc(fp); 1113 if (ch == EOF || ch == (int)'\n') break; 1114 } 1115 } 1116 fclose(fp); 1117 } 1118 return false; 1119} 1120 1121// Locate initial thread stack. This special handling of initial thread stack 1122// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1123// bogus value for initial thread. 1124void os::Linux::capture_initial_stack(size_t max_size) { 1125 // stack size is the easy part, get it from RLIMIT_STACK 1126 size_t stack_size; 1127 struct rlimit rlim; 1128 getrlimit(RLIMIT_STACK, &rlim); 1129 stack_size = rlim.rlim_cur; 1130 1131 // 6308388: a bug in ld.so will relocate its own .data section to the 1132 // lower end of primordial stack; reduce ulimit -s value a little bit 1133 // so we won't install guard page on ld.so's data section. 1134 stack_size -= 2 * page_size(); 1135 1136 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1137 // 7.1, in both cases we will get 2G in return value. 1138 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1139 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1140 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1141 // in case other parts in glibc still assumes 2M max stack size. 1142 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1143 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1144 if (stack_size > 2 * K * K IA64_ONLY(*2)) { 1145 stack_size = 2 * K * K IA64_ONLY(*2); 1146 } 1147 // Try to figure out where the stack base (top) is. This is harder. 1148 // 1149 // When an application is started, glibc saves the initial stack pointer in 1150 // a global variable "__libc_stack_end", which is then used by system 1151 // libraries. __libc_stack_end should be pretty close to stack top. The 1152 // variable is available since the very early days. However, because it is 1153 // a private interface, it could disappear in the future. 1154 // 1155 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1156 // to __libc_stack_end, it is very close to stack top, but isn't the real 1157 // stack top. Note that /proc may not exist if VM is running as a chroot 1158 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1159 // /proc/<pid>/stat could change in the future (though unlikely). 1160 // 1161 // We try __libc_stack_end first. If that doesn't work, look for 1162 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1163 // as a hint, which should work well in most cases. 1164 1165 uintptr_t stack_start; 1166 1167 // try __libc_stack_end first 1168 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1169 if (p && *p) { 1170 stack_start = *p; 1171 } else { 1172 // see if we can get the start_stack field from /proc/self/stat 1173 FILE *fp; 1174 int pid; 1175 char state; 1176 int ppid; 1177 int pgrp; 1178 int session; 1179 int nr; 1180 int tpgrp; 1181 unsigned long flags; 1182 unsigned long minflt; 1183 unsigned long cminflt; 1184 unsigned long majflt; 1185 unsigned long cmajflt; 1186 unsigned long utime; 1187 unsigned long stime; 1188 long cutime; 1189 long cstime; 1190 long prio; 1191 long nice; 1192 long junk; 1193 long it_real; 1194 uintptr_t start; 1195 uintptr_t vsize; 1196 intptr_t rss; 1197 uintptr_t rsslim; 1198 uintptr_t scodes; 1199 uintptr_t ecode; 1200 int i; 1201 1202 // Figure what the primordial thread stack base is. Code is inspired 1203 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1204 // followed by command name surrounded by parentheses, state, etc. 1205 char stat[2048]; 1206 int statlen; 1207 1208 fp = fopen("/proc/self/stat", "r"); 1209 if (fp) { 1210 statlen = fread(stat, 1, 2047, fp); 1211 stat[statlen] = '\0'; 1212 fclose(fp); 1213 1214 // Skip pid and the command string. Note that we could be dealing with 1215 // weird command names, e.g. user could decide to rename java launcher 1216 // to "java 1.4.2 :)", then the stat file would look like 1217 // 1234 (java 1.4.2 :)) R ... ... 1218 // We don't really need to know the command string, just find the last 1219 // occurrence of ")" and then start parsing from there. See bug 4726580. 1220 char * s = strrchr(stat, ')'); 1221 1222 i = 0; 1223 if (s) { 1224 // Skip blank chars 1225 do s++; while (isspace(*s)); 1226 1227#define _UFM UINTX_FORMAT 1228#define _DFM INTX_FORMAT 1229 1230 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1231 // 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 1232 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, 1233 &state, // 3 %c 1234 &ppid, // 4 %d 1235 &pgrp, // 5 %d 1236 &session, // 6 %d 1237 &nr, // 7 %d 1238 &tpgrp, // 8 %d 1239 &flags, // 9 %lu 1240 &minflt, // 10 %lu 1241 &cminflt, // 11 %lu 1242 &majflt, // 12 %lu 1243 &cmajflt, // 13 %lu 1244 &utime, // 14 %lu 1245 &stime, // 15 %lu 1246 &cutime, // 16 %ld 1247 &cstime, // 17 %ld 1248 &prio, // 18 %ld 1249 &nice, // 19 %ld 1250 &junk, // 20 %ld 1251 &it_real, // 21 %ld 1252 &start, // 22 UINTX_FORMAT 1253 &vsize, // 23 UINTX_FORMAT 1254 &rss, // 24 INTX_FORMAT 1255 &rsslim, // 25 UINTX_FORMAT 1256 &scodes, // 26 UINTX_FORMAT 1257 &ecode, // 27 UINTX_FORMAT 1258 &stack_start); // 28 UINTX_FORMAT 1259 } 1260 1261#undef _UFM 1262#undef _DFM 1263 1264 if (i != 28 - 2) { 1265 assert(false, "Bad conversion from /proc/self/stat"); 1266 // product mode - assume we are the initial thread, good luck in the 1267 // embedded case. 1268 warning("Can't detect initial thread stack location - bad conversion"); 1269 stack_start = (uintptr_t) &rlim; 1270 } 1271 } else { 1272 // For some reason we can't open /proc/self/stat (for example, running on 1273 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1274 // most cases, so don't abort: 1275 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1276 stack_start = (uintptr_t) &rlim; 1277 } 1278 } 1279 1280 // Now we have a pointer (stack_start) very close to the stack top, the 1281 // next thing to do is to figure out the exact location of stack top. We 1282 // can find out the virtual memory area that contains stack_start by 1283 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1284 // and its upper limit is the real stack top. (again, this would fail if 1285 // running inside chroot, because /proc may not exist.) 1286 1287 uintptr_t stack_top; 1288 address low, high; 1289 if (find_vma((address)stack_start, &low, &high)) { 1290 // success, "high" is the true stack top. (ignore "low", because initial 1291 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1292 stack_top = (uintptr_t)high; 1293 } else { 1294 // failed, likely because /proc/self/maps does not exist 1295 warning("Can't detect initial thread stack location - find_vma failed"); 1296 // best effort: stack_start is normally within a few pages below the real 1297 // stack top, use it as stack top, and reduce stack size so we won't put 1298 // guard page outside stack. 1299 stack_top = stack_start; 1300 stack_size -= 16 * page_size(); 1301 } 1302 1303 // stack_top could be partially down the page so align it 1304 stack_top = align_size_up(stack_top, page_size()); 1305 1306 if (max_size && stack_size > max_size) { 1307 _initial_thread_stack_size = max_size; 1308 } else { 1309 _initial_thread_stack_size = stack_size; 1310 } 1311 1312 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1313 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1314} 1315 1316//////////////////////////////////////////////////////////////////////////////// 1317// time support 1318 1319// Time since start-up in seconds to a fine granularity. 1320// Used by VMSelfDestructTimer and the MemProfiler. 1321double os::elapsedTime() { 1322 1323 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1324} 1325 1326jlong os::elapsed_counter() { 1327 return javaTimeNanos() - initial_time_count; 1328} 1329 1330jlong os::elapsed_frequency() { 1331 return NANOSECS_PER_SEC; // nanosecond resolution 1332} 1333 1334bool os::supports_vtime() { return true; } 1335bool os::enable_vtime() { return false; } 1336bool os::vtime_enabled() { return false; } 1337 1338double os::elapsedVTime() { 1339 struct rusage usage; 1340 int retval = getrusage(RUSAGE_THREAD, &usage); 1341 if (retval == 0) { 1342 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); 1343 } else { 1344 // better than nothing, but not much 1345 return elapsedTime(); 1346 } 1347} 1348 1349jlong os::javaTimeMillis() { 1350 timeval time; 1351 int status = gettimeofday(&time, NULL); 1352 assert(status != -1, "linux error"); 1353 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1354} 1355 1356#ifndef CLOCK_MONOTONIC 1357 #define CLOCK_MONOTONIC (1) 1358#endif 1359 1360void os::Linux::clock_init() { 1361 // we do dlopen's in this particular order due to bug in linux 1362 // dynamical loader (see 6348968) leading to crash on exit 1363 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1364 if (handle == NULL) { 1365 handle = dlopen("librt.so", RTLD_LAZY); 1366 } 1367 1368 if (handle) { 1369 int (*clock_getres_func)(clockid_t, struct timespec*) = 1370 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1371 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1372 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1373 if (clock_getres_func && clock_gettime_func) { 1374 // See if monotonic clock is supported by the kernel. Note that some 1375 // early implementations simply return kernel jiffies (updated every 1376 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1377 // for nano time (though the monotonic property is still nice to have). 1378 // It's fixed in newer kernels, however clock_getres() still returns 1379 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1380 // resolution for now. Hopefully as people move to new kernels, this 1381 // won't be a problem. 1382 struct timespec res; 1383 struct timespec tp; 1384 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1385 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1386 // yes, monotonic clock is supported 1387 _clock_gettime = clock_gettime_func; 1388 return; 1389 } else { 1390 // close librt if there is no monotonic clock 1391 dlclose(handle); 1392 } 1393 } 1394 } 1395 warning("No monotonic clock was available - timed services may " \ 1396 "be adversely affected if the time-of-day clock changes"); 1397} 1398 1399#ifndef SYS_clock_getres 1400 #if defined(IA32) || defined(AMD64) 1401 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1402 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1403 #else 1404 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1405 #define sys_clock_getres(x,y) -1 1406 #endif 1407#else 1408 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1409#endif 1410 1411void os::Linux::fast_thread_clock_init() { 1412 if (!UseLinuxPosixThreadCPUClocks) { 1413 return; 1414 } 1415 clockid_t clockid; 1416 struct timespec tp; 1417 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1418 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1419 1420 // Switch to using fast clocks for thread cpu time if 1421 // the sys_clock_getres() returns 0 error code. 1422 // Note, that some kernels may support the current thread 1423 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1424 // returned by the pthread_getcpuclockid(). 1425 // If the fast Posix clocks are supported then the sys_clock_getres() 1426 // must return at least tp.tv_sec == 0 which means a resolution 1427 // better than 1 sec. This is extra check for reliability. 1428 1429 if (pthread_getcpuclockid_func && 1430 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1431 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1432 _supports_fast_thread_cpu_time = true; 1433 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1434 } 1435} 1436 1437jlong os::javaTimeNanos() { 1438 if (os::supports_monotonic_clock()) { 1439 struct timespec tp; 1440 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1441 assert(status == 0, "gettime error"); 1442 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1443 return result; 1444 } else { 1445 timeval time; 1446 int status = gettimeofday(&time, NULL); 1447 assert(status != -1, "linux error"); 1448 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1449 return 1000 * usecs; 1450 } 1451} 1452 1453void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1454 if (os::supports_monotonic_clock()) { 1455 info_ptr->max_value = ALL_64_BITS; 1456 1457 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1458 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1459 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1460 } else { 1461 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1462 info_ptr->max_value = ALL_64_BITS; 1463 1464 // gettimeofday is a real time clock so it skips 1465 info_ptr->may_skip_backward = true; 1466 info_ptr->may_skip_forward = true; 1467 } 1468 1469 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1470} 1471 1472// Return the real, user, and system times in seconds from an 1473// arbitrary fixed point in the past. 1474bool os::getTimesSecs(double* process_real_time, 1475 double* process_user_time, 1476 double* process_system_time) { 1477 struct tms ticks; 1478 clock_t real_ticks = times(&ticks); 1479 1480 if (real_ticks == (clock_t) (-1)) { 1481 return false; 1482 } else { 1483 double ticks_per_second = (double) clock_tics_per_sec; 1484 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1485 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1486 *process_real_time = ((double) real_ticks) / ticks_per_second; 1487 1488 return true; 1489 } 1490} 1491 1492 1493char * os::local_time_string(char *buf, size_t buflen) { 1494 struct tm t; 1495 time_t long_time; 1496 time(&long_time); 1497 localtime_r(&long_time, &t); 1498 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1499 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1500 t.tm_hour, t.tm_min, t.tm_sec); 1501 return buf; 1502} 1503 1504struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1505 return localtime_r(clock, res); 1506} 1507 1508//////////////////////////////////////////////////////////////////////////////// 1509// runtime exit support 1510 1511// Note: os::shutdown() might be called very early during initialization, or 1512// called from signal handler. Before adding something to os::shutdown(), make 1513// sure it is async-safe and can handle partially initialized VM. 1514void os::shutdown() { 1515 1516 // allow PerfMemory to attempt cleanup of any persistent resources 1517 perfMemory_exit(); 1518 1519 // needs to remove object in file system 1520 AttachListener::abort(); 1521 1522 // flush buffered output, finish log files 1523 ostream_abort(); 1524 1525 // Check for abort hook 1526 abort_hook_t abort_hook = Arguments::abort_hook(); 1527 if (abort_hook != NULL) { 1528 abort_hook(); 1529 } 1530 1531} 1532 1533// Note: os::abort() might be called very early during initialization, or 1534// called from signal handler. Before adding something to os::abort(), make 1535// sure it is async-safe and can handle partially initialized VM. 1536void os::abort(bool dump_core) { 1537 os::shutdown(); 1538 if (dump_core) { 1539#ifndef PRODUCT 1540 fdStream out(defaultStream::output_fd()); 1541 out.print_raw("Current thread is "); 1542 char buf[16]; 1543 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1544 out.print_raw_cr(buf); 1545 out.print_raw_cr("Dumping core ..."); 1546#endif 1547 ::abort(); // dump core 1548 } 1549 1550 ::exit(1); 1551} 1552 1553// Die immediately, no exit hook, no abort hook, no cleanup. 1554void os::die() { 1555 // _exit() on LinuxThreads only kills current thread 1556 ::abort(); 1557} 1558 1559 1560// This method is a copy of JDK's sysGetLastErrorString 1561// from src/solaris/hpi/src/system_md.c 1562 1563size_t os::lasterror(char *buf, size_t len) { 1564 if (errno == 0) return 0; 1565 1566 const char *s = ::strerror(errno); 1567 size_t n = ::strlen(s); 1568 if (n >= len) { 1569 n = len - 1; 1570 } 1571 ::strncpy(buf, s, n); 1572 buf[n] = '\0'; 1573 return n; 1574} 1575 1576intx os::current_thread_id() { return (intx)pthread_self(); } 1577int os::current_process_id() { 1578 1579 // Under the old linux thread library, linux gives each thread 1580 // its own process id. Because of this each thread will return 1581 // a different pid if this method were to return the result 1582 // of getpid(2). Linux provides no api that returns the pid 1583 // of the launcher thread for the vm. This implementation 1584 // returns a unique pid, the pid of the launcher thread 1585 // that starts the vm 'process'. 1586 1587 // Under the NPTL, getpid() returns the same pid as the 1588 // launcher thread rather than a unique pid per thread. 1589 // Use gettid() if you want the old pre NPTL behaviour. 1590 1591 // if you are looking for the result of a call to getpid() that 1592 // returns a unique pid for the calling thread, then look at the 1593 // OSThread::thread_id() method in osThread_linux.hpp file 1594 1595 return (int)(_initial_pid ? _initial_pid : getpid()); 1596} 1597 1598// DLL functions 1599 1600const char* os::dll_file_extension() { return ".so"; } 1601 1602// This must be hard coded because it's the system's temporary 1603// directory not the java application's temp directory, ala java.io.tmpdir. 1604const char* os::get_temp_directory() { return "/tmp"; } 1605 1606static bool file_exists(const char* filename) { 1607 struct stat statbuf; 1608 if (filename == NULL || strlen(filename) == 0) { 1609 return false; 1610 } 1611 return os::stat(filename, &statbuf) == 0; 1612} 1613 1614bool os::dll_build_name(char* buffer, size_t buflen, 1615 const char* pname, const char* fname) { 1616 bool retval = false; 1617 // Copied from libhpi 1618 const size_t pnamelen = pname ? strlen(pname) : 0; 1619 1620 // Return error on buffer overflow. 1621 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1622 return retval; 1623 } 1624 1625 if (pnamelen == 0) { 1626 snprintf(buffer, buflen, "lib%s.so", fname); 1627 retval = true; 1628 } else if (strchr(pname, *os::path_separator()) != NULL) { 1629 int n; 1630 char** pelements = split_path(pname, &n); 1631 if (pelements == NULL) { 1632 return false; 1633 } 1634 for (int i = 0; i < n; i++) { 1635 // Really shouldn't be NULL, but check can't hurt 1636 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1637 continue; // skip the empty path values 1638 } 1639 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1640 if (file_exists(buffer)) { 1641 retval = true; 1642 break; 1643 } 1644 } 1645 // release the storage 1646 for (int i = 0; i < n; i++) { 1647 if (pelements[i] != NULL) { 1648 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1649 } 1650 } 1651 if (pelements != NULL) { 1652 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1653 } 1654 } else { 1655 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1656 retval = true; 1657 } 1658 return retval; 1659} 1660 1661// check if addr is inside libjvm.so 1662bool os::address_is_in_vm(address addr) { 1663 static address libjvm_base_addr; 1664 Dl_info dlinfo; 1665 1666 if (libjvm_base_addr == NULL) { 1667 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1668 libjvm_base_addr = (address)dlinfo.dli_fbase; 1669 } 1670 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1671 } 1672 1673 if (dladdr((void *)addr, &dlinfo) != 0) { 1674 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1675 } 1676 1677 return false; 1678} 1679 1680bool os::dll_address_to_function_name(address addr, char *buf, 1681 int buflen, int *offset) { 1682 // buf is not optional, but offset is optional 1683 assert(buf != NULL, "sanity check"); 1684 1685 Dl_info dlinfo; 1686 1687 if (dladdr((void*)addr, &dlinfo) != 0) { 1688 // see if we have a matching symbol 1689 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1690 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1691 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1692 } 1693 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1694 return true; 1695 } 1696 // no matching symbol so try for just file info 1697 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1698 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1699 buf, buflen, offset, dlinfo.dli_fname)) { 1700 return true; 1701 } 1702 } 1703 } 1704 1705 buf[0] = '\0'; 1706 if (offset != NULL) *offset = -1; 1707 return false; 1708} 1709 1710struct _address_to_library_name { 1711 address addr; // input : memory address 1712 size_t buflen; // size of fname 1713 char* fname; // output: library name 1714 address base; // library base addr 1715}; 1716 1717static int address_to_library_name_callback(struct dl_phdr_info *info, 1718 size_t size, void *data) { 1719 int i; 1720 bool found = false; 1721 address libbase = NULL; 1722 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1723 1724 // iterate through all loadable segments 1725 for (i = 0; i < info->dlpi_phnum; i++) { 1726 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1727 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1728 // base address of a library is the lowest address of its loaded 1729 // segments. 1730 if (libbase == NULL || libbase > segbase) { 1731 libbase = segbase; 1732 } 1733 // see if 'addr' is within current segment 1734 if (segbase <= d->addr && 1735 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1736 found = true; 1737 } 1738 } 1739 } 1740 1741 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1742 // so dll_address_to_library_name() can fall through to use dladdr() which 1743 // can figure out executable name from argv[0]. 