duke@435: /* duke@435: * Copyright 1999-2007 Sun Microsystems, Inc. All Rights Reserved. duke@435: * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. duke@435: * duke@435: * This code is free software; you can redistribute it and/or modify it duke@435: * under the terms of the GNU General Public License version 2 only, as duke@435: * published by the Free Software Foundation. duke@435: * duke@435: * This code is distributed in the hope that it will be useful, but WITHOUT duke@435: * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or duke@435: * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License duke@435: * version 2 for more details (a copy is included in the LICENSE file that duke@435: * accompanied this code). duke@435: * duke@435: * You should have received a copy of the GNU General Public License version duke@435: * 2 along with this work; if not, write to the Free Software Foundation, duke@435: * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. duke@435: * duke@435: * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, duke@435: * CA 95054 USA or visit www.sun.com if you need additional information or duke@435: * have any questions. duke@435: * duke@435: */ duke@435: duke@435: // do not include precompiled header file duke@435: # include "incls/_os_linux.cpp.incl" duke@435: duke@435: // put OS-includes here duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: # include duke@435: duke@435: #define MAX_PATH (2 * K) duke@435: duke@435: // for timer info max values which include all bits duke@435: #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) duke@435: #define SEC_IN_NANOSECS 1000000000LL duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // global variables duke@435: julong os::Linux::_physical_memory = 0; duke@435: duke@435: address os::Linux::_initial_thread_stack_bottom = NULL; duke@435: uintptr_t os::Linux::_initial_thread_stack_size = 0; duke@435: duke@435: int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; duke@435: int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; duke@435: Mutex* os::Linux::_createThread_lock = NULL; duke@435: pthread_t os::Linux::_main_thread; duke@435: int os::Linux::_page_size = -1; duke@435: bool os::Linux::_is_floating_stack = false; duke@435: bool os::Linux::_is_NPTL = false; duke@435: bool os::Linux::_supports_fast_thread_cpu_time = false; duke@435: char * os::Linux::_glibc_version = NULL; duke@435: char * os::Linux::_libpthread_version = NULL; duke@435: duke@435: static jlong initial_time_count=0; duke@435: duke@435: static int clock_tics_per_sec = 100; duke@435: duke@435: // For diagnostics to print a message once. see run_periodic_checks duke@435: static sigset_t check_signal_done; duke@435: static bool check_signals = true;; duke@435: duke@435: static pid_t _initial_pid = 0; duke@435: duke@435: /* Signal number used to suspend/resume a thread */ duke@435: duke@435: /* do not use any signal number less than SIGSEGV, see 4355769 */ duke@435: static int SR_signum = SIGUSR2; duke@435: sigset_t SR_sigset; duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // utility functions duke@435: duke@435: static int SR_initialize(); duke@435: static int SR_finalize(); duke@435: duke@435: julong os::available_memory() { duke@435: return Linux::available_memory(); duke@435: } duke@435: duke@435: julong os::Linux::available_memory() { duke@435: // values in struct sysinfo are "unsigned long" duke@435: struct sysinfo si; duke@435: sysinfo(&si); duke@435: duke@435: return (julong)si.freeram * si.mem_unit; duke@435: } duke@435: duke@435: julong os::physical_memory() { duke@435: return Linux::physical_memory(); duke@435: } duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // environment support duke@435: duke@435: bool os::getenv(const char* name, char* buf, int len) { duke@435: const char* val = ::getenv(name); duke@435: if (val != NULL && strlen(val) < (size_t)len) { duke@435: strcpy(buf, val); duke@435: return true; duke@435: } duke@435: if (len > 0) buf[0] = 0; // return a null string duke@435: return false; duke@435: } duke@435: duke@435: duke@435: // Return true if user is running as root. duke@435: duke@435: bool os::have_special_privileges() { duke@435: static bool init = false; duke@435: static bool privileges = false; duke@435: if (!init) { duke@435: privileges = (getuid() != geteuid()) || (getgid() != getegid()); duke@435: init = true; duke@435: } duke@435: return privileges; duke@435: } duke@435: duke@435: duke@435: #ifndef SYS_gettid duke@435: // i386: 224, ia64: 1105, amd64: 186, sparc 143 duke@435: #ifdef __ia64__ duke@435: #define SYS_gettid 1105 duke@435: #elif __i386__ duke@435: #define SYS_gettid 224 duke@435: #elif __amd64__ duke@435: #define SYS_gettid 186 duke@435: #elif __sparc__ duke@435: #define SYS_gettid 143 duke@435: #else duke@435: #error define gettid for the arch duke@435: #endif duke@435: #endif duke@435: duke@435: // Cpu architecture string duke@435: #if defined(IA64) duke@435: static char cpu_arch[] = "ia64"; duke@435: #elif defined(IA32) duke@435: static char cpu_arch[] = "i386"; duke@435: #elif defined(AMD64) duke@435: static char cpu_arch[] = "amd64"; duke@435: #elif defined(SPARC) duke@435: # ifdef _LP64 duke@435: static char cpu_arch[] = "sparcv9"; duke@435: # else duke@435: static char cpu_arch[] = "sparc"; duke@435: # endif duke@435: #else duke@435: #error Add appropriate cpu_arch setting duke@435: #endif duke@435: duke@435: duke@435: // pid_t gettid() duke@435: // duke@435: // Returns the kernel thread id of the currently running thread. Kernel duke@435: // thread id is used to access /proc. duke@435: // duke@435: // (Note that getpid() on LinuxThreads returns kernel thread id too; but duke@435: // on NPTL, it returns the same pid for all threads, as required by POSIX.) duke@435: // duke@435: pid_t os::Linux::gettid() { duke@435: int rslt = syscall(SYS_gettid); duke@435: if (rslt == -1) { duke@435: // old kernel, no NPTL support duke@435: return getpid(); duke@435: } else { duke@435: return (pid_t)rslt; duke@435: } duke@435: } duke@435: duke@435: // Most versions of linux have a bug where the number of processors are duke@435: // determined by looking at the /proc file system. In a chroot environment, duke@435: // the system call returns 1. This causes the VM to act as if it is duke@435: // a single processor and elide locking (see is_MP() call). duke@435: static bool unsafe_chroot_detected = false; duke@435: static char *unstable_chroot_error = "/proc file system not found.\n" duke@435: "Java may be unstable running multithreaded in a chroot " duke@435: "environment on Linux when /proc filesystem is not mounted."; duke@435: duke@435: void os::Linux::initialize_system_info() { duke@435: _processor_count = sysconf(_SC_NPROCESSORS_CONF); duke@435: if (_processor_count == 1) { duke@435: pid_t pid = os::Linux::gettid(); duke@435: char fname[32]; duke@435: jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); duke@435: FILE *fp = fopen(fname, "r"); duke@435: if (fp == NULL) { duke@435: unsafe_chroot_detected = true; duke@435: } else { duke@435: fclose(fp); duke@435: } duke@435: } duke@435: _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); duke@435: assert(_processor_count > 0, "linux error"); duke@435: } duke@435: duke@435: void os::init_system_properties_values() { duke@435: // char arch[12]; duke@435: // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); duke@435: duke@435: // The next steps are taken in the product version: duke@435: // duke@435: // Obtain the JAVA_HOME value from the location of libjvm[_g].so. duke@435: // This library should be located at: duke@435: // /jre/lib//{client|server}/libjvm[_g].so. duke@435: // duke@435: // If "/jre/lib/" appears at the right place in the path, then we duke@435: // assume libjvm[_g].so is installed in a JDK and we use this path. duke@435: // duke@435: // Otherwise exit with message: "Could not create the Java virtual machine." duke@435: // duke@435: // The following extra steps are taken in the debugging version: duke@435: // duke@435: // If "/jre/lib/" does NOT appear at the right place in the path duke@435: // instead of exit check for $JAVA_HOME environment variable. duke@435: // duke@435: // If it is defined and we are able to locate $JAVA_HOME/jre/lib/, duke@435: // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so duke@435: // it looks like libjvm[_g].so is installed there duke@435: // /jre/lib//hotspot/libjvm[_g].so. duke@435: // duke@435: // Otherwise exit. duke@435: // duke@435: // Important note: if the location of libjvm.so changes this duke@435: // code needs to be changed accordingly. duke@435: duke@435: // The next few definitions allow the code to be verbatim: duke@435: #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n)) duke@435: #define getenv(n) ::getenv(n) duke@435: duke@435: /* duke@435: * See ld(1): duke@435: * The linker uses the following search paths to locate required duke@435: * shared libraries: duke@435: * 1: ... duke@435: * ... duke@435: * 7: The default directories, normally /lib and /usr/lib. duke@435: */ duke@435: #define DEFAULT_LIBPATH "/lib:/usr/lib" duke@435: duke@435: #define EXTENSIONS_DIR "/lib/ext" duke@435: #define ENDORSED_DIR "/lib/endorsed" duke@435: #define REG_DIR "/usr/java/packages" duke@435: duke@435: { duke@435: /* sysclasspath, java_home, dll_dir */ duke@435: { duke@435: char *home_path; duke@435: char *dll_path; duke@435: char *pslash; duke@435: char buf[MAXPATHLEN]; duke@435: os::jvm_path(buf, sizeof(buf)); duke@435: duke@435: // Found the full path to libjvm.so. duke@435: // Now cut the path to /jre if we can. duke@435: *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ duke@435: pslash = strrchr(buf, '/'); duke@435: if (pslash != NULL) duke@435: *pslash = '\0'; /* get rid of /{client|server|hotspot} */ duke@435: dll_path = malloc(strlen(buf) + 1); duke@435: if (dll_path == NULL) duke@435: return; duke@435: strcpy(dll_path, buf); duke@435: Arguments::set_dll_dir(dll_path); duke@435: duke@435: if (pslash != NULL) { duke@435: pslash = strrchr(buf, '/'); duke@435: if (pslash != NULL) { duke@435: *pslash = '\0'; /* get rid of / */ duke@435: pslash = strrchr(buf, '/'); duke@435: if (pslash != NULL) duke@435: *pslash = '\0'; /* get rid of /lib */ duke@435: } duke@435: } duke@435: duke@435: home_path = malloc(strlen(buf) + 1); duke@435: if (home_path == NULL) duke@435: return; duke@435: strcpy(home_path, buf); duke@435: Arguments::set_java_home(home_path); duke@435: duke@435: if (!set_boot_path('/', ':')) duke@435: return; duke@435: } duke@435: duke@435: /* duke@435: * Where to look for native libraries duke@435: * duke@435: * Note: Due to a legacy implementation, most of the library path duke@435: * is set in the launcher. This was to accomodate linking restrictions duke@435: * on legacy Linux implementations (which are no longer supported). duke@435: * Eventually, all the library path setting will be done here. duke@435: * duke@435: * However, to prevent the proliferation of improperly built native duke@435: * libraries, the new path component /usr/java/packages is added here. duke@435: * Eventually, all the library path setting will be done here. duke@435: */ duke@435: { duke@435: char *ld_library_path; duke@435: duke@435: /* duke@435: * Construct the invariant part of ld_library_path. Note that the duke@435: * space for the colon and the trailing null are provided by the duke@435: * nulls included by the sizeof operator (so actually we allocate duke@435: * a byte more than necessary). duke@435: */ duke@435: ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + duke@435: strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); duke@435: sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); duke@435: duke@435: /* duke@435: * Get the user setting of LD_LIBRARY_PATH, and prepended it. It duke@435: * should always exist (until the legacy problem cited above is duke@435: * addressed). duke@435: */ duke@435: char *v = getenv("LD_LIBRARY_PATH"); duke@435: if (v != NULL) { duke@435: char *t = ld_library_path; duke@435: /* That's +1 for the colon and +1 for the trailing '\0' */ duke@435: ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); duke@435: sprintf(ld_library_path, "%s:%s", v, t); duke@435: } duke@435: Arguments::set_library_path(ld_library_path); duke@435: } duke@435: duke@435: /* duke@435: * Extensions directories. duke@435: * duke@435: * Note that the space for the colon and the trailing null are provided duke@435: * by the nulls included by the sizeof operator (so actually one byte more duke@435: * than necessary is allocated). duke@435: */ duke@435: { duke@435: char *buf = malloc(strlen(Arguments::get_java_home()) + duke@435: sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); duke@435: sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, duke@435: Arguments::get_java_home()); duke@435: Arguments::set_ext_dirs(buf); duke@435: } duke@435: duke@435: /* Endorsed standards default directory. */ duke@435: { duke@435: char * buf; duke@435: buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); duke@435: sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); duke@435: Arguments::set_endorsed_dirs(buf); duke@435: } duke@435: } duke@435: duke@435: #undef malloc duke@435: #undef getenv duke@435: #undef EXTENSIONS_DIR duke@435: #undef ENDORSED_DIR duke@435: duke@435: // Done duke@435: return; duke@435: } duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // breakpoint support duke@435: duke@435: void os::breakpoint() { duke@435: BREAKPOINT; duke@435: } duke@435: duke@435: extern "C" void breakpoint() { duke@435: // use debugger to set breakpoint here duke@435: } duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // signal support duke@435: duke@435: debug_only(static bool signal_sets_initialized = false); duke@435: static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; duke@435: duke@435: bool os::Linux::is_sig_ignored(int sig) { duke@435: struct sigaction oact; duke@435: sigaction(sig, (struct sigaction*)NULL, &oact); duke@435: void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) duke@435: : CAST_FROM_FN_PTR(void*, oact.sa_handler); duke@435: if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) duke@435: return true; duke@435: else duke@435: return false; duke@435: } duke@435: duke@435: void os::Linux::signal_sets_init() { duke@435: // Should also have an assertion stating we are still single-threaded. duke@435: assert(!signal_sets_initialized, "Already initialized"); duke@435: // Fill in signals that are necessarily unblocked for all threads in duke@435: // the VM. Currently, we unblock the following signals: duke@435: // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden duke@435: // by -Xrs (=ReduceSignalUsage)); duke@435: // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all duke@435: // other threads. The "ReduceSignalUsage" boolean tells us not to alter duke@435: // the dispositions or masks wrt these signals. duke@435: // Programs embedding the VM that want to use the above signals for their duke@435: // own purposes must, at this time, use the "-Xrs" option to prevent duke@435: // interference with shutdown hooks and BREAK_SIGNAL thread dumping. duke@435: // (See bug 4345157, and other related bugs). duke@435: // In reality, though, unblocking these signals is really a nop, since duke@435: // these signals are not blocked by default. duke@435: sigemptyset(&unblocked_sigs); duke@435: sigemptyset(&allowdebug_blocked_sigs); duke@435: sigaddset(&unblocked_sigs, SIGILL); duke@435: sigaddset(&unblocked_sigs, SIGSEGV); duke@435: sigaddset(&unblocked_sigs, SIGBUS); duke@435: sigaddset(&unblocked_sigs, SIGFPE); duke@435: sigaddset(&unblocked_sigs, SR_signum); duke@435: duke@435: if (!ReduceSignalUsage) { duke@435: if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { duke@435: sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); duke@435: sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); duke@435: } duke@435: if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { duke@435: sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); duke@435: sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); duke@435: } duke@435: if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { duke@435: sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); duke@435: sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); duke@435: } duke@435: } duke@435: // Fill in signals that are blocked by all but the VM thread. duke@435: sigemptyset(&vm_sigs); duke@435: if (!ReduceSignalUsage) duke@435: sigaddset(&vm_sigs, BREAK_SIGNAL); duke@435: debug_only(signal_sets_initialized = true); duke@435: duke@435: } duke@435: duke@435: // These are signals that are unblocked while a thread is running Java. duke@435: // (For some reason, they get blocked by default.) duke@435: sigset_t* os::Linux::unblocked_signals() { duke@435: assert(signal_sets_initialized, "Not initialized"); duke@435: return &unblocked_sigs; duke@435: } duke@435: duke@435: // These are the signals that are blocked while a (non-VM) thread is duke@435: // running Java. Only the VM thread handles these signals. duke@435: sigset_t* os::Linux::vm_signals() { duke@435: assert(signal_sets_initialized, "Not initialized"); duke@435: return &vm_sigs; duke@435: } duke@435: duke@435: // These are signals that are blocked during cond_wait to allow debugger in duke@435: sigset_t* os::Linux::allowdebug_blocked_signals() { duke@435: assert(signal_sets_initialized, "Not initialized"); duke@435: return &allowdebug_blocked_sigs; duke@435: } duke@435: duke@435: void os::Linux::hotspot_sigmask(Thread* thread) { duke@435: duke@435: //Save caller's signal mask before setting VM signal mask duke@435: sigset_t caller_sigmask; duke@435: pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); duke@435: duke@435: OSThread* osthread = thread->osthread(); duke@435: osthread->set_caller_sigmask(caller_sigmask); duke@435: duke@435: pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); duke@435: duke@435: if (!ReduceSignalUsage) { duke@435: if (thread->is_VM_thread()) { duke@435: // Only the VM thread handles BREAK_SIGNAL ... duke@435: pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); duke@435: } else { duke@435: // ... all other threads block BREAK_SIGNAL duke@435: pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); duke@435: } duke@435: } duke@435: } duke@435: duke@435: ////////////////////////////////////////////////////////////////////////////// duke@435: // detecting pthread library duke@435: duke@435: void os::Linux::libpthread_init() { duke@435: // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION duke@435: // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a duke@435: // generic name for earlier versions. duke@435: // Define macros here so we can build HotSpot on old systems. duke@435: # ifndef _CS_GNU_LIBC_VERSION duke@435: # define _CS_GNU_LIBC_VERSION 2 duke@435: # endif duke@435: # ifndef _CS_GNU_LIBPTHREAD_VERSION duke@435: # define _CS_GNU_LIBPTHREAD_VERSION 3 duke@435: # endif duke@435: duke@435: size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); duke@435: if (n > 0) { duke@435: char *str = (char *)malloc(n); duke@435: confstr(_CS_GNU_LIBC_VERSION, str, n); duke@435: os::Linux::set_glibc_version(str); duke@435: } else { duke@435: // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() duke@435: static char _gnu_libc_version[32]; duke@435: jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), duke@435: "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); duke@435: os::Linux::set_glibc_version(_gnu_libc_version); duke@435: } duke@435: duke@435: n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); duke@435: if (n > 0) { duke@435: char *str = (char *)malloc(n); duke@435: confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); duke@435: duke@435: // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells duke@435: // us "NPTL-0.29" even we are running with LinuxThreads. Check if this duke@435: // is the case: duke@435: if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && duke@435: strstr(str, "NPTL")) { duke@435: // LinuxThreads has a hard limit on max number of threads. So duke@435: // sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. duke@435: // On the other hand, NPTL does not have such a limit, sysconf() duke@435: // will return -1 and errno is not changed. Check if it is really duke@435: // NPTL: duke@435: if (sysconf(_SC_THREAD_THREADS_MAX) > 0) { duke@435: free(str); duke@435: str = "linuxthreads"; duke@435: } duke@435: } duke@435: os::Linux::set_libpthread_version(str); duke@435: } else { duke@435: // glibc before 2.3.2 only has LinuxThreads. duke@435: os::Linux::set_libpthread_version("linuxthreads"); duke@435: } duke@435: duke@435: if (strstr(libpthread_version(), "NPTL")) { duke@435: os::Linux::set_is_NPTL(); duke@435: } else { duke@435: os::Linux::set_is_LinuxThreads(); duke@435: } duke@435: duke@435: // LinuxThreads have two flavors: floating-stack mode, which allows variable duke@435: // stack size; and fixed-stack mode. NPTL is always floating-stack. duke@435: if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { duke@435: os::Linux::set_is_floating_stack(); duke@435: } duke@435: } duke@435: duke@435: ///////////////////////////////////////////////////////////////////////////// duke@435: // thread stack duke@435: duke@435: // Force Linux kernel to expand current thread stack. If "bottom" is close duke@435: // to the stack guard, caller should block all signals. duke@435: // duke@435: // MAP_GROWSDOWN: duke@435: // A special mmap() flag that is used to implement thread stacks. It tells duke@435: // kernel that the memory region should extend downwards when needed. This duke@435: // allows early versions of LinuxThreads to only mmap the first few pages duke@435: // when creating a new thread. Linux kernel will automatically expand thread duke@435: // stack as needed (on page faults). duke@435: // duke@435: // However, because the memory region of a MAP_GROWSDOWN stack can grow on duke@435: // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN duke@435: // region, it's hard to tell if the fault is due