1744 if (found && info->dlpi_name && info->dlpi_name[0]) { 1745 d->base = libbase; 1746 if (d->fname) { 1747 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1748 } 1749 return 1; 1750 } 1751 return 0; 1752} 1753 1754bool os::dll_address_to_library_name(address addr, char* buf, 1755 int buflen, int* offset) { 1756 // buf is not optional, but offset is optional 1757 assert(buf != NULL, "sanity check"); 1758 1759 Dl_info dlinfo; 1760 struct _address_to_library_name data; 1761 1762 // There is a bug in old glibc dladdr() implementation that it could resolve 1763 // to wrong library name if the .so file has a base address != NULL. Here 1764 // we iterate through the program headers of all loaded libraries to find 1765 // out which library 'addr' really belongs to. This workaround can be 1766 // removed once the minimum requirement for glibc is moved to 2.3.x. 1767 data.addr = addr; 1768 data.fname = buf; 1769 data.buflen = buflen; 1770 data.base = NULL; 1771 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1772 1773 if (rslt) { 1774 // buf already contains library name 1775 if (offset) *offset = addr - data.base; 1776 return true; 1777 } 1778 if (dladdr((void*)addr, &dlinfo) != 0) { 1779 if (dlinfo.dli_fname != NULL) { 1780 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1781 } 1782 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1783 *offset = addr - (address)dlinfo.dli_fbase; 1784 } 1785 return true; 1786 } 1787 1788 buf[0] = '\0'; 1789 if (offset) *offset = -1; 1790 return false; 1791} 1792 1793// Loads .dll/.so and 1794// in case of error it checks if .dll/.so was built for the 1795// same architecture as Hotspot is running on 1796 1797 1798// Remember the stack's state. The Linux dynamic linker will change 1799// the stack to 'executable' at most once, so we must safepoint only once. 1800bool os::Linux::_stack_is_executable = false; 1801 1802// VM operation that loads a library. This is necessary if stack protection 1803// of the Java stacks can be lost during loading the library. If we 1804// do not stop the Java threads, they can stack overflow before the stacks 1805// are protected again. 1806class VM_LinuxDllLoad: public VM_Operation { 1807 private: 1808 const char *_filename; 1809 char *_ebuf; 1810 int _ebuflen; 1811 void *_lib; 1812 public: 1813 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1814 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1815 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1816 void doit() { 1817 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1818 os::Linux::_stack_is_executable = true; 1819 } 1820 void* loaded_library() { return _lib; } 1821}; 1822 1823void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1824 void * result = NULL; 1825 bool load_attempted = false; 1826 1827 // Check whether the library to load might change execution rights 1828 // of the stack. If they are changed, the protection of the stack 1829 // guard pages will be lost. We need a safepoint to fix this. 1830 // 1831 // See Linux man page execstack(8) for more info. 1832 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1833 ElfFile ef(filename); 1834 if (!ef.specifies_noexecstack()) { 1835 if (!is_init_completed()) { 1836 os::Linux::_stack_is_executable = true; 1837 // This is OK - No Java threads have been created yet, and hence no 1838 // stack guard pages to fix. 1839 // 1840 // This should happen only when you are building JDK7 using a very 1841 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1842 // 1843 // Dynamic loader will make all stacks executable after 1844 // this function returns, and will not do that again. 1845 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1846 } else { 1847 warning("You have loaded library %s which might have disabled stack guard. " 1848 "The VM will try to fix the stack guard now.\n" 1849 "It's highly recommended that you fix the library with " 1850 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1851 filename); 1852 1853 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1854 JavaThread *jt = JavaThread::current(); 1855 if (jt->thread_state() != _thread_in_native) { 1856 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1857 // that requires ExecStack. Cannot enter safe point. Let's give up. 1858 warning("Unable to fix stack guard. Giving up."); 1859 } else { 1860 if (!LoadExecStackDllInVMThread) { 1861 // This is for the case where the DLL has an static 1862 // constructor function that executes JNI code. We cannot 1863 // load such DLLs in the VMThread. 1864 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1865 } 1866 1867 ThreadInVMfromNative tiv(jt); 1868 debug_only(VMNativeEntryWrapper vew;) 1869 1870 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1871 VMThread::execute(&op); 1872 if (LoadExecStackDllInVMThread) { 1873 result = op.loaded_library(); 1874 } 1875 load_attempted = true; 1876 } 1877 } 1878 } 1879 } 1880 1881 if (!load_attempted) { 1882 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1883 } 1884 1885 if (result != NULL) { 1886 // Successful loading 1887 return result; 1888 } 1889 1890 Elf32_Ehdr elf_head; 1891 int diag_msg_max_length=ebuflen-strlen(ebuf); 1892 char* diag_msg_buf=ebuf+strlen(ebuf); 1893 1894 if (diag_msg_max_length==0) { 1895 // No more space in ebuf for additional diagnostics message 1896 return NULL; 1897 } 1898 1899 1900 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1901 1902 if (file_descriptor < 0) { 1903 // Can't open library, report dlerror() message 1904 return NULL; 1905 } 1906 1907 bool failed_to_read_elf_head= 1908 (sizeof(elf_head)!= 1909 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1910 1911 ::close(file_descriptor); 1912 if (failed_to_read_elf_head) { 1913 // file i/o error - report dlerror() msg 1914 return NULL; 1915 } 1916 1917 typedef struct { 1918 Elf32_Half code; // Actual value as defined in elf.h 1919 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1920 char elf_class; // 32 or 64 bit 1921 char endianess; // MSB or LSB 1922 char* name; // String representation 1923 } arch_t; 1924 1925#ifndef EM_486 1926 #define EM_486 6 /* Intel 80486 */ 1927#endif 1928 1929 static const arch_t arch_array[]={ 1930 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1931 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1932 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1933 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1934 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1935 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1936 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1937 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1938#if defined(VM_LITTLE_ENDIAN) 1939 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"}, 1940#else 1941 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1942#endif 1943 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1944 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1945 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1946 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1947 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1948 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1949 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1950 }; 1951 1952#if (defined IA32) 1953 static Elf32_Half running_arch_code=EM_386; 1954#elif (defined AMD64) 1955 static Elf32_Half running_arch_code=EM_X86_64; 1956#elif (defined IA64) 1957 static Elf32_Half running_arch_code=EM_IA_64; 1958#elif (defined __sparc) && (defined _LP64) 1959 static Elf32_Half running_arch_code=EM_SPARCV9; 1960#elif (defined __sparc) && (!defined _LP64) 1961 static Elf32_Half running_arch_code=EM_SPARC; 1962#elif (defined __powerpc64__) 1963 static Elf32_Half running_arch_code=EM_PPC64; 1964#elif (defined __powerpc__) 1965 static Elf32_Half running_arch_code=EM_PPC; 1966#elif (defined ARM) 1967 static Elf32_Half running_arch_code=EM_ARM; 1968#elif (defined S390) 1969 static Elf32_Half running_arch_code=EM_S390; 1970#elif (defined ALPHA) 1971 static Elf32_Half running_arch_code=EM_ALPHA; 1972#elif (defined MIPSEL) 1973 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1974#elif (defined PARISC) 1975 static Elf32_Half running_arch_code=EM_PARISC; 1976#elif (defined MIPS) 1977 static Elf32_Half running_arch_code=EM_MIPS; 1978#elif (defined M68K) 1979 static Elf32_Half running_arch_code=EM_68K; 1980#else 1981 #error Method os::dll_load requires that one of following is defined:\ 1982 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 1983#endif 1984 1985 // Identify compatability class for VM's architecture and library's architecture 1986 // Obtain string descriptions for architectures 1987 1988 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1989 int running_arch_index=-1; 1990 1991 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1992 if (running_arch_code == arch_array[i].code) { 1993 running_arch_index = i; 1994 } 1995 if (lib_arch.code == arch_array[i].code) { 1996 lib_arch.compat_class = arch_array[i].compat_class; 1997 lib_arch.name = arch_array[i].name; 1998 } 1999 } 2000 2001 assert(running_arch_index != -1, 2002 "Didn't find running architecture code (running_arch_code) in arch_array"); 2003 if (running_arch_index == -1) { 2004 // Even though running architecture detection failed 2005 // we may still continue with reporting dlerror() message 2006 return NULL; 2007 } 2008 2009 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 2010 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 2011 return NULL; 2012 } 2013 2014#ifndef S390 2015 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 2016 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 2017 return NULL; 2018 } 2019#endif // !S390 2020 2021 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2022 if (lib_arch.name!=NULL) { 2023 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2024 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2025 lib_arch.name, arch_array[running_arch_index].name); 2026 } else { 2027 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2028 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2029 lib_arch.code, 2030 arch_array[running_arch_index].name); 2031 } 2032 } 2033 2034 return NULL; 2035} 2036 2037void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 2038 int ebuflen) { 2039 void * result = ::dlopen(filename, RTLD_LAZY); 2040 if (result == NULL) { 2041 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2042 ebuf[ebuflen-1] = '\0'; 2043 } 2044 return result; 2045} 2046 2047void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 2048 int ebuflen) { 2049 void * result = NULL; 2050 if (LoadExecStackDllInVMThread) { 2051 result = dlopen_helper(filename, ebuf, ebuflen); 2052 } 2053 2054 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2055 // library that requires an executable stack, or which does not have this 2056 // stack attribute set, dlopen changes the stack attribute to executable. The 2057 // read protection of the guard pages gets lost. 2058 // 2059 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2060 // may have been queued at the same time. 2061 2062 if (!_stack_is_executable) { 2063 JavaThread *jt = Threads::first(); 2064 2065 while (jt) { 2066 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2067 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2068 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2069 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2070 warning("Attempt to reguard stack yellow zone failed."); 2071 } 2072 } 2073 jt = jt->next(); 2074 } 2075 } 2076 2077 return result; 2078} 2079 2080// glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2081// chances are you might want to run the generated bits against glibc-2.0 2082// libdl.so, so always use locking for any version of glibc. 2083// 2084void* os::dll_lookup(void* handle, const char* name) { 2085 pthread_mutex_lock(&dl_mutex); 2086 void* res = dlsym(handle, name); 2087 pthread_mutex_unlock(&dl_mutex); 2088 return res; 2089} 2090 2091void* os::get_default_process_handle() { 2092 return (void*)::dlopen(NULL, RTLD_LAZY); 2093} 2094 2095static bool _print_ascii_file(const char* filename, outputStream* st) { 2096 int fd = ::open(filename, O_RDONLY); 2097 if (fd == -1) { 2098 return false; 2099 } 2100 2101 char buf[32]; 2102 int bytes; 2103 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2104 st->print_raw(buf, bytes); 2105 } 2106 2107 ::close(fd); 2108 2109 return true; 2110} 2111 2112void os::print_dll_info(outputStream *st) { 2113 st->print_cr("Dynamic libraries:"); 2114 2115 char fname[32]; 2116 pid_t pid = os::Linux::gettid(); 2117 2118 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2119 2120 if (!_print_ascii_file(fname, st)) { 2121 st->print("Can not get library information for pid = %d\n", pid); 2122 } 2123} 2124 2125void os::print_os_info_brief(outputStream* st) { 2126 os::Linux::print_distro_info(st); 2127 2128 os::Posix::print_uname_info(st); 2129 2130 os::Linux::print_libversion_info(st); 2131 2132} 2133 2134void os::print_os_info(outputStream* st) { 2135 st->print("OS:"); 2136 2137 os::Linux::print_distro_info(st); 2138 2139 os::Posix::print_uname_info(st); 2140 2141 // Print warning if unsafe chroot environment detected 2142 if (unsafe_chroot_detected) { 2143 st->print("WARNING!! "); 2144 st->print_cr("%s", unstable_chroot_error); 2145 } 2146 2147 os::Linux::print_libversion_info(st); 2148 2149 os::Posix::print_rlimit_info(st); 2150 2151 os::Posix::print_load_average(st); 2152 2153 os::Linux::print_full_memory_info(st); 2154} 2155 2156// Try to identify popular distros. 2157// Most Linux distributions have a /etc/XXX-release file, which contains 2158// the OS version string. Newer Linux distributions have a /etc/lsb-release 2159// file that also contains the OS version string. Some have more than one 2160// /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2161// /etc/redhat-release.), so the order is important. 2162// Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2163// their own specific XXX-release file as well as a redhat-release file. 2164// Because of this the XXX-release file needs to be searched for before the 2165// redhat-release file. 2166// Since Red Hat has a lsb-release file that is not very descriptive the 2167// search for redhat-release needs to be before lsb-release. 2168// Since the lsb-release file is the new standard it needs to be searched 2169// before the older style release files. 2170// Searching system-release (Red Hat) and os-release (other Linuxes) are a 2171// next to last resort. The os-release file is a new standard that contains 2172// distribution information and the system-release file seems to be an old 2173// standard that has been replaced by the lsb-release and os-release files. 2174// Searching for the debian_version file is the last resort. It contains 2175// an informative string like "6.0.6" or "wheezy/sid". Because of this 2176// "Debian " is printed before the contents of the debian_version file. 2177void os::Linux::print_distro_info(outputStream* st) { 2178 if (!_print_ascii_file("/etc/oracle-release", st) && 2179 !_print_ascii_file("/etc/mandriva-release", st) && 2180 !_print_ascii_file("/etc/mandrake-release", st) && 2181 !_print_ascii_file("/etc/sun-release", st) && 2182 !_print_ascii_file("/etc/redhat-release", st) && 2183 !_print_ascii_file("/etc/lsb-release", st) && 2184 !_print_ascii_file("/etc/SuSE-release", st) && 2185 !_print_ascii_file("/etc/turbolinux-release", st) && 2186 !_print_ascii_file("/etc/gentoo-release", st) && 2187 !_print_ascii_file("/etc/ltib-release", st) && 2188 !_print_ascii_file("/etc/angstrom-version", st) && 2189 !_print_ascii_file("/etc/system-release", st) && 2190 !_print_ascii_file("/etc/os-release", st)) { 2191 2192 if (file_exists("/etc/debian_version")) { 2193 st->print("Debian "); 2194 _print_ascii_file("/etc/debian_version", st); 2195 } else { 2196 st->print("Linux"); 2197 } 2198 } 2199 st->cr(); 2200} 2201 2202void os::Linux::print_libversion_info(outputStream* st) { 2203 // libc, pthread 2204 st->print("libc:"); 2205 st->print("%s ", os::Linux::glibc_version()); 2206 st->print("%s ", os::Linux::libpthread_version()); 2207 if (os::Linux::is_LinuxThreads()) { 2208 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2209 } 2210 st->cr(); 2211} 2212 2213void os::Linux::print_full_memory_info(outputStream* st) { 2214 st->print("\n/proc/meminfo:\n"); 2215 _print_ascii_file("/proc/meminfo", st); 2216 st->cr(); 2217} 2218 2219void os::print_memory_info(outputStream* st) { 2220 2221 st->print("Memory:"); 2222 st->print(" %dk page", os::vm_page_size()>>10); 2223 2224 // values in struct sysinfo are "unsigned long" 2225 struct sysinfo si; 2226 sysinfo(&si); 2227 2228 st->print(", physical " UINT64_FORMAT "k", 2229 os::physical_memory() >> 10); 2230 st->print("(" UINT64_FORMAT "k free)", 2231 os::available_memory() >> 10); 2232 st->print(", swap " UINT64_FORMAT "k", 2233 ((jlong)si.totalswap * si.mem_unit) >> 10); 2234 st->print("(" UINT64_FORMAT "k free)", 2235 ((jlong)si.freeswap * si.mem_unit) >> 10); 2236 st->cr(); 2237} 2238 2239void os::pd_print_cpu_info(outputStream* st) { 2240 st->print("\n/proc/cpuinfo:\n"); 2241 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2242 st->print(" <Not Available>"); 2243 } 2244 st->cr(); 2245} 2246 2247void os::print_siginfo(outputStream* st, void* siginfo) { 2248 const siginfo_t* si = (const siginfo_t*)siginfo; 2249 2250 os::Posix::print_siginfo_brief(st, si); 2251#if INCLUDE_CDS 2252 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2253 UseSharedSpaces) { 2254 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2255 if (mapinfo->is_in_shared_space(si->si_addr)) { 2256 st->print("\n\nError accessing class data sharing archive." \ 2257 " Mapped file inaccessible during execution, " \ 2258 " possible disk/network problem."); 2259 } 2260 } 2261#endif 2262 st->cr(); 2263} 2264 2265 2266static void print_signal_handler(outputStream* st, int sig, 2267 char* buf, size_t buflen); 2268 2269void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2270 st->print_cr("Signal Handlers:"); 2271 print_signal_handler(st, SIGSEGV, buf, buflen); 2272 print_signal_handler(st, SIGBUS , buf, buflen); 2273 print_signal_handler(st, SIGFPE , buf, buflen); 2274 print_signal_handler(st, SIGPIPE, buf, buflen); 2275 print_signal_handler(st, SIGXFSZ, buf, buflen); 2276 print_signal_handler(st, SIGILL , buf, buflen); 2277 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2278 print_signal_handler(st, SR_signum, buf, buflen); 2279 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2280 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2281 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2282 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2283#if defined(PPC64) 2284 print_signal_handler(st, SIGTRAP, buf, buflen); 2285#endif 2286} 2287 2288static char saved_jvm_path[MAXPATHLEN] = {0}; 2289 2290// Find the full path to the current module, libjvm.so 2291void os::jvm_path(char *buf, jint buflen) { 2292 // Error checking. 2293 if (buflen < MAXPATHLEN) { 2294 assert(false, "must use a large-enough buffer"); 2295 buf[0] = '\0'; 2296 return; 2297 } 2298 // Lazy resolve the path to current module. 2299 if (saved_jvm_path[0] != 0) { 2300 strcpy(buf, saved_jvm_path); 2301 return; 2302 } 2303 2304 char dli_fname[MAXPATHLEN]; 2305 bool ret = dll_address_to_library_name( 2306 CAST_FROM_FN_PTR(address, os::jvm_path), 2307 dli_fname, sizeof(dli_fname), NULL); 2308 assert(ret, "cannot locate libjvm"); 2309 char *rp = NULL; 2310 if (ret && dli_fname[0] != '\0') { 2311 rp = realpath(dli_fname, buf); 2312 } 2313 if (rp == NULL) { 2314 return; 2315 } 2316 2317 if (Arguments::sun_java_launcher_is_altjvm()) { 2318 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2319 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2320 // If "/jre/lib/" appears at the right place in the string, then 2321 // assume we are installed in a JDK and we're done. Otherwise, check 2322 // for a JAVA_HOME environment variable and fix up the path so it 2323 // looks like libjvm.so is installed there (append a fake suffix 2324 // hotspot/libjvm.so). 2325 const char *p = buf + strlen(buf) - 1; 2326 for (int count = 0; p > buf && count < 5; ++count) { 2327 for (--p; p > buf && *p != '/'; --p) 2328 /* empty */ ; 2329 } 2330 2331 if (strncmp(p, "/jre/lib/", 9) != 0) { 2332 // Look for JAVA_HOME in the environment. 2333 char* java_home_var = ::getenv("JAVA_HOME"); 2334 if (java_home_var != NULL && java_home_var[0] != 0) { 2335 char* jrelib_p; 2336 int len; 2337 2338 // Check the current module name "libjvm.so". 2339 p = strrchr(buf, '/'); 2340 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2341 2342 rp = realpath(java_home_var, buf); 2343 if (rp == NULL) { 2344 return; 2345 } 2346 2347 // determine if this is a legacy image or modules image 2348 // modules image doesn't have "jre" subdirectory 2349 len = strlen(buf); 2350 assert(len < buflen, "Ran out of buffer room"); 2351 jrelib_p = buf + len; 2352 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2353 if (0 != access(buf, F_OK)) { 2354 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2355 } 2356 2357 if (0 == access(buf, F_OK)) { 2358 // Use current module name "libjvm.so" 2359 len = strlen(buf); 2360 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2361 } else { 2362 // Go back to path of .so 2363 rp = realpath(dli_fname, buf); 2364 if (rp == NULL) { 2365 return; 2366 } 2367 } 2368 } 2369 } 2370 } 2371 2372 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2373} 2374 2375void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2376 // no prefix required, not even "_" 2377} 2378 2379void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2380 // no suffix required 2381} 2382 2383//////////////////////////////////////////////////////////////////////////////// 2384// sun.misc.Signal support 2385 2386static volatile jint sigint_count = 0; 2387 2388static void UserHandler(int sig, void *siginfo, void *context) { 2389 // 4511530 - sem_post is serialized and handled by the manager thread. When 2390 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2391 // don't want to flood the manager thread with sem_post requests. 2392 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2393 return; 2394 } 2395 2396 // Ctrl-C is pressed during error reporting, likely because the error 2397 // handler fails to abort. Let VM die immediately. 2398 if (sig == SIGINT && is_error_reported()) { 2399 os::die(); 2400 } 2401 2402 os::signal_notify(sig); 2403} 2404 2405void* os::user_handler() { 2406 return CAST_FROM_FN_PTR(void*, UserHandler); 2407} 2408 2409class Semaphore : public StackObj { 2410 public: 2411 Semaphore(); 2412 ~Semaphore(); 2413 void signal(); 2414 void wait(); 2415 bool trywait(); 2416 bool timedwait(unsigned int sec, int nsec); 2417 private: 2418 sem_t _semaphore; 2419}; 2420 2421Semaphore::Semaphore() { 2422 sem_init(&_semaphore, 0, 0); 2423} 2424 2425Semaphore::~Semaphore() { 2426 sem_destroy(&_semaphore); 2427} 2428 2429void Semaphore::signal() { 2430 sem_post(&_semaphore); 2431} 2432 2433void Semaphore::wait() { 2434 sem_wait(&_semaphore); 2435} 2436 2437bool Semaphore::trywait() { 2438 return sem_trywait(&_semaphore) == 0; 2439} 2440 2441bool Semaphore::timedwait(unsigned int sec, int nsec) { 2442 2443 struct timespec ts; 2444 // Semaphore's are always associated with CLOCK_REALTIME 2445 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2446 // see unpackTime for discussion on overflow checking 2447 if (sec >= MAX_SECS) { 2448 ts.tv_sec += MAX_SECS; 2449 ts.tv_nsec = 0; 2450 } else { 2451 ts.tv_sec += sec; 2452 ts.tv_nsec += nsec; 2453 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2454 ts.tv_nsec -= NANOSECS_PER_SEC; 2455 ++ts.tv_sec; // note: this must be <= max_secs 2456 } 2457 } 2458 2459 while (1) { 2460 int result = sem_timedwait(&_semaphore, &ts); 2461 if (result == 0) { 2462 return true; 2463 } else if (errno == EINTR) { 2464 continue; 2465 } else if (errno == ETIMEDOUT) { 2466 return false; 2467 } else { 2468 return false; 2469 } 2470 } 2471} 2472 2473extern "C" { 2474 typedef void (*sa_handler_t)(int); 2475 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2476} 2477 2478void* os::signal(int signal_number, void* handler) { 2479 struct sigaction sigAct, oldSigAct; 2480 2481 sigfillset(&(sigAct.sa_mask)); 2482 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2483 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2484 2485 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2486 // -1 means registration failed 2487 return (void *)-1; 2488 } 2489 2490 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2491} 2492 2493void os::signal_raise(int signal_number) { 2494 ::raise(signal_number); 2495} 2496 2497// The following code is moved from os.cpp for making this 2498// code platform specific, which it is by its very nature. 2499 2500// Will be modified when max signal is changed to be dynamic 2501int os::sigexitnum_pd() { 2502 return NSIG; 2503} 2504 2505// a counter for each possible signal value 2506static volatile jint pending_signals[NSIG+1] = { 0 }; 2507 2508// Linux(POSIX) specific hand shaking semaphore. 2509static sem_t sig_sem; 2510static Semaphore sr_semaphore; 2511 2512void os::signal_init_pd() { 2513 // Initialize signal structures 2514 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2515 2516 // Initialize signal semaphore 2517 ::sem_init(&sig_sem, 0, 0); 2518} 2519 2520void os::signal_notify(int sig) { 2521 Atomic::inc(&pending_signals[sig]); 2522 ::sem_post(&sig_sem); 2523} 2524 2525static int check_pending_signals(bool wait) { 2526 Atomic::store(0, &sigint_count); 2527 for (;;) { 2528 for (int i = 0; i < NSIG + 1; i++) { 2529 jint n = pending_signals[i]; 2530 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2531 return i; 2532 } 2533 } 2534 if (!wait) { 2535 return -1; 2536 } 2537 JavaThread *thread = JavaThread::current(); 2538 ThreadBlockInVM tbivm(thread); 2539 2540 bool threadIsSuspended; 2541 do { 2542 thread->set_suspend_equivalent(); 2543 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2544 ::sem_wait(&sig_sem); 2545 2546 // were we externally suspended while we were waiting? 2547 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2548 if (threadIsSuspended) { 2549 // The semaphore has been incremented, but while we were waiting 2550 // another thread suspended us. We don't want to continue running 2551 // while suspended because that would surprise the thread that 2552 // suspended us. 2553 ::sem_post(&sig_sem); 2554 2555 thread->java_suspend_self(); 2556 } 2557 } while (threadIsSuspended); 2558 } 2559} 2560 2561int os::signal_lookup() { 2562 return check_pending_signals(false); 2563} 2564 2565int os::signal_wait() { 2566 return check_pending_signals(true); 2567} 2568 2569//////////////////////////////////////////////////////////////////////////////// 2570// Virtual Memory 2571 2572int os::vm_page_size() { 2573 // Seems redundant as all get out 2574 assert(os::Linux::page_size() != -1, "must call os::init"); 2575 return os::Linux::page_size(); 2576} 2577 2578// Solaris allocates memory by pages. 2579int os::vm_allocation_granularity() { 2580 assert(os::Linux::page_size() != -1, "must call os::init"); 2581 return os::Linux::page_size(); 2582} 2583 2584// Rationale behind this function: 2585// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2586// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2587// samples for JITted code. Here we create private executable mapping over the code cache 2588// and then we can use standard (well, almost, as mapping can change) way to provide 2589// info for the reporting script by storing timestamp and location of symbol 2590void linux_wrap_code(char* base, size_t size) { 2591 static volatile jint cnt = 0; 2592 2593 if (!UseOprofile) { 2594 return; 2595 } 2596 2597 char buf[PATH_MAX+1]; 2598 int num = Atomic::add(1, &cnt); 2599 2600 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2601 os::get_temp_directory(), os::current_process_id(), num); 2602 unlink(buf); 2603 2604 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2605 2606 if (fd != -1) { 2607 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2608 if (rv != (off_t)-1) { 2609 if (::write(fd, "", 1) == 1) { 2610 mmap(base, size, 2611 PROT_READ|PROT_WRITE|PROT_EXEC, 2612 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2613 } 2614 } 2615 ::close(fd); 2616 unlink(buf); 2617 } 2618} 2619 2620static bool recoverable_mmap_error(int err) { 2621 // See if the error is one we can let the caller handle. This 2622 // list of errno values comes from JBS-6843484. I can't find a 2623 // Linux man page that documents this specific set of errno 2624 // values so while this list currently matches Solaris, it may 2625 // change as we gain experience with this failure mode. 2626 switch (err) { 2627 case EBADF: 2628 case EINVAL: 2629 case ENOTSUP: 2630 // let the caller deal with these errors 2631 return true; 2632 2633 default: 2634 // Any remaining errors on this OS can cause our reserved mapping 2635 // to be lost. That can cause confusion where different data 2636 // structures think they have the same memory mapped. The worst 2637 // scenario is if both the VM and a library think they have the 2638 // same memory mapped. 2639 return false; 2640 } 2641} 2642 2643static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2644 int err) { 2645 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2646 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2647 strerror(err), err); 2648} 2649 2650static void warn_fail_commit_memory(char* addr, size_t size, 2651 size_t alignment_hint, bool exec, 2652 int err) { 2653 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2654 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2655 alignment_hint, exec, strerror(err), err); 2656} 2657 2658// NOTE: Linux kernel does not really reserve the pages for us. 2659// All it does is to check if there are enough free pages 2660// left at the time of mmap(). This could be a potential 2661// problem. 2662int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2663 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2664 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2665 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2666 if (res != (uintptr_t) MAP_FAILED) { 2667 if (UseNUMAInterleaving) { 2668 numa_make_global(addr, size); 2669 } 2670 return 0; 2671 } 2672 2673 int err = errno; // save errno from mmap() call above 2674 2675 if (!recoverable_mmap_error(err)) { 2676 warn_fail_commit_memory(addr, size, exec, err); 2677 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2678 } 2679 2680 return err; 2681} 2682 2683bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2684 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2685} 2686 2687void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2688 const char* mesg) { 2689 assert(mesg != NULL, "mesg must be specified"); 2690 int err = os::Linux::commit_memory_impl(addr, size, exec); 2691 if (err != 0) { 2692 // the caller wants all commit errors to exit with the specified mesg: 2693 warn_fail_commit_memory(addr, size, exec, err); 2694 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2695 } 2696} 2697 2698// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2699#ifndef MAP_HUGETLB 2700 #define MAP_HUGETLB 0x40000 2701#endif 2702 2703// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2704#ifndef MADV_HUGEPAGE 2705 #define MADV_HUGEPAGE 14 2706#endif 2707 2708int os::Linux::commit_memory_impl(char* addr, size_t size, 2709 size_t alignment_hint, bool exec) { 2710 int err = os::Linux::commit_memory_impl(addr, size, exec); 2711 if (err == 0) { 2712 realign_memory(addr, size, alignment_hint); 2713 } 2714 return err; 2715} 2716 2717bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2718 bool exec) { 2719 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2720} 2721 2722void os::pd_commit_memory_or_exit(char* addr, size_t size, 2723 size_t alignment_hint, bool exec, 2724 const char* mesg) { 2725 assert(mesg != NULL, "mesg must be specified"); 2726 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2727 if (err != 0) { 2728 // the caller wants all commit errors to exit with the specified mesg: 2729 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2730 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2731 } 2732} 2733 2734void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2735 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2736 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2737 // be supported or the memory may already be backed by huge pages. 2738 ::madvise(addr, bytes, MADV_HUGEPAGE); 2739 } 2740} 2741 2742void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2743 // This method works by doing an mmap over an existing mmaping and effectively discarding 2744 // the existing pages. However it won't work for SHM-based large pages that cannot be 2745 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2746 // small pages on top of the SHM segment. This method always works for small pages, so we 2747 // allow that in any case. 2748 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2749 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2750 } 2751} 2752 2753void os::numa_make_global(char *addr, size_t bytes) { 2754 Linux::numa_interleave_memory(addr, bytes); 2755} 2756 2757// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2758// bind policy to MPOL_PREFERRED for the current thread. 2759#define USE_MPOL_PREFERRED 0 2760 2761void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2762 // To make NUMA and large pages more robust when both enabled, we need to ease 2763 // the requirements on where the memory should be allocated. MPOL_BIND is the 2764 // default policy and it will force memory to be allocated on the specified 2765 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2766 // the specified node, but will not force it. Using this policy will prevent 2767 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2768 // free large pages. 2769 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2770 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2771} 2772 2773bool os::numa_topology_changed() { return false; } 2774 2775size_t os::numa_get_groups_num() { 2776 int max_node = Linux::numa_max_node(); 2777 return max_node > 0 ? max_node + 1 : 1; 2778} 2779 2780int os::numa_get_group_id() { 2781 int cpu_id = Linux::sched_getcpu(); 2782 if (cpu_id != -1) { 2783 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2784 if (lgrp_id != -1) { 2785 return lgrp_id; 2786 } 2787 } 2788 return 0; 2789} 2790 2791size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2792 for (size_t i = 0; i < size; i++) { 2793 ids[i] = i; 2794 } 2795 return size; 2796} 2797 2798bool os::get_page_info(char *start, page_info* info) { 2799 return false; 2800} 2801 2802char *os::scan_pages(char *start, char* end, page_info* page_expected, 2803 page_info* page_found) { 2804 return end; 2805} 2806 2807 2808int os::Linux::sched_getcpu_syscall(void) { 2809 unsigned int cpu; 2810 int retval = -1; 2811 2812#if defined(IA32) 2813 #ifndef SYS_getcpu 2814 #define SYS_getcpu 318 2815 #endif 2816 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2817#elif defined(AMD64) 2818// Unfortunately we have to bring all these macros here from vsyscall.h 2819// to be able to compile on old linuxes. 2820 #define __NR_vgetcpu 2 2821 #define VSYSCALL_START (-10UL << 20) 2822 #define VSYSCALL_SIZE 1024 2823 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2824 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2825 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2826 retval = vgetcpu(&cpu, NULL, NULL); 2827#endif 2828 2829 return (retval == -1) ? retval : cpu; 2830} 2831 2832// Something to do with the numa-aware allocator needs these symbols 2833extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2834extern "C" JNIEXPORT void numa_error(char *where) { } 2835extern "C" JNIEXPORT int fork1() { return fork(); } 2836 2837 2838// If we are running with libnuma version > 2, then we should 2839// be trying to use symbols with versions 1.1 2840// If we are running with earlier version, which did not have symbol versions, 2841// we should use the base version. 2842void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2843 void *f = dlvsym(handle, name, "libnuma_1.1"); 2844 if (f == NULL) { 2845 f = dlsym(handle, name); 2846 } 2847 return f; 2848} 2849 2850bool os::Linux::libnuma_init() { 2851 // sched_getcpu() should be in libc. 2852 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2853 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2854 2855 // If it's not, try a direct syscall. 2856 if (sched_getcpu() == -1) { 2857 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2858 (void*)&sched_getcpu_syscall)); 2859 } 2860 2861 if (sched_getcpu() != -1) { // Does it work? 2862 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2863 if (handle != NULL) { 2864 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2865 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2866 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2867 libnuma_dlsym(handle, "numa_max_node"))); 2868 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2869 libnuma_dlsym(handle, "numa_available"))); 2870 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2871 libnuma_dlsym(handle, "numa_tonode_memory"))); 2872 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2873 libnuma_dlsym(handle, "numa_interleave_memory"))); 2874 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2875 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2876 2877 2878 if (numa_available() != -1) { 2879 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2880 // Create a cpu -> node mapping 2881 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2882 rebuild_cpu_to_node_map(); 2883 return true; 2884 } 2885 } 2886 } 2887 return false; 2888} 2889 2890// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2891// The table is later used in get_node_by_cpu(). 2892void os::Linux::rebuild_cpu_to_node_map() { 2893 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2894 // in libnuma (possible values are starting from 16, 2895 // and continuing up with every other power of 2, but less 2896 // than the maximum number of CPUs supported by kernel), and 2897 // is a subject to change (in libnuma version 2 the requirements 2898 // are more reasonable) we'll just hardcode the number they use 2899 // in the library. 2900 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2901 2902 size_t cpu_num = os::active_processor_count(); 2903 size_t cpu_map_size = NCPUS / BitsPerCLong; 2904 size_t cpu_map_valid_size = 2905 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2906 2907 cpu_to_node()->clear(); 2908 cpu_to_node()->at_grow(cpu_num - 1); 2909 size_t node_num = numa_get_groups_num(); 2910 2911 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2912 for (size_t i = 0; i < node_num; i++) { 2913 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2914 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2915 if (cpu_map[j] != 0) { 2916 for (size_t k = 0; k < BitsPerCLong; k++) { 2917 if (cpu_map[j] & (1UL << k)) { 2918 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2919 } 2920 } 2921 } 2922 } 2923 } 2924 } 2925 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2926} 2927 2928int os::Linux::get_node_by_cpu(int cpu_id) { 2929 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2930 return cpu_to_node()->at(cpu_id); 2931 } 2932 return -1; 2933} 2934 2935GrowableArray<int>* os::Linux::_cpu_to_node; 2936os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2937os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2938os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2939os::Linux::numa_available_func_t os::Linux::_numa_available; 2940os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2941os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2942os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2943unsigned long* os::Linux::_numa_all_nodes; 2944 2945bool os::pd_uncommit_memory(char* addr, size_t size) { 2946 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2947 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2948 return res != (uintptr_t) MAP_FAILED; 2949} 2950 2951static address get_stack_commited_bottom(address bottom, size_t size) { 2952 address nbot = bottom; 2953 address ntop = bottom + size; 2954 2955 size_t page_sz = os::vm_page_size(); 2956 unsigned pages = size / page_sz; 2957 2958 unsigned char vec[1]; 2959 unsigned imin = 1, imax = pages + 1, imid; 2960 int mincore_return_value = 0; 2961 2962 assert(imin <= imax, "Unexpected page size"); 2963 2964 while (imin < imax) { 2965 imid = (imax + imin) / 2; 2966 nbot = ntop - (imid * page_sz); 2967 2968 // Use a trick with mincore to check whether the page is mapped or not. 2969 // mincore sets vec to 1 if page resides in memory and to 0 if page 2970 // is swapped output but if page we are asking for is unmapped 2971 // it returns -1,ENOMEM 2972 mincore_return_value = mincore(nbot, page_sz, vec); 2973 2974 if (mincore_return_value == -1) { 2975 // Page is not mapped go up 2976 // to find first mapped page 2977 if (errno != EAGAIN) { 2978 assert(errno == ENOMEM, "Unexpected mincore errno"); 2979 imax = imid; 2980 } 2981 } else { 2982 // Page is mapped go down 2983 // to find first not mapped page 2984 imin = imid + 1; 2985 } 2986 } 2987 2988 nbot = nbot + page_sz; 2989 2990 // Adjust stack bottom one page up if last checked page is not mapped 2991 if (mincore_return_value == -1) { 2992 nbot = nbot + page_sz; 2993 } 2994 2995 return nbot; 2996} 2997 2998 2999// Linux uses a growable mapping for the stack, and if the mapping for 3000// the stack guard pages is not removed when we detach a thread the 3001// stack cannot grow beyond the pages where the stack guard was 3002// mapped. If at some point later in the process the stack expands to 3003// that point, the Linux kernel cannot expand the stack any further 3004// because the guard pages are in the way, and a segfault occurs. 