to a legitimate stack duke@435: // access or because of reading/writing non-exist memory (e.g. buffer duke@435: // overrun). As a rule, if the fault happens below current stack pointer, duke@435: // Linux kernel does not expand stack, instead a SIGSEGV is sent to the duke@435: // application (see Linux kernel fault.c). duke@435: // duke@435: // This Linux feature can cause SIGSEGV when VM bangs thread stack for duke@435: // stack overflow detection. duke@435: // duke@435: // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do duke@435: // not use this flag. However, the stack of initial thread is not created duke@435: // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though duke@435: // unlikely) that user code can create a thread with MAP_GROWSDOWN stack duke@435: // and then attach the thread to JVM. duke@435: // duke@435: // To get around the problem and allow stack banging on Linux, we need to duke@435: // manually expand thread stack after receiving the SIGSEGV. duke@435: // duke@435: // There are two ways to expand thread stack to address "bottom", we used duke@435: // both of them in JVM before 1.5: duke@435: // 1. adjust stack pointer first so that it is below "bottom", and then duke@435: // touch "bottom" duke@435: // 2. mmap() the page in question duke@435: // duke@435: // Now alternate signal stack is gone, it's harder to use 2. For instance, duke@435: // if current sp is already near the lower end of page 101, and we need to duke@435: // call mmap() to map page 100, it is possible that part of the mmap() frame duke@435: // will be placed in page 100. When page 100 is mapped, it is zero-filled. duke@435: // That will destroy the mmap() frame and cause VM to crash. duke@435: // duke@435: // The following code works by adjusting sp first, then accessing the "bottom" duke@435: // page to force a page fault. Linux kernel will then automatically expand the duke@435: // stack mapping. duke@435: // duke@435: // _expand_stack_to() assumes its frame size is less than page size, which duke@435: // should always be true if the function is not inlined. duke@435: duke@435: #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute duke@435: #define NOINLINE duke@435: #else duke@435: #define NOINLINE __attribute__ ((noinline)) duke@435: #endif duke@435: duke@435: static void _expand_stack_to(address bottom) NOINLINE; duke@435: duke@435: static void _expand_stack_to(address bottom) { duke@435: address sp; duke@435: size_t size; duke@435: volatile char *p; duke@435: duke@435: // Adjust bottom to point to the largest address within the same page, it duke@435: // gives us a one-page buffer if alloca() allocates slightly more memory. duke@435: bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); duke@435: bottom += os::Linux::page_size() - 1; duke@435: duke@435: // sp might be slightly above current stack pointer; if that's the case, we duke@435: // will alloca() a little more space than necessary, which is OK. Don't use duke@435: // os::current_stack_pointer(), as its result can be slightly below current duke@435: // stack pointer, causing us to not alloca enough to reach "bottom". duke@435: sp = (address)&sp; duke@435: duke@435: if (sp > bottom) { duke@435: size = sp - bottom; duke@435: p = (volatile char *)alloca(size); duke@435: assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); duke@435: p[0] = '\0'; duke@435: } duke@435: } duke@435: duke@435: bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { duke@435: assert(t!=NULL, "just checking"); duke@435: assert(t->osthread()->expanding_stack(), "expand should be set"); duke@435: assert(t->stack_base() != NULL, "stack_base was not initialized"); duke@435: duke@435: if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { duke@435: sigset_t mask_all, old_sigset; duke@435: sigfillset(&mask_all); duke@435: pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); duke@435: _expand_stack_to(addr); duke@435: pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); duke@435: return true; duke@435: } duke@435: return false; duke@435: } duke@435: duke@435: ////////////////////////////////////////////////////////////////////////////// duke@435: // create new thread duke@435: duke@435: static address highest_vm_reserved_address(); duke@435: duke@435: // check if it's safe to start a new thread duke@435: static bool _thread_safety_check(Thread* thread) { duke@435: if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { duke@435: // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) duke@435: // Heap is mmap'ed at lower end of memory space. Thread stacks are duke@435: // allocated (MAP_FIXED) from high address space. Every thread stack duke@435: // occupies a fixed size slot (usually 2Mbytes, but user can change duke@435: // it to other values if they rebuild LinuxThreads). duke@435: // duke@435: // Problem with MAP_FIXED is that mmap() can still succeed even part of duke@435: // the memory region has already been mmap'ed. That means if we have too duke@435: // many threads and/or very large heap, eventually thread stack will duke@435: // collide with heap. duke@435: // duke@435: // Here we try to prevent heap/stack collision by comparing current duke@435: // stack bottom with the highest address that has been mmap'ed by JVM duke@435: // plus a safety margin for memory maps created by native code. duke@435: // duke@435: // This feature can be disabled by setting ThreadSafetyMargin to 0 duke@435: // duke@435: if (ThreadSafetyMargin > 0) { duke@435: address stack_bottom = os::current_stack_base() - os::current_stack_size(); duke@435: duke@435: // not safe if our stack extends below the safety margin duke@435: return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); duke@435: } else { duke@435: return true; duke@435: } duke@435: } else { duke@435: // Floating stack LinuxThreads or NPTL: duke@435: // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When duke@435: // there's not enough space left, pthread_create() will fail. If we come duke@435: // here, that means enough space has been reserved for stack. duke@435: return true; duke@435: } duke@435: } duke@435: duke@435: // Thread start routine for all newly created threads duke@435: static void *java_start(Thread *thread) { duke@435: // Try to randomize the cache line index of hot stack frames. duke@435: // This helps when threads of the same stack traces evict each other's duke@435: // cache lines. The threads can be either from the same JVM instance, or duke@435: // from different JVM instances. The benefit is especially true for duke@435: // processors with hyperthreading technology. duke@435: static int counter = 0; duke@435: int pid = os::current_process_id(); duke@435: alloca(((pid ^ counter++) & 7) * 128); duke@435: duke@435: ThreadLocalStorage::set_thread(thread); duke@435: duke@435: OSThread* osthread = thread->osthread(); duke@435: Monitor* sync = osthread->startThread_lock(); duke@435: duke@435: // non floating stack LinuxThreads needs extra check, see above duke@435: if (!_thread_safety_check(thread)) { duke@435: // notify parent thread duke@435: MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); duke@435: osthread->set_state(ZOMBIE); duke@435: sync->notify_all(); duke@435: return NULL; duke@435: } duke@435: duke@435: // thread_id is kernel thread id (similar to Solaris LWP id) duke@435: osthread->set_thread_id(os::Linux::gettid()); duke@435: duke@435: if (UseNUMA) { duke@435: int lgrp_id = os::numa_get_group_id(); duke@435: if (lgrp_id != -1) { duke@435: thread->set_lgrp_id(lgrp_id); duke@435: } duke@435: } duke@435: // initialize signal mask for this thread duke@435: os::Linux::hotspot_sigmask(thread); duke@435: duke@435: // initialize floating point control register duke@435: os::Linux::init_thread_fpu_state(); duke@435: duke@435: // handshaking with parent thread duke@435: { duke@435: MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); duke@435: duke@435: // notify parent thread duke@435: osthread->set_state(INITIALIZED); duke@435: sync->notify_all(); duke@435: duke@435: // wait until os::start_thread() duke@435: while (osthread->get_state() == INITIALIZED) { duke@435: sync->wait(Mutex::_no_safepoint_check_flag); duke@435: } duke@435: } duke@435: duke@435: // call one more level start routine duke@435: thread->run(); duke@435: duke@435: return 0; duke@435: } duke@435: duke@435: bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { duke@435: assert(thread->osthread() == NULL, "caller responsible"); duke@435: duke@435: // Allocate the OSThread object duke@435: OSThread* osthread = new OSThread(NULL, NULL); duke@435: if (osthread == NULL) { duke@435: return false; duke@435: } duke@435: duke@435: // set the correct thread state duke@435: osthread->set_thread_type(thr_type); duke@435: duke@435: // Initial state is ALLOCATED but not INITIALIZED duke@435: osthread->set_state(ALLOCATED); duke@435: duke@435: thread->set_osthread(osthread); duke@435: duke@435: // init thread attributes duke@435: pthread_attr_t attr; duke@435: pthread_attr_init(&attr); duke@435: pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); duke@435: duke@435: // stack size duke@435: if (os::Linux::supports_variable_stack_size()) { duke@435: // calculate stack size if it's not specified by caller duke@435: if (stack_size == 0) { duke@435: stack_size = os::Linux::default_stack_size(thr_type); duke@435: duke@435: switch (thr_type) { duke@435: case os::java_thread: duke@435: // Java threads use ThreadStackSize which default value can be changed with the flag -Xss duke@435: if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create(); duke@435: break; duke@435: case os::compiler_thread: duke@435: if (CompilerThreadStackSize > 0) { duke@435: stack_size = (size_t)(CompilerThreadStackSize * K); duke@435: break; duke@435: } // else fall through: duke@435: // use VMThreadStackSize if CompilerThreadStackSize is not defined duke@435: case os::vm_thread: duke@435: case os::pgc_thread: duke@435: case os::cgc_thread: duke@435: case os::watcher_thread: duke@435: if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); duke@435: break; duke@435: } duke@435: } duke@435: duke@435: stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); duke@435: pthread_attr_setstacksize(&attr, stack_size); duke@435: } else { duke@435: // let pthread_create() pick the default value. duke@435: } duke@435: duke@435: // glibc guard page duke@435: pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); duke@435: duke@435: ThreadState state; duke@435: duke@435: { duke@435: // Serialize thread creation if we are running with fixed stack LinuxThreads duke@435: bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); duke@435: if (lock) { duke@435: os::Linux::createThread_lock()->lock_without_safepoint_check(); duke@435: } duke@435: duke@435: pthread_t tid; duke@435: int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); duke@435: duke@435: pthread_attr_destroy(&attr); duke@435: duke@435: if (ret != 0) { duke@435: if (PrintMiscellaneous && (Verbose || WizardMode)) { duke@435: perror("pthread_create()"); duke@435: } duke@435: // Need to clean up stuff we've allocated so far duke@435: thread->set_osthread(NULL); duke@435: delete osthread; duke@435: if (lock) os::Linux::createThread_lock()->unlock(); duke@435: return false; duke@435: } duke@435: duke@435: // Store pthread info into the OSThread duke@435: osthread->set_pthread_id(tid); duke@435: duke@435: // Wait until child thread is either initialized or aborted duke@435: { duke@435: Monitor* sync_with_child = osthread->startThread_lock(); duke@435: MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); duke@435: while ((state = osthread->get_state()) == ALLOCATED) { duke@435: sync_with_child->wait(Mutex::_no_safepoint_check_flag); duke@435: } duke@435: } duke@435: duke@435: if (lock) { duke@435: os::Linux::createThread_lock()->unlock(); duke@435: } duke@435: } duke@435: duke@435: // Aborted due to thread limit being reached duke@435: if (state == ZOMBIE) { duke@435: thread->set_osthread(NULL); duke@435: delete osthread; duke@435: return false; duke@435: } duke@435: duke@435: // The thread is returned suspended (in state INITIALIZED), duke@435: // and is started higher up in the call chain duke@435: assert(state == INITIALIZED, "race condition"); duke@435: return true; duke@435: } duke@435: duke@435: ///////////////////////////////////////////////////////////////////////////// duke@435: // attach existing thread duke@435: duke@435: // bootstrap the main thread duke@435: bool os::create_main_thread(JavaThread* thread) { duke@435: assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); duke@435: return create_attached_thread(thread); duke@435: } duke@435: duke@435: bool os::create_attached_thread(JavaThread* thread) { duke@435: #ifdef ASSERT duke@435: thread->verify_not_published(); duke@435: #endif duke@435: duke@435: // Allocate the OSThread object duke@435: OSThread* osthread = new OSThread(NULL, NULL); duke@435: duke@435: if (osthread == NULL) { duke@435: return false; duke@435: } duke@435: duke@435: // Store pthread info into the OSThread duke@435: osthread->set_thread_id(os::Linux::gettid()); duke@435: osthread->set_pthread_id(::pthread_self()); duke@435: duke@435: // initialize floating point control register duke@435: os::Linux::init_thread_fpu_state(); duke@435: duke@435: // Initial thread state is RUNNABLE duke@435: osthread->set_state(RUNNABLE); duke@435: duke@435: thread->set_osthread(osthread); duke@435: duke@435: if (UseNUMA) { duke@435: int lgrp_id = os::numa_get_group_id(); duke@435: if (lgrp_id != -1) { duke@435: thread->set_lgrp_id(lgrp_id); duke@435: } duke@435: } duke@435: duke@435: if (os::Linux::is_initial_thread()) { duke@435: // If current thread is initial thread, its stack is mapped on demand, duke@435: // see notes about MAP_GROWSDOWN. Here we try to force kernel to map duke@435: // the entire stack region to avoid SEGV in stack banging. duke@435: // It is also useful to get around the heap-stack-gap problem on SuSE duke@435: // kernel (see 4821821 for details). We first expand stack to the top duke@435: // of yellow zone, then enable stack yellow zone (order is significant, duke@435: // enabling yellow zone first will crash JVM on SuSE Linux), so there duke@435: // is no gap between the last two virtual memory regions. duke@435: duke@435: JavaThread *jt = (JavaThread *)thread; duke@435: address addr = jt->stack_yellow_zone_base(); duke@435: assert(addr != NULL, "initialization problem?"); duke@435: assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); duke@435: duke@435: osthread->set_expanding_stack(); duke@435: os::Linux::manually_expand_stack(jt, addr); duke@435: osthread->clear_expanding_stack(); duke@435: } duke@435: duke@435: // initialize signal mask for this thread duke@435: // and save the caller's signal mask duke@435: os::Linux::hotspot_sigmask(thread); duke@435: duke@435: return true; duke@435: } duke@435: duke@435: void os::pd_start_thread(Thread* thread) { duke@435: OSThread * osthread = thread->osthread(); duke@435: assert(osthread->get_state() != INITIALIZED, "just checking"); duke@435: Monitor* sync_with_child = osthread->startThread_lock(); duke@435: MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); duke@435: sync_with_child->notify(); duke@435: } duke@435: duke@435: // Free Linux resources related to the OSThread duke@435: void os::free_thread(OSThread* osthread) { duke@435: assert(osthread != NULL, "osthread not set"); duke@435: duke@435: if (Thread::current()->osthread() == osthread) { duke@435: // Restore caller's signal mask duke@435: sigset_t sigmask = osthread->caller_sigmask(); duke@435: pthread_sigmask(SIG_SETMASK, &sigmask, NULL); duke@435: } duke@435: duke@435: delete osthread; duke@435: } duke@435: duke@435: ////////////////////////////////////////////////////////////////////////////// duke@435: // thread local storage duke@435: duke@435: int os::allocate_thread_local_storage() { duke@435: pthread_key_t key; duke@435: int rslt = pthread_key_create(&key, NULL); duke@435: assert(rslt == 0, "cannot allocate thread local storage"); duke@435: return (int)key; duke@435: } duke@435: duke@435: // Note: This is currently not used by VM, as we don't destroy TLS key duke@435: // on VM exit. duke@435: void os::free_thread_local_storage(int index) { duke@435: int rslt = pthread_key_delete((pthread_key_t)index); duke@435: assert(rslt == 0, "invalid index"); duke@435: } duke@435: duke@435: void os::thread_local_storage_at_put(int index, void* value) { duke@435: int rslt = pthread_setspecific((pthread_key_t)index, value); duke@435: assert(rslt == 0, "pthread_setspecific failed"); duke@435: } duke@435: duke@435: extern "C" Thread* get_thread() { duke@435: return ThreadLocalStorage::thread(); duke@435: } duke@435: duke@435: ////////////////////////////////////////////////////////////////////////////// duke@435: // initial thread duke@435: duke@435: // Check if current thread is the initial thread, similar to Solaris thr_main. duke@435: bool os::Linux::is_initial_thread(void) { duke@435: char dummy; duke@435: // If called before init complete, thread stack bottom will be null. duke@435: // Can be called if fatal error occurs before initialization. duke@435: if (initial_thread_stack_bottom() == NULL) return false; duke@435: assert(initial_thread_stack_bottom() != NULL && duke@435: initial_thread_stack_size() != 0, duke@435: "os::init did not locate initial thread's stack region"); duke@435: if ((address)&dummy >= initial_thread_stack_bottom() && duke@435: (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) duke@435: return true; duke@435: else return false; duke@435: } duke@435: duke@435: // Find the virtual memory area that contains addr duke@435: static bool find_vma(address addr, address* vma_low, address* vma_high) { duke@435: FILE *fp = fopen("/proc/self/maps", "r"); duke@435: if (fp) { duke@435: address low, high; duke@435: while (!feof(fp)) { duke@435: if (fscanf(fp, "%p-%p", &low, &high) == 2) { duke@435: if (low <= addr && addr < high) { duke@435: if (vma_low) *vma_low = low; duke@435: if (vma_high) *vma_high = high; duke@435: fclose (fp); duke@435: return true; duke@435: } duke@435: } duke@435: for (;;) { duke@435: int ch = fgetc(fp); duke@435: if (ch == EOF || ch == (int)'\n') break; duke@435: } duke@435: } duke@435: fclose(fp); duke@435: } duke@435: return false; duke@435: } duke@435: duke@435: // Locate initial thread stack. This special handling of initial thread stack duke@435: // is needed because pthread_getattr_np() on most (all?) Linux distros returns duke@435: // bogus value for initial thread. duke@435: void os::Linux::capture_initial_stack(size_t max_size) { duke@435: // stack size is the easy part, get it from RLIMIT_STACK duke@435: size_t stack_size; duke@435: struct rlimit rlim; duke@435: getrlimit(RLIMIT_STACK, &rlim); duke@435: stack_size = rlim.rlim_cur; duke@435: duke@435: // 6308388: a bug in ld.so will relocate its own .data section to the duke@435: // lower end of primordial stack; reduce ulimit -s value a little bit duke@435: // so we won't install guard page on ld.so's data section. duke@435: stack_size -= 2 * page_size(); duke@435: duke@435: // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat duke@435: // 7.1, in both cases we will get 2G in return value. duke@435: // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, duke@435: // SuSE 7.2, Debian) can not handle alternate signal stack correctly duke@435: // for initial thread if its stack size exceeds 6M. Cap it at 2M, duke@435: // in case other parts in glibc still assumes 2M max stack size. duke@435: // FIXME: alt signal stack is gone, maybe we can relax this constraint? duke@435: #ifndef IA64 duke@435: if (stack_size > 2 * K * K) stack_size = 2 * K * K; duke@435: #else duke@435: // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small duke@435: if (stack_size > 4 * K * K) stack_size = 4 * K * K; duke@435: #endif duke@435: duke@435: // Try to figure out where the stack base (top) is. This is harder. duke@435: // duke@435: // When an application is started, glibc saves the initial stack pointer in duke@435: // a global variable "__libc_stack_end", which is then used by system duke@435: // libraries. __libc_stack_end should be pretty close to stack top. The duke@435: // variable is available since the very early days. However, because it is duke@435: // a private interface, it could disappear in the future. duke@435: // duke@435: // Linux kernel saves start_stack information in /proc//stat. Similar duke@435: // to __libc_stack_end, it is very close to stack top, but isn't the real duke@435: // stack top. Note that /proc may not exist if VM is running as a chroot duke@435: // program, so reading /proc//stat could fail. Also the contents of duke@435: // /proc//stat could change in the future (though unlikely). duke@435: // duke@435: // We try __libc_stack_end first. If that doesn't work, look for duke@435: // /proc//stat. If neither of them works, we use current stack pointer duke@435: // as a hint, which should work well in most cases. duke@435: duke@435: uintptr_t stack_start; duke@435: duke@435: // try __libc_stack_end first duke@435: uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); duke@435: if (p && *p) { duke@435: stack_start = *p; duke@435: } else { duke@435: // see if we can get the start_stack field from /proc/self/stat duke@435: FILE *fp; duke@435: int pid; duke@435: char state; duke@435: int ppid; duke@435: int pgrp; duke@435: int session; duke@435: int nr; duke@435: int tpgrp; duke@435: unsigned long flags; duke@435: unsigned long minflt; duke@435: unsigned long cminflt; duke@435: unsigned long majflt; duke@435: unsigned long cmajflt; duke@435: unsigned long utime; duke@435: unsigned long stime; duke@435: long cutime; duke@435: long cstime; duke@435: long prio; duke@435: long nice; duke@435: long junk; duke@435: long it_real; duke@435: uintptr_t start; duke@435: uintptr_t vsize; duke@435: uintptr_t rss; duke@435: unsigned long rsslim; duke@435: uintptr_t scodes; duke@435: uintptr_t ecode; duke@435: int i; duke@435: duke@435: // Figure what the primordial thread stack base is. Code is inspired duke@435: // by email from Hans Boehm. /proc/self/stat begins with current pid, duke@435: // followed by command name surrounded by parentheses, state, etc. duke@435: char stat[2048]; duke@435: int statlen; duke@435: duke@435: fp = fopen("/proc/self/stat", "r"); duke@435: if (fp) { duke@435: statlen = fread(stat, 1, 2047, fp); duke@435: stat[statlen] = '\0'; duke@435: fclose(fp); duke@435: duke@435: // Skip pid and the command string. Note that we could be dealing with duke@435: // weird command names, e.g. user could decide to rename java launcher duke@435: // to "java 1.4.2 :)", then the stat file would look like duke@435: // 1234 (java 1.4.2 :)) R ... ... duke@435: // We don't really need to know the command string, just find the last duke@435: // occurrence of ")" and then start parsing from there. See bug 4726580. duke@435: char * s = strrchr(stat, ')'); duke@435: duke@435: i = 0; duke@435: if (s) { duke@435: // Skip blank chars duke@435: do s++; while (isspace(*s)); duke@435: duke@435: /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ duke@435: /* 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 */ duke@435: i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld %lu %lu %ld %lu %lu %lu %lu", duke@435: &state, /* 3 %c */ duke@435: &ppid, /* 4 %d */ duke@435: &pgrp, /* 5 %d */ duke@435: &session, /* 6 %d */ duke@435: &nr, /* 7 %d */ duke@435: &tpgrp, /* 8 %d */ duke@435: &flags, /* 9 %lu */ duke@435: &minflt, /* 10 %lu */ duke@435: &cminflt, /* 11 %lu */ duke@435: &majflt, /* 12 %lu */ duke@435: &cmajflt, /* 13 %lu */ duke@435: &utime, /* 14 %lu */ duke@435: &stime, /* 15 %lu */ duke@435: &cutime, /* 16 %ld */ duke@435: &cstime, /* 17 %ld */ duke@435: &prio, /* 18 %ld */ duke@435: &nice, /* 19 %ld */ duke@435: &junk, /* 20 %ld */ duke@435: &it_real, /* 21 %ld */ duke@435: &start, /* 22 %lu */ duke@435: &vsize, /* 23 %lu */ duke@435: &rss, /* 24 %ld */ duke@435: &rsslim, /* 25 %lu */ duke@435: &scodes, /* 26 %lu */ duke@435: &ecode, /* 27 %lu */ duke@435: &stack_start); /* 28 %lu */ duke@435: } duke@435: duke@435: if (i != 28 - 2) { duke@435: assert(false, "Bad conversion from /proc/self/stat"); duke@435: // product mode - assume we are the initial thread, good luck in the duke@435: // embedded case. duke@435: warning("Can't detect initial thread stack location - bad conversion"); duke@435: stack_start = (uintptr_t) &rlim; duke@435: } duke@435: } else { duke@435: // For some reason we can't open /proc/self/stat (for example, running on duke@435: // FreeBSD with a Linux emulator, or inside chroot), this should work for duke@435: // most cases, so don't abort: duke@435: warning("Can't detect initial thread stack location - no /proc/self/stat"); duke@435: stack_start = (uintptr_t) &rlim; duke@435: } duke@435: } duke@435: duke@435: // Now we have a pointer (stack_start) very close to the stack top, the duke@435: // next thing to do is to figure out the exact location of stack top. We duke@435: // can find out the virtual memory area that contains stack_start by duke@435: // reading /proc/self/maps, it should be the last vma in /proc/self/maps, duke@435: // and its upper limit is the real stack top. (again, this would fail if duke@435: // running inside chroot, because /proc may not exist.) duke@435: duke@435: uintptr_t stack_top; duke@435: address low, high; duke@435: if (find_vma((address)stack_start, &low, &high)) { duke@435: // success, "high" is the true stack top. (ignore "low", because initial duke@435: // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) duke@435: stack_top = (uintptr_t)high; duke@435: } else { duke@435: // failed, likely because /proc/self/maps does not exist duke@435: warning("Can't detect initial thread stack location - find_vma failed"); duke@435: // best effort: stack_start is normally within a few pages below the real duke@435: // stack top, use it as stack top, and reduce stack size so we won't put duke@435: // guard page outside stack. duke@435: stack_top = stack_start; duke@435: stack_size -= 16 * page_size(); duke@435: } duke@435: duke@435: // stack_top could be partially down the page so align it duke@435: stack_top = align_size_up(stack_top, page_size()); duke@435: duke@435: if (max_size && stack_size > max_size) { duke@435: _initial_thread_stack_size = max_size; duke@435: } else { duke@435: _initial_thread_stack_size = stack_size; duke@435: } duke@435: duke@435: _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); duke@435: _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; duke@435: } duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // time support duke@435: duke@435: // Time since start-up in seconds to a fine granularity. duke@435: // Used by VMSelfDestructTimer and the MemProfiler. duke@435: double os::elapsedTime() { duke@435: duke@435: return (double)(os::elapsed_counter()) * 0.000001; duke@435: } duke@435: duke@435: jlong os::elapsed_counter() { duke@435: timeval time; duke@435: int status = gettimeofday(&time, NULL); duke@435: return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; duke@435: } duke@435: duke@435: jlong os::elapsed_frequency() { duke@435: return (1000 * 1000); duke@435: } duke@435: sbohne@496: jlong os::javaTimeMillis() { duke@435: timeval time; duke@435: int status = gettimeofday(&time, NULL); duke@435: assert(status != -1, "linux error"); duke@435: return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); duke@435: } duke@435: duke@435: #ifndef CLOCK_MONOTONIC duke@435: #define CLOCK_MONOTONIC (1) duke@435: #endif duke@435: duke@435: void os::Linux::clock_init() { duke@435: // we do dlopen's in this particular order due to bug in linux duke@435: // dynamical loader (see 6348968) leading to crash on exit duke@435: void* handle = dlopen("librt.so.1", RTLD_LAZY); duke@435: if (handle == NULL) { duke@435: handle = dlopen("librt.so", RTLD_LAZY); duke@435: } duke@435: duke@435: if (handle) { duke@435: int (*clock_getres_func)(clockid_t, struct timespec*) = duke@435: (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); duke@435: int (*clock_gettime_func)(clockid_t, struct timespec*) = duke@435: (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); duke@435: if (clock_getres_func && clock_gettime_func) { duke@435: // See if monotonic clock is supported by the kernel. Note that some duke@435: // early implementations simply return kernel jiffies (updated every duke@435: // 1/100 or 1/1000 second). It would be bad to use such a low res clock duke@435: // for nano time (though the monotonic property is still nice to have). duke@435: // It's fixed in newer kernels, however clock_getres() still returns duke@435: // 1/HZ. We check if clock_getres() works, but will ignore its reported duke@435: // resolution for now. Hopefully as people move to new kernels, this duke@435: // won't be a problem. duke@435: struct timespec res; duke@435: struct timespec tp; duke@435: if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && duke@435: clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { duke@435: // yes, monotonic clock is supported duke@435: _clock_gettime = clock_gettime_func; duke@435: } else { duke@435: // close librt if there is no monotonic clock duke@435: dlclose(handle); duke@435: } duke@435: } duke@435: } duke@435: } duke@435: duke@435: #ifndef SYS_clock_getres duke@435: duke@435: #if defined(IA32) || defined(AMD64) duke@435: #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) duke@435: #else duke@435: #error Value of SYS_clock_getres not known on this platform duke@435: #endif duke@435: duke@435: #endif duke@435: duke@435: #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) duke@435: duke@435: void os::Linux::fast_thread_clock_init() { duke@435: if (!UseLinuxPosixThreadCPUClocks) { duke@435: return; duke@435: } duke@435: clockid_t clockid; duke@435: struct timespec tp; duke@435: int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = duke@435: (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); duke@435: duke@435: // Switch to using fast clocks for thread cpu time if duke@435: // the sys_clock_getres() returns 0 error code. duke@435: // Note, that some kernels may support the current thread duke@435: // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks duke@435: // returned by the pthread_getcpuclockid(). duke@435: // If the fast Posix clocks are supported then the sys_clock_getres() duke@435: // must return at least tp.tv_sec == 0 which means a resolution duke@435: // better than 1 sec. This is extra check for reliability. duke@435: duke@435: if(pthread_getcpuclockid_func && duke@435: pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && duke@435: sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { duke@435: duke@435: _supports_fast_thread_cpu_time = true; duke@435: _pthread_getcpuclockid = pthread_getcpuclockid_func; duke@435: } duke@435: } duke@435: duke@435: jlong os::javaTimeNanos() { duke@435: if (Linux::supports_monotonic_clock()) { duke@435: struct timespec tp; duke@435: int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); duke@435: assert(status == 0, "gettime error"); duke@435: jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); duke@435: return result; duke@435: } else { duke@435: timeval time; duke@435: int status = gettimeofday(&time, NULL); duke@435: assert(status != -1, "linux error"); duke@435: jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); duke@435: return 1000 * usecs; duke@435: } duke@435: } duke@435: duke@435: void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { duke@435: if (Linux::supports_monotonic_clock()) { duke@435: info_ptr->max_value = ALL_64_BITS; duke@435: duke@435: // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past duke@435: info_ptr->may_skip_backward = false; // not subject to resetting or drifting duke@435: info_ptr->may_skip_forward = false; // not subject to resetting or drifting duke@435: } else { duke@435: // gettimeofday - based on time in seconds since the Epoch thus does not wrap duke@435: info_ptr->max_value = ALL_64_BITS; duke@435: duke@435: // gettimeofday is a real time clock so it skips duke@435: info_ptr->may_skip_backward = true; duke@435: info_ptr->may_skip_forward = true; duke@435: } duke@435: duke@435: info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time duke@435: } duke@435: duke@435: // Return the real, user, and system times in seconds from an duke@435: // arbitrary fixed point in the past. duke@435: bool os::getTimesSecs(double* process_real_time, duke@435: double* process_user_time, duke@435: double* process_system_time) { duke@435: struct tms ticks; duke@435: clock_t real_ticks = times(&ticks); duke@435: duke@435: if (real_ticks == (clock_t) (-1)) { duke@435: return false; duke@435: } else { duke@435: double ticks_per_second = (double) clock_tics_per_sec; duke@435: *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; duke@435: *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; duke@435: *process_real_time = ((double) real_ticks) / ticks_per_second; duke@435: duke@435: return true; duke@435: } duke@435: } duke@435: duke@435: duke@435: char * os::local_time_string(char *buf, size_t buflen) { duke@435: struct tm t; duke@435: time_t long_time; duke@435: time(&long_time); duke@435: localtime_r(&long_time, &t); duke@435: jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", duke@435: t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, duke@435: t.tm_hour, t.tm_min, t.tm_sec); duke@435: return buf; duke@435: } duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // runtime exit support duke@435: duke@435: // Note: os::shutdown() might be called very early during initialization, or duke@435: // called from signal handler. Before adding something to os::shutdown(), make duke@435: // sure it is async-safe and can handle partially initialized VM. duke@435: void os::shutdown() { duke@435: duke@435: // allow PerfMemory to attempt cleanup of any persistent resources duke@435: perfMemory_exit(); duke@435: duke@435: // needs to remove object in file system duke@435: AttachListener::abort(); duke@435: duke@435: // flush buffered output, finish log files duke@435: ostream_abort(); duke@435: duke@435: // Check for abort hook duke@435: abort_hook_t abort_hook = Arguments::abort_hook(); duke@435: if (abort_hook != NULL) { duke@435: abort_hook(); duke@435: } duke@435: duke@435: } duke@435: duke@435: // Note: os::abort() might be called very early during initialization, or duke@435: // called from signal handler. Before adding something to os::abort(), make duke@435: // sure it is async-safe and can handle partially initialized VM. duke@435: void os::abort(bool dump_core) { duke@435: os::shutdown(); duke@435: if (dump_core) { duke@435: #ifndef PRODUCT duke@435: fdStream out(defaultStream::output_fd()); duke@435: out.print_raw("Current thread is "); duke@435: char buf[16]; duke@435: jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); duke@435: out.print_raw_cr(buf); duke@435: out.print_raw_cr("Dumping core ..."); duke@435: #endif duke@435: ::abort(); // dump core duke@435: } duke@435: duke@435: ::exit(1); duke@435: } duke@435: duke@435: // Die immediately, no exit hook, no abort hook, no cleanup. duke@435: void os::die() { duke@435: // _exit() on LinuxThreads only kills current thread duke@435: ::abort(); duke@435: } duke@435: duke@435: // unused on linux for now. duke@435: void os::set_error_file(const char *logfile) {} duke@435: duke@435: intx os::current_thread_id() { return (intx)pthread_self(); } duke@435: int os::current_process_id() { duke@435: duke@435: // Under the old linux thread library, linux gives each thread duke@435: // its own process id. Because of this each thread will return duke@435: // a different pid if this method were to return the result duke@435: // of getpid(2). Linux provides no api that returns the pid duke@435: // of the launcher thread for the vm. This implementation duke@435: // returns a unique pid, the pid of the launcher thread duke@435: // that starts the vm 'process'. duke@435: duke@435: // Under the NPTL, getpid() returns the same pid as the duke@435: // launcher thread rather than a unique pid per thread. duke@435: // Use gettid() if you want the old pre NPTL behaviour. duke@435: duke@435: // if you are looking for the result of a call to getpid() that duke@435: // returns a unique pid for the calling thread, then look at the duke@435: // OSThread::thread_id() method in osThread_linux.hpp file duke@435: duke@435: return (int)(_initial_pid ? _initial_pid : getpid()); duke@435: } duke@435: duke@435: // DLL functions duke@435: duke@435: const char* os::dll_file_extension() { return ".so"; } duke@435: duke@435: const char* os::get_temp_directory() { return "/tmp/"; } duke@435: duke@435: const char* os::get_current_directory(char *buf, int buflen) { duke@435: return getcwd(buf, buflen); duke@435: } duke@435: duke@435: // check if addr is inside libjvm[_g].so duke@435: bool os::address_is_in_vm(address addr) { duke@435: static address libjvm_base_addr; duke@435: Dl_info dlinfo; duke@435: duke@435: if (libjvm_base_addr == NULL) { duke@435: dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); duke@435: libjvm_base_addr = (address)dlinfo.dli_fbase; duke@435: assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); duke@435: } duke@435: duke@435: if (dladdr((void *)addr, &dlinfo)) { duke@435: if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; duke@435: } duke@435: duke@435: return false; duke@435: } duke@435: duke@435: bool os::dll_address_to_function_name(address addr, char *buf, duke@435: int buflen, int *offset) { duke@435: Dl_info dlinfo; duke@435: duke@435: if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { duke@435: if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); duke@435: if (offset) *offset = addr - (address)dlinfo.dli_saddr; duke@435: return true; duke@435: } else { duke@435: if (buf) buf[0] = '\0'; duke@435: if (offset) *offset = -1; duke@435: return false; duke@435: } duke@435: } duke@435: duke@435: struct _address_to_library_name { duke@435: address addr; // input : memory address duke@435: size_t buflen; // size of fname duke@435: char* fname; // output: library name duke@435: address base; // library base addr duke@435: }; duke@435: duke@435: static int address_to_library_name_callback(struct dl_phdr_info *info, duke@435: size_t size, void *data) { duke@435: int i; duke@435: bool found = false; duke@435: address libbase = NULL; duke@435: struct _address_to_library_name * d = (struct _address_to_library_name *)data; duke@435: duke@435: // iterate through all loadable segments duke@435: for (i = 0; i < info->dlpi_phnum; i++) { duke@435: address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); duke@435: if (info->dlpi_phdr[i].p_type == PT_LOAD) { duke@435: // base address of a library is the lowest address of its loaded duke@435: // segments. duke@435: if (libbase == NULL || libbase > segbase) { duke@435: libbase = segbase; duke@435: } duke@435: // see if 'addr' is within current segment duke@435: if (segbase <= d->addr && duke@435: d->addr < segbase + info->dlpi_phdr[i].p_memsz) { duke@435: found = true; duke@435: } duke@435: } duke@435: } duke@435: duke@435: // dlpi_name is NULL or empty if the ELF file is executable, return 0 duke@435: // so dll_address_to_library_name() can fall through to use dladdr() which duke@435: // can figure out executable name from argv[0]. duke@435: if (found && info->dlpi_name && info->dlpi_name[0]) { duke@435: d->base = libbase; duke@435: if (d->fname) { duke@435: jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); duke@435: } duke@435: return 1; duke@435: } duke@435: return 0; duke@435: } duke@435: duke@435: bool os::dll_address_to_library_name(address addr, char* buf, duke@435: int buflen, int* offset) { duke@435: Dl_info dlinfo; duke@435: struct _address_to_library_name data; duke@435: duke@435: // There is a bug in old glibc dladdr() implementation that it could resolve duke@435: // to wrong library name if the .so file has a base address != NULL. Here duke@435: // we iterate through the program headers of all loaded libraries to find duke@435: // out which library 'addr' really belongs to. This workaround can be duke@435: // removed once the minimum requirement for glibc is moved to 2.3.x. duke@435: data.addr = addr; duke@435: data.fname = buf; duke@435: data.buflen = buflen; duke@435: data.base = NULL; duke@435: int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); duke@435: duke@435: if (rslt) { duke@435: // buf already contains library name duke@435: if (offset) *offset = addr - data.base; duke@435: return true; duke@435: } else if (dladdr((void*)addr, &dlinfo)){ duke@435: if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); duke@435: if (offset) *offset = addr - (address)dlinfo.dli_fbase; duke@435: return true; duke@435: } else { duke@435: if (buf) buf[0] = '\0'; duke@435: if (offset) *offset = -1; duke@435: return false; duke@435: } duke@435: } duke@435: duke@435: // Loads .dll/.so and duke@435: // in case of error it checks if .dll/.so was built for the duke@435: // same architecture as Hotspot is running on duke@435: duke@435: void * os::dll_load(const char *filename, char *ebuf, int ebuflen) duke@435: { duke@435: void * result= ::dlopen(filename, RTLD_LAZY); duke@435: if (result != NULL) { duke@435: // Successful loading duke@435: return result; duke@435: } duke@435: duke@435: Elf32_Ehdr elf_head; duke@435: duke@435: // Read system error message into ebuf duke@435: // It may or may not be overwritten below duke@435: ::strncpy(ebuf, ::dlerror(), ebuflen-1); duke@435: ebuf[ebuflen-1]='\0'; duke@435: int diag_msg_max_length=ebuflen-strlen(ebuf); duke@435: char* diag_msg_buf=ebuf+strlen(ebuf); duke@435: duke@435: if (diag_msg_max_length==0) { duke@435: // No more space in ebuf for additional diagnostics message duke@435: return NULL; duke@435: } duke@435: duke@435: duke@435: int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); duke@435: duke@435: if (file_descriptor < 0) { duke@435: // Can't open library, report dlerror() message duke@435: return NULL; duke@435: } duke@435: duke@435: bool failed_to_read_elf_head= duke@435: (sizeof(elf_head)!= duke@435: (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; duke@435: duke@435: ::close(file_descriptor); duke@435: if (failed_to_read_elf_head) { duke@435: // file i/o error - report dlerror() msg duke@435: return NULL; duke@435: } duke@435: duke@435: typedef struct { duke@435: Elf32_Half code; // Actual value as defined in elf.h duke@435: Elf32_Half compat_class; // Compatibility of archs at VM's sense duke@435: char elf_class; // 32 or 64 bit duke@435: char endianess; // MSB or LSB duke@435: char* name; // String representation duke@435: } arch_t; duke@435: duke@435: #ifndef EM_486 duke@435: #define EM_486 6 /* Intel 80486 */ duke@435: #endif duke@435: duke@435: static const arch_t arch_array[]={ duke@435: {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, duke@435: {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, duke@435: {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, duke@435: {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, duke@435: {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, duke@435: {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, duke@435: {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, duke@435: {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, duke@435: {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"} duke@435: }; duke@435: duke@435: #if (defined IA32) duke@435: static Elf32_Half running_arch_code=EM_386; duke@435: #elif (defined AMD64) duke@435: static Elf32_Half running_arch_code=EM_X86_64; duke@435: #elif (defined IA64) duke@435: static Elf32_Half running_arch_code=EM_IA_64; duke@435: #elif (defined __sparc) && (defined _LP64) duke@435: static Elf32_Half running_arch_code=EM_SPARCV9; duke@435: #elif (defined __sparc) && (!defined _LP64) duke@435: static Elf32_Half running_arch_code=EM_SPARC; duke@435: #elif (defined __powerpc64__) duke@435: static Elf32_Half running_arch_code=EM_PPC64; duke@435: #elif (defined __powerpc__) duke@435: static Elf32_Half running_arch_code=EM_PPC; duke@435: #else duke@435: #error Method os::dll_load requires that one of following is defined:\ duke@435: IA32, AMD64, IA64, __sparc, __powerpc__ duke@435: #endif duke@435: duke@435: // Identify compatability class for VM's architecture and library's architecture duke@435: // Obtain string descriptions for architectures duke@435: duke@435: arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; duke@435: int running_arch_index=-1; duke@435: duke@435: for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { duke@435: if (running_arch_code == arch_array[i].code) { duke@435: running_arch_index = i; duke@435: } duke@435: if (lib_arch.code == arch_array[i].code) { duke@435: lib_arch.compat_class = arch_array[i].compat_class; duke@435: lib_arch.name = arch_array[i].name; duke@435: } duke@435: } duke@435: duke@435: assert(running_arch_index != -1, duke@435: "Didn't find running architecture code (running_arch_code) in arch_array"); duke@435: if (running_arch_index == -1) { duke@435: // Even though running architecture detection failed duke@435: // we may still continue with reporting dlerror() message duke@435: return NULL; duke@435: } duke@435: duke@435: if (lib_arch.endianess != arch_array[running_arch_index].endianess) { duke@435: ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); duke@435: return NULL; duke@435: } duke@435: duke@435: if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { duke@435: ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); duke@435: return NULL; duke@435: } duke@435: duke@435: if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { duke@435: if ( lib_arch.name!