3005// 3006// However, it's essential not to split the stack region by unmapping 3007// a region (leaving a hole) that's already part of the stack mapping, 3008// so if the stack mapping has already grown beyond the guard pages at 3009// the time we create them, we have to truncate the stack mapping. 3010// So, we need to know the extent of the stack mapping when 3011// create_stack_guard_pages() is called. 3012 3013// We only need this for stacks that are growable: at the time of 3014// writing thread stacks don't use growable mappings (i.e. those 3015// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3016// only applies to the main thread. 3017 3018// If the (growable) stack mapping already extends beyond the point 3019// where we're going to put our guard pages, truncate the mapping at 3020// that point by munmap()ping it. This ensures that when we later 3021// munmap() the guard pages we don't leave a hole in the stack 3022// mapping. This only affects the main/initial thread 3023 3024bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3025 if (os::Linux::is_initial_thread()) { 3026 // As we manually grow stack up to bottom inside create_attached_thread(), 3027 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3028 // we don't need to do anything special. 3029 // Check it first, before calling heavy function. 3030 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3031 unsigned char vec[1]; 3032 3033 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3034 // Fallback to slow path on all errors, including EAGAIN 3035 stack_extent = (uintptr_t) get_stack_commited_bottom( 3036 os::Linux::initial_thread_stack_bottom(), 3037 (size_t)addr - stack_extent); 3038 } 3039 3040 if (stack_extent < (uintptr_t)addr) { 3041 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3042 } 3043 } 3044 3045 return os::commit_memory(addr, size, !ExecMem); 3046} 3047 3048// If this is a growable mapping, remove the guard pages entirely by 3049// munmap()ping them. If not, just call uncommit_memory(). This only 3050// affects the main/initial thread, but guard against future OS changes 3051// It's safe to always unmap guard pages for initial thread because we 3052// always place it right after end of the mapped region 3053 3054bool os::remove_stack_guard_pages(char* addr, size_t size) { 3055 uintptr_t stack_extent, stack_base; 3056 3057 if (os::Linux::is_initial_thread()) { 3058 return ::munmap(addr, size) == 0; 3059 } 3060 3061 return os::uncommit_memory(addr, size); 3062} 3063 3064static address _highest_vm_reserved_address = NULL; 3065 3066// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3067// at 'requested_addr'. If there are existing memory mappings at the same 3068// location, however, they will be overwritten. If 'fixed' is false, 3069// 'requested_addr' is only treated as a hint, the return value may or 3070// may not start from the requested address. Unlike Linux mmap(), this 3071// function returns NULL to indicate failure. 3072static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3073 char * addr; 3074 int flags; 3075 3076 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3077 if (fixed) { 3078 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3079 flags |= MAP_FIXED; 3080 } 3081 3082 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3083 // touch an uncommitted page. Otherwise, the read/write might 3084 // succeed if we have enough swap space to back the physical page. 3085 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3086 flags, -1, 0); 3087 3088 if (addr != MAP_FAILED) { 3089 // anon_mmap() should only get called during VM initialization, 3090 // don't need lock (actually we can skip locking even it can be called 3091 // from multiple threads, because _highest_vm_reserved_address is just a 3092 // hint about the upper limit of non-stack memory regions.) 3093 if ((address)addr + bytes > _highest_vm_reserved_address) { 3094 _highest_vm_reserved_address = (address)addr + bytes; 3095 } 3096 } 3097 3098 return addr == MAP_FAILED ? NULL : addr; 3099} 3100 3101// Don't update _highest_vm_reserved_address, because there might be memory 3102// regions above addr + size. If so, releasing a memory region only creates 3103// a hole in the address space, it doesn't help prevent heap-stack collision. 3104// 3105static int anon_munmap(char * addr, size_t size) { 3106 return ::munmap(addr, size) == 0; 3107} 3108 3109char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3110 size_t alignment_hint) { 3111 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3112} 3113 3114bool os::pd_release_memory(char* addr, size_t size) { 3115 return anon_munmap(addr, size); 3116} 3117 3118static address highest_vm_reserved_address() { 3119 return _highest_vm_reserved_address; 3120} 3121 3122static bool linux_mprotect(char* addr, size_t size, int prot) { 3123 // Linux wants the mprotect address argument to be page aligned. 3124 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3125 3126 // According to SUSv3, mprotect() should only be used with mappings 3127 // established by mmap(), and mmap() always maps whole pages. Unaligned 3128 // 'addr' likely indicates problem in the VM (e.g. trying to change 3129 // protection of malloc'ed or statically allocated memory). Check the 3130 // caller if you hit this assert. 3131 assert(addr == bottom, "sanity check"); 3132 3133 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3134 return ::mprotect(bottom, size, prot) == 0; 3135} 3136 3137// Set protections specified 3138bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3139 bool is_committed) { 3140 unsigned int p = 0; 3141 switch (prot) { 3142 case MEM_PROT_NONE: p = PROT_NONE; break; 3143 case MEM_PROT_READ: p = PROT_READ; break; 3144 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3145 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3146 default: 3147 ShouldNotReachHere(); 3148 } 3149 // is_committed is unused. 3150 return linux_mprotect(addr, bytes, p); 3151} 3152 3153bool os::guard_memory(char* addr, size_t size) { 3154 return linux_mprotect(addr, size, PROT_NONE); 3155} 3156 3157bool os::unguard_memory(char* addr, size_t size) { 3158 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3159} 3160 3161bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3162 size_t page_size) { 3163 bool result = false; 3164 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3165 MAP_ANONYMOUS|MAP_PRIVATE, 3166 -1, 0); 3167 if (p != MAP_FAILED) { 3168 void *aligned_p = align_ptr_up(p, page_size); 3169 3170 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3171 3172 munmap(p, page_size * 2); 3173 } 3174 3175 if (warn && !result) { 3176 warning("TransparentHugePages is not supported by the operating system."); 3177 } 3178 3179 return result; 3180} 3181 3182bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3183 bool result = false; 3184 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3185 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3186 -1, 0); 3187 3188 if (p != MAP_FAILED) { 3189 // We don't know if this really is a huge page or not. 3190 FILE *fp = fopen("/proc/self/maps", "r"); 3191 if (fp) { 3192 while (!feof(fp)) { 3193 char chars[257]; 3194 long x = 0; 3195 if (fgets(chars, sizeof(chars), fp)) { 3196 if (sscanf(chars, "%lx-%*x", &x) == 1 3197 && x == (long)p) { 3198 if (strstr (chars, "hugepage")) { 3199 result = true; 3200 break; 3201 } 3202 } 3203 } 3204 } 3205 fclose(fp); 3206 } 3207 munmap(p, page_size); 3208 } 3209 3210 if (warn && !result) { 3211 warning("HugeTLBFS is not supported by the operating system."); 3212 } 3213 3214 return result; 3215} 3216 3217// Set the coredump_filter bits to include largepages in core dump (bit 6) 3218// 3219// From the coredump_filter documentation: 3220// 3221// - (bit 0) anonymous private memory 3222// - (bit 1) anonymous shared memory 3223// - (bit 2) file-backed private memory 3224// - (bit 3) file-backed shared memory 3225// - (bit 4) ELF header pages in file-backed private memory areas (it is 3226// effective only if the bit 2 is cleared) 3227// - (bit 5) hugetlb private memory 3228// - (bit 6) hugetlb shared memory 3229// 3230static void set_coredump_filter(void) { 3231 FILE *f; 3232 long cdm; 3233 3234 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3235 return; 3236 } 3237 3238 if (fscanf(f, "%lx", &cdm) != 1) { 3239 fclose(f); 3240 return; 3241 } 3242 3243 rewind(f); 3244 3245 if ((cdm & LARGEPAGES_BIT) == 0) { 3246 cdm |= LARGEPAGES_BIT; 3247 fprintf(f, "%#lx", cdm); 3248 } 3249 3250 fclose(f); 3251} 3252 3253// Large page support 3254 3255static size_t _large_page_size = 0; 3256 3257size_t os::Linux::find_large_page_size() { 3258 size_t large_page_size = 0; 3259 3260 // large_page_size on Linux is used to round up heap size. x86 uses either 3261 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3262 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3263 // page as large as 256M. 3264 // 3265 // Here we try to figure out page size by parsing /proc/meminfo and looking 3266 // for a line with the following format: 3267 // Hugepagesize: 2048 kB 3268 // 3269 // If we can't determine the value (e.g. /proc is not mounted, or the text 3270 // format has been changed), we'll use the largest page size supported by 3271 // the processor. 3272 3273#ifndef ZERO 3274 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3275 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3276#endif // ZERO 3277 3278 FILE *fp = fopen("/proc/meminfo", "r"); 3279 if (fp) { 3280 while (!feof(fp)) { 3281 int x = 0; 3282 char buf[16]; 3283 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3284 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3285 large_page_size = x * K; 3286 break; 3287 } 3288 } else { 3289 // skip to next line 3290 for (;;) { 3291 int ch = fgetc(fp); 3292 if (ch == EOF || ch == (int)'\n') break; 3293 } 3294 } 3295 } 3296 fclose(fp); 3297 } 3298 3299 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3300 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3301 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3302 proper_unit_for_byte_size(large_page_size)); 3303 } 3304 3305 return large_page_size; 3306} 3307 3308size_t os::Linux::setup_large_page_size() { 3309 _large_page_size = Linux::find_large_page_size(); 3310 const size_t default_page_size = (size_t)Linux::page_size(); 3311 if (_large_page_size > default_page_size) { 3312 _page_sizes[0] = _large_page_size; 3313 _page_sizes[1] = default_page_size; 3314 _page_sizes[2] = 0; 3315 } 3316 3317 return _large_page_size; 3318} 3319 3320bool os::Linux::setup_large_page_type(size_t page_size) { 3321 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3322 FLAG_IS_DEFAULT(UseSHM) && 3323 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3324 3325 // The type of large pages has not been specified by the user. 3326 3327 // Try UseHugeTLBFS and then UseSHM. 3328 UseHugeTLBFS = UseSHM = true; 3329 3330 // Don't try UseTransparentHugePages since there are known 3331 // performance issues with it turned on. This might change in the future. 3332 UseTransparentHugePages = false; 3333 } 3334 3335 if (UseTransparentHugePages) { 3336 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3337 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3338 UseHugeTLBFS = false; 3339 UseSHM = false; 3340 return true; 3341 } 3342 UseTransparentHugePages = false; 3343 } 3344 3345 if (UseHugeTLBFS) { 3346 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3347 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3348 UseSHM = false; 3349 return true; 3350 } 3351 UseHugeTLBFS = false; 3352 } 3353 3354 return UseSHM; 3355} 3356 3357void os::large_page_init() { 3358 if (!UseLargePages && 3359 !UseTransparentHugePages && 3360 !UseHugeTLBFS && 3361 !UseSHM) { 3362 // Not using large pages. 3363 return; 3364 } 3365 3366 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3367 // The user explicitly turned off large pages. 3368 // Ignore the rest of the large pages flags. 3369 UseTransparentHugePages = false; 3370 UseHugeTLBFS = false; 3371 UseSHM = false; 3372 return; 3373 } 3374 3375 size_t large_page_size = Linux::setup_large_page_size(); 3376 UseLargePages = Linux::setup_large_page_type(large_page_size); 3377 3378 set_coredump_filter(); 3379} 3380 3381#ifndef SHM_HUGETLB 3382 #define SHM_HUGETLB 04000 3383#endif 3384 3385char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3386 char* req_addr, bool exec) { 3387 // "exec" is passed in but not used. Creating the shared image for 3388 // the code cache doesn't have an SHM_X executable permission to check. 3389 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3390 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3391 3392 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3393 return NULL; // Fallback to small pages. 3394 } 3395 3396 key_t key = IPC_PRIVATE; 3397 char *addr; 3398 3399 bool warn_on_failure = UseLargePages && 3400 (!FLAG_IS_DEFAULT(UseLargePages) || 3401 !FLAG_IS_DEFAULT(UseSHM) || 3402 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3403 char msg[128]; 3404 3405 // Create a large shared memory region to attach to based on size. 3406 // Currently, size is the total size of the heap 3407 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3408 if (shmid == -1) { 3409 // Possible reasons for shmget failure: 3410 // 1. shmmax is too small for Java heap. 3411 // > check shmmax value: cat /proc/sys/kernel/shmmax 3412 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3413 // 2. not enough large page memory. 3414 // > check available large pages: cat /proc/meminfo 3415 // > increase amount of large pages: 3416 // echo new_value > /proc/sys/vm/nr_hugepages 3417 // Note 1: different Linux may use different name for this property, 3418 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3419 // Note 2: it's possible there's enough physical memory available but 3420 // they are so fragmented after a long run that they can't 3421 // coalesce into large pages. Try to reserve large pages when 3422 // the system is still "fresh". 3423 if (warn_on_failure) { 3424 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3425 warning("%s", msg); 3426 } 3427 return NULL; 3428 } 3429 3430 // attach to the region 3431 addr = (char*)shmat(shmid, req_addr, 0); 3432 int err = errno; 3433 3434 // Remove shmid. If shmat() is successful, the actual shared memory segment 3435 // will be deleted when it's detached by shmdt() or when the process 3436 // terminates. If shmat() is not successful this will remove the shared 3437 // segment immediately. 3438 shmctl(shmid, IPC_RMID, NULL); 3439 3440 if ((intptr_t)addr == -1) { 3441 if (warn_on_failure) { 3442 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3443 warning("%s", msg); 3444 } 3445 return NULL; 3446 } 3447 3448 return addr; 3449} 3450 3451static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3452 int error) { 3453 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3454 3455 bool warn_on_failure = UseLargePages && 3456 (!FLAG_IS_DEFAULT(UseLargePages) || 3457 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3458 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3459 3460 if (warn_on_failure) { 3461 char msg[128]; 3462 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3463 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3464 warning("%s", msg); 3465 } 3466} 3467 3468char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3469 char* req_addr, 3470 bool exec) { 3471 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3472 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3473 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3474 3475 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3476 char* addr = (char*)::mmap(req_addr, bytes, prot, 3477 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3478 -1, 0); 3479 3480 if (addr == MAP_FAILED) { 3481 warn_on_large_pages_failure(req_addr, bytes, errno); 3482 return NULL; 3483 } 3484 3485 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3486 3487 return addr; 3488} 3489 3490char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3491 size_t alignment, 3492 char* req_addr, 3493 bool exec) { 3494 size_t large_page_size = os::large_page_size(); 3495 3496 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3497 3498 // Allocate small pages. 3499 3500 char* start; 3501 if (req_addr != NULL) { 3502 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3503 assert(is_size_aligned(bytes, alignment), "Must be"); 3504 start = os::reserve_memory(bytes, req_addr); 3505 assert(start == NULL || start == req_addr, "Must be"); 3506 } else { 3507 start = os::reserve_memory_aligned(bytes, alignment); 3508 } 3509 3510 if (start == NULL) { 3511 return NULL; 3512 } 3513 3514 assert(is_ptr_aligned(start, alignment), "Must be"); 3515 3516 if (MemTracker::tracking_level() > NMT_minimal) { 3517 // os::reserve_memory_special will record this memory area. 3518 // Need to release it here to prevent overlapping reservations. 3519 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3520 tkr.record((address)start, bytes); 3521 } 3522 3523 char* end = start + bytes; 3524 3525 // Find the regions of the allocated chunk that can be promoted to large pages. 3526 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3527 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3528 3529 size_t lp_bytes = lp_end - lp_start; 3530 3531 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3532 3533 if (lp_bytes == 0) { 3534 // The mapped region doesn't even span the start and the end of a large page. 3535 // Fall back to allocate a non-special area. 3536 ::munmap(start, end - start); 3537 return NULL; 3538 } 3539 3540 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3541 3542 3543 void* result; 3544 3545 if (start != lp_start) { 3546 result = ::mmap(start, lp_start - start, prot, 3547 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3548 -1, 0); 3549 if (result == MAP_FAILED) { 3550 ::munmap(lp_start, end - lp_start); 3551 return NULL; 3552 } 3553 } 3554 3555 result = ::mmap(lp_start, lp_bytes, prot, 3556 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3557 -1, 0); 3558 if (result == MAP_FAILED) { 3559 warn_on_large_pages_failure(req_addr, bytes, errno); 3560 // If the mmap above fails, the large pages region will be unmapped and we 3561 // have regions before and after with small pages. Release these regions. 3562 // 3563 // | mapped | unmapped | mapped | 3564 // ^ ^ ^ ^ 3565 // start lp_start lp_end end 3566 // 3567 ::munmap(start, lp_start - start); 3568 ::munmap(lp_end, end - lp_end); 3569 return NULL; 3570 } 3571 3572 if (lp_end != end) { 3573 result = ::mmap(lp_end, end - lp_end, prot, 3574 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3575 -1, 0); 3576 if (result == MAP_FAILED) { 3577 ::munmap(start, lp_end - start); 3578 return NULL; 3579 } 3580 } 3581 3582 return start; 3583} 3584 3585char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3586 size_t alignment, 3587 char* req_addr, 3588 bool exec) { 3589 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3590 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3591 assert(is_power_of_2(alignment), "Must be"); 3592 assert(is_power_of_2(os::large_page_size()), "Must be"); 3593 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3594 3595 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3596 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3597 } else { 3598 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3599 } 3600} 3601 3602char* os::reserve_memory_special(size_t bytes, size_t alignment, 3603 char* req_addr, bool exec) { 3604 assert(UseLargePages, "only for large pages"); 3605 3606 char* addr; 3607 if (UseSHM) { 3608 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3609 } else { 3610 assert(UseHugeTLBFS, "must be"); 3611 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3612 } 3613 3614 if (addr != NULL) { 3615 if (UseNUMAInterleaving) { 3616 numa_make_global(addr, bytes); 3617 } 3618 3619 // The memory is committed 3620 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3621 } 3622 3623 return addr; 3624} 3625 3626bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3627 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3628 return shmdt(base) == 0; 3629} 3630 3631bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3632 return pd_release_memory(base, bytes); 3633} 3634 3635bool os::release_memory_special(char* base, size_t bytes) { 3636 bool res; 3637 if (MemTracker::tracking_level() > NMT_minimal) { 3638 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3639 res = os::Linux::release_memory_special_impl(base, bytes); 3640 if (res) { 3641 tkr.record((address)base, bytes); 3642 } 3643 3644 } else { 3645 res = os::Linux::release_memory_special_impl(base, bytes); 3646 } 3647 return res; 3648} 3649 3650bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3651 assert(UseLargePages, "only for large pages"); 3652 bool res; 3653 3654 if (UseSHM) { 3655 res = os::Linux::release_memory_special_shm(base, bytes); 3656 } else { 3657 assert(UseHugeTLBFS, "must be"); 3658 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3659 } 3660 return res; 3661} 3662 3663size_t os::large_page_size() { 3664 return _large_page_size; 3665} 3666 3667// With SysV SHM the entire memory region must be allocated as shared 3668// memory. 