=NULL ) { duke@435: ::snprintf(diag_msg_buf, diag_msg_max_length-1, duke@435: " (Possible cause: can't load %s-bit .so on a %s-bit platform)", duke@435: lib_arch.name, arch_array[running_arch_index].name); duke@435: } else { duke@435: ::snprintf(diag_msg_buf, diag_msg_max_length-1, duke@435: " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", duke@435: lib_arch.code, duke@435: arch_array[running_arch_index].name); duke@435: } duke@435: } duke@435: duke@435: return NULL; duke@435: } duke@435: duke@435: duke@435: duke@435: duke@435: bool _print_ascii_file(const char* filename, outputStream* st) { duke@435: int fd = open(filename, O_RDONLY); duke@435: if (fd == -1) { duke@435: return false; duke@435: } duke@435: duke@435: char buf[32]; duke@435: int bytes; duke@435: while ((bytes = read(fd, buf, sizeof(buf))) > 0) { duke@435: st->print_raw(buf, bytes); duke@435: } duke@435: duke@435: close(fd); duke@435: duke@435: return true; duke@435: } duke@435: duke@435: void os::print_dll_info(outputStream *st) { duke@435: st->print_cr("Dynamic libraries:"); duke@435: duke@435: char fname[32]; duke@435: pid_t pid = os::Linux::gettid(); duke@435: duke@435: jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); duke@435: duke@435: if (!_print_ascii_file(fname, st)) { duke@435: st->print("Can not get library information for pid = %d\n", pid); duke@435: } duke@435: } duke@435: duke@435: duke@435: void os::print_os_info(outputStream* st) { duke@435: st->print("OS:"); duke@435: duke@435: // Try to identify popular distros. duke@435: // Most Linux distributions have /etc/XXX-release file, which contains duke@435: // the OS version string. Some have more than one /etc/XXX-release file duke@435: // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), duke@435: // so the order is important. duke@435: if (!_print_ascii_file("/etc/mandrake-release", st) && duke@435: !_print_ascii_file("/etc/sun-release", st) && duke@435: !_print_ascii_file("/etc/redhat-release", st) && duke@435: !_print_ascii_file("/etc/SuSE-release", st) && duke@435: !_print_ascii_file("/etc/turbolinux-release", st) && duke@435: !_print_ascii_file("/etc/gentoo-release", st) && duke@435: !_print_ascii_file("/etc/debian_version", st)) { duke@435: st->print("Linux"); duke@435: } duke@435: st->cr(); duke@435: duke@435: // kernel duke@435: st->print("uname:"); duke@435: struct utsname name; duke@435: uname(&name); duke@435: st->print(name.sysname); st->print(" "); duke@435: st->print(name.release); st->print(" "); duke@435: st->print(name.version); st->print(" "); duke@435: st->print(name.machine); duke@435: st->cr(); duke@435: duke@435: // Print warning if unsafe chroot environment detected duke@435: if (unsafe_chroot_detected) { duke@435: st->print("WARNING!! "); duke@435: st->print_cr(unstable_chroot_error); duke@435: } duke@435: duke@435: // libc, pthread duke@435: st->print("libc:"); duke@435: st->print(os::Linux::glibc_version()); st->print(" "); duke@435: st->print(os::Linux::libpthread_version()); st->print(" "); duke@435: if (os::Linux::is_LinuxThreads()) { duke@435: st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); duke@435: } duke@435: st->cr(); duke@435: duke@435: // rlimit duke@435: st->print("rlimit:"); duke@435: struct rlimit rlim; duke@435: duke@435: st->print(" STACK "); duke@435: getrlimit(RLIMIT_STACK, &rlim); duke@435: if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); duke@435: else st->print("%uk", rlim.rlim_cur >> 10); duke@435: duke@435: st->print(", CORE "); duke@435: getrlimit(RLIMIT_CORE, &rlim); duke@435: if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); duke@435: else st->print("%uk", rlim.rlim_cur >> 10); duke@435: duke@435: st->print(", NPROC "); duke@435: getrlimit(RLIMIT_NPROC, &rlim); duke@435: if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); duke@435: else st->print("%d", rlim.rlim_cur); duke@435: duke@435: st->print(", NOFILE "); duke@435: getrlimit(RLIMIT_NOFILE, &rlim); duke@435: if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); duke@435: else st->print("%d", rlim.rlim_cur); duke@435: duke@435: st->print(", AS "); duke@435: getrlimit(RLIMIT_AS, &rlim); duke@435: if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); duke@435: else st->print("%uk", rlim.rlim_cur >> 10); duke@435: st->cr(); duke@435: duke@435: // load average duke@435: st->print("load average:"); duke@435: double loadavg[3]; duke@435: os::loadavg(loadavg, 3); duke@435: st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]); duke@435: st->cr(); duke@435: } duke@435: duke@435: void os::print_memory_info(outputStream* st) { duke@435: duke@435: st->print("Memory:"); duke@435: st->print(" %dk page", os::vm_page_size()>>10); duke@435: duke@435: // values in struct sysinfo are "unsigned long" duke@435: struct sysinfo si; duke@435: sysinfo(&si); duke@435: duke@435: st->print(", physical " UINT64_FORMAT "k", duke@435: os::physical_memory() >> 10); duke@435: st->print("(" UINT64_FORMAT "k free)", duke@435: os::available_memory() >> 10); duke@435: st->print(", swap " UINT64_FORMAT "k", duke@435: ((jlong)si.totalswap * si.mem_unit) >> 10); duke@435: st->print("(" UINT64_FORMAT "k free)", duke@435: ((jlong)si.freeswap * si.mem_unit) >> 10); duke@435: st->cr(); duke@435: } duke@435: duke@435: // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific duke@435: // but they're the same for all the linux arch that we support duke@435: // and they're the same for solaris but there's no common place to put this. duke@435: const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", duke@435: "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", duke@435: "ILL_COPROC", "ILL_BADSTK" }; duke@435: duke@435: const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", duke@435: "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", duke@435: "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; duke@435: duke@435: const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; duke@435: duke@435: const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; duke@435: duke@435: void os::print_siginfo(outputStream* st, void* siginfo) { duke@435: st->print("siginfo:"); duke@435: duke@435: const int buflen = 100; duke@435: char buf[buflen]; duke@435: siginfo_t *si = (siginfo_t*)siginfo; duke@435: st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); duke@435: if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { duke@435: st->print("si_errno=%s", buf); duke@435: } else { duke@435: st->print("si_errno=%d", si->si_errno); duke@435: } duke@435: const int c = si->si_code; duke@435: assert(c > 0, "unexpected si_code"); duke@435: switch (si->si_signo) { duke@435: case SIGILL: duke@435: st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); duke@435: st->print(", si_addr=" PTR_FORMAT, si->si_addr); duke@435: break; duke@435: case SIGFPE: duke@435: st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); duke@435: st->print(", si_addr=" PTR_FORMAT, si->si_addr); duke@435: break; duke@435: case SIGSEGV: duke@435: st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); duke@435: st->print(", si_addr=" PTR_FORMAT, si->si_addr); duke@435: break; duke@435: case SIGBUS: duke@435: st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); duke@435: st->print(", si_addr=" PTR_FORMAT, si->si_addr); duke@435: break; duke@435: default: duke@435: st->print(", si_code=%d", si->si_code); duke@435: // no si_addr duke@435: } duke@435: duke@435: if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && duke@435: UseSharedSpaces) { duke@435: FileMapInfo* mapinfo = FileMapInfo::current_info(); duke@435: if (mapinfo->is_in_shared_space(si->si_addr)) { duke@435: st->print("\n\nError accessing class data sharing archive." \ duke@435: " Mapped file inaccessible during execution, " \ duke@435: " possible disk/network problem."); duke@435: } duke@435: } duke@435: st->cr(); duke@435: } duke@435: duke@435: duke@435: static void print_signal_handler(outputStream* st, int sig, duke@435: char* buf, size_t buflen); duke@435: duke@435: void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { duke@435: st->print_cr("Signal Handlers:"); duke@435: print_signal_handler(st, SIGSEGV, buf, buflen); duke@435: print_signal_handler(st, SIGBUS , buf, buflen); duke@435: print_signal_handler(st, SIGFPE , buf, buflen); duke@435: print_signal_handler(st, SIGPIPE, buf, buflen); duke@435: print_signal_handler(st, SIGXFSZ, buf, buflen); duke@435: print_signal_handler(st, SIGILL , buf, buflen); duke@435: print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); duke@435: print_signal_handler(st, SR_signum, buf, buflen); duke@435: print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); duke@435: print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); duke@435: print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); duke@435: print_signal_handler(st, BREAK_SIGNAL, buf, buflen); duke@435: } duke@435: duke@435: static char saved_jvm_path[MAXPATHLEN] = {0}; duke@435: duke@435: // Find the full path to the current module, libjvm.so or libjvm_g.so duke@435: void os::jvm_path(char *buf, jint len) { duke@435: // Error checking. duke@435: if (len < MAXPATHLEN) { duke@435: assert(false, "must use a large-enough buffer"); duke@435: buf[0] = '\0'; duke@435: return; duke@435: } duke@435: // Lazy resolve the path to current module. duke@435: if (saved_jvm_path[0] != 0) { duke@435: strcpy(buf, saved_jvm_path); duke@435: return; duke@435: } duke@435: duke@435: char dli_fname[MAXPATHLEN]; duke@435: bool ret = dll_address_to_library_name( duke@435: CAST_FROM_FN_PTR(address, os::jvm_path), duke@435: dli_fname, sizeof(dli_fname), NULL); duke@435: assert(ret != 0, "cannot locate libjvm"); duke@435: realpath(dli_fname, buf); duke@435: duke@435: if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) { duke@435: // Support for the gamma launcher. Typical value for buf is duke@435: // "/jre/lib///libjvm.so". If "/jre/lib/" appears at duke@435: // the right place in the string, then assume we are installed in a JDK and duke@435: // we're done. Otherwise, check for a JAVA_HOME environment variable and fix duke@435: // up the path so it looks like libjvm.so is installed there (append a duke@435: // fake suffix hotspot/libjvm.so). duke@435: const char *p = buf + strlen(buf) - 1; duke@435: for (int count = 0; p > buf && count < 5; ++count) { duke@435: for (--p; p > buf && *p != '/'; --p) duke@435: /* empty */ ; duke@435: } duke@435: duke@435: if (strncmp(p, "/jre/lib/", 9) != 0) { duke@435: // Look for JAVA_HOME in the environment. duke@435: char* java_home_var = ::getenv("JAVA_HOME"); duke@435: if (java_home_var != NULL && java_home_var[0] != 0) { duke@435: // Check the current module name "libjvm.so" or "libjvm_g.so". duke@435: p = strrchr(buf, '/'); duke@435: assert(strstr(p, "/libjvm") == p, "invalid library name"); duke@435: p = strstr(p, "_g") ? "_g" : ""; duke@435: duke@435: realpath(java_home_var, buf); duke@435: sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch); duke@435: if (0 == access(buf, F_OK)) { duke@435: // Use current module name "libjvm[_g].so" instead of duke@435: // "libjvm"debug_only("_g")".so" since for fastdebug version duke@435: // we should have "libjvm.so" but debug_only("_g") adds "_g"! duke@435: // It is used when we are choosing the HPI library's name duke@435: // "libhpi[_g].so" in hpi::initialize_get_interface(). duke@435: sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p); duke@435: } else { duke@435: // Go back to path of .so duke@435: realpath(dli_fname, buf); duke@435: } duke@435: } duke@435: } duke@435: } duke@435: duke@435: strcpy(saved_jvm_path, buf); duke@435: } duke@435: duke@435: void os::print_jni_name_prefix_on(outputStream* st, int args_size) { duke@435: // no prefix required, not even "_" duke@435: } duke@435: duke@435: void os::print_jni_name_suffix_on(outputStream* st, int args_size) { duke@435: // no suffix required duke@435: } duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // sun.misc.Signal support duke@435: duke@435: static volatile jint sigint_count = 0; duke@435: duke@435: static void duke@435: UserHandler(int sig, void *siginfo, void *context) { duke@435: // 4511530 - sem_post is serialized and handled by the manager thread. When duke@435: // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We duke@435: // don't want to flood the manager thread with sem_post requests. duke@435: if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) duke@435: return; duke@435: duke@435: // Ctrl-C is pressed during error reporting, likely because the error duke@435: // handler fails to abort. Let VM die immediately. duke@435: if (sig == SIGINT && is_error_reported()) { duke@435: os::die(); duke@435: } duke@435: duke@435: os::signal_notify(sig); duke@435: } duke@435: duke@435: void* os::user_handler() { duke@435: return CAST_FROM_FN_PTR(void*, UserHandler); duke@435: } duke@435: duke@435: extern "C" { duke@435: typedef void (*sa_handler_t)(int); duke@435: typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); duke@435: } duke@435: duke@435: void* os::signal(int signal_number, void* handler) { duke@435: struct sigaction sigAct, oldSigAct; duke@435: duke@435: sigfillset(&(sigAct.sa_mask)); duke@435: sigAct.sa_flags = SA_RESTART|SA_SIGINFO; duke@435: sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); duke@435: duke@435: if (sigaction(signal_number, &sigAct, &oldSigAct)) { duke@435: // -1 means registration failed duke@435: return (void *)-1; duke@435: } duke@435: duke@435: return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); duke@435: } duke@435: duke@435: void os::signal_raise(int signal_number) { duke@435: ::raise(signal_number); duke@435: } duke@435: duke@435: /* duke@435: * The following code is moved from os.cpp for making this duke@435: * code platform specific, which it is by its very nature. duke@435: */ duke@435: duke@435: // Will be modified when max signal is changed to be dynamic duke@435: int os::sigexitnum_pd() { duke@435: return NSIG; duke@435: } duke@435: duke@435: // a counter for each possible signal value duke@435: static volatile jint pending_signals[NSIG+1] = { 0 }; duke@435: duke@435: // Linux(POSIX) specific hand shaking semaphore. duke@435: static sem_t sig_sem; duke@435: duke@435: void os::signal_init_pd() { duke@435: // Initialize signal structures duke@435: ::memset((void*)pending_signals, 0, sizeof(pending_signals)); duke@435: duke@435: // Initialize signal semaphore duke@435: ::sem_init(&sig_sem, 0, 0); duke@435: } duke@435: duke@435: void os::signal_notify(int sig) { duke@435: Atomic::inc(&pending_signals[sig]); duke@435: ::sem_post(&sig_sem); duke@435: } duke@435: duke@435: static int check_pending_signals(bool wait) { duke@435: Atomic::store(0, &sigint_count); duke@435: for (;;) { duke@435: for (int i = 0; i < NSIG + 1; i++) { duke@435: jint n = pending_signals[i]; duke@435: if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { duke@435: return i; duke@435: } duke@435: } duke@435: if (!wait) { duke@435: return -1; duke@435: } duke@435: JavaThread *thread = JavaThread::current(); duke@435: ThreadBlockInVM tbivm(thread); duke@435: duke@435: bool threadIsSuspended; duke@435: do { duke@435: thread->set_suspend_equivalent(); duke@435: // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() duke@435: ::sem_wait(&sig_sem); duke@435: duke@435: // were we externally suspended while we were waiting? duke@435: threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); duke@435: if (threadIsSuspended) { duke@435: // duke@435: // The semaphore has been incremented, but while we were waiting duke@435: // another thread suspended us. We don't want to continue running duke@435: // while suspended because that would surprise the thread that duke@435: // suspended us. duke@435: // duke@435: ::sem_post(&sig_sem); duke@435: duke@435: thread->java_suspend_self(); duke@435: } duke@435: } while (threadIsSuspended); duke@435: } duke@435: } duke@435: duke@435: int os::signal_lookup() { duke@435: return check_pending_signals(false); duke@435: } duke@435: duke@435: int os::signal_wait() { duke@435: return check_pending_signals(true); duke@435: } duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // Virtual Memory duke@435: duke@435: int os::vm_page_size() { duke@435: // Seems redundant as all get out duke@435: assert(os::Linux::page_size() != -1, "must call os::init"); duke@435: return os::Linux::page_size(); duke@435: } duke@435: duke@435: // Solaris allocates memory by pages. duke@435: int os::vm_allocation_granularity() { duke@435: assert(os::Linux::page_size() != -1, "must call os::init"); duke@435: return os::Linux::page_size(); duke@435: } duke@435: duke@435: // Rationale behind this function: duke@435: // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable duke@435: // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get duke@435: // samples for JITted code. Here we create private executable mapping over the code cache duke@435: // and then we can use standard (well, almost, as mapping can change) way to provide duke@435: // info for the reporting script by storing timestamp and location of symbol duke@435: void linux_wrap_code(char* base, size_t size) { duke@435: static volatile jint cnt = 0; duke@435: duke@435: if (!UseOprofile) { duke@435: return; duke@435: } duke@435: duke@435: char buf[40]; duke@435: int num = Atomic::add(1, &cnt); duke@435: duke@435: sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num); duke@435: unlink(buf); duke@435: duke@435: int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU); duke@435: duke@435: if (fd != -1) { duke@435: off_t rv = lseek(fd, size-2, SEEK_SET); duke@435: if (rv != (off_t)-1) { duke@435: if (write(fd, "", 1) == 1) { duke@435: mmap(base, size, duke@435: PROT_READ|PROT_WRITE|PROT_EXEC, duke@435: MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); duke@435: } duke@435: } duke@435: close(fd); duke@435: unlink(buf); duke@435: } duke@435: } duke@435: duke@435: // NOTE: Linux kernel does not really reserve the pages for us. duke@435: // All it does is to check if there are enough free pages duke@435: // left at the time of mmap(). This could be a potential duke@435: // problem. duke@435: bool os::commit_memory(char* addr, size_t size) { duke@435: uintptr_t res = (uintptr_t) ::mmap(addr, size, duke@435: PROT_READ|PROT_WRITE|PROT_EXEC, duke@435: MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); duke@435: return res != (uintptr_t) MAP_FAILED; duke@435: } duke@435: duke@435: bool os::commit_memory(char* addr, size_t size, size_t alignment_hint) { duke@435: return commit_memory(addr, size); duke@435: } duke@435: duke@435: void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { } duke@435: void os::free_memory(char *addr, size_t bytes) { } duke@435: void os::numa_make_global(char *addr, size_t bytes) { } duke@435: void os::numa_make_local(char *addr, size_t bytes) { } duke@435: bool os::numa_topology_changed() { return false; } duke@435: size_t os::numa_get_groups_num() { return 1; } duke@435: int os::numa_get_group_id() { return 0; } duke@435: size_t os::numa_get_leaf_groups(int *ids, size_t size) { duke@435: if (size > 0) { duke@435: ids[0] = 0; duke@435: return 1; duke@435: } duke@435: return 0; duke@435: } duke@435: duke@435: bool os::get_page_info(char *start, page_info* info) { duke@435: return false; duke@435: } duke@435: duke@435: char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { duke@435: return end; duke@435: } duke@435: duke@435: bool os::uncommit_memory(char* addr, size_t size) { duke@435: return ::mmap(addr, size, duke@435: PROT_READ|PROT_WRITE|PROT_EXEC, duke@435: MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0) duke@435: != MAP_FAILED; duke@435: } duke@435: duke@435: static address _highest_vm_reserved_address = NULL; duke@435: duke@435: // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory duke@435: // at 'requested_addr'. If there are existing memory mappings at the same duke@435: // location, however, they will be overwritten. If 'fixed' is false, duke@435: // 'requested_addr' is only treated as a hint, the return value may or duke@435: // may not start from the requested address. Unlike Linux mmap(), this duke@435: // function returns NULL to indicate failure. duke@435: static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { duke@435: char * addr; duke@435: int flags; duke@435: duke@435: flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; duke@435: if (fixed) { duke@435: assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); duke@435: flags |= MAP_FIXED; duke@435: } duke@435: duke@435: addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE|PROT_EXEC, duke@435: flags, -1, 0); duke@435: duke@435: if (addr != MAP_FAILED) { duke@435: // anon_mmap() should only get called during VM initialization, duke@435: // don't need lock (actually we can skip locking even it can be called duke@435: // from multiple threads, because _highest_vm_reserved_address is just a duke@435: // hint about the upper limit of non-stack memory regions.) duke@435: if ((address)addr + bytes > _highest_vm_reserved_address) { duke@435: _highest_vm_reserved_address = (address)addr + bytes; duke@435: } duke@435: } duke@435: duke@435: return addr == MAP_FAILED ? NULL : addr; duke@435: } duke@435: duke@435: // Don't update _highest_vm_reserved_address, because there might be memory duke@435: // regions above addr + size. If so, releasing a memory region only creates duke@435: // a hole in the address space, it doesn't help prevent heap-stack collision. duke@435: // duke@435: static int anon_munmap(char * addr, size_t size) { duke@435: return ::munmap(addr, size) == 0; duke@435: } duke@435: duke@435: char* os::reserve_memory(size_t bytes, char* requested_addr, duke@435: size_t alignment_hint) { duke@435: return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); duke@435: } duke@435: duke@435: bool os::release_memory(char* addr, size_t size) { duke@435: return anon_munmap(addr, size); duke@435: } duke@435: duke@435: static address highest_vm_reserved_address() { duke@435: return _highest_vm_reserved_address; duke@435: } duke@435: duke@435: static bool linux_mprotect(char* addr, size_t size, int prot) { duke@435: // Linux wants the mprotect address argument to be page aligned. duke@435: char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); duke@435: duke@435: // According to SUSv3, mprotect() should only be used with mappings duke@435: // established by mmap(), and mmap() always maps whole pages. Unaligned duke@435: // 'addr' likely indicates problem in the VM (e.g. trying to change duke@435: // protection of malloc'ed or statically allocated memory). Check the duke@435: // caller if you hit this assert. duke@435: assert(addr == bottom, "sanity check"); duke@435: duke@435: size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); duke@435: return ::mprotect(bottom, size, prot) == 0; duke@435: } duke@435: duke@435: bool os::protect_memory(char* addr, size_t size) { duke@435: return linux_mprotect(addr, size, PROT_READ); duke@435: } duke@435: duke@435: bool os::guard_memory(char* addr, size_t size) { duke@435: return linux_mprotect(addr, size, PROT_NONE); duke@435: } duke@435: duke@435: bool os::unguard_memory(char* addr, size_t size) { duke@435: return linux_mprotect(addr, size, PROT_READ|PROT_WRITE|PROT_EXEC); duke@435: } duke@435: duke@435: // Large page support duke@435: duke@435: static size_t _large_page_size = 0; duke@435: duke@435: bool os::large_page_init() { duke@435: if (!UseLargePages) return false; duke@435: duke@435: if (LargePageSizeInBytes) { duke@435: _large_page_size = LargePageSizeInBytes; duke@435: } else { duke@435: // large_page_size on Linux is used to round up heap size. x86 uses either duke@435: // 2M or 4M page, depending on whether PAE (Physical Address Extensions) duke@435: // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use duke@435: // page as large as 256M. duke@435: // duke@435: // Here we try to figure out page size by parsing /proc/meminfo and looking duke@435: // for a line with the following format: duke@435: // Hugepagesize: 2048 kB duke@435: // duke@435: // If we can't determine the value (e.g. /proc is not mounted, or the text duke@435: // format has been changed), we'll use the largest page size supported by duke@435: // the processor. duke@435: duke@435: _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M); duke@435: duke@435: FILE *fp = fopen("/proc/meminfo", "r"); duke@435: if (fp) { duke@435: while (!feof(fp)) { duke@435: int x = 0; duke@435: char buf[16]; duke@435: if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { duke@435: if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { duke@435: _large_page_size = x * K; duke@435: break; duke@435: } duke@435: } else { duke@435: // skip to next line duke@435: for (;;) { duke@435: int ch = fgetc(fp); duke@435: if (ch == EOF || ch == (int)'\n') break; duke@435: } duke@435: } duke@435: } duke@435: fclose(fp); duke@435: } duke@435: } duke@435: duke@435: const size_t default_page_size = (size_t)Linux::page_size(); duke@435: if (_large_page_size > default_page_size) { duke@435: _page_sizes[0] = _large_page_size; duke@435: _page_sizes[1] = default_page_size; duke@435: _page_sizes[2] = 0; duke@435: } duke@435: duke@435: // Large page support is available on 2.6 or newer kernel, some vendors duke@435: // (e.g. Redhat) have backported it to their 2.4 based distributions. duke@435: // We optimistically assume the support is available. If later it turns out duke@435: // not true, VM will automatically switch to use regular page size. duke@435: return true; duke@435: } duke@435: duke@435: #ifndef SHM_HUGETLB duke@435: #define SHM_HUGETLB 04000 duke@435: #endif duke@435: duke@435: char* os::reserve_memory_special(size_t bytes) { duke@435: assert(UseLargePages, "only for large pages"); duke@435: duke@435: key_t key = IPC_PRIVATE; duke@435: char *addr; duke@435: duke@435: bool warn_on_failure = UseLargePages && duke@435: (!FLAG_IS_DEFAULT(UseLargePages) || duke@435: !FLAG_IS_DEFAULT(LargePageSizeInBytes) duke@435: ); duke@435: char msg[128]; duke@435: duke@435: // Create a large shared memory region to attach to based on size. duke@435: // Currently, size is the total size of the heap duke@435: int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); duke@435: if (shmid == -1) { duke@435: // Possible reasons for shmget failure: duke@435: // 1. shmmax is too small for Java heap. duke@435: // > check shmmax value: cat /proc/sys/kernel/shmmax duke@435: // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax duke@435: // 2. not enough large page memory. duke@435: // > check available large pages: cat /proc/meminfo duke@435: // > increase amount of large pages: duke@435: // echo new_value > /proc/sys/vm/nr_hugepages duke@435: // Note 1: different Linux may use different name for this property, duke@435: // e.g. on Redhat AS-3 it is "hugetlb_pool". duke@435: // Note 2: it's possible there's enough physical memory available but duke@435: // they are so fragmented after a long run that they can't duke@435: // coalesce into large pages. Try to reserve large pages when duke@435: // the system is still "fresh". duke@435: if (warn_on_failure) { duke@435: jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); duke@435: warning(msg); duke@435: } duke@435: return NULL; duke@435: } duke@435: duke@435: // attach to the region duke@435: addr = (char*)shmat(shmid, NULL, 0); duke@435: int err = errno; duke@435: duke@435: // Remove shmid. If shmat() is successful, the actual shared memory segment duke@435: // will be deleted when it's detached by shmdt() or when the process duke@435: // terminates. If shmat() is not successful this will remove the shared duke@435: // segment immediately. duke@435: shmctl(shmid, IPC_RMID, NULL); duke@435: duke@435: if ((intptr_t)addr == -1) { duke@435: if (warn_on_failure) { duke@435: jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); duke@435: warning(msg); duke@435: } duke@435: return NULL; duke@435: } duke@435: duke@435: return addr; duke@435: } duke@435: duke@435: bool os::release_memory_special(char* base, size_t bytes) { duke@435: // detaching the SHM segment will also delete it, see reserve_memory_special() duke@435: int rslt = shmdt(base); duke@435: return rslt == 0; duke@435: } duke@435: duke@435: size_t os::large_page_size() { duke@435: return _large_page_size; duke@435: } duke@435: duke@435: // Linux does not support anonymous mmap with large page memory. The only way duke@435: // to reserve large page memory without file backing is through SysV shared duke@435: // memory API. The entire memory region is committed and pinned upfront. duke@435: // Hopefully this will change in the future... duke@435: bool os::can_commit_large_page_memory() { duke@435: return false; duke@435: } duke@435: duke@435: // Reserve memory at an arbitrary address, only if that area is duke@435: // available (and not reserved for something else). duke@435: duke@435: char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) { duke@435: const int max_tries = 10; duke@435: char* base[max_tries]; duke@435: size_t size[max_tries]; duke@435: const size_t gap = 0x000000; duke@435: duke@435: // Assert only that the size is a multiple of the page size, since duke@435: // that's all that mmap requires, and since that's all we really know duke@435: // about at this low abstraction level. If we need higher alignment, duke@435: // we can either pass an alignment to this method or verify alignment duke@435: // in one of the methods further up the call chain. See bug 5044738. duke@435: assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); duke@435: duke@435: // Repeatedly allocate blocks until the block is allocated at the duke@435: // right spot. Give up after max_tries. Note that reserve_memory() will duke@435: // automatically update _highest_vm_reserved_address if the call is duke@435: // successful. The variable tracks the highest memory address every reserved duke@435: // by JVM. It is used to detect heap-stack collision if running with duke@435: // fixed-stack LinuxThreads. Because here we may attempt to reserve more duke@435: // space than needed, it could confuse the collision detecting code. To duke@435: // solve the problem, save current _highest_vm_reserved_address and duke@435: // calculate the correct value before return. duke@435: address old_highest = _highest_vm_reserved_address; duke@435: duke@435: // Linux mmap allows caller to pass an address as hint; give it a try first, duke@435: // if kernel honors the hint then we can return immediately. duke@435: char * addr = anon_mmap(requested_addr, bytes, false); duke@435: if (addr == requested_addr) { duke@435: return requested_addr; duke@435: } duke@435: duke@435: if (addr != NULL) { duke@435: // mmap() is successful but it fails to reserve at the requested address duke@435: anon_munmap(addr, bytes); duke@435: } duke@435: duke@435: int i; duke@435: for (i = 0; i < max_tries; ++i) { duke@435: base[i] = reserve_memory(bytes); duke@435: duke@435: if (base[i] != NULL) { duke@435: // Is this the block we wanted? duke@435: if (base[i] == requested_addr) { duke@435: size[i] = bytes; duke@435: break; duke@435: } duke@435: duke@435: // Does this overlap the block we wanted? Give back the overlapped duke@435: // parts and try again. duke@435: duke@435: size_t top_overlap = requested_addr + (bytes + gap) - base[i]; duke@435: if (top_overlap >= 0 && top_overlap < bytes) { duke@435: unmap_memory(base[i], top_overlap); duke@435: base[i] += top_overlap; duke@435: size[i] = bytes - top_overlap; duke@435: } else { duke@435: size_t bottom_overlap = base[i] + bytes - requested_addr; duke@435: if (bottom_overlap >= 0 && bottom_overlap < bytes) { duke@435: unmap_memory(requested_addr, bottom_overlap); duke@435: size[i] = bytes - bottom_overlap; duke@435: } else { duke@435: size[i] = bytes; duke@435: } duke@435: } duke@435: } duke@435: } duke@435: duke@435: // Give back the unused reserved pieces. duke@435: duke@435: for (int j = 0; j < i; ++j) { duke@435: if (base[j] != NULL) { duke@435: unmap_memory(base[j], size[j]); duke@435: } duke@435: } duke@435: duke@435: if (i < max_tries) { duke@435: _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); duke@435: return requested_addr; duke@435: } else { duke@435: _highest_vm_reserved_address = old_highest; duke@435: return NULL; duke@435: } duke@435: } duke@435: duke@435: size_t os::read(int fd, void *buf, unsigned int nBytes) { duke@435: return ::read(fd, buf, nBytes); duke@435: } duke@435: duke@435: // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. duke@435: // Solaris uses poll(), linux uses park(). duke@435: // Poll() is likely a better choice, assuming that Thread.interrupt() duke@435: // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with duke@435: // SIGSEGV, see 4355769. duke@435: duke@435: const int NANOSECS_PER_MILLISECS = 1000000; duke@435: duke@435: int os::sleep(Thread* thread, jlong millis, bool interruptible) { duke@435: assert(thread == Thread::current(), "thread consistency check"); duke@435: duke@435: ParkEvent * const slp = thread->_SleepEvent ; duke@435: slp->reset() ; duke@435: OrderAccess::fence() ; duke@435: duke@435: if (interruptible) { duke@435: jlong prevtime = javaTimeNanos(); duke@435: duke@435: for (;;) { duke@435: if (os::is_interrupted(thread, true)) { duke@435: return OS_INTRPT; duke@435: } duke@435: duke@435: jlong newtime = javaTimeNanos(); duke@435: duke@435: if (newtime - prevtime < 0) { duke@435: // time moving backwards, should only happen if no monotonic clock duke@435: // not a guarantee() because JVM should not abort on kernel/glibc bugs duke@435: assert(!Linux::supports_monotonic_clock(), "time moving backwards"); duke@435: } else { duke@435: millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; duke@435: } duke@435: duke@435: if(millis <= 0) { duke@435: return OS_OK; duke@435: } duke@435: duke@435: prevtime = newtime; duke@435: duke@435: { duke@435: assert(thread->is_Java_thread(), "sanity check"); duke@435: JavaThread *jt = (JavaThread *) thread; duke@435: ThreadBlockInVM tbivm(jt); duke@435: OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); duke@435: duke@435: jt->set_suspend_equivalent(); duke@435: // cleared by handle_special_suspend_equivalent_condition() or duke@435: // java_suspend_self() via check_and_wait_while_suspended() duke@435: duke@435: slp->park(millis); duke@435: duke@435: // were we externally suspended while we were waiting? duke@435: jt->check_and_wait_while_suspended(); duke@435: } duke@435: } duke@435: } else { duke@435: OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); duke@435: jlong prevtime = javaTimeNanos(); duke@435: duke@435: for (;;) { duke@435: // It'd be nice to avoid the back-to-back javaTimeNanos() calls on duke@435: // the 1st iteration ... duke@435: jlong newtime = javaTimeNanos(); duke@435: duke@435: if (newtime - prevtime < 0) { duke@435: // time moving backwards, should only happen if no monotonic clock duke@435: // not a guarantee() because JVM should not abort on kernel/glibc bugs duke@435: assert(!Linux::supports_monotonic_clock(), "time moving backwards"); duke@435: } else { duke@435: millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; duke@435: } duke@435: duke@435: if(millis <= 0) break ; duke@435: duke@435: prevtime = newtime; duke@435: slp->park(millis); duke@435: } duke@435: return OS_OK ; duke@435: } duke@435: } duke@435: duke@435: int os::naked_sleep() { duke@435: // %% make the sleep time an integer flag. for now use 1 millisec. duke@435: return os::sleep(Thread::current(), 1, false); duke@435: } duke@435: duke@435: // Sleep forever; naked call to OS-specific sleep; use with CAUTION duke@435: void os::infinite_sleep() { duke@435: while (true) { // sleep forever ... duke@435: ::sleep(100); // ... 100 seconds at a time duke@435: } duke@435: } duke@435: duke@435: // Used to convert frequent JVM_Yield() to nops duke@435: bool os::dont_yield() { duke@435: return DontYieldALot; duke@435: } duke@435: duke@435: void os::yield() { duke@435: sched_yield(); duke@435: } duke@435: duke@435: os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} duke@435: duke@435: void os::yield_all(int attempts) { duke@435: // Yields to all threads, including threads with lower priorities duke@435: // Threads on Linux are all with same priority. The Solaris style duke@435: // os::yield_all() with nanosleep(1ms) is not necessary. duke@435: sched_yield(); duke@435: } duke@435: duke@435: // Called from the tight loops to possibly influence time-sharing heuristics duke@435: void os::loop_breaker(int attempts) { duke@435: os::yield_all(attempts); duke@435: } duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // thread priority support duke@435: duke@435: // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER duke@435: // only supports dynamic priority, static priority must be zero. For real-time duke@435: // applications, Linux supports SCHED_RR which allows static priority (1-99). duke@435: // However, for large multi-threaded applications, SCHED_RR is not only slower duke@435: // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out duke@435: // of 5 runs - Sep 2005). duke@435: // duke@435: // The following code actually changes the niceness of kernel-thread/LWP. It duke@435: // has an assumption that setpriority() only modifies one kernel-thread/LWP, duke@435: // not the entire user process, and user level threads are 1:1 mapped to kernel duke@435: // threads. It has always been the case, but could change in the future. For duke@435: // this reason, the code should not be used as default (ThreadPriorityPolicy=0). duke@435: // It is only used when ThreadPriorityPolicy=1 and requires root privilege. duke@435: duke@435: int os::java_to_os_priority[MaxPriority + 1] = { duke@435: 19, // 0 Entry should never be used duke@435: duke@435: 4, // 1 MinPriority duke@435: 3, // 2 duke@435: 2, // 3 duke@435: duke@435: 1, // 4 duke@435: 0, // 5 NormPriority duke@435: -1, // 6 duke@435: duke@435: -2, // 7 duke@435: -3, // 8 duke@435: -4, // 9 NearMaxPriority duke@435: duke@435: -5 // 10 MaxPriority duke@435: }; duke@435: duke@435: static int prio_init() { duke@435: if (ThreadPriorityPolicy == 1) { duke@435: // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 duke@435: // if effective uid is not root. Perhaps, a more elegant way of doing duke@435: // this is to test CAP_SYS_NICE capability, but that will require libcap.so duke@435: if (geteuid() != 0) { duke@435: if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { duke@435: warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); duke@435: } duke@435: ThreadPriorityPolicy = 0; duke@435: } duke@435: } duke@435: return 0; duke@435: } duke@435: duke@435: OSReturn os::set_native_priority(Thread* thread, int newpri) { duke@435: if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; duke@435: duke@435: int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); duke@435: return (ret == 0) ? OS_OK : OS_ERR; duke@435: } duke@435: duke@435: OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { duke@435: if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { duke@435: *priority_ptr = java_to_os_priority[NormPriority]; duke@435: return OS_OK; duke@435: } duke@435: duke@435: errno = 0; duke@435: *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); duke@435: return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); duke@435: } duke@435: duke@435: // Hint to the underlying OS that a task switch would not be good. duke@435: // Void return because it's a hint and can fail. duke@435: void os::hint_no_preempt() {} duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // suspend/resume support duke@435: duke@435: // the low-level signal-based suspend/resume support is a remnant from the duke@435: // old VM-suspension that used to be for java-suspension, safepoints etc, duke@435: // within hotspot. Now there is a single use-case for this: duke@435: // - calling get_thread_pc() on the VMThread by the flat-profiler task duke@435: // that runs in the watcher thread. duke@435: // The remaining code is greatly simplified from the more general suspension duke@435: // code that used to be used. duke@435: // duke@435: // The protocol is quite simple: duke@435: // - suspend: duke@435: // - sends a signal to the target thread duke@435: // - polls the suspend state of the osthread using a yield loop duke@435: // - target thread signal handler (SR_handler) sets suspend state duke@435: // and blocks in sigsuspend until continued duke@435: // - resume: duke@435: // - sets target osthread state to continue duke@435: // - sends signal to end the sigsuspend loop in the SR_handler duke@435: // duke@435: // Note that the SR_lock plays no role in this suspend/resume protocol. duke@435: // duke@435: duke@435: static void resume_clear_context(OSThread *osthread) { duke@435: osthread->set_ucontext(NULL); duke@435: osthread->set_siginfo(NULL); duke@435: duke@435: // notify the suspend action is completed, we have now resumed duke@435: osthread->sr.clear_suspended(); duke@435: } duke@435: duke@435: static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { duke@435: osthread->set_ucontext(context); duke@435: osthread->set_siginfo(siginfo); duke@435: } duke@435: duke@435: // duke@435: // Handler function invoked when a thread's execution is suspended or duke@435: // resumed. We have to be careful that only async-safe functions are duke@435: // called here (Note: most pthread functions are not async safe and duke@435: // should be avoided.) duke@435: // duke@435: // Note: sigwait() is a more natural fit than sigsuspend() from an duke@435: // interface point of view, but sigwait() prevents the signal hander duke@435: // from being run. libpthread would get very confused by not having duke@435: // its signal handlers run and prevents sigwait()'s use with the duke@435: // mutex granting granting signal. duke@435: // duke@435: // Currently only ever called on the VMThread duke@435: // duke@435: static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { duke@435: // Save and restore errno to avoid confusing native code with EINTR duke@435: // after sigsuspend. duke@435: int old_errno = errno; duke@435: duke@435: Thread* thread = Thread::current(); duke@435: OSThread* osthread = thread->osthread(); duke@435: assert(thread->is_VM_thread(), "Must be VMThread"); duke@435: // read current suspend action duke@435: int action = osthread->sr.suspend_action(); duke@435: if (action == SR_SUSPEND) { duke@435: suspend_save_context(osthread, siginfo, context); duke@435: duke@435: // Notify the suspend action is about to be completed. do_suspend() duke@435: // waits until SR_SUSPENDED is set and then returns. We will wait duke@435: // here for a resume signal and that completes the suspend-other