3669// HugeTLBFS allows application to commit large page memory on demand. 3670// However, when committing memory with HugeTLBFS fails, the region 3671// that was supposed to be committed will lose the old reservation 3672// and allow other threads to steal that memory region. Because of this 3673// behavior we can't commit HugeTLBFS memory. 3674bool os::can_commit_large_page_memory() { 3675 return UseTransparentHugePages; 3676} 3677 3678bool os::can_execute_large_page_memory() { 3679 return UseTransparentHugePages || UseHugeTLBFS; 3680} 3681 3682// Reserve memory at an arbitrary address, only if that area is 3683// available (and not reserved for something else). 3684 3685char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3686 const int max_tries = 10; 3687 char* base[max_tries]; 3688 size_t size[max_tries]; 3689 const size_t gap = 0x000000; 3690 3691 // Assert only that the size is a multiple of the page size, since 3692 // that's all that mmap requires, and since that's all we really know 3693 // about at this low abstraction level. If we need higher alignment, 3694 // we can either pass an alignment to this method or verify alignment 3695 // in one of the methods further up the call chain. See bug 5044738. 3696 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3697 3698 // Repeatedly allocate blocks until the block is allocated at the 3699 // right spot. Give up after max_tries. Note that reserve_memory() will 3700 // automatically update _highest_vm_reserved_address if the call is 3701 // successful. The variable tracks the highest memory address every reserved 3702 // by JVM. It is used to detect heap-stack collision if running with 3703 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3704 // space than needed, it could confuse the collision detecting code. To 3705 // solve the problem, save current _highest_vm_reserved_address and 3706 // calculate the correct value before return. 3707 address old_highest = _highest_vm_reserved_address; 3708 3709 // Linux mmap allows caller to pass an address as hint; give it a try first, 3710 // if kernel honors the hint then we can return immediately. 3711 char * addr = anon_mmap(requested_addr, bytes, false); 3712 if (addr == requested_addr) { 3713 return requested_addr; 3714 } 3715 3716 if (addr != NULL) { 3717 // mmap() is successful but it fails to reserve at the requested address 3718 anon_munmap(addr, bytes); 3719 } 3720 3721 int i; 3722 for (i = 0; i < max_tries; ++i) { 3723 base[i] = reserve_memory(bytes); 3724 3725 if (base[i] != NULL) { 3726 // Is this the block we wanted? 3727 if (base[i] == requested_addr) { 3728 size[i] = bytes; 3729 break; 3730 } 3731 3732 // Does this overlap the block we wanted? Give back the overlapped 3733 // parts and try again. 3734 3735 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3736 if (top_overlap >= 0 && top_overlap < bytes) { 3737 unmap_memory(base[i], top_overlap); 3738 base[i] += top_overlap; 3739 size[i] = bytes - top_overlap; 3740 } else { 3741 size_t bottom_overlap = base[i] + bytes - requested_addr; 3742 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3743 unmap_memory(requested_addr, bottom_overlap); 3744 size[i] = bytes - bottom_overlap; 3745 } else { 3746 size[i] = bytes; 3747 } 3748 } 3749 } 3750 } 3751 3752 // Give back the unused reserved pieces. 3753 3754 for (int j = 0; j < i; ++j) { 3755 if (base[j] != NULL) { 3756 unmap_memory(base[j], size[j]); 3757 } 3758 } 3759 3760 if (i < max_tries) { 3761 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3762 return requested_addr; 3763 } else { 3764 _highest_vm_reserved_address = old_highest; 3765 return NULL; 3766 } 3767} 3768 3769size_t os::read(int fd, void *buf, unsigned int nBytes) { 3770 return ::read(fd, buf, nBytes); 3771} 3772 3773// Short sleep, direct OS call. 3774// 3775// Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3776// sched_yield(2) will actually give up the CPU: 3777// 3778// * Alone on this pariticular CPU, keeps running. 3779// * Before the introduction of "skip_buddy" with "compat_yield" disabled 3780// (pre 2.6.39). 3781// 3782// So calling this with 0 is an alternative. 3783// 3784void os::naked_short_sleep(jlong ms) { 3785 struct timespec req; 3786 3787 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3788 req.tv_sec = 0; 3789 if (ms > 0) { 3790 req.tv_nsec = (ms % 1000) * 1000000; 3791 } else { 3792 req.tv_nsec = 1; 3793 } 3794 3795 nanosleep(&req, NULL); 3796 3797 return; 3798} 3799 3800// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3801void os::infinite_sleep() { 3802 while (true) { // sleep forever ... 3803 ::sleep(100); // ... 100 seconds at a time 3804 } 3805} 3806 3807// Used to convert frequent JVM_Yield() to nops 3808bool os::dont_yield() { 3809 return DontYieldALot; 3810} 3811 3812void os::naked_yield() { 3813 sched_yield(); 3814} 3815 3816//////////////////////////////////////////////////////////////////////////////// 3817// thread priority support 3818 3819// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3820// only supports dynamic priority, static priority must be zero. For real-time 3821// applications, Linux supports SCHED_RR which allows static priority (1-99). 3822// However, for large multi-threaded applications, SCHED_RR is not only slower 3823// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3824// of 5 runs - Sep 2005). 3825// 3826// The following code actually changes the niceness of kernel-thread/LWP. It 3827// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3828// not the entire user process, and user level threads are 1:1 mapped to kernel 3829// threads. It has always been the case, but could change in the future. For 3830// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3831// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3832 3833int os::java_to_os_priority[CriticalPriority + 1] = { 3834 19, // 0 Entry should never be used 3835 3836 4, // 1 MinPriority 3837 3, // 2 3838 2, // 3 3839 3840 1, // 4 3841 0, // 5 NormPriority 3842 -1, // 6 3843 3844 -2, // 7 3845 -3, // 8 3846 -4, // 9 NearMaxPriority 3847 3848 -5, // 10 MaxPriority 3849 3850 -5 // 11 CriticalPriority 3851}; 3852 3853static int prio_init() { 3854 if (ThreadPriorityPolicy == 1) { 3855 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3856 // if effective uid is not root. Perhaps, a more elegant way of doing 3857 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3858 if (geteuid() != 0) { 3859 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3860 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3861 } 3862 ThreadPriorityPolicy = 0; 3863 } 3864 } 3865 if (UseCriticalJavaThreadPriority) { 3866 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3867 } 3868 return 0; 3869} 3870 3871OSReturn os::set_native_priority(Thread* thread, int newpri) { 3872 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 3873 3874 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3875 return (ret == 0) ? OS_OK : OS_ERR; 3876} 3877 3878OSReturn os::get_native_priority(const Thread* const thread, 3879 int *priority_ptr) { 3880 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 3881 *priority_ptr = java_to_os_priority[NormPriority]; 3882 return OS_OK; 3883 } 3884 3885 errno = 0; 3886 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3887 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3888} 3889 3890// Hint to the underlying OS that a task switch would not be good. 3891// Void return because it's a hint and can fail. 3892void os::hint_no_preempt() {} 3893 3894//////////////////////////////////////////////////////////////////////////////// 3895// suspend/resume support 3896 3897// the low-level signal-based suspend/resume support is a remnant from the 3898// old VM-suspension that used to be for java-suspension, safepoints etc, 3899// within hotspot. Now there is a single use-case for this: 3900// - calling get_thread_pc() on the VMThread by the flat-profiler task 3901// that runs in the watcher thread. 3902// The remaining code is greatly simplified from the more general suspension 3903// code that used to be used. 3904// 3905// The protocol is quite simple: 3906// - suspend: 3907// - sends a signal to the target thread 3908// - polls the suspend state of the osthread using a yield loop 3909// - target thread signal handler (SR_handler) sets suspend state 3910// and blocks in sigsuspend until continued 3911// - resume: 3912// - sets target osthread state to continue 3913// - sends signal to end the sigsuspend loop in the SR_handler 3914// 3915// Note that the SR_lock plays no role in this suspend/resume protocol. 3916 3917static void resume_clear_context(OSThread *osthread) { 3918 osthread->set_ucontext(NULL); 3919 osthread->set_siginfo(NULL); 3920} 3921 3922static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 3923 ucontext_t* context) { 3924 osthread->set_ucontext(context); 3925 osthread->set_siginfo(siginfo); 3926} 3927 3928// Handler function invoked when a thread's execution is suspended or 3929// resumed. We have to be careful that only async-safe functions are 3930// called here (Note: most pthread functions are not async safe and 3931// should be avoided.) 3932// 3933// Note: sigwait() is a more natural fit than sigsuspend() from an 3934// interface point of view, but sigwait() prevents the signal hander 3935// from being run. libpthread would get very confused by not having 3936// its signal handlers run and prevents sigwait()'s use with the 3937// mutex granting granting signal. 3938// 3939// Currently only ever called on the VMThread and JavaThreads (PC sampling) 3940// 3941static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3942 // Save and restore errno to avoid confusing native code with EINTR 3943 // after sigsuspend. 3944 int old_errno = errno; 3945 3946 Thread* thread = Thread::current(); 3947 OSThread* osthread = thread->osthread(); 3948 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3949 3950 os::SuspendResume::State current = osthread->sr.state(); 3951 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3952 suspend_save_context(osthread, siginfo, context); 3953 3954 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 3955 os::SuspendResume::State state = osthread->sr.suspended(); 3956 if (state == os::SuspendResume::SR_SUSPENDED) { 3957 sigset_t suspend_set; // signals for sigsuspend() 3958 3959 // get current set of blocked signals and unblock resume signal 3960 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3961 sigdelset(&suspend_set, SR_signum); 3962 3963 sr_semaphore.signal(); 3964 // wait here until we are resumed 3965 while (1) { 3966 sigsuspend(&suspend_set); 3967 3968 os::SuspendResume::State result = osthread->sr.running(); 3969 if (result == os::SuspendResume::SR_RUNNING) { 3970 sr_semaphore.signal(); 3971 break; 3972 } 3973 } 3974 3975 } else if (state == os::SuspendResume::SR_RUNNING) { 3976 // request was cancelled, continue 3977 } else { 3978 ShouldNotReachHere(); 3979 } 3980 3981 resume_clear_context(osthread); 3982 } else if (current == os::SuspendResume::SR_RUNNING) { 3983 // request was cancelled, continue 3984 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 3985 // ignore 3986 } else { 3987 // ignore 3988 } 3989 3990 errno = old_errno; 3991} 3992 3993 3994static int SR_initialize() { 3995 struct sigaction act; 3996 char *s; 3997 // Get signal number to use for suspend/resume 3998 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3999 int sig = ::strtol(s, 0, 10); 4000 if (sig > 0 || sig < _NSIG) { 4001 SR_signum = sig; 4002 } 4003 } 4004 4005 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4006 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4007 4008 sigemptyset(&SR_sigset); 4009 sigaddset(&SR_sigset, SR_signum); 4010 4011 // Set up signal handler for suspend/resume 4012 act.sa_flags = SA_RESTART|SA_SIGINFO; 4013 act.sa_handler = (void (*)(int)) SR_handler; 4014 4015 // SR_signum is blocked by default. 4016 // 4528190 - We also need to block pthread restart signal (32 on all 4017 // supported Linux platforms). Note that LinuxThreads need to block 4018 // this signal for all threads to work properly. So we don't have 4019 // to use hard-coded signal number when setting up the mask. 4020 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4021 4022 if (sigaction(SR_signum, &act, 0) == -1) { 4023 return -1; 4024 } 4025 4026 // Save signal flag 4027 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4028 return 0; 4029} 4030 4031static int sr_notify(OSThread* osthread) { 4032 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4033 assert_status(status == 0, status, "pthread_kill"); 4034 return status; 4035} 4036 4037// "Randomly" selected value for how long we want to spin 4038// before bailing out on suspending a thread, also how often 4039// we send a signal to a thread we want to resume 4040static const int RANDOMLY_LARGE_INTEGER = 1000000; 4041static const int RANDOMLY_LARGE_INTEGER2 = 100; 4042 4043// returns true on success and false on error - really an error is fatal 4044// but this seems the normal response to library errors 4045static bool do_suspend(OSThread* osthread) { 4046 assert(osthread->sr.is_running(), "thread should be running"); 4047 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4048 4049 // mark as suspended and send signal 4050 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4051 // failed to switch, state wasn't running? 4052 ShouldNotReachHere(); 4053 return false; 4054 } 4055 4056 if (sr_notify(osthread) != 0) { 4057 ShouldNotReachHere(); 4058 } 4059 4060 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4061 while (true) { 4062 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4063 break; 4064 } else { 4065 // timeout 4066 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4067 if (cancelled == os::SuspendResume::SR_RUNNING) { 4068 return false; 4069 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4070 // make sure that we consume the signal on the semaphore as well 4071 sr_semaphore.wait(); 4072 break; 4073 } else { 4074 ShouldNotReachHere(); 4075 return false; 4076 } 4077 } 4078 } 4079 4080 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4081 return true; 4082} 4083 4084static void do_resume(OSThread* osthread) { 4085 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4086 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4087 4088 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4089 // failed to switch to WAKEUP_REQUEST 4090 ShouldNotReachHere(); 4091 return; 4092 } 4093 4094 while (true) { 4095 if (sr_notify(osthread) == 0) { 4096 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4097 if (osthread->sr.is_running()) { 4098 return; 4099 } 4100 } 4101 } else { 4102 ShouldNotReachHere(); 4103 } 4104 } 4105 4106 guarantee(osthread->sr.is_running(), "Must be running!"); 4107} 4108 4109/////////////////////////////////////////////////////////////////////////////////// 4110// signal handling (except suspend/resume) 4111 4112// This routine may be used by user applications as a "hook" to catch signals. 4113// The user-defined signal handler must pass unrecognized signals to this 4114// routine, and if it returns true (non-zero), then the signal handler must 4115// return immediately. If the flag "abort_if_unrecognized" is true, then this 4116// routine will never retun false (zero), but instead will execute a VM panic 4117// routine kill the process. 4118// 4119// If this routine returns false, it is OK to call it again. This allows 4120// the user-defined signal handler to perform checks either before or after 4121// the VM performs its own checks. Naturally, the user code would be making 4122// a serious error if it tried to handle an exception (such as a null check 4123// or breakpoint) that the VM was generating for its own correct operation. 4124// 4125// This routine may recognize any of the following kinds of signals: 4126// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4127// It should be consulted by handlers for any of those signals. 4128// 4129// The caller of this routine must pass in the three arguments supplied 4130// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4131// field of the structure passed to sigaction(). This routine assumes that 4132// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4133// 4134// Note that the VM will print warnings if it detects conflicting signal 4135// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4136// 4137extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4138 siginfo_t* siginfo, 4139 void* ucontext, 4140 int abort_if_unrecognized); 4141 4142void signalHandler(int sig, siginfo_t* info, void* uc) { 4143 assert(info != NULL && uc != NULL, "it must be old kernel"); 4144 int orig_errno = errno; // Preserve errno value over signal handler. 4145 JVM_handle_linux_signal(sig, info, uc, true); 4146 errno = orig_errno; 4147} 4148 4149 4150// This boolean allows users to forward their own non-matching signals 4151// to JVM_handle_linux_signal, harmlessly. 4152bool os::Linux::signal_handlers_are_installed = false; 4153 4154// For signal-chaining 4155struct sigaction os::Linux::sigact[MAXSIGNUM]; 4156unsigned int os::Linux::sigs = 0; 4157bool os::Linux::libjsig_is_loaded = false; 4158typedef struct sigaction *(*get_signal_t)(int); 4159get_signal_t os::Linux::get_signal_action = NULL; 4160 4161struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4162 struct sigaction *actp = NULL; 4163 4164 if (libjsig_is_loaded) { 4165 // Retrieve the old signal handler from libjsig 4166 actp = (*get_signal_action)(sig); 4167 } 4168 if (actp == NULL) { 4169 // Retrieve the preinstalled signal handler from jvm 4170 actp = get_preinstalled_handler(sig); 4171 } 4172 4173 return actp; 4174} 4175 4176static bool call_chained_handler(struct sigaction *actp, int sig, 4177 siginfo_t *siginfo, void *context) { 4178 // Call the old signal handler 4179 if (actp->sa_handler == SIG_DFL) { 4180 // It's more reasonable to let jvm treat it as an unexpected exception 4181 // instead of taking the default action. 4182 return false; 4183 } else if (actp->sa_handler != SIG_IGN) { 4184 if ((actp->sa_flags & SA_NODEFER) == 0) { 4185 // automaticlly block the signal 4186 sigaddset(&(actp->sa_mask), sig); 4187 } 4188 4189 sa_handler_t hand; 4190 sa_sigaction_t sa; 4191 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4192 // retrieve the chained handler 4193 if (siginfo_flag_set) { 4194 sa = actp->sa_sigaction; 4195 } else { 4196 hand = actp->sa_handler; 4197 } 4198 4199 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4200 actp->sa_handler = SIG_DFL; 4201 } 4202 4203 // try to honor the signal mask 4204 sigset_t oset; 4205 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4206 4207 // call into the chained handler 4208 if (siginfo_flag_set) { 4209 (*sa)(sig, siginfo, context); 4210 } else { 4211 (*hand)(sig); 4212 } 4213 4214 // restore the signal mask 4215 pthread_sigmask(SIG_SETMASK, &oset, 0); 4216 } 4217 // Tell jvm's signal handler the signal is taken care of. 4218 return true; 4219} 4220 4221bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4222 bool chained = false; 4223 // signal-chaining 4224 if (UseSignalChaining) { 4225 struct sigaction *actp = get_chained_signal_action(sig); 4226 if (actp != NULL) { 4227 chained = call_chained_handler(actp, sig, siginfo, context); 4228 } 4229 } 4230 return chained; 4231} 4232 4233struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4234 if ((((unsigned int)1 << sig) & sigs) != 0) { 4235 return &sigact[sig]; 4236 } 4237 return NULL; 4238} 4239 4240void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4241 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4242 sigact[sig] = oldAct; 4243 sigs |= (unsigned int)1 << sig; 4244} 4245 4246// for diagnostic 4247int os::Linux::sigflags[MAXSIGNUM]; 4248 4249int os::Linux::get_our_sigflags(int sig) { 4250 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4251 return sigflags[sig]; 4252} 4253 4254void os::Linux::set_our_sigflags(int sig, int flags) { 4255 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4256 sigflags[sig] = flags; 4257} 4258 4259void os::Linux::set_signal_handler(int sig, bool set_installed) { 4260 // Check for overwrite. 4261 struct sigaction oldAct; 4262 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4263 4264 void* oldhand = oldAct.sa_sigaction 4265 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4266 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4267 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4268 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4269 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4270 if (AllowUserSignalHandlers || !set_installed) { 4271 // Do not overwrite; user takes responsibility to forward to us. 4272 return; 4273 } else if (UseSignalChaining) { 4274 // save the old handler in jvm 4275 save_preinstalled_handler(sig, oldAct); 4276 // libjsig also interposes the sigaction() call below and saves the 4277 // old sigaction on it own. 4278 } else { 4279 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4280 "%#lx for signal %d.", (long)oldhand, sig)); 4281 } 4282 } 4283 4284 struct sigaction sigAct; 4285 sigfillset(&(sigAct.sa_mask)); 4286 sigAct.sa_handler = SIG_DFL; 4287 if (!set_installed) { 4288 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4289 } else { 4290 sigAct.sa_sigaction = signalHandler; 4291 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4292 } 4293 // Save flags, which are set by ours 4294 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4295 sigflags[sig] = sigAct.sa_flags; 4296 4297 int ret = sigaction(sig, &sigAct, &oldAct); 4298 assert(ret == 0, "check"); 4299 4300 void* oldhand2 = oldAct.sa_sigaction 4301 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4302 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4303 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4304} 4305 4306// install signal handlers for signals that HotSpot needs to 4307// handle in order to support Java-level exception handling. 4308 4309void os::Linux::install_signal_handlers() { 4310 if (!signal_handlers_are_installed) { 4311 signal_handlers_are_installed = true; 4312 4313 // signal-chaining 4314 typedef void (*signal_setting_t)(); 4315 signal_setting_t begin_signal_setting = NULL; 4316 signal_setting_t end_signal_setting = NULL; 4317 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4318 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4319 if (begin_signal_setting != NULL) { 4320 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4321 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4322 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4323 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4324 libjsig_is_loaded = true; 4325 assert(UseSignalChaining, "should enable signal-chaining"); 4326 } 4327 if (libjsig_is_loaded) { 4328 // Tell libjsig jvm is setting signal handlers 4329 (*begin_signal_setting)(); 4330 } 4331 4332 set_signal_handler(SIGSEGV, true); 4333 set_signal_handler(SIGPIPE, true); 4334 set_signal_handler(SIGBUS, true); 4335 set_signal_handler(SIGILL, true); 4336 set_signal_handler(SIGFPE, true); 4337#if defined(PPC64) 4338 set_signal_handler(SIGTRAP, true); 4339#endif 4340 set_signal_handler(SIGXFSZ, true); 4341 4342 if (libjsig_is_loaded) { 4343 // Tell libjsig jvm finishes setting signal handlers 4344 (*end_signal_setting)(); 4345 } 4346 4347 // We don't activate signal checker if libjsig is in place, we trust ourselves 4348 // and if UserSignalHandler is installed all bets are off. 4349 // Log that signal checking is off only if -verbose:jni is specified. 4350 if (CheckJNICalls) { 4351 if (libjsig_is_loaded) { 4352 if (PrintJNIResolving) { 4353 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4354 } 4355 check_signals = false; 4356 } 4357 if (AllowUserSignalHandlers) { 4358 if (PrintJNIResolving) { 4359 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4360 } 4361 check_signals = false; 4362 } 4363 } 4364 } 4365} 4366 4367// This is the fastest way to get thread cpu time on Linux. 4368// Returns cpu time (user+sys) for any thread, not only for current. 4369// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4370// It might work on 2.6.10+ with a special kernel/glibc patch. 4371// For reference, please, see IEEE Std 1003.1-2004: 4372// http://www.unix.org/single_unix_specification 4373 4374jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4375 struct timespec tp; 4376 int rc = os::Linux::clock_gettime(clockid, &tp); 4377 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4378 4379 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4380} 4381 4382///// 4383// glibc on Linux platform uses non-documented flag 4384// to indicate, that some special sort of signal 4385// trampoline is used. 4386// We will never set this flag, and we should 4387// ignore this flag in our diagnostic 4388#ifdef SIGNIFICANT_SIGNAL_MASK 4389 #undef SIGNIFICANT_SIGNAL_MASK 4390#endif 4391#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4392 4393static const char* get_signal_handler_name(address handler, 4394 char* buf, int buflen) { 4395 int offset; 4396 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4397 if (found) { 4398 // skip directory names 4399 const char *p1, *p2; 4400 p1 = buf; 4401 size_t len = strlen(os::file_separator()); 4402 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4403 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4404 } else { 4405 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4406 } 4407 return buf; 4408} 4409 4410static void print_signal_handler(outputStream* st, int sig, 4411 char* buf, size_t buflen) { 4412 struct sigaction sa; 4413 4414 sigaction(sig, NULL, &sa); 4415 4416 // See comment for SIGNIFICANT_SIGNAL_MASK define 4417 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4418 4419 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4420 4421 address handler = (sa.sa_flags & SA_SIGINFO) 4422 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4423 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4424 4425 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4426 st->print("SIG_DFL"); 4427 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4428 st->print("SIG_IGN"); 4429 } else { 4430 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4431 } 4432 4433 st->print(", sa_mask[0]="); 4434 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4435 4436 address rh = VMError::get_resetted_sighandler(sig); 4437 // May be, handler was resetted by VMError? 4438 if (rh != NULL) { 4439 handler = rh; 4440 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4441 } 4442 4443 st->print(", sa_flags="); 4444 os::Posix::print_sa_flags(st, sa.sa_flags); 4445 4446 // Check: is it our handler? 4447 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4448 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4449 // It is our signal handler 4450 // check for flags, reset system-used one! 4451 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4452 st->print( 4453 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4454 os::Linux::get_our_sigflags(sig)); 4455 } 4456 } 4457 st->cr(); 4458} 4459 4460 4461#define DO_SIGNAL_CHECK(sig) \ 4462 do { \ 4463 if (!sigismember(&check_signal_done, sig)) { \ 4464 os::Linux::check_signal_handler(sig); \ 4465 } \ 4466 } while (0) 4467 4468// This method is a periodic task to check for misbehaving JNI applications 4469// under CheckJNI, we can add any periodic checks here 4470 4471void os::run_periodic_checks() { 4472 if (check_signals == false) return; 4473 4474 // SEGV and BUS if overridden could potentially prevent 4475 // generation of hs*.log in the event of a crash, debugging 4476 // such a case can be very challenging, so we absolutely 4477 // check the following for a good measure: 4478 DO_SIGNAL_CHECK(SIGSEGV); 4479 DO_SIGNAL_CHECK(SIGILL); 4480 DO_SIGNAL_CHECK(SIGFPE); 4481 DO_SIGNAL_CHECK(SIGBUS); 4482 DO_SIGNAL_CHECK(SIGPIPE); 4483 DO_SIGNAL_CHECK(SIGXFSZ); 4484#if defined(PPC64) 4485 DO_SIGNAL_CHECK(SIGTRAP); 4486#endif 4487 4488 // ReduceSignalUsage allows the user to override these handlers 4489 // see comments at the very top and jvm_solaris.h 4490 if (!ReduceSignalUsage) { 4491 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4492 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4493 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4494 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4495 } 4496 4497 DO_SIGNAL_CHECK(SR_signum); 4498 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4499} 4500 4501typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4502 4503static os_sigaction_t os_sigaction = NULL; 4504 4505void os::Linux::check_signal_handler(int sig) { 4506 char buf[O_BUFLEN]; 4507 address jvmHandler = NULL; 4508 4509 4510 struct sigaction act; 4511 if (os_sigaction == NULL) { 4512 // only trust the default sigaction, in case it has been interposed 4513 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4514 if (os_sigaction == NULL) return; 4515 } 4516 4517 os_sigaction(sig, (struct sigaction*)NULL, &act); 4518 4519 4520 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4521 4522 address thisHandler = (act.sa_flags & SA_SIGINFO) 4523 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4524 : CAST_FROM_FN_PTR(address, act.sa_handler); 4525 4526 4527 switch (sig) { 4528 case SIGSEGV: 4529 case SIGBUS: 4530 case SIGFPE: 4531 case SIGPIPE: 4532 case SIGILL: 4533 case SIGXFSZ: 4534 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4535 break; 4536 4537 case SHUTDOWN1_SIGNAL: 4538 case SHUTDOWN2_SIGNAL: 4539 case SHUTDOWN3_SIGNAL: 4540 case BREAK_SIGNAL: 4541 jvmHandler = (address)user_handler(); 4542 break; 4543 4544 case INTERRUPT_SIGNAL: 4545 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4546 break; 4547 4548 default: 4549 if (sig == SR_signum) { 4550 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4551 } else { 4552 return; 4553 } 4554 break; 4555 } 4556 4557 if (thisHandler != jvmHandler) { 4558 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4559 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4560 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4561 // No need to check this sig any longer 4562 sigaddset(&check_signal_done, sig); 4563 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4564 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4565 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4566 exception_name(sig, buf, O_BUFLEN)); 4567 } 4568 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4569 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4570 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4571 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4572 // No need to check this sig any longer 4573 sigaddset(&check_signal_done, sig); 4574 } 4575 4576 // Dump all the signal 4577 if (sigismember(&check_signal_done, sig)) { 4578 print_signal_handlers(tty, buf, O_BUFLEN); 4579 } 4580} 4581 4582extern void report_error(char* file_name, int line_no, char* title, 4583 char* format, ...); 4584 4585extern bool signal_name(int signo, char* buf, size_t len); 4586 4587const char* os::exception_name(int exception_code, char* buf, size_t size) { 4588 if (0 < exception_code && exception_code <= SIGRTMAX) { 4589 // signal 4590 if (!signal_name(exception_code, buf, size)) { 4591 jio_snprintf(buf, size, "SIG%d", exception_code); 4592 } 4593 return buf; 4594 } else { 4595 return NULL; 4596 } 4597} 4598 4599// this is called _before_ the most of global arguments have been parsed 4600void os::init(void) { 4601 char dummy; // used to get a guess on initial stack address 4602// first_hrtime = gethrtime(); 4603 4604 // With LinuxThreads the JavaMain thread pid (primordial thread) 4605 // is different than the pid of the java launcher thread. 4606 // So, on Linux, the launcher thread pid is passed to the VM 4607 // via the sun.java.launcher.pid property. 4608 // Use this property instead of getpid() if it was correctly passed. 4609 // See bug 6351349. 4610 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4611 4612 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4613 4614 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4615 4616 init_random(1234567); 4617 4618 ThreadCritical::initialize(); 4619 4620 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4621 if (Linux::page_size() == -1) { 4622 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4623 strerror(errno))); 4624 } 4625 init_page_sizes((size_t) Linux::page_size()); 4626 4627 Linux::initialize_system_info(); 4628 4629 // main_thread points to the aboriginal thread 4630 Linux::_main_thread = pthread_self(); 4631 4632 Linux::clock_init(); 4633 initial_time_count = javaTimeNanos(); 4634 4635 // pthread_condattr initialization for monotonic clock 4636 int status; 4637 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4638 if ((status = pthread_condattr_init(_condattr)) != 0) { 4639 fatal(err_msg("pthread_condattr_init: %s", strerror(status))); 4640 } 4641 // Only set the clock if CLOCK_MONOTONIC is available 4642 if (os::supports_monotonic_clock()) { 4643 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4644 if (status == EINVAL) { 4645 warning("Unable to use monotonic clock with relative timed-waits" \ 4646 " - changes to the time-of-day clock may have adverse affects"); 4647 } else { 4648 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status))); 4649 } 4650 } 4651 } 4652 // else it defaults to CLOCK_REALTIME 4653 4654 pthread_mutex_init(&dl_mutex, NULL); 4655 4656 // If the pagesize of the VM is greater than 8K determine the appropriate 4657 // number of initial guard pages. The user can change this with the 4658 // command line arguments, if needed. 4659 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4660 StackYellowPages = 1; 4661 StackRedPages = 1; 4662 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4663 } 4664} 4665 4666// To install functions for atexit system call 4667extern "C" { 4668 static void perfMemory_exit_helper() { 4669 perfMemory_exit(); 4670 } 4671} 4672 4673// this is called _after_ the global arguments have been parsed 4674jint os::init_2(void) { 4675 Linux::fast_thread_clock_init(); 4676 4677 // Allocate a single page and mark it as readable for safepoint polling 4678 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4679 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page"); 4680 4681 os::set_polling_page(polling_page); 4682 4683#ifndef PRODUCT 4684 if (Verbose && PrintMiscellaneous) { 4685 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", 4686 (intptr_t)polling_page); 4687 } 4688#endif 4689 4690 if (!UseMembar) { 4691 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4692 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4693 os::set_memory_serialize_page(mem_serialize_page); 4694 4695#ifndef PRODUCT 4696 if (Verbose && PrintMiscellaneous) { 4697 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", 4698 (intptr_t)mem_serialize_page); 4699 } 4700#endif 4701 } 4702 4703 // initialize suspend/resume support - must do this before signal_sets_init() 4704 if (SR_initialize() != 0) { 4705 perror("SR_initialize failed"); 4706 return JNI_ERR; 4707 } 4708 4709 Linux::signal_sets_init(); 4710 Linux::install_signal_handlers(); 4711 4712 // Check minimum allowable stack size for thread creation and to initialize 4713 // the java system classes, including StackOverflowError - depends on page 4714 // size. Add a page for compiler2 recursion in main thread. 4715 // Add in 2*BytesPerWord times page size to account for VM stack during 4716 // class initialization depending on 32 or 64 bit VM. 4717 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4718 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4719 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4720 4721 size_t threadStackSizeInBytes = ThreadStackSize * K; 4722 if (threadStackSizeInBytes != 0 && 4723 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4724 tty->print_cr("\nThe stack size specified is too small, " 4725 "Specify at least %dk", 4726 os::Linux::min_stack_allowed/ K); 4727 return JNI_ERR; 4728 } 4729 4730 // Make the stack size a multiple of the page size so that 4731 // the yellow/red zones can be guarded. 4732 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4733 vm_page_size())); 4734 4735 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4736 4737#if defined(IA32) 4738 workaround_expand_exec_shield_cs_limit(); 4739#endif 4740 4741 Linux::libpthread_init(); 4742 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4743 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4744 Linux::glibc_version(), Linux::libpthread_version(), 4745 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4746 } 4747 4748 if (UseNUMA) { 4749 if (!Linux::libnuma_init()) { 4750 UseNUMA = false; 4751 } else { 4752 if ((Linux::numa_max_node() < 1)) { 4753 // There's only one node(they start from 0), disable NUMA. 4754 UseNUMA = false; 4755 } 4756 } 4757 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4758 // we can make the adaptive lgrp chunk resizing work. If the user specified 4759 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4760 // disable adaptive resizing. 4761 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4762 if (FLAG_IS_DEFAULT(UseNUMA)) { 4763 UseNUMA = false; 4764 } else { 4765 if (FLAG_IS_DEFAULT(UseLargePages) && 4766 FLAG_IS_DEFAULT(UseSHM) && 4767 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4768 UseLargePages = false; 4769 } else { 4770 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing"); 4771 UseAdaptiveSizePolicy = false; 4772 UseAdaptiveNUMAChunkSizing = false; 4773 } 4774 } 4775 } 4776 if (!UseNUMA && ForceNUMA) { 4777 UseNUMA = true; 4778 } 4779 } 4780 4781 if (MaxFDLimit) { 4782 // set the number of file descriptors to max. print out error 4783 // if getrlimit/setrlimit fails but continue regardless. 4784 struct rlimit nbr_files; 4785 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4786 if (status != 0) { 4787 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4788 perror("os::init_2 getrlimit failed"); 4789 } 4790 } else { 4791 nbr_files.rlim_cur = nbr_files.rlim_max; 4792 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4793 if (status != 0) { 4794 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4795 perror("os::init_2 setrlimit failed"); 4796 } 4797 } 4798 } 4799 } 4800 4801 // Initialize lock used to serialize thread creation (see os::create_thread) 4802 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4803 4804 // at-exit methods are called in the reverse order of their registration. 4805 // atexit functions are called on return from main or as a result of a 4806 // call to exit(3C). There can be only 32 of these functions registered 4807 // and atexit() does not set errno. 4808 4809 if (PerfAllowAtExitRegistration) { 4810 // only register atexit functions if PerfAllowAtExitRegistration is set. 4811 // atexit functions can be delayed until process exit time, which 4812 // can be problematic for embedded VM situations. Embedded VMs should 4813 // call DestroyJavaVM() to assure that VM resources are released. 4814 4815 // note: perfMemory_exit_helper atexit function may be removed in 4816 // the future if the appropriate cleanup code can be added to the 4817 // VM_Exit VMOperation's doit method. 4818 if (atexit(perfMemory_exit_helper) != 0) { 4819 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4820 } 4821 } 4822 4823 // initialize thread priority policy 4824 prio_init(); 4825 4826 return JNI_OK; 4827} 4828 4829// this is called at the end of vm_initialization 4830void os::init_3(void) { 4831#ifdef JAVASE_EMBEDDED 4832 // Start the MemNotifyThread 4833 if (LowMemoryProtection) { 4834 MemNotifyThread::start(); 4835 } 4836 return; 4837#endif 4838} 4839 4840// Mark the polling page as unreadable 4841void os::make_polling_page_unreadable(void) { 4842 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 4843 fatal("Could not disable polling page"); 4844 } 4845} 4846 4847// Mark the polling page as readable 4848void os::make_polling_page_readable(void) { 4849 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4850 fatal("Could not enable polling page"); 4851 } 4852} 4853 4854int os::active_processor_count() { 4855 // Linux doesn't yet have a (official) notion of processor sets, 4856 // so just return the number of online processors. 4857 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4858 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4859 return online_cpus; 4860} 4861 4862void os::set_native_thread_name(const char *name) { 4863 // Not yet implemented. 4864 return; 4865} 4866 4867bool os::distribute_processes(uint length, uint* distribution) { 4868 // Not yet implemented. 4869 return false; 4870} 4871 4872bool os::bind_to_processor(uint processor_id) { 4873 // Not yet implemented. 4874 return false; 4875} 4876 4877/// 4878 4879void os::SuspendedThreadTask::internal_do_task() { 4880 if (do_suspend(_thread->osthread())) { 4881 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4882 do_task(context); 4883 do_resume(_thread->osthread()); 4884 } 4885} 4886 4887class PcFetcher : public os::SuspendedThreadTask { 4888 public: 4889 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4890 ExtendedPC result(); 4891 protected: 4892 void do_task(const os::SuspendedThreadTaskContext& context); 4893 private: 4894 ExtendedPC _epc; 4895}; 4896 4897ExtendedPC PcFetcher::result() { 4898 guarantee(is_done(), "task is not done yet."); 4899 return _epc; 4900} 4901 4902void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4903 Thread* thread = context.thread(); 4904 OSThread* osthread = thread->osthread(); 4905 if (osthread->ucontext() != NULL) { 4906 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4907 } else { 4908 // NULL context is unexpected, double-check this is the VMThread 4909 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4910 } 4911} 4912 4913// Suspends the target using the signal mechanism and then grabs the PC before 4914// resuming the target. Used by the flat-profiler only 4915ExtendedPC os::get_thread_pc(Thread* thread) { 4916 // Make sure that it is called by the watcher for the VMThread 4917 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4918 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4919 4920 PcFetcher fetcher(thread); 4921 fetcher.run(); 4922 return fetcher.result(); 4923} 4924 4925int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, 4926 pthread_mutex_t *_mutex, 4927 const struct timespec *_abstime) { 4928 if (is_NPTL()) { 4929 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4930 } else { 4931 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4932 // word back to default 64bit precision if condvar is signaled. Java 4933 // wants 53bit precision. Save and restore current value. 