duke@435: // action. do_suspend/do_resume is always called as a pair from duke@435: // the same thread - so there are no races duke@435: duke@435: // notify the caller duke@435: osthread->sr.set_suspended(); duke@435: duke@435: sigset_t suspend_set; // signals for sigsuspend() duke@435: duke@435: // get current set of blocked signals and unblock resume signal duke@435: pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); duke@435: sigdelset(&suspend_set, SR_signum); duke@435: duke@435: // wait here until we are resumed duke@435: do { duke@435: sigsuspend(&suspend_set); duke@435: // ignore all returns until we get a resume signal duke@435: } while (osthread->sr.suspend_action() != SR_CONTINUE); duke@435: duke@435: resume_clear_context(osthread); duke@435: duke@435: } else { duke@435: assert(action == SR_CONTINUE, "unexpected sr action"); duke@435: // nothing special to do - just leave the handler duke@435: } duke@435: duke@435: errno = old_errno; duke@435: } duke@435: duke@435: duke@435: static int SR_initialize() { duke@435: struct sigaction act; duke@435: char *s; duke@435: /* Get signal number to use for suspend/resume */ duke@435: if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { duke@435: int sig = ::strtol(s, 0, 10); duke@435: if (sig > 0 || sig < _NSIG) { duke@435: SR_signum = sig; duke@435: } duke@435: } duke@435: duke@435: assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, duke@435: "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); duke@435: duke@435: sigemptyset(&SR_sigset); duke@435: sigaddset(&SR_sigset, SR_signum); duke@435: duke@435: /* Set up signal handler for suspend/resume */ duke@435: act.sa_flags = SA_RESTART|SA_SIGINFO; duke@435: act.sa_handler = (void (*)(int)) SR_handler; duke@435: duke@435: // SR_signum is blocked by default. duke@435: // 4528190 - We also need to block pthread restart signal (32 on all duke@435: // supported Linux platforms). Note that LinuxThreads need to block duke@435: // this signal for all threads to work properly. So we don't have duke@435: // to use hard-coded signal number when setting up the mask. duke@435: pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); duke@435: duke@435: if (sigaction(SR_signum, &act, 0) == -1) { duke@435: return -1; duke@435: } duke@435: duke@435: // Save signal flag duke@435: os::Linux::set_our_sigflags(SR_signum, act.sa_flags); duke@435: return 0; duke@435: } duke@435: duke@435: static int SR_finalize() { duke@435: return 0; duke@435: } duke@435: duke@435: duke@435: // returns true on success and false on error - really an error is fatal duke@435: // but this seems the normal response to library errors duke@435: static bool do_suspend(OSThread* osthread) { duke@435: // mark as suspended and send signal duke@435: osthread->sr.set_suspend_action(SR_SUSPEND); duke@435: int status = pthread_kill(osthread->pthread_id(), SR_signum); duke@435: assert_status(status == 0, status, "pthread_kill"); duke@435: duke@435: // check status and wait until notified of suspension duke@435: if (status == 0) { duke@435: for (int i = 0; !osthread->sr.is_suspended(); i++) { duke@435: os::yield_all(i); duke@435: } duke@435: osthread->sr.set_suspend_action(SR_NONE); duke@435: return true; duke@435: } duke@435: else { duke@435: osthread->sr.set_suspend_action(SR_NONE); duke@435: return false; duke@435: } duke@435: } duke@435: duke@435: static void do_resume(OSThread* osthread) { duke@435: assert(osthread->sr.is_suspended(), "thread should be suspended"); duke@435: osthread->sr.set_suspend_action(SR_CONTINUE); duke@435: duke@435: int status = pthread_kill(osthread->pthread_id(), SR_signum); duke@435: assert_status(status == 0, status, "pthread_kill"); duke@435: // check status and wait unit notified of resumption duke@435: if (status == 0) { duke@435: for (int i = 0; osthread->sr.is_suspended(); i++) { duke@435: os::yield_all(i); duke@435: } duke@435: } duke@435: osthread->sr.set_suspend_action(SR_NONE); duke@435: } duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // interrupt support duke@435: duke@435: void os::interrupt(Thread* thread) { duke@435: assert(Thread::current() == thread || Threads_lock->owned_by_self(), duke@435: "possibility of dangling Thread pointer"); duke@435: duke@435: OSThread* osthread = thread->osthread(); duke@435: duke@435: if (!osthread->interrupted()) { duke@435: osthread->set_interrupted(true); duke@435: // More than one thread can get here with the same value of osthread, duke@435: // resulting in multiple notifications. We do, however, want the store duke@435: // to interrupted() to be visible to other threads before we execute unpark(). duke@435: OrderAccess::fence(); duke@435: ParkEvent * const slp = thread->_SleepEvent ; duke@435: if (slp != NULL) slp->unpark() ; duke@435: } duke@435: duke@435: // For JSR166. Unpark even if interrupt status already was set duke@435: if (thread->is_Java_thread()) duke@435: ((JavaThread*)thread)->parker()->unpark(); duke@435: duke@435: ParkEvent * ev = thread->_ParkEvent ; duke@435: if (ev != NULL) ev->unpark() ; duke@435: duke@435: } duke@435: duke@435: bool os::is_interrupted(Thread* thread, bool clear_interrupted) { duke@435: assert(Thread::current() == thread || Threads_lock->owned_by_self(), duke@435: "possibility of dangling Thread pointer"); duke@435: duke@435: OSThread* osthread = thread->osthread(); duke@435: duke@435: bool interrupted = osthread->interrupted(); duke@435: duke@435: if (interrupted && clear_interrupted) { duke@435: osthread->set_interrupted(false); duke@435: // consider thread->_SleepEvent->reset() ... optional optimization duke@435: } duke@435: duke@435: return interrupted; duke@435: } duke@435: duke@435: /////////////////////////////////////////////////////////////////////////////////// duke@435: // signal handling (except suspend/resume) duke@435: duke@435: // This routine may be used by user applications as a "hook" to catch signals. duke@435: // The user-defined signal handler must pass unrecognized signals to this duke@435: // routine, and if it returns true (non-zero), then the signal handler must duke@435: // return immediately. If the flag "abort_if_unrecognized" is true, then this duke@435: // routine will never retun false (zero), but instead will execute a VM panic duke@435: // routine kill the process. duke@435: // duke@435: // If this routine returns false, it is OK to call it again. This allows duke@435: // the user-defined signal handler to perform checks either before or after duke@435: // the VM performs its own checks. Naturally, the user code would be making duke@435: // a serious error if it tried to handle an exception (such as a null check duke@435: // or breakpoint) that the VM was generating for its own correct operation. duke@435: // duke@435: // This routine may recognize any of the following kinds of signals: duke@435: // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. duke@435: // It should be consulted by handlers for any of those signals. duke@435: // duke@435: // The caller of this routine must pass in the three arguments supplied duke@435: // to the function referred to in the "sa_sigaction" (not the "sa_handler") duke@435: // field of the structure passed to sigaction(). This routine assumes that duke@435: // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. duke@435: // duke@435: // Note that the VM will print warnings if it detects conflicting signal duke@435: // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". duke@435: // duke@435: extern "C" int duke@435: JVM_handle_linux_signal(int signo, siginfo_t* siginfo, duke@435: void* ucontext, int abort_if_unrecognized); duke@435: duke@435: void signalHandler(int sig, siginfo_t* info, void* uc) { duke@435: assert(info != NULL && uc != NULL, "it must be old kernel"); duke@435: JVM_handle_linux_signal(sig, info, uc, true); duke@435: } duke@435: duke@435: duke@435: // This boolean allows users to forward their own non-matching signals duke@435: // to JVM_handle_linux_signal, harmlessly. duke@435: bool os::Linux::signal_handlers_are_installed = false; duke@435: duke@435: // For signal-chaining duke@435: struct sigaction os::Linux::sigact[MAXSIGNUM]; duke@435: unsigned int os::Linux::sigs = 0; duke@435: bool os::Linux::libjsig_is_loaded = false; duke@435: typedef struct sigaction *(*get_signal_t)(int); duke@435: get_signal_t os::Linux::get_signal_action = NULL; duke@435: duke@435: struct sigaction* os::Linux::get_chained_signal_action(int sig) { duke@435: struct sigaction *actp = NULL; duke@435: duke@435: if (libjsig_is_loaded) { duke@435: // Retrieve the old signal handler from libjsig duke@435: actp = (*get_signal_action)(sig); duke@435: } duke@435: if (actp == NULL) { duke@435: // Retrieve the preinstalled signal handler from jvm duke@435: actp = get_preinstalled_handler(sig); duke@435: } duke@435: duke@435: return actp; duke@435: } duke@435: duke@435: static bool call_chained_handler(struct sigaction *actp, int sig, duke@435: siginfo_t *siginfo, void *context) { duke@435: // Call the old signal handler duke@435: if (actp->sa_handler == SIG_DFL) { duke@435: // It's more reasonable to let jvm treat it as an unexpected exception duke@435: // instead of taking the default action. duke@435: return false; duke@435: } else if (actp->sa_handler != SIG_IGN) { duke@435: if ((actp->sa_flags & SA_NODEFER) == 0) { duke@435: // automaticlly block the signal duke@435: sigaddset(&(actp->sa_mask), sig); duke@435: } duke@435: duke@435: sa_handler_t hand; duke@435: sa_sigaction_t sa; duke@435: bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; duke@435: // retrieve the chained handler duke@435: if (siginfo_flag_set) { duke@435: sa = actp->sa_sigaction; duke@435: } else { duke@435: hand = actp->sa_handler; duke@435: } duke@435: duke@435: if ((actp->sa_flags & SA_RESETHAND) != 0) { duke@435: actp->sa_handler = SIG_DFL; duke@435: } duke@435: duke@435: // try to honor the signal mask duke@435: sigset_t oset; duke@435: pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); duke@435: duke@435: // call into the chained handler duke@435: if (siginfo_flag_set) { duke@435: (*sa)(sig, siginfo, context); duke@435: } else { duke@435: (*hand)(sig); duke@435: } duke@435: duke@435: // restore the signal mask duke@435: pthread_sigmask(SIG_SETMASK, &oset, 0); duke@435: } duke@435: // Tell jvm's signal handler the signal is taken care of. duke@435: return true; duke@435: } duke@435: duke@435: bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { duke@435: bool chained = false; duke@435: // signal-chaining duke@435: if (UseSignalChaining) { duke@435: struct sigaction *actp = get_chained_signal_action(sig); duke@435: if (actp != NULL) { duke@435: chained = call_chained_handler(actp, sig, siginfo, context); duke@435: } duke@435: } duke@435: return chained; duke@435: } duke@435: duke@435: struct sigaction* os::Linux::get_preinstalled_handler(int sig) { duke@435: if ((( (unsigned int)1 << sig ) & sigs) != 0) { duke@435: return &sigact[sig]; duke@435: } duke@435: return NULL; duke@435: } duke@435: duke@435: void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { duke@435: assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); duke@435: sigact[sig] = oldAct; duke@435: sigs |= (unsigned int)1 << sig; duke@435: } duke@435: duke@435: // for diagnostic duke@435: int os::Linux::sigflags[MAXSIGNUM]; duke@435: duke@435: int os::Linux::get_our_sigflags(int sig) { duke@435: assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); duke@435: return sigflags[sig]; duke@435: } duke@435: duke@435: void os::Linux::set_our_sigflags(int sig, int flags) { duke@435: assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); duke@435: sigflags[sig] = flags; duke@435: } duke@435: duke@435: void os::Linux::set_signal_handler(int sig, bool set_installed) { duke@435: // Check for overwrite. duke@435: struct sigaction oldAct; duke@435: sigaction(sig, (struct sigaction*)NULL, &oldAct); duke@435: duke@435: void* oldhand = oldAct.sa_sigaction duke@435: ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) duke@435: : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); duke@435: if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && duke@435: oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && duke@435: oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { duke@435: if (AllowUserSignalHandlers || !set_installed) { duke@435: // Do not overwrite; user takes responsibility to forward to us. duke@435: return; duke@435: } else if (UseSignalChaining) { duke@435: // save the old handler in jvm duke@435: save_preinstalled_handler(sig, oldAct); duke@435: // libjsig also interposes the sigaction() call below and saves the duke@435: // old sigaction on it own. duke@435: } else { duke@435: fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig); duke@435: } duke@435: } duke@435: duke@435: struct sigaction sigAct; duke@435: sigfillset(&(sigAct.sa_mask)); duke@435: sigAct.sa_handler = SIG_DFL; duke@435: if (!set_installed) { duke@435: sigAct.sa_flags = SA_SIGINFO|SA_RESTART; duke@435: } else { duke@435: sigAct.sa_sigaction = signalHandler; duke@435: sigAct.sa_flags = SA_SIGINFO|SA_RESTART; duke@435: } duke@435: // Save flags, which are set by ours duke@435: assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); duke@435: sigflags[sig] = sigAct.sa_flags; duke@435: duke@435: int ret = sigaction(sig, &sigAct, &oldAct); duke@435: assert(ret == 0, "check"); duke@435: duke@435: void* oldhand2 = oldAct.sa_sigaction duke@435: ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) duke@435: : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); duke@435: assert(oldhand2 == oldhand, "no concurrent signal handler installation"); duke@435: } duke@435: duke@435: // install signal handlers for signals that HotSpot needs to duke@435: // handle in order to support Java-level exception handling. duke@435: duke@435: void os::Linux::install_signal_handlers() { duke@435: if (!signal_handlers_are_installed) { duke@435: signal_handlers_are_installed = true; duke@435: duke@435: // signal-chaining duke@435: typedef void (*signal_setting_t)(); duke@435: signal_setting_t begin_signal_setting = NULL; duke@435: signal_setting_t end_signal_setting = NULL; duke@435: begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, duke@435: dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); duke@435: if (begin_signal_setting != NULL) { duke@435: end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, duke@435: dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); duke@435: get_signal_action = CAST_TO_FN_PTR(get_signal_t, duke@435: dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); duke@435: libjsig_is_loaded = true; duke@435: assert(UseSignalChaining, "should enable signal-chaining"); duke@435: } duke@435: if (libjsig_is_loaded) { duke@435: // Tell libjsig jvm is setting signal handlers duke@435: (*begin_signal_setting)(); duke@435: } duke@435: duke@435: set_signal_handler(SIGSEGV, true); duke@435: set_signal_handler(SIGPIPE, true); duke@435: set_signal_handler(SIGBUS, true); duke@435: set_signal_handler(SIGILL, true); duke@435: set_signal_handler(SIGFPE, true); duke@435: set_signal_handler(SIGXFSZ, true); duke@435: duke@435: if (libjsig_is_loaded) { duke@435: // Tell libjsig jvm finishes setting signal handlers duke@435: (*end_signal_setting)(); duke@435: } duke@435: duke@435: // We don't activate signal checker if libjsig is in place, we trust ourselves duke@435: // and if UserSignalHandler is installed all bets are off duke@435: if (CheckJNICalls) { duke@435: if (libjsig_is_loaded) { duke@435: tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); duke@435: check_signals = false; duke@435: } duke@435: if (AllowUserSignalHandlers) { duke@435: tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); duke@435: check_signals = false; duke@435: } duke@435: } duke@435: } duke@435: } duke@435: duke@435: // This is the fastest way to get thread cpu time on Linux. duke@435: // Returns cpu time (user+sys) for any thread, not only for current. duke@435: // POSIX compliant clocks are implemented in the kernels 2.6.16+. duke@435: // It might work on 2.6.10+ with a special kernel/glibc patch. duke@435: // For reference, please, see IEEE Std 1003.1-2004: duke@435: // http://www.unix.org/single_unix_specification duke@435: duke@435: jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { duke@435: struct timespec tp; duke@435: int rc = os::Linux::clock_gettime(clockid, &tp); duke@435: assert(rc == 0, "clock_gettime is expected to return 0 code"); duke@435: duke@435: return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec; duke@435: } duke@435: duke@435: ///// duke@435: // glibc on Linux platform uses non-documented flag duke@435: // to indicate, that some special sort of signal duke@435: // trampoline is used. duke@435: // We will never set this flag, and we should duke@435: // ignore this flag in our diagnostic duke@435: #ifdef SIGNIFICANT_SIGNAL_MASK duke@435: #undef SIGNIFICANT_SIGNAL_MASK duke@435: #endif duke@435: #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) duke@435: duke@435: static const char* get_signal_handler_name(address handler, duke@435: char* buf, int buflen) { duke@435: int offset; duke@435: bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); duke@435: if (found) { duke@435: // skip directory names duke@435: const char *p1, *p2; duke@435: p1 = buf; duke@435: size_t len = strlen(os::file_separator()); duke@435: while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; duke@435: jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); duke@435: } else { duke@435: jio_snprintf(buf, buflen, PTR_FORMAT, handler); duke@435: } duke@435: return buf; duke@435: } duke@435: duke@435: static void print_signal_handler(outputStream* st, int sig, duke@435: char* buf, size_t buflen) { duke@435: struct sigaction sa; duke@435: duke@435: sigaction(sig, NULL, &sa); duke@435: duke@435: // See comment for SIGNIFICANT_SIGNAL_MASK define duke@435: sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; duke@435: duke@435: st->print("%s: ", os::exception_name(sig, buf, buflen)); duke@435: duke@435: address handler = (sa.sa_flags & SA_SIGINFO) duke@435: ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) duke@435: : CAST_FROM_FN_PTR(address, sa.sa_handler); duke@435: duke@435: if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { duke@435: st->print("SIG_DFL"); duke@435: } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { duke@435: st->print("SIG_IGN"); duke@435: } else { duke@435: st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); duke@435: } duke@435: duke@435: st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); duke@435: duke@435: address rh = VMError::get_resetted_sighandler(sig); duke@435: // May be, handler was resetted by VMError? duke@435: if(rh != NULL) { duke@435: handler = rh; duke@435: sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; duke@435: } duke@435: duke@435: st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); duke@435: duke@435: // Check: is it our handler? duke@435: if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || duke@435: handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { duke@435: // It is our signal handler duke@435: // check for flags, reset system-used one! duke@435: if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { duke@435: st->print( duke@435: ", flags was changed from " PTR32_FORMAT ", consider using jsig library", duke@435: os::Linux::get_our_sigflags(sig)); duke@435: } duke@435: } duke@435: st->cr(); duke@435: } duke@435: duke@435: duke@435: #define DO_SIGNAL_CHECK(sig) \ duke@435: if (!sigismember(&check_signal_done, sig)) \ duke@435: os::Linux::check_signal_handler(sig) duke@435: duke@435: // This method is a periodic task to check for misbehaving JNI applications duke@435: // under CheckJNI, we can add any periodic checks here duke@435: duke@435: void os::run_periodic_checks() { duke@435: duke@435: if (check_signals == false) return; duke@435: duke@435: // SEGV and BUS if overridden could potentially prevent duke@435: // generation of hs*.log in the event of a crash, debugging duke@435: // such a case can be very challenging, so we absolutely duke@435: // check the following for a good measure: duke@435: DO_SIGNAL_CHECK(SIGSEGV); duke@435: DO_SIGNAL_CHECK(SIGILL); duke@435: DO_SIGNAL_CHECK(SIGFPE); duke@435: DO_SIGNAL_CHECK(SIGBUS); duke@435: DO_SIGNAL_CHECK(SIGPIPE); duke@435: DO_SIGNAL_CHECK(SIGXFSZ); duke@435: duke@435: duke@435: // ReduceSignalUsage allows the user to override these handlers duke@435: // see comments at the very top and jvm_solaris.h duke@435: if (!ReduceSignalUsage) { duke@435: DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); duke@435: DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); duke@435: DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); duke@435: DO_SIGNAL_CHECK(BREAK_SIGNAL); duke@435: } duke@435: duke@435: DO_SIGNAL_CHECK(SR_signum); duke@435: DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); duke@435: } duke@435: duke@435: typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); duke@435: duke@435: static os_sigaction_t os_sigaction = NULL; duke@435: duke@435: void os::Linux::check_signal_handler(int sig) { duke@435: char buf[O_BUFLEN]; duke@435: address jvmHandler = NULL; duke@435: duke@435: duke@435: struct sigaction act; duke@435: if (os_sigaction == NULL) { duke@435: // only trust the default sigaction, in case it has been interposed duke@435: os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); duke@435: if (os_sigaction == NULL) return; duke@435: } duke@435: duke@435: os_sigaction(sig, (struct sigaction*)NULL, &act); duke@435: duke@435: duke@435: act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; duke@435: duke@435: address thisHandler = (act.sa_flags & SA_SIGINFO) duke@435: ? CAST_FROM_FN_PTR(address, act.sa_sigaction) duke@435: : CAST_FROM_FN_PTR(address, act.sa_handler) ; duke@435: duke@435: duke@435: switch(sig) { duke@435: case SIGSEGV: duke@435: case SIGBUS: duke@435: case SIGFPE: duke@435: case SIGPIPE: duke@435: case SIGILL: duke@435: case SIGXFSZ: duke@435: jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); duke@435: break; duke@435: duke@435: case SHUTDOWN1_SIGNAL: duke@435: case SHUTDOWN2_SIGNAL: duke@435: case SHUTDOWN3_SIGNAL: duke@435: case BREAK_SIGNAL: duke@435: jvmHandler = (address)user_handler(); duke@435: break; duke@435: duke@435: case INTERRUPT_SIGNAL: duke@435: jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); duke@435: break; duke@435: duke@435: default: duke@435: if (sig == SR_signum) { duke@435: jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); duke@435: } else { duke@435: return; duke@435: } duke@435: break; duke@435: } duke@435: duke@435: if (thisHandler != jvmHandler) { duke@435: tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); duke@435: tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); duke@435: tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); duke@435: // No need to check this sig any longer duke@435: sigaddset(&check_signal_done, sig); duke@435: } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { duke@435: tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); duke@435: tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); duke@435: tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); duke@435: // No need to check this sig any longer duke@435: sigaddset(&check_signal_done, sig); duke@435: } duke@435: duke@435: // Dump all the signal duke@435: if (sigismember(&check_signal_done, sig)) { duke@435: print_signal_handlers(tty, buf, O_BUFLEN); duke@435: } duke@435: } duke@435: duke@435: extern void report_error(char* file_name, int line_no, char* title, char* format, ...); duke@435: duke@435: extern bool signal_name(int signo, char* buf, size_t len); duke@435: duke@435: const char* os::exception_name(int exception_code, char* buf, size_t size) { duke@435: if (0 < exception_code && exception_code <= SIGRTMAX) { duke@435: // signal duke@435: if (!signal_name(exception_code, buf, size)) { duke@435: jio_snprintf(buf, size, "SIG%d", exception_code); duke@435: } duke@435: return buf; duke@435: } else { duke@435: return NULL; duke@435: } duke@435: } duke@435: duke@435: // this is called _before_ the most of global arguments have been parsed duke@435: void os::init(void) { duke@435: char dummy; /* used to get a guess on initial stack address */ duke@435: // first_hrtime = gethrtime(); duke@435: duke@435: // With LinuxThreads the JavaMain thread pid (primordial thread) duke@435: // is different than the pid of the java launcher thread. duke@435: // So, on Linux, the launcher thread pid is passed to the VM duke@435: // via the sun.java.launcher.pid property. duke@435: // Use this property instead of getpid() if it was correctly passed. duke@435: // See bug 6351349. duke@435: pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); duke@435: duke@435: _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); duke@435: duke@435: clock_tics_per_sec = sysconf(_SC_CLK_TCK); duke@435: duke@435: init_random(1234567); duke@435: duke@435: ThreadCritical::initialize(); duke@435: duke@435: Linux::set_page_size(sysconf(_SC_PAGESIZE)); duke@435: if (Linux::page_size() == -1) { duke@435: fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno)); duke@435: } duke@435: init_page_sizes((size_t) Linux::page_size()); duke@435: duke@435: Linux::initialize_system_info(); duke@435: duke@435: // main_thread points to the aboriginal thread duke@435: Linux::_main_thread = pthread_self(); duke@435: duke@435: Linux::clock_init(); duke@435: initial_time_count = os::elapsed_counter(); duke@435: } duke@435: duke@435: // To install functions for atexit system call duke@435: extern "C" { duke@435: static void perfMemory_exit_helper() { duke@435: perfMemory_exit(); duke@435: } duke@435: } duke@435: duke@435: // this is called _after_ the global arguments have been parsed duke@435: jint os::init_2(void) duke@435: { duke@435: Linux::fast_thread_clock_init(); duke@435: duke@435: // Allocate a single page and mark it as readable for safepoint polling duke@435: address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); duke@435: guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); duke@435: duke@435: os::set_polling_page( polling_page ); duke@435: duke@435: #ifndef PRODUCT duke@435: if(Verbose && PrintMiscellaneous) duke@435: tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); duke@435: #endif duke@435: duke@435: if (!UseMembar) { duke@435: address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); duke@435: guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); duke@435: os::set_memory_serialize_page( mem_serialize_page ); duke@435: duke@435: #ifndef PRODUCT duke@435: if(Verbose && PrintMiscellaneous) duke@435: tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); duke@435: #endif duke@435: } duke@435: duke@435: FLAG_SET_DEFAULT(UseLargePages, os::large_page_init()); duke@435: duke@435: // initialize suspend/resume support - must do this before signal_sets_init() duke@435: if (SR_initialize() != 0) { duke@435: perror("SR_initialize failed"); duke@435: return JNI_ERR; duke@435: } duke@435: duke@435: Linux::signal_sets_init(); duke@435: Linux::install_signal_handlers(); duke@435: duke@435: size_t threadStackSizeInBytes = ThreadStackSize * K; duke@435: if (threadStackSizeInBytes != 0 && duke@435: threadStackSizeInBytes < Linux::min_stack_allowed) { duke@435: tty->print_cr("\nThe stack size specified is too small, " duke@435: "Specify at least %dk", duke@435: Linux::min_stack_allowed / K); duke@435: return JNI_ERR; duke@435: } duke@435: duke@435: // Make the stack size a multiple of the page size so that duke@435: // the yellow/red zones can be guarded. duke@435: JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, duke@435: vm_page_size())); duke@435: duke@435: Linux::capture_initial_stack(JavaThread::stack_size_at_create()); duke@435: duke@435: Linux::libpthread_init(); duke@435: if (PrintMiscellaneous && (Verbose || WizardMode)) { duke@435: tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", duke@435: Linux::glibc_version(), Linux::libpthread_version(), duke@435: Linux::is_floating_stack() ? "floating stack" : "fixed stack"); duke@435: } duke@435: duke@435: if (MaxFDLimit) { duke@435: // set the number of file descriptors to max. print out error duke@435: // if getrlimit/setrlimit fails but continue regardless. duke@435: struct rlimit nbr_files; duke@435: int status = getrlimit(RLIMIT_NOFILE, &nbr_files); duke@435: if (status != 0) { duke@435: if (PrintMiscellaneous && (Verbose || WizardMode)) duke@435: perror("os::init_2 getrlimit failed"); duke@435: } else { duke@435: nbr_files.rlim_cur = nbr_files.rlim_max; duke@435: status = setrlimit(RLIMIT_NOFILE, &nbr_files); duke@435: if (status != 0) { duke@435: if (PrintMiscellaneous && (Verbose || WizardMode)) duke@435: perror("os::init_2 setrlimit failed"); duke@435: } duke@435: } duke@435: } duke@435: duke@435: // Initialize lock used to serialize thread creation (see os::create_thread) duke@435: Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); duke@435: duke@435: // Initialize HPI. duke@435: jint hpi_result = hpi::initialize(); duke@435: if (hpi_result != JNI_OK) { duke@435: tty->print_cr("There was an error trying to initialize the HPI library."); duke@435: return hpi_result; duke@435: } duke@435: duke@435: // at-exit methods are called in the reverse order of their registration. duke@435: // atexit functions are called on return from main or as a result of a duke@435: // call to exit(3C). There can be only 32 of these functions registered duke@435: // and atexit() does not set errno. duke@435: duke@435: if (PerfAllowAtExitRegistration) { duke@435: // only register atexit functions if PerfAllowAtExitRegistration is set. duke@435: // atexit functions can be delayed until process exit time, which duke@435: // can be problematic for embedded VM situations. Embedded VMs should duke@435: // call DestroyJavaVM() to assure that VM resources are released. duke@435: duke@435: // note: perfMemory_exit_helper atexit function may be removed in duke@435: // the future if the appropriate cleanup code can be added to the duke@435: // VM_Exit VMOperation's doit method. duke@435: if (atexit(perfMemory_exit_helper) != 0) { duke@435: warning("os::init2 atexit(perfMemory_exit_helper) failed"); duke@435: } duke@435: } duke@435: duke@435: // initialize thread priority policy duke@435: prio_init(); duke@435: duke@435: return JNI_OK; duke@435: } duke@435: duke@435: // Mark the polling page as unreadable duke@435: void os::make_polling_page_unreadable(void) { duke@435: if( !guard_memory((char*)_polling_page, Linux::page_size()) ) duke@435: fatal("Could not disable polling page"); duke@435: }; duke@435: duke@435: // Mark the polling page as readable duke@435: void os::make_polling_page_readable(void) { duke@435: if( !protect_memory((char *)_polling_page, Linux::page_size()) ) duke@435: fatal("Could not enable polling page"); duke@435: }; duke@435: duke@435: int os::active_processor_count() { duke@435: // Linux doesn't yet have a (official) notion of processor sets, duke@435: // so just return the number of online processors. duke@435: int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); duke@435: assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); duke@435: return online_cpus; duke@435: } duke@435: duke@435: bool os::distribute_processes(uint length, uint* distribution) { duke@435: // Not yet implemented. duke@435: return false; duke@435: } duke@435: duke@435: bool os::bind_to_processor(uint processor_id) { duke@435: // Not yet implemented. duke@435: return false; duke@435: } duke@435: duke@435: /// duke@435: duke@435: // Suspends the target using the signal mechanism and then grabs the PC before duke@435: // resuming the target. Used by the flat-profiler only duke@435: ExtendedPC os::get_thread_pc(Thread* thread) { duke@435: // Make sure that it is called by the watcher for the VMThread duke@435: assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); duke@435: assert(thread->is_VM_thread(), "Can only be called for VMThread"); duke@435: duke@435: ExtendedPC epc; duke@435: duke@435: OSThread* osthread = thread->osthread(); duke@435: if (do_suspend(osthread)) { duke@435: if (osthread->ucontext() != NULL) { duke@435: epc = os::Linux::ucontext_get_pc(osthread->ucontext()); duke@435: } else { duke@435: // NULL context is unexpected, double-check this is the VMThread duke@435: guarantee(thread->is_VM_thread(), "can only be called for VMThread"); duke@435: } duke@435: do_resume(osthread); duke@435: } duke@435: // failure means pthread_kill failed for some reason - arguably this is duke@435: // a fatal problem, but such problems are ignored elsewhere duke@435: duke@435: return epc; duke@435: } duke@435: duke@435: int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) duke@435: { duke@435: if (is_NPTL()) { duke@435: return pthread_cond_timedwait(_cond, _mutex, _abstime); duke@435: } else { duke@435: #ifndef IA64 duke@435: // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control duke@435: // word back to default 64bit precision if condvar is signaled. Java duke@435: // wants 53bit precision. Save and restore current value. duke@435: int fpu = get_fpu_control_word(); duke@435: #endif // IA64 duke@435: int status = pthread_cond_timedwait(_cond, _mutex, _abstime); duke@435: #ifndef IA64 duke@435: set_fpu_control_word(fpu); duke@435: #endif // IA64 duke@435: return status; duke@435: } duke@435: } duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // debug support duke@435: duke@435: #ifndef PRODUCT duke@435: static address same_page(address x, address y) { duke@435: int page_bits = -os::vm_page_size(); duke@435: if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) duke@435: return x; duke@435: else if (x > y) duke@435: return (address)(intptr_t(y) | ~page_bits) + 1; duke@435: else duke@435: return (address)(intptr_t(y) & page_bits); duke@435: } duke@435: duke@435: bool os::find(address addr) { duke@435: Dl_info dlinfo; duke@435: memset(&dlinfo, 0, sizeof(dlinfo)); duke@435: if (dladdr(addr, &dlinfo)) { duke@435: tty->print(PTR_FORMAT ": ", addr); duke@435: if (dlinfo.dli_sname != NULL) { duke@435: tty->print("%s+%#x", dlinfo.dli_sname, duke@435: addr - (intptr_t)dlinfo.dli_saddr); duke@435: } else if (dlinfo.dli_fname) { duke@435: tty->print("", addr - (intptr_t)dlinfo.dli_fbase); duke@435: } else { duke@435: tty->print(""); duke@435: } duke@435: if (dlinfo.dli_fname) { duke@435: tty->print(" in %s", dlinfo.dli_fname); duke@435: } duke@435: if (dlinfo.dli_fbase) { duke@435: tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase); duke@435: } duke@435: tty->cr(); duke@435: duke@435: if (Verbose) { duke@435: // decode some bytes around the PC duke@435: address begin = same_page(addr-40, addr); duke@435: address end = same_page(addr+40, addr); duke@435: address lowest = (address) dlinfo.dli_sname; duke@435: if (!lowest) lowest = (address) dlinfo.dli_fbase; duke@435: if (begin < lowest) begin = lowest; duke@435: Dl_info dlinfo2; duke@435: if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr duke@435: && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) duke@435: end = (address) dlinfo2.dli_saddr; duke@435: Disassembler::decode(begin, end); duke@435: } duke@435: return true; duke@435: } duke@435: return false; duke@435: } duke@435: duke@435: #endif duke@435: duke@435: //////////////////////////////////////////////////////////////////////////////// duke@435: // misc duke@435: duke@435: // This does not do anything on Linux. This is basically a hook for being duke@435: // able to use structured exception handling (thread-local exception filters) duke@435: // on, e.g., Win32. duke@435: void duke@435: os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, duke@435: JavaCallArguments* args, Thread* thread) { duke@435: f(value, method, args, thread); duke@435: } duke@435: duke@435: void os::print_statistics() { duke@435: } duke@435: duke@435: int os::message_box(const char* title, const char* message) { duke@435: int i; duke@435: fdStream err(defaultStream::error_fd()); duke@435: for (i = 0; i < 78; i++) err.print_raw("="); duke@435: err.cr(); duke@435: err.print_raw_cr(title); duke@435: for (i = 0; i < 78; i++) err.print_raw("-"); duke@435: err.cr(); duke@435: err.print_raw_cr(message); duke@435: for (i = 0; i < 78; i++) err.print_raw("="); duke@435: err.cr(); duke@435: duke@435: char buf[16]; duke@435: // Prevent process from exiting upon "read error" without consuming all CPU duke@435: while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } duke@435: duke@435: return buf[0] == 'y' || buf[0] == 'Y'; duke@435: } duke@435: duke@435: int os::stat(const char *path, struct stat *sbuf) { duke@435: char pathbuf[MAX_PATH]; duke@435: if (strlen(path) > MAX_PATH - 1) { duke@435: errno = ENAMETOOLONG; duke@435: return -1; duke@435: } duke@435: hpi::native_path(strcpy(pathbuf, path)); duke@435: return ::stat(pathbuf, sbuf); duke@435: } duke@435: duke@435: bool os::check_heap(bool force) { duke@435: return true; duke@435: } duke@435: duke@435: int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { duke@435: return ::vsnprintf(buf, count, format, args); duke@435: } duke@435: duke@435: // Is a (classpath) directory empty? duke@435: bool os::dir_is_empty(const char* path) { duke@435: DIR *dir = NULL; duke@435: struct dirent *ptr; duke@435: duke@435: dir = opendir(path); duke@435: if (dir == NULL) return true; duke@435: duke@435: /* Scan the directory */ duke@435: bool result = true; duke@435: char buf[sizeof(struct dirent) + MAX_PATH]; duke@435: while (result && (ptr = ::readdir(dir)) != NULL) { duke@435: if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { duke@435: result = false; duke@435: } duke@435: } duke@435: closedir(dir); duke@435: return result; duke@435: } duke@435: duke@435: // create binary file, rewriting existing file if required duke@435: int os::create_binary_file(const char* path, bool rewrite_existing) { duke@435: int oflags = O_WRONLY | O_CREAT; duke@435: if (!rewrite_existing) { duke@435: oflags |= O_EXCL; duke@435: } duke@435: return ::open64(path, oflags, S_IREAD | S_IWRITE); duke@435: } duke@435: duke@435: // return current position of file pointer duke@435: jlong os::current_file_offset(int fd) { duke@435: return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); duke@435: } duke@435: duke@435: // move file pointer to the specified offset duke@435: jlong os::seek_to_file_offset(int fd, jlong offset) { duke@435: return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); duke@435: } duke@435: duke@435: // Map a block of memory. duke@435: char* os::map_memory(int fd, const char* file_name, size_t file_offset, duke@435: char *addr, size_t bytes, bool read_only, duke@435: bool allow_exec) { duke@435: int prot; duke@435: int flags; duke@435: duke@435: if (read_only) { duke@435: prot = PROT_READ; duke@435: flags = MAP_SHARED; duke@435: } else { duke@435: prot = PROT_READ | PROT_WRITE; duke@435: flags = MAP_PRIVATE; duke@435: } duke@435: duke@435: if (allow_exec) { duke@435: prot |= PROT_EXEC; duke@435: } duke@435: duke@435: if (addr != NULL) { duke@435: flags |= MAP_FIXED; duke@435: } duke@435: duke@435: char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, duke@435: fd, file_offset); duke@435: if (mapped_address == MAP_FAILED) { duke@435: return NULL; duke@435: } duke@435: return mapped_address; duke@435: } duke@435: duke@435: duke@435: // Remap a block of memory. duke@435: char* os::remap_memory(int fd, const char* file_name, size_t file_offset, duke@435: char *addr, size_t bytes, bool read_only, duke@435: bool allow_exec) { duke@435: // same as map_memory() on this OS duke@435: return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, duke@435: allow_exec); duke@435: } duke@435: duke@435: duke@435: // Unmap a block of memory. duke@435: bool os::unmap_memory(char* addr, size_t bytes) { duke@435: return munmap(addr, bytes) == 0; duke@435: } duke@435: duke@435: static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); duke@435: duke@435: static clockid_t thread_cpu_clockid(Thread* thread) { duke@435: pthread_t tid = thread->osthread()->pthread_id(); duke@435: clockid_t clockid; duke@435: duke@435: // Get thread clockid duke@435: int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); duke@435: assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); duke@435: return clockid; duke@435: } duke@435: duke@435: // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) duke@435: // are used by JVM M&M and JVMTI to get user+sys or user CPU time duke@435: // of a thread. duke@435: // duke@435: // current_thread_cpu_time() and thread_cpu_time(Thread*) returns duke@435: // the fast estimate available on the platform. duke@435: duke@435: jlong os::current_thread_cpu_time() { duke@435: if (os::Linux::supports_fast_thread_cpu_time()) { duke@435: return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); duke@435: } else { duke@435: // return user + sys since the cost is the same duke@435: return slow_thread_cpu_time(Thread::current(), true /* user + sys */); duke@435: } duke@435: } duke@435: duke@435: jlong os::thread_cpu_time(Thread* thread) { duke@435: // consistent with what current_thread_cpu_time() returns duke@435: if (os::Linux::supports_fast_thread_cpu_time()) { duke@435: return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); duke@435: } else { duke@435: return slow_thread_cpu_time(thread, true /* user + sys */); duke@435: } duke@435: } duke@435: duke@435: jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { duke@435: if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { duke@435: return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); duke@435: } else { duke@435: return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); duke@435: } duke@435: } duke@435: duke@435: jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { duke@435: if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { duke@435: return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); duke@435: } else { duke@435: return slow_thread_cpu_time(thread, user_sys_cpu_time); duke@435: } duke@435: } duke@435: duke@435: // duke@435: // -1 on error. duke@435: // duke@435: duke@435: static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { duke@435: static bool proc_pid_cpu_avail = true; duke@435: static bool proc_task_unchecked = true; duke@435: static const char *proc_stat_path = "/proc/%d/stat"; duke@435: pid_t tid = thread->osthread()->thread_id(); duke@435: int i; duke@435: char *s; duke@435: char stat[2048]; duke@435: int statlen; duke@435: char proc_name[64]; duke@435: int count; duke@435: long sys_time, user_time; duke@435: char string[64]; duke@435: int idummy; duke@435: long ldummy; duke@435: FILE *fp; duke@435: duke@435: // We first try accessing /proc//cpu since this is faster to duke@435: // process. If this file is not present (linux kernels 2.5 and above) duke@435: // then we open /proc//stat. duke@435: if ( proc_pid_cpu_avail ) { duke@435: sprintf(proc_name, "/proc/%d/cpu", tid); duke@435: fp = fopen(proc_name, "r"); duke@435: if ( fp != NULL ) { duke@435: count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time); duke@435: fclose(fp); duke@435: if ( count != 3 ) return -1; duke@435: duke@435: if (user_sys_cpu_time) { duke@435: return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); duke@435: } else { duke@435: return (jlong)user_time * (1000000000 / clock_tics_per_sec); duke@435: } duke@435: } duke@435: else proc_pid_cpu_avail = false; duke@435: } duke@435: duke@435: // The /proc//stat aggregates per-process usage on duke@435: // new Linux kernels 2.6+ where NPTL is supported. duke@435: // The /proc/self/task//stat still has the per-thread usage. duke@435: // See bug 6328462. duke@435: // There can be no directory /proc/self/task on kernels 2.4 with NPTL duke@435: // and possibly in some other cases, so we check its availability. duke@435: if (proc_task_unchecked && os::Linux::is_NPTL()) { duke@435: // This is executed only once duke@435: proc_task_unchecked = false; duke@435: fp = fopen("/proc/self/task", "r"); duke@435: if (fp != NULL) { duke@435: proc_stat_path = "/proc/self/task/%d/stat"; duke@435: fclose(fp); duke@435: } duke@435: } duke@435: duke@435: sprintf(proc_name, proc_stat_path, tid); duke@435: fp = fopen(proc_name, "r"); duke@435: if ( fp == NULL ) return -1; duke@435: statlen = fread(stat, 1, 2047, fp); duke@435: stat[statlen] = '\0'; duke@435: fclose(fp); duke@435: duke@435: // Skip pid and the command string. Note that we could be dealing with duke@435: // weird command names, e.g. user could decide to rename java launcher duke@435: // to "java 1.4.2 :)", then the stat file would look like duke@435: // 1234 (java 1.4.2 :)) R ... ... duke@435: // We don't really need to know the command string, just find the last duke@435: // occurrence of ")" and then start parsing from there. See bug 4726580. duke@435: s = strrchr(stat, ')'); duke@435: i = 0; duke@435: if (s == NULL ) return -1; duke@435: duke@435: // Skip blank chars duke@435: do s++; while (isspace(*s)); duke@435: duke@435: count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", duke@435: &idummy, &idummy, &idummy, &idummy, &idummy, &idummy, duke@435: &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, duke@435: &user_time, &sys_time); duke@435: if ( count != 13 ) return -1; duke@435: if (user_sys_cpu_time) { duke@435: return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); duke@435: } else { duke@435: return (jlong)user_time * (1000000000 / clock_tics_per_sec); duke@435: } duke@435: } duke@435: duke@435: void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { duke@435: info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits duke@435: info_ptr->may_skip_backward = false; // elapsed time not wall time duke@435: info_ptr->may_skip_forward = false; // elapsed time not wall time duke@435: info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned duke@435: } duke@435: duke@435: void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { duke@435: info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits duke@435: info_ptr->may_skip_backward = false; // elapsed time not wall time duke@435: info_ptr->may_skip_forward = false; // elapsed time not wall time duke@435: info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned duke@435: } duke@435: duke@435: bool os::is_thread_cpu_time_supported() { duke@435: return true; duke@435: } duke@435: duke@435: // System loadavg support. Returns -1 if load average cannot be obtained. duke@435: // Linux doesn't yet have a (official) notion of processor sets, duke@435: // so just return the system wide load average. duke@435: int os::loadavg(double loadavg[], int nelem) { duke@435: return ::getloadavg(loadavg, nelem); duke@435: } duke@435: duke@435: void os::pause() { duke@435: char filename[MAX_PATH]; duke@435: if (PauseAtStartupFile && PauseAtStartupFile[0]) { duke@435: jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); duke@435: } else { duke@435: jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); duke@435: } duke@435: duke@435: int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); duke@435: if (fd != -1) { duke@435: struct stat buf; duke@435: close(fd); duke@435: while (::stat(filename, &buf) == 0) { duke@435: (void)::poll(NULL, 0, 100); duke@435: } duke@435: } else { duke@435: jio_fprintf(stderr, duke@435: "Could not open pause file '%s', continuing immediately.