4934 int fpu = get_fpu_control_word(); 4935 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4936 set_fpu_control_word(fpu); 4937 return status; 4938 } 4939} 4940 4941//////////////////////////////////////////////////////////////////////////////// 4942// debug support 4943 4944bool os::find(address addr, outputStream* st) { 4945 Dl_info dlinfo; 4946 memset(&dlinfo, 0, sizeof(dlinfo)); 4947 if (dladdr(addr, &dlinfo) != 0) { 4948 st->print(PTR_FORMAT ": ", addr); 4949 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 4950 st->print("%s+%#x", dlinfo.dli_sname, 4951 addr - (intptr_t)dlinfo.dli_saddr); 4952 } else if (dlinfo.dli_fbase != NULL) { 4953 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4954 } else { 4955 st->print("<absolute address>"); 4956 } 4957 if (dlinfo.dli_fname != NULL) { 4958 st->print(" in %s", dlinfo.dli_fname); 4959 } 4960 if (dlinfo.dli_fbase != NULL) { 4961 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4962 } 4963 st->cr(); 4964 4965 if (Verbose) { 4966 // decode some bytes around the PC 4967 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 4968 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 4969 address lowest = (address) dlinfo.dli_sname; 4970 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4971 if (begin < lowest) begin = lowest; 4972 Dl_info dlinfo2; 4973 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 4974 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 4975 end = (address) dlinfo2.dli_saddr; 4976 } 4977 Disassembler::decode(begin, end, st); 4978 } 4979 return true; 4980 } 4981 return false; 4982} 4983 4984//////////////////////////////////////////////////////////////////////////////// 4985// misc 4986 4987// This does not do anything on Linux. This is basically a hook for being 4988// able to use structured exception handling (thread-local exception filters) 4989// on, e.g., Win32. 4990void 4991os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4992 JavaCallArguments* args, Thread* thread) { 4993 f(value, method, args, thread); 4994} 4995 4996void os::print_statistics() { 4997} 4998 4999int os::message_box(const char* title, const char* message) { 5000 int i; 5001 fdStream err(defaultStream::error_fd()); 5002 for (i = 0; i < 78; i++) err.print_raw("="); 5003 err.cr(); 5004 err.print_raw_cr(title); 5005 for (i = 0; i < 78; i++) err.print_raw("-"); 5006 err.cr(); 5007 err.print_raw_cr(message); 5008 for (i = 0; i < 78; i++) err.print_raw("="); 5009 err.cr(); 5010 5011 char buf[16]; 5012 // Prevent process from exiting upon "read error" without consuming all CPU 5013 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5014 5015 return buf[0] == 'y' || buf[0] == 'Y'; 5016} 5017 5018int os::stat(const char *path, struct stat *sbuf) { 5019 char pathbuf[MAX_PATH]; 5020 if (strlen(path) > MAX_PATH - 1) { 5021 errno = ENAMETOOLONG; 5022 return -1; 5023 } 5024 os::native_path(strcpy(pathbuf, path)); 5025 return ::stat(pathbuf, sbuf); 5026} 5027 5028bool os::check_heap(bool force) { 5029 return true; 5030} 5031 5032int local_vsnprintf(char* buf, size_t count, const char* format, 5033 va_list args) { 5034 return ::vsnprintf(buf, count, format, args); 5035} 5036 5037// Is a (classpath) directory empty? 5038bool os::dir_is_empty(const char* path) { 5039 DIR *dir = NULL; 5040 struct dirent *ptr; 5041 5042 dir = opendir(path); 5043 if (dir == NULL) return true; 5044 5045 // Scan the directory 5046 bool result = true; 5047 char buf[sizeof(struct dirent) + MAX_PATH]; 5048 while (result && (ptr = ::readdir(dir)) != NULL) { 5049 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5050 result = false; 5051 } 5052 } 5053 closedir(dir); 5054 return result; 5055} 5056 5057// This code originates from JDK's sysOpen and open64_w 5058// from src/solaris/hpi/src/system_md.c 5059 5060#ifndef O_DELETE 5061 #define O_DELETE 0x10000 5062#endif 5063 5064// Open a file. Unlink the file immediately after open returns 5065// if the specified oflag has the O_DELETE flag set. 5066// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 5067 5068int os::open(const char *path, int oflag, int mode) { 5069 if (strlen(path) > MAX_PATH - 1) { 5070 errno = ENAMETOOLONG; 5071 return -1; 5072 } 5073 int fd; 5074 int o_delete = (oflag & O_DELETE); 5075 oflag = oflag & ~O_DELETE; 5076 5077 fd = ::open64(path, oflag, mode); 5078 if (fd == -1) return -1; 5079 5080 //If the open succeeded, the file might still be a directory 5081 { 5082 struct stat64 buf64; 5083 int ret = ::fstat64(fd, &buf64); 5084 int st_mode = buf64.st_mode; 5085 5086 if (ret != -1) { 5087 if ((st_mode & S_IFMT) == S_IFDIR) { 5088 errno = EISDIR; 5089 ::close(fd); 5090 return -1; 5091 } 5092 } else { 5093 ::close(fd); 5094 return -1; 5095 } 5096 } 5097 5098 // All file descriptors that are opened in the JVM and not 5099 // specifically destined for a subprocess should have the 5100 // close-on-exec flag set. If we don't set it, then careless 3rd 5101 // party native code might fork and exec without closing all 5102 // appropriate file descriptors (e.g. as we do in closeDescriptors in 5103 // UNIXProcess.c), and this in turn might: 5104 // 5105 // - cause end-of-file to fail to be detected on some file 5106 // descriptors, resulting in mysterious hangs, or 5107 // 5108 // - might cause an fopen in the subprocess to fail on a system 5109 // suffering from bug 1085341. 5110 // 5111 // (Yes, the default setting of the close-on-exec flag is a Unix 5112 // design flaw) 5113 // 5114 // See: 5115 // 1085341: 32-bit stdio routines should support file descriptors >255 5116 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5117 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5118 // 5119#ifdef FD_CLOEXEC 5120 { 5121 int flags = ::fcntl(fd, F_GETFD); 5122 if (flags != -1) { 5123 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5124 } 5125 } 5126#endif 5127 5128 if (o_delete != 0) { 5129 ::unlink(path); 5130 } 5131 return fd; 5132} 5133 5134 5135// create binary file, rewriting existing file if required 5136int os::create_binary_file(const char* path, bool rewrite_existing) { 5137 int oflags = O_WRONLY | O_CREAT; 5138 if (!rewrite_existing) { 5139 oflags |= O_EXCL; 5140 } 5141 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5142} 5143 5144// return current position of file pointer 5145jlong os::current_file_offset(int fd) { 5146 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5147} 5148 5149// move file pointer to the specified offset 5150jlong os::seek_to_file_offset(int fd, jlong offset) { 5151 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5152} 5153 5154// This code originates from JDK's sysAvailable 5155// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5156 5157int os::available(int fd, jlong *bytes) { 5158 jlong cur, end; 5159 int mode; 5160 struct stat64 buf64; 5161 5162 if (::fstat64(fd, &buf64) >= 0) { 5163 mode = buf64.st_mode; 5164 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5165 // XXX: is the following call interruptible? If so, this might 5166 // need to go through the INTERRUPT_IO() wrapper as for other 5167 // blocking, interruptible calls in this file. 5168 int n; 5169 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5170 *bytes = n; 5171 return 1; 5172 } 5173 } 5174 } 5175 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5176 return 0; 5177 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5178 return 0; 5179 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5180 return 0; 5181 } 5182 *bytes = end - cur; 5183 return 1; 5184} 5185 5186int os::socket_available(int fd, jint *pbytes) { 5187 // Linux doc says EINTR not returned, unlike Solaris 5188 int ret = ::ioctl(fd, FIONREAD, pbytes); 5189 5190 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 5191 // is expected to return 0 on failure and 1 on success to the jdk. 5192 return (ret < 0) ? 0 : 1; 5193} 5194 5195// Map a block of memory. 5196char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5197 char *addr, size_t bytes, bool read_only, 5198 bool allow_exec) { 5199 int prot; 5200 int flags = MAP_PRIVATE; 5201 5202 if (read_only) { 5203 prot = PROT_READ; 5204 } else { 5205 prot = PROT_READ | PROT_WRITE; 5206 } 5207 5208 if (allow_exec) { 5209 prot |= PROT_EXEC; 5210 } 5211 5212 if (addr != NULL) { 5213 flags |= MAP_FIXED; 5214 } 5215 5216 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5217 fd, file_offset); 5218 if (mapped_address == MAP_FAILED) { 5219 return NULL; 5220 } 5221 return mapped_address; 5222} 5223 5224 5225// Remap a block of memory. 5226char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5227 char *addr, size_t bytes, bool read_only, 5228 bool allow_exec) { 5229 // same as map_memory() on this OS 5230 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5231 allow_exec); 5232} 5233 5234 5235// Unmap a block of memory. 5236bool os::pd_unmap_memory(char* addr, size_t bytes) { 5237 return munmap(addr, bytes) == 0; 5238} 5239 5240static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5241 5242static clockid_t thread_cpu_clockid(Thread* thread) { 5243 pthread_t tid = thread->osthread()->pthread_id(); 5244 clockid_t clockid; 5245 5246 // Get thread clockid 5247 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5248 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5249 return clockid; 5250} 5251 5252// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5253// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5254// of a thread. 5255// 5256// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5257// the fast estimate available on the platform. 5258 5259jlong os::current_thread_cpu_time() { 5260 if (os::Linux::supports_fast_thread_cpu_time()) { 5261 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5262 } else { 5263 // return user + sys since the cost is the same 5264 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5265 } 5266} 5267 5268jlong os::thread_cpu_time(Thread* thread) { 5269 // consistent with what current_thread_cpu_time() returns 5270 if (os::Linux::supports_fast_thread_cpu_time()) { 5271 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5272 } else { 5273 return slow_thread_cpu_time(thread, true /* user + sys */); 5274 } 5275} 5276 5277jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5278 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5279 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5280 } else { 5281 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5282 } 5283} 5284 5285jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5286 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5287 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5288 } else { 5289 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5290 } 5291} 5292 5293// -1 on error. 5294static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5295 pid_t tid = thread->osthread()->thread_id(); 5296 char *s; 5297 char stat[2048]; 5298 int statlen; 5299 char proc_name[64]; 5300 int count; 5301 long sys_time, user_time; 5302 char cdummy; 5303 int idummy; 5304 long ldummy; 5305 FILE *fp; 5306 5307 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5308 fp = fopen(proc_name, "r"); 5309 if (fp == NULL) return -1; 5310 statlen = fread(stat, 1, 2047, fp); 5311 stat[statlen] = '\0'; 5312 fclose(fp); 5313 5314 // Skip pid and the command string. Note that we could be dealing with 5315 // weird command names, e.g. user could decide to rename java launcher 5316 // to "java 1.4.2 :)", then the stat file would look like 5317 // 1234 (java 1.4.2 :)) R ... ... 5318 // We don't really need to know the command string, just find the last 5319 // occurrence of ")" and then start parsing from there. See bug 4726580. 5320 s = strrchr(stat, ')'); 5321 if (s == NULL) return -1; 5322 5323 // Skip blank chars 5324 do s++; while (isspace(*s)); 5325 5326 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5327 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5328 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5329 &user_time, &sys_time); 5330 if (count != 13) return -1; 5331 if (user_sys_cpu_time) { 5332 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5333 } else { 5334 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5335 } 5336} 5337 5338void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5339 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5340 info_ptr->may_skip_backward = false; // elapsed time not wall time 5341 info_ptr->may_skip_forward = false; // elapsed time not wall time 5342 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5343} 5344 5345void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5346 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5347 info_ptr->may_skip_backward = false; // elapsed time not wall time 5348 info_ptr->may_skip_forward = false; // elapsed time not wall time 5349 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5350} 5351 5352bool os::is_thread_cpu_time_supported() { 5353 return true; 5354} 5355 5356// System loadavg support. Returns -1 if load average cannot be obtained. 5357// Linux doesn't yet have a (official) notion of processor sets, 5358// so just return the system wide load average. 5359int os::loadavg(double loadavg[], int nelem) { 5360 return ::getloadavg(loadavg, nelem); 5361} 5362 5363void os::pause() { 5364 char filename[MAX_PATH]; 5365 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5366 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5367 } else { 5368 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5369 } 5370 5371 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5372 if (fd != -1) { 5373 struct stat buf; 5374 ::close(fd); 5375 while (::stat(filename, &buf) == 0) { 5376 (void)::poll(NULL, 0, 100); 5377 } 5378 } else { 5379 jio_fprintf(stderr, 5380 "Could not open pause file '%s', continuing immediately.\n", filename); 5381 } 5382} 5383 5384 5385// Refer to the comments in os_solaris.cpp park-unpark. 5386// 5387// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5388// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5389// For specifics regarding the bug see GLIBC BUGID 261237 : 5390// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5391// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5392// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5393// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5394// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5395// and monitorenter when we're using 1-0 locking. All those operations may result in 5396// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5397// of libpthread avoids the problem, but isn't practical. 5398// 5399// Possible remedies: 5400// 5401// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5402// This is palliative and probabilistic, however. If the thread is preempted 5403// between the call to compute_abstime() and pthread_cond_timedwait(), more 5404// than the minimum period may have passed, and the abstime may be stale (in the 5405// past) resultin in a hang. Using this technique reduces the odds of a hang 5406// but the JVM is still vulnerable, particularly on heavily loaded systems. 5407// 5408// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5409// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5410// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5411// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5412// thread. 5413// 5414// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5415// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5416// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5417// This also works well. In fact it avoids kernel-level scalability impediments 5418// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5419// timers in a graceful fashion. 5420// 5421// 4. When the abstime value is in the past it appears that control returns 5422// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5423// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5424// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5425// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5426// It may be possible to avoid reinitialization by checking the return 5427// value from pthread_cond_timedwait(). In addition to reinitializing the 5428// condvar we must establish the invariant that cond_signal() is only called 5429// within critical sections protected by the adjunct mutex. This prevents 5430// cond_signal() from "seeing" a condvar that's in the midst of being 5431// reinitialized or that is corrupt. Sadly, this invariant obviates the 5432// desirable signal-after-unlock optimization that avoids futile context switching. 5433// 5434// I'm also concerned that some versions of NTPL might allocate an auxilliary 5435// structure when a condvar is used or initialized. cond_destroy() would 5436// release the helper structure. Our reinitialize-after-timedwait fix 5437// put excessive stress on malloc/free and locks protecting the c-heap. 5438// 5439// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5440// It may be possible to refine (4) by checking the kernel and NTPL verisons 5441// and only enabling the work-around for vulnerable environments. 5442 5443// utility to compute the abstime argument to timedwait: 5444// millis is the relative timeout time 5445// abstime will be the absolute timeout time 5446// TODO: replace compute_abstime() with unpackTime() 5447 5448static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5449 if (millis < 0) millis = 0; 5450 5451 jlong seconds = millis / 1000; 5452 millis %= 1000; 5453 if (seconds > 50000000) { // see man cond_timedwait(3T) 5454 seconds = 50000000; 5455 } 5456 5457 if (os::supports_monotonic_clock()) { 5458 struct timespec now; 5459 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5460 assert_status(status == 0, status, "clock_gettime"); 5461 abstime->tv_sec = now.tv_sec + seconds; 5462 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5463 if (nanos >= NANOSECS_PER_SEC) { 5464 abstime->tv_sec += 1; 5465 nanos -= NANOSECS_PER_SEC; 5466 } 5467 abstime->tv_nsec = nanos; 5468 } else { 5469 struct timeval now; 5470 int status = gettimeofday(&now, NULL); 5471 assert(status == 0, "gettimeofday"); 5472 abstime->tv_sec = now.tv_sec + seconds; 5473 long usec = now.tv_usec + millis * 1000; 5474 if (usec >= 1000000) { 5475 abstime->tv_sec += 1; 5476 usec -= 1000000; 5477 } 5478 abstime->tv_nsec = usec * 1000; 5479 } 5480 return abstime; 5481} 5482 5483void os::PlatformEvent::park() { // AKA "down()" 5484 // Invariant: Only the thread associated with the Event/PlatformEvent 5485 // may call park(). 5486 // TODO: assert that _Assoc != NULL or _Assoc == Self 5487 assert(_nParked == 0, "invariant"); 5488 5489 int v; 5490 for (;;) { 5491 v = _Event; 5492 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5493 } 5494 guarantee(v >= 0, "invariant"); 5495 if (v == 0) { 5496 // Do this the hard way by blocking ... 5497 int status = pthread_mutex_lock(_mutex); 5498 assert_status(status == 0, status, "mutex_lock"); 5499 guarantee(_nParked == 0, "invariant"); 5500 ++_nParked; 5501 while (_Event < 0) { 5502 status = pthread_cond_wait(_cond, _mutex); 5503 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5504 // Treat this the same as if the wait was interrupted 5505 if (status == ETIME) { status = EINTR; } 5506 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5507 } 5508 --_nParked; 5509 5510 _Event = 0; 5511 status = pthread_mutex_unlock(_mutex); 5512 assert_status(status == 0, status, "mutex_unlock"); 5513 // Paranoia to ensure our locked and lock-free paths interact 5514 // correctly with each other. 5515 OrderAccess::fence(); 5516 } 5517 guarantee(_Event >= 0, "invariant"); 5518} 5519 5520int os::PlatformEvent::park(jlong millis) { 5521 guarantee(_nParked == 0, "invariant"); 5522 5523 int v; 5524 for (;;) { 5525 v = _Event; 5526 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5527 } 5528 guarantee(v >= 0, "invariant"); 5529 if (v != 0) return OS_OK; 5530 5531 // We do this the hard way, by blocking the thread. 5532 // Consider enforcing a minimum timeout value. 5533 struct timespec abst; 5534 compute_abstime(&abst, millis); 5535 5536 int ret = OS_TIMEOUT; 5537 int status = pthread_mutex_lock(_mutex); 5538 assert_status(status == 0, status, "mutex_lock"); 5539 guarantee(_nParked == 0, "invariant"); 5540 ++_nParked; 5541 5542 // Object.wait(timo) will return because of 5543 // (a) notification 5544 // (b) timeout 5545 // (c) thread.interrupt 5546 // 5547 // Thread.interrupt and object.notify{All} both call Event::set. 5548 // That is, we treat thread.interrupt as a special case of notification. 5549 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false. 5550 // We assume all ETIME returns are valid. 5551 // 5552 // TODO: properly differentiate simultaneous notify+interrupt. 5553 // In that case, we should propagate the notify to another waiter. 5554 5555 while (_Event < 0) { 5556 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5557 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5558 pthread_cond_destroy(_cond); 5559 pthread_cond_init(_cond, os::Linux::condAttr()); 5560 } 5561 assert_status(status == 0 || status == EINTR || 5562 status == ETIME || status == ETIMEDOUT, 5563 status, "cond_timedwait"); 5564 if (!FilterSpuriousWakeups) break; // previous semantics 5565 if (status == ETIME || status == ETIMEDOUT) break; 5566 // We consume and ignore EINTR and spurious wakeups. 5567 } 5568 --_nParked; 5569 if (_Event >= 0) { 5570 ret = OS_OK; 5571 } 5572 _Event = 0; 5573 status = pthread_mutex_unlock(_mutex); 5574 assert_status(status == 0, status, "mutex_unlock"); 5575 assert(_nParked == 0, "invariant"); 5576 // Paranoia to ensure our locked and lock-free paths interact 5577 // correctly with each other. 5578 OrderAccess::fence(); 5579 return ret; 5580} 5581 5582void os::PlatformEvent::unpark() { 5583 // Transitions for _Event: 5584 // 0 :=> 1 5585 // 1 :=> 1 5586 // -1 :=> either 0 or 1; must signal target thread 5587 // That is, we can safely transition _Event from -1 to either 5588 // 0 or 1. 5589 // See also: "Semaphores in Plan 9" by Mullender & Cox 5590 // 5591 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5592 // that it will take two back-to-back park() calls for the owning 5593 // thread to block. This has the benefit of forcing a spurious return 5594 // from the first park() call after an unpark() call which will help 5595 // shake out uses of park() and unpark() without condition variables. 5596 5597 if (Atomic::xchg(1, &_Event) >= 0) return; 5598 5599 // Wait for the thread associated with the event to vacate 5600 int status = pthread_mutex_lock(_mutex); 5601 assert_status(status == 0, status, "mutex_lock"); 5602 int AnyWaiters = _nParked; 5603 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5604 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5605 AnyWaiters = 0; 5606 pthread_cond_signal(_cond); 5607 } 5608 status = pthread_mutex_unlock(_mutex); 5609 assert_status(status == 0, status, "mutex_unlock"); 5610 if (AnyWaiters != 0) { 5611 status = pthread_cond_signal(_cond); 5612 assert_status(status == 0, status, "cond_signal"); 5613 } 5614 5615 // Note that we signal() _after dropping the lock for "immortal" Events. 5616 // This is safe and avoids a common class of futile wakeups. In rare 5617 // circumstances this can cause a thread to return prematurely from 5618 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5619 // simply re-test the condition and re-park itself. 5620} 5621 5622 5623// JSR166 5624// ------------------------------------------------------- 5625 5626// The solaris and linux implementations of park/unpark are fairly 5627// conservative for now, but can be improved. They currently use a 5628// mutex/condvar pair, plus a a count. 5629// Park decrements count if > 0, else does a condvar wait. Unpark 5630// sets count to 1 and signals condvar. Only one thread ever waits 5631// on the condvar. Contention seen when trying to park implies that someone 5632// is unparking you, so don't wait. And spurious returns are fine, so there 5633// is no need to track notifications. 5634 5635// This code is common to linux and solaris and will be moved to a 5636// common place in dolphin. 5637// 5638// The passed in time value is either a relative time in nanoseconds 5639// or an absolute time in milliseconds. Either way it has to be unpacked 5640// into suitable seconds and nanoseconds components and stored in the 5641// given timespec structure. 5642// Given time is a 64-bit value and the time_t used in the timespec is only 5643// a signed-32-bit value (except on 64-bit Linux) we have to watch for 5644// overflow if times way in the future are given. Further on Solaris versions 5645// prior to 10 there is a restriction (see cond_timedwait) that the specified 5646// number of seconds, in abstime, is less than current_time + 100,000,000. 5647// As it will be 28 years before "now + 100000000" will overflow we can 5648// ignore overflow and just impose a hard-limit on seconds using the value 5649// of "now + 100,000,000". This places a limit on the timeout of about 3.17 5650// years from "now". 5651 5652static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5653 assert(time > 0, "convertTime"); 5654 time_t max_secs = 0; 5655 5656 if (!os::supports_monotonic_clock() || isAbsolute) { 5657 struct timeval now; 5658 int status = gettimeofday(&now, NULL); 5659 assert(status == 0, "gettimeofday"); 5660 5661 max_secs = now.tv_sec + MAX_SECS; 5662 5663 if (isAbsolute) { 5664 jlong secs = time / 1000; 5665 if (secs > max_secs) { 5666 absTime->tv_sec = max_secs; 5667 } else { 5668 absTime->tv_sec = secs; 5669 } 5670 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5671 } else { 5672 jlong secs = time / NANOSECS_PER_SEC; 5673 if (secs >= MAX_SECS) { 5674 absTime->tv_sec = max_secs; 5675 absTime->tv_nsec = 0; 5676 } else { 5677 absTime->tv_sec = now.tv_sec + secs; 5678 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5679 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5680 absTime->tv_nsec -= NANOSECS_PER_SEC; 5681 ++absTime->tv_sec; // note: this must be <= max_secs 5682 } 5683 } 5684 } 5685 } else { 5686 // must be relative using monotonic clock 5687 struct timespec now; 5688 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5689 assert_status(status == 0, status, "clock_gettime"); 5690 max_secs = now.tv_sec + MAX_SECS; 5691 jlong secs = time / NANOSECS_PER_SEC; 5692 if (secs >= MAX_SECS) { 5693 absTime->tv_sec = max_secs; 5694 absTime->tv_nsec = 0; 5695 } else { 5696 absTime->tv_sec = now.tv_sec + secs; 5697 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5698 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5699 absTime->tv_nsec -= NANOSECS_PER_SEC; 5700 ++absTime->tv_sec; // note: this must be <= max_secs 5701 } 5702 } 5703 } 5704 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5705 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5706 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5707 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5708} 5709 5710void Parker::park(bool isAbsolute, jlong time) { 5711 // Ideally we'd do something useful while spinning, such 5712 // as calling unpackTime(). 5713 5714 // Optional fast-path check: 5715 // Return immediately if a permit is available. 5716 // We depend on Atomic::xchg() having full barrier semantics 5717 // since we are doing a lock-free update to _counter. 5718 if (Atomic::xchg(0, &_counter) > 0) return; 5719 5720 Thread* thread = Thread::current(); 5721 assert(thread->is_Java_thread(), "Must be JavaThread"); 5722 JavaThread *jt = (JavaThread *)thread; 5723 5724 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5725 // Check interrupt before trying to wait 5726 if (Thread::is_interrupted(thread, false)) { 5727 return; 5728 } 5729 5730 // Next, demultiplex/decode time arguments 5731 timespec absTime; 5732 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all 5733 return; 5734 } 5735 if (time > 0) { 5736 unpackTime(&absTime, isAbsolute, time); 5737 } 5738 5739 5740 // Enter safepoint region 5741 // Beware of deadlocks such as 6317397. 5742 // The per-thread Parker:: mutex is a classic leaf-lock. 5743 // In particular a thread must never block on the Threads_lock while 5744 // holding the Parker:: mutex. If safepoints are pending both the 5745 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5746 ThreadBlockInVM tbivm(jt); 5747 5748 // Don't wait if cannot get lock since interference arises from 5749 // unblocking. Also. check interrupt before trying wait 5750 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5751 return; 5752 } 5753 5754 int status; 5755 if (_counter > 0) { // no wait needed 5756 _counter = 0; 5757 status = pthread_mutex_unlock(_mutex); 5758 assert(status == 0, "invariant"); 5759 // Paranoia to ensure our locked and lock-free paths interact 5760 // correctly with each other and Java-level accesses. 5761 OrderAccess::fence(); 5762 return; 5763 } 5764 5765#ifdef ASSERT 5766 // Don't catch signals while blocked; let the running threads have the signals. 5767 // (This allows a debugger to break into the running thread.) 5768 sigset_t oldsigs; 5769 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5770 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5771#endif 5772 5773 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5774 jt->set_suspend_equivalent(); 5775 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5776 5777 assert(_cur_index == -1, "invariant"); 5778 if (time == 0) { 5779 _cur_index = REL_INDEX; // arbitrary choice when not timed 5780 status = pthread_cond_wait(&_cond[_cur_index], _mutex); 5781 } else { 5782 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5783 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime); 5784 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5785 pthread_cond_destroy(&_cond[_cur_index]); 5786 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr()); 5787 } 5788 } 5789 _cur_index = -1; 5790 assert_status(status == 0 || status == EINTR || 5791 status == ETIME || status == ETIMEDOUT, 5792 status, "cond_timedwait"); 5793 5794#ifdef ASSERT 5795 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5796#endif 5797 5798 _counter = 0; 5799 status = pthread_mutex_unlock(_mutex); 5800 assert_status(status == 0, status, "invariant"); 5801 // Paranoia to ensure our locked and lock-free paths interact 5802 // correctly with each other and Java-level accesses. 5803 OrderAccess::fence(); 5804 5805 // If externally suspended while waiting, re-suspend 5806 if (jt->handle_special_suspend_equivalent_condition()) { 5807 jt->java_suspend_self(); 5808 } 5809} 5810 5811void Parker::unpark() { 5812 int status = pthread_mutex_lock(_mutex); 5813 assert(status == 0, "invariant"); 5814 const int s = _counter; 5815 _counter = 1; 5816 if (s < 1) { 5817 // thread might be parked 5818 if (_cur_index != -1) { 5819 // thread is definitely parked 5820 if (WorkAroundNPTLTimedWaitHang) { 5821 status = pthread_cond_signal(&_cond[_cur_index]); 5822 assert(status == 0, "invariant"); 5823 status = pthread_mutex_unlock(_mutex); 5824 assert(status == 0, "invariant"); 5825 } else { 5826 status = pthread_mutex_unlock(_mutex); 5827 assert(status == 0, "invariant"); 5828 status = pthread_cond_signal(&_cond[_cur_index]); 5829 assert(status == 0, "invariant"); 5830 } 5831 } else { 5832 pthread_mutex_unlock(_mutex); 5833 assert(status == 0, "invariant"); 5834 } 5835 } else { 5836 pthread_mutex_unlock(_mutex); 5837 assert(status == 0, "invariant"); 5838 } 5839} 5840 5841 5842extern char** environ; 5843 5844#ifndef __NR_fork 5845 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5846#endif 5847 5848#ifndef __NR_execve 5849 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5850#endif 5851 5852// Run the specified command in a separate process. Return its exit value, 5853// or -1 on failure (e.g. can't fork a new process). 5854// Unlike system(), this function can be called from signal handler. It 5855// doesn't block SIGINT et al. 5856int os::fork_and_exec(char* cmd) { 5857 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5858 5859 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5860 // pthread_atfork handlers and reset pthread library. All we need is a 5861 // separate process to execve. Make a direct syscall to fork process. 5862 // On IA64 there's no fork syscall, we have to use fork() and hope for 5863 // the best... 5864 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5865 IA64_ONLY(fork();) 5866 5867 if (pid < 0) { 5868 // fork failed 5869 return -1; 5870 5871 } else if (pid == 0) { 5872 // child process 5873 5874 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5875 // first to kill every thread on the thread list. Because this list is 5876 // not reset by fork() (see notes above), execve() will instead kill 5877 // every thread in the parent process. We know this is the only thread 5878 // in the new process, so make a system call directly. 5879 // IA64 should use normal execve() from glibc to match the glibc fork() 5880 // above. 5881 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5882 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5883 5884 // execve failed 5885 _exit(-1); 5886 5887 } else { 5888 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5889 // care about the actual exit code, for now. 5890 5891 int status; 5892 5893 // Wait for the child process to exit. This returns immediately if 5894 // the child has already exited. */ 5895 while (waitpid(pid, &status, 0) < 0) { 5896 switch (errno) { 5897 case ECHILD: return 0; 5898 case EINTR: break; 5899 default: return -1; 5900 } 5901 } 5902 5903 if (WIFEXITED(status)) { 5904 // The child exited normally; get its exit code. 5905 return WEXITSTATUS(status); 5906 } else if (WIFSIGNALED(status)) { 5907 // The child exited because of a signal 5908 // The best value to return is 0x80 + signal number, 5909 // because that is what all Unix shells do, and because 5910 // it allows callers to distinguish between process exit and 5911 // process death by signal. 5912 return 0x80 + WTERMSIG(status); 5913 } else { 5914 // Unknown exit code; pass it through 5915 return status; 5916 } 5917 } 5918} 5919 5920// is_headless_jre() 5921// 5922// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5923// in order to report if we are running in a headless jre 5924// 5925// Since JDK8 xawt/libmawt.so was moved into the same directory 5926// as libawt.so, and renamed libawt_xawt.so 5927// 5928bool os::is_headless_jre() { 5929 struct stat statbuf; 5930 char buf[MAXPATHLEN]; 5931 char libmawtpath[MAXPATHLEN]; 5932 const char *xawtstr = "/xawt/libmawt.so"; 5933 const char *new_xawtstr = "/libawt_xawt.so"; 5934 char *p; 5935 5936 // Get path to libjvm.so 5937 os::jvm_path(buf, sizeof(buf)); 5938 5939 // Get rid of libjvm.so 5940 p = strrchr(buf, '/'); 5941 if (p == NULL) { 5942 return false; 5943 } else { 5944 *p = '\0'; 5945 } 5946 5947 // Get rid of client or server 5948 p = strrchr(buf, '/'); 5949 if (p == NULL) { 5950 return false; 5951 } else { 5952 *p = '\0'; 5953 } 5954 5955 // check xawt/libmawt.so 5956 strcpy(libmawtpath, buf); 5957 strcat(libmawtpath, xawtstr); 5958 if (::stat(libmawtpath, &statbuf) == 0) return false; 5959 5960 // check libawt_xawt.so 5961 strcpy(libmawtpath, buf); 5962 strcat(libmawtpath, new_xawtstr); 5963 if (::stat(libmawtpath, &statbuf) == 0) return false; 5964 5965 return true; 5966} 5967 5968// Get the default path to the core file 5969// Returns the length of the string 5970int os::get_core_path(char* buffer, size_t bufferSize) { 5971 const char* p = get_current_directory(buffer, bufferSize); 5972 5973 if (p == NULL) { 5974 assert(p != NULL, "failed to get current directory"); 5975 return 0; 5976 } 5977 5978 return strlen(buffer); 5979} 5980 5981#ifdef JAVASE_EMBEDDED 5982// 5983// A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 5984// 5985MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 5986 5987// ctor 5988// 5989MemNotifyThread::MemNotifyThread(int fd): Thread() { 5990 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 5991 _fd = fd; 5992 5993 if (os::create_thread(this, os::os_thread)) { 5994 _memnotify_thread = this; 5995 os::set_priority(this, NearMaxPriority); 5996 os::start_thread(this); 5997 } 5998} 5999 6000// Where all the work gets done 6001// 6002void MemNotifyThread::run() { 6003 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 6004 6005 // Set up the select arguments 6006 fd_set rfds; 6007 if (_fd != -1) { 6008 FD_ZERO(&rfds); 6009 FD_SET(_fd, &rfds); 6010 } 6011 6012 // Now wait for the mem_notify device to wake up 6013 while (1) { 6014 // Wait for the mem_notify device to signal us.. 6015 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 6016 if (rc == -1) { 6017 perror("select!\n"); 6018 break; 6019 } else if (rc) { 6020 //ssize_t free_before = os::available_memory(); 6021 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 6022 6023 // The kernel is telling us there is not much memory left... 6024 // try to do something about that 6025 6026 // If we are not already in a GC, try one. 6027 if (!Universe::heap()->is_gc_active()) { 6028 Universe::heap()->collect(GCCause::_allocation_failure); 6029 6030 //ssize_t free_after = os::available_memory(); 6031 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 6032 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 6033 } 6034 // We might want to do something like the following if we find the GC's are not helping... 6035 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 6036 } 6037 } 6038} 6039 6040// See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 6041// 6042void MemNotifyThread::start() { 6043 int fd; 6044 fd = open("/dev/mem_notify", O_RDONLY, 0); 6045 if (fd < 0) { 6046 return; 6047 } 6048 6049 if (memnotify_thread() == NULL) { 6050 new MemNotifyThread(fd); 6051 } 6052} 6053 6054#endif // JAVASE_EMBEDDED 6055 6056 6057/////////////// Unit tests /////////////// 6058 6059#ifndef PRODUCT 6060 6061#define test_log(...) \ 6062 do { \ 6063 if (VerboseInternalVMTests) { \ 6064 tty->print_cr(__VA_ARGS__); \ 6065 tty->flush(); \ 6066 } \ 6067 } while (false) 6068 6069class TestReserveMemorySpecial : AllStatic { 6070 public: 6071 static void small_page_write(void* addr, size_t size) { 6072 size_t page_size = os::vm_page_size(); 6073 6074 char* end = (char*)addr + size; 6075 for (char* p = (char*)addr; p < end; p += page_size) { 6076 *p = 1; 6077 } 6078 } 6079 6080 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6081 if (!UseHugeTLBFS) { 6082 return; 6083 } 6084 6085 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6086 6087 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6088 6089 if (addr != NULL) { 6090 small_page_write(addr, size); 6091 6092 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6093 } 6094 } 6095 6096 static void test_reserve_memory_special_huge_tlbfs_only() { 6097 if (!UseHugeTLBFS) { 6098 return; 6099 } 6100 6101 size_t lp = os::large_page_size(); 6102 6103 for (size_t size = lp; size <= lp * 10; size += lp) { 6104 test_reserve_memory_special_huge_tlbfs_only(size); 6105 } 6106 } 6107 6108 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) { 6109 if (!UseHugeTLBFS) { 6110 return; 6111 } 6112 6113 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")", 6114 size, alignment); 6115 6116 assert(size >= os::large_page_size(), "Incorrect input to test"); 6117 6118 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6119 6120 if (addr != NULL) { 6121 small_page_write(addr, size); 6122 6123 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6124 } 6125 } 6126 6127 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) { 6128 size_t lp = os::large_page_size(); 6129 size_t ag = os::vm_allocation_granularity(); 6130 6131 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6132 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment); 6133 } 6134 } 6135 6136 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6137 size_t lp = os::large_page_size(); 6138 size_t ag = os::vm_allocation_granularity(); 6139 6140 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp); 6141 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag); 6142 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2); 6143 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2); 6144 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag); 6145 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag); 6146 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2); 6147 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10); 6148 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2); 6149 } 6150 6151 static void test_reserve_memory_special_huge_tlbfs() { 6152 if (!UseHugeTLBFS) { 6153 return; 6154 } 6155 6156 test_reserve_memory_special_huge_tlbfs_only(); 6157 test_reserve_memory_special_huge_tlbfs_mixed(); 6158 } 6159 6160 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6161 if (!UseSHM) { 6162 return; 6163 } 6164 6165 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6166 6167 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6168 6169 if (addr != NULL) { 6170 assert(is_ptr_aligned(addr, alignment), "Check"); 6171 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6172 6173 small_page_write(addr, size); 6174 6175 os::Linux::release_memory_special_shm(addr, size); 6176 } 6177 } 6178 6179 static void test_reserve_memory_special_shm() { 6180 size_t lp = os::large_page_size(); 6181 size_t ag = os::vm_allocation_granularity(); 6182 6183 for (size_t size = ag; size < lp * 3; size += ag) { 6184 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6185 test_reserve_memory_special_shm(size, alignment); 6186 } 6187 } 6188 } 6189 6190 static void test() { 6191 test_reserve_memory_special_huge_tlbfs(); 6192 test_reserve_memory_special_shm(); 6193 } 6194}; 6195 6196void TestReserveMemorySpecial_test() { 6197 TestReserveMemorySpecial::test(); 6198} 6199 6200#endif 6201