\n", filename); duke@435: } duke@435: } duke@435: duke@435: extern "C" { duke@435: duke@435: /** duke@435: * NOTE: the following code is to keep the green threads code duke@435: * in the libjava.so happy. Once the green threads is removed, duke@435: * these code will no longer be needed. duke@435: */ duke@435: int duke@435: jdk_waitpid(pid_t pid, int* status, int options) { duke@435: return waitpid(pid, status, options); duke@435: } duke@435: duke@435: int duke@435: fork1() { duke@435: return fork(); duke@435: } duke@435: duke@435: int duke@435: jdk_sem_init(sem_t *sem, int pshared, unsigned int value) { duke@435: return sem_init(sem, pshared, value); duke@435: } duke@435: duke@435: int duke@435: jdk_sem_post(sem_t *sem) { duke@435: return sem_post(sem); duke@435: } duke@435: duke@435: int duke@435: jdk_sem_wait(sem_t *sem) { duke@435: return sem_wait(sem); duke@435: } duke@435: duke@435: int duke@435: jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) { duke@435: return pthread_sigmask(how , newmask, oldmask); duke@435: } duke@435: duke@435: } duke@435: duke@435: // Refer to the comments in os_solaris.cpp park-unpark. duke@435: // duke@435: // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can duke@435: // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. duke@435: // For specifics regarding the bug see GLIBC BUGID 261237 : duke@435: // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. duke@435: // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future duke@435: // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar duke@435: // is used. (The simple C test-case provided in the GLIBC bug report manifests the duke@435: // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() duke@435: // and monitorenter when we're using 1-0 locking. All those operations may result in duke@435: // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version duke@435: // of libpthread avoids the problem, but isn't practical. duke@435: // duke@435: // Possible remedies: duke@435: // duke@435: // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. duke@435: // This is palliative and probabilistic, however. If the thread is preempted duke@435: // between the call to compute_abstime() and pthread_cond_timedwait(), more duke@435: // than the minimum period may have passed, and the abstime may be stale (in the duke@435: // past) resultin in a hang. Using this technique reduces the odds of a hang duke@435: // but the JVM is still vulnerable, particularly on heavily loaded systems. duke@435: // duke@435: // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead duke@435: // of the usual flag-condvar-mutex idiom. The write side of the pipe is set duke@435: // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) duke@435: // reduces to poll()+read(). This works well, but consumes 2 FDs per extant duke@435: // thread. duke@435: // duke@435: // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread duke@435: // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing duke@435: // a timeout request to the chron thread and then blocking via pthread_cond_wait(). duke@435: // This also works well. In fact it avoids kernel-level scalability impediments duke@435: // on certain platforms that don't handle lots of active pthread_cond_timedwait() duke@435: // timers in a graceful fashion. duke@435: // duke@435: // 4. When the abstime value is in the past it appears that control returns duke@435: // correctly from pthread_cond_timedwait(), but the condvar is left corrupt. duke@435: // Subsequent timedwait/wait calls may hang indefinitely. Given that, we duke@435: // can avoid the problem by reinitializing the condvar -- by cond_destroy() duke@435: // followed by cond_init() -- after all calls to pthread_cond_timedwait(). duke@435: // It may be possible to avoid reinitialization by checking the return duke@435: // value from pthread_cond_timedwait(). In addition to reinitializing the duke@435: // condvar we must establish the invariant that cond_signal() is only called duke@435: // within critical sections protected by the adjunct mutex. This prevents duke@435: // cond_signal() from "seeing" a condvar that's in the midst of being duke@435: // reinitialized or that is corrupt. Sadly, this invariant obviates the duke@435: // desirable signal-after-unlock optimization that avoids futile context switching. duke@435: // duke@435: // I'm also concerned that some versions of NTPL might allocate an auxilliary duke@435: // structure when a condvar is used or initialized. cond_destroy() would duke@435: // release the helper structure. Our reinitialize-after-timedwait fix duke@435: // put excessive stress on malloc/free and locks protecting the c-heap. duke@435: // duke@435: // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. duke@435: // It may be possible to refine (4) by checking the kernel and NTPL verisons duke@435: // and only enabling the work-around for vulnerable environments. duke@435: duke@435: // utility to compute the abstime argument to timedwait: duke@435: // millis is the relative timeout time duke@435: // abstime will be the absolute timeout time duke@435: // TODO: replace compute_abstime() with unpackTime() duke@435: duke@435: static struct timespec* compute_abstime(timespec* abstime, jlong millis) { duke@435: if (millis < 0) millis = 0; duke@435: struct timeval now; duke@435: int status = gettimeofday(&now, NULL); duke@435: assert(status == 0, "gettimeofday"); duke@435: jlong seconds = millis / 1000; duke@435: millis %= 1000; duke@435: if (seconds > 50000000) { // see man cond_timedwait(3T) duke@435: seconds = 50000000; duke@435: } duke@435: abstime->tv_sec = now.tv_sec + seconds; duke@435: long usec = now.tv_usec + millis * 1000; duke@435: if (usec >= 1000000) { duke@435: abstime->tv_sec += 1; duke@435: usec -= 1000000; duke@435: } duke@435: abstime->tv_nsec = usec * 1000; duke@435: return abstime; duke@435: } duke@435: duke@435: duke@435: // Test-and-clear _Event, always leaves _Event set to 0, returns immediately. duke@435: // Conceptually TryPark() should be equivalent to park(0). duke@435: duke@435: int os::PlatformEvent::TryPark() { duke@435: for (;;) { duke@435: const int v = _Event ; duke@435: guarantee ((v == 0) || (v == 1), "invariant") ; duke@435: if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; duke@435: } duke@435: } duke@435: duke@435: void os::PlatformEvent::park() { // AKA "down()" duke@435: // Invariant: Only the thread associated with the Event/PlatformEvent duke@435: // may call park(). duke@435: // TODO: assert that _Assoc != NULL or _Assoc == Self duke@435: int v ; duke@435: for (;;) { duke@435: v = _Event ; duke@435: if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; duke@435: } duke@435: guarantee (v >= 0, "invariant") ; duke@435: if (v == 0) { duke@435: // Do this the hard way by blocking ... duke@435: int status = pthread_mutex_lock(_mutex); duke@435: assert_status(status == 0, status, "mutex_lock"); duke@435: guarantee (_nParked == 0, "invariant") ; duke@435: ++ _nParked ; duke@435: while (_Event < 0) { duke@435: status = pthread_cond_wait(_cond, _mutex); duke@435: // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... duke@435: // Treat this the same as if the wait was interrupted duke@435: if (status == ETIME) { status = EINTR; } duke@435: assert_status(status == 0 || status == EINTR, status, "cond_wait"); duke@435: } duke@435: -- _nParked ; duke@435: duke@435: // In theory we could move the ST of 0 into _Event past the unlock(), duke@435: // but then we'd need a MEMBAR after the ST. duke@435: _Event = 0 ; duke@435: status = pthread_mutex_unlock(_mutex); duke@435: assert_status(status == 0, status, "mutex_unlock"); duke@435: } duke@435: guarantee (_Event >= 0, "invariant") ; duke@435: } duke@435: duke@435: int os::PlatformEvent::park(jlong millis) { duke@435: guarantee (_nParked == 0, "invariant") ; duke@435: duke@435: int v ; duke@435: for (;;) { duke@435: v = _Event ; duke@435: if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; duke@435: } duke@435: guarantee (v >= 0, "invariant") ; duke@435: if (v != 0) return OS_OK ; duke@435: duke@435: // We do this the hard way, by blocking the thread. duke@435: // Consider enforcing a minimum timeout value. duke@435: struct timespec abst; duke@435: compute_abstime(&abst, millis); duke@435: duke@435: int ret = OS_TIMEOUT; duke@435: int status = pthread_mutex_lock(_mutex); duke@435: assert_status(status == 0, status, "mutex_lock"); duke@435: guarantee (_nParked == 0, "invariant") ; duke@435: ++_nParked ; duke@435: duke@435: // Object.wait(timo) will return because of duke@435: // (a) notification duke@435: // (b) timeout duke@435: // (c) thread.interrupt duke@435: // duke@435: // Thread.interrupt and object.notify{All} both call Event::set. duke@435: // That is, we treat thread.interrupt as a special case of notification. duke@435: // The underlying Solaris implementation, cond_timedwait, admits duke@435: // spurious/premature wakeups, but the JLS/JVM spec prevents the duke@435: // JVM from making those visible to Java code. As such, we must duke@435: // filter out spurious wakeups. We assume all ETIME returns are valid. duke@435: // duke@435: // TODO: properly differentiate simultaneous notify+interrupt. duke@435: // In that case, we should propagate the notify to another waiter. duke@435: duke@435: while (_Event < 0) { duke@435: status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); duke@435: if (status != 0 && WorkAroundNPTLTimedWaitHang) { duke@435: pthread_cond_destroy (_cond); duke@435: pthread_cond_init (_cond, NULL) ; duke@435: } duke@435: assert_status(status == 0 || status == EINTR || duke@435: status == ETIME || status == ETIMEDOUT, duke@435: status, "cond_timedwait"); duke@435: if (!FilterSpuriousWakeups) break ; // previous semantics duke@435: if (status == ETIME || status == ETIMEDOUT) break ; duke@435: // We consume and ignore EINTR and spurious wakeups. duke@435: } duke@435: --_nParked ; duke@435: if (_Event >= 0) { duke@435: ret = OS_OK; duke@435: } duke@435: _Event = 0 ; duke@435: status = pthread_mutex_unlock(_mutex); duke@435: assert_status(status == 0, status, "mutex_unlock"); duke@435: assert (_nParked == 0, "invariant") ; duke@435: return ret; duke@435: } duke@435: duke@435: void os::PlatformEvent::unpark() { duke@435: int v, AnyWaiters ; duke@435: for (;;) { duke@435: v = _Event ; duke@435: if (v > 0) { duke@435: // The LD of _Event could have reordered or be satisfied duke@435: // by a read-aside from this processor's write buffer. duke@435: // To avoid problems execute a barrier and then duke@435: // ratify the value. duke@435: OrderAccess::fence() ; duke@435: if (_Event == v) return ; duke@435: continue ; duke@435: } duke@435: if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ; duke@435: } duke@435: if (v < 0) { duke@435: // Wait for the thread associated with the event to vacate duke@435: int status = pthread_mutex_lock(_mutex); duke@435: assert_status(status == 0, status, "mutex_lock"); duke@435: AnyWaiters = _nParked ; duke@435: assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ; duke@435: if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { duke@435: AnyWaiters = 0 ; duke@435: pthread_cond_signal (_cond); duke@435: } duke@435: status = pthread_mutex_unlock(_mutex); duke@435: assert_status(status == 0, status, "mutex_unlock"); duke@435: if (AnyWaiters != 0) { duke@435: status = pthread_cond_signal(_cond); duke@435: assert_status(status == 0, status, "cond_signal"); duke@435: } duke@435: } duke@435: duke@435: // Note that we signal() _after dropping the lock for "immortal" Events. duke@435: // This is safe and avoids a common class of futile wakeups. In rare duke@435: // circumstances this can cause a thread to return prematurely from duke@435: // cond_{timed}wait() but the spurious wakeup is benign and the victim will duke@435: // simply re-test the condition and re-park itself. duke@435: } duke@435: duke@435: duke@435: // JSR166 duke@435: // ------------------------------------------------------- duke@435: duke@435: /* duke@435: * The solaris and linux implementations of park/unpark are fairly duke@435: * conservative for now, but can be improved. They currently use a duke@435: * mutex/condvar pair, plus a a count. duke@435: * Park decrements count if > 0, else does a condvar wait. Unpark duke@435: * sets count to 1 and signals condvar. Only one thread ever waits duke@435: * on the condvar. Contention seen when trying to park implies that someone duke@435: * is unparking you, so don't wait. And spurious returns are fine, so there duke@435: * is no need to track notifications. duke@435: */ duke@435: duke@435: duke@435: #define NANOSECS_PER_SEC 1000000000 duke@435: #define NANOSECS_PER_MILLISEC 1000000 duke@435: #define MAX_SECS 100000000 duke@435: /* duke@435: * This code is common to linux and solaris and will be moved to a duke@435: * common place in dolphin. duke@435: * duke@435: * The passed in time value is either a relative time in nanoseconds duke@435: * or an absolute time in milliseconds. Either way it has to be unpacked duke@435: * into suitable seconds and nanoseconds components and stored in the duke@435: * given timespec structure. duke@435: * Given time is a 64-bit value and the time_t used in the timespec is only duke@435: * a signed-32-bit value (except on 64-bit Linux) we have to watch for duke@435: * overflow if times way in the future are given. Further on Solaris versions duke@435: * prior to 10 there is a restriction (see cond_timedwait) that the specified duke@435: * number of seconds, in abstime, is less than current_time + 100,000,000. duke@435: * As it will be 28 years before "now + 100000000" will overflow we can duke@435: * ignore overflow and just impose a hard-limit on seconds using the value duke@435: * of "now + 100,000,000". This places a limit on the timeout of about 3.17 duke@435: * years from "now". duke@435: */ duke@435: duke@435: static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { duke@435: assert (time > 0, "convertTime"); duke@435: duke@435: struct timeval now; duke@435: int status = gettimeofday(&now, NULL); duke@435: assert(status == 0, "gettimeofday"); duke@435: duke@435: time_t max_secs = now.tv_sec + MAX_SECS; duke@435: duke@435: if (isAbsolute) { duke@435: jlong secs = time / 1000; duke@435: if (secs > max_secs) { duke@435: absTime->tv_sec = max_secs; duke@435: } duke@435: else { duke@435: absTime->tv_sec = secs; duke@435: } duke@435: absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; duke@435: } duke@435: else { duke@435: jlong secs = time / NANOSECS_PER_SEC; duke@435: if (secs >= MAX_SECS) { duke@435: absTime->tv_sec = max_secs; duke@435: absTime->tv_nsec = 0; duke@435: } duke@435: else { duke@435: absTime->tv_sec = now.tv_sec + secs; duke@435: absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; duke@435: if (absTime->tv_nsec >= NANOSECS_PER_SEC) { duke@435: absTime->tv_nsec -= NANOSECS_PER_SEC; duke@435: ++absTime->tv_sec; // note: this must be <= max_secs duke@435: } duke@435: } duke@435: } duke@435: assert(absTime->tv_sec >= 0, "tv_sec < 0"); duke@435: assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); duke@435: assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); duke@435: assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); duke@435: } duke@435: duke@435: void Parker::park(bool isAbsolute, jlong time) { duke@435: // Optional fast-path check: duke@435: // Return immediately if a permit is available. duke@435: if (_counter > 0) { duke@435: _counter = 0 ; duke@435: return ; duke@435: } duke@435: duke@435: Thread* thread = Thread::current(); duke@435: assert(thread->is_Java_thread(), "Must be JavaThread"); duke@435: JavaThread *jt = (JavaThread *)thread; duke@435: duke@435: // Optional optimization -- avoid state transitions if there's an interrupt pending. duke@435: // Check interrupt before trying to wait duke@435: if (Thread::is_interrupted(thread, false)) { duke@435: return; duke@435: } duke@435: duke@435: // Next, demultiplex/decode time arguments duke@435: timespec absTime; duke@435: if (time < 0) { // don't wait at all duke@435: return; duke@435: } duke@435: if (time > 0) { duke@435: unpackTime(&absTime, isAbsolute, time); duke@435: } duke@435: duke@435: duke@435: // Enter safepoint region duke@435: // Beware of deadlocks such as 6317397. duke@435: // The per-thread Parker:: mutex is a classic leaf-lock. duke@435: // In particular a thread must never block on the Threads_lock while duke@435: // holding the Parker:: mutex. If safepoints are pending both the duke@435: // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. duke@435: ThreadBlockInVM tbivm(jt); duke@435: duke@435: // Don't wait if cannot get lock since interference arises from duke@435: // unblocking. Also. check interrupt before trying wait duke@435: if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { duke@435: return; duke@435: } duke@435: duke@435: int status ; duke@435: if (_counter > 0) { // no wait needed duke@435: _counter = 0; duke@435: status = pthread_mutex_unlock(_mutex); duke@435: assert (status == 0, "invariant") ; duke@435: return; duke@435: } duke@435: duke@435: #ifdef ASSERT duke@435: // Don't catch signals while blocked; let the running threads have the signals. duke@435: // (This allows a debugger to break into the running thread.) duke@435: sigset_t oldsigs; duke@435: sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); duke@435: pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); duke@435: #endif duke@435: duke@435: OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); duke@435: jt->set_suspend_equivalent(); duke@435: // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() duke@435: duke@435: if (time == 0) { duke@435: status = pthread_cond_wait (_cond, _mutex) ; duke@435: } else { duke@435: status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; duke@435: if (status != 0 && WorkAroundNPTLTimedWaitHang) { duke@435: pthread_cond_destroy (_cond) ; duke@435: pthread_cond_init (_cond, NULL); duke@435: } duke@435: } duke@435: assert_status(status == 0 || status == EINTR || duke@435: status == ETIME || status == ETIMEDOUT, duke@435: status, "cond_timedwait"); duke@435: duke@435: #ifdef ASSERT duke@435: pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); duke@435: #endif duke@435: duke@435: _counter = 0 ; duke@435: status = pthread_mutex_unlock(_mutex) ; duke@435: assert_status(status == 0, status, "invariant") ; duke@435: // If externally suspended while waiting, re-suspend duke@435: if (jt->handle_special_suspend_equivalent_condition()) { duke@435: jt->java_suspend_self(); duke@435: } duke@435: duke@435: } duke@435: duke@435: void Parker::unpark() { duke@435: int s, status ; duke@435: status = pthread_mutex_lock(_mutex); duke@435: assert (status == 0, "invariant") ; duke@435: s = _counter; duke@435: _counter = 1; duke@435: if (s < 1) { duke@435: if (WorkAroundNPTLTimedWaitHang) { duke@435: status = pthread_cond_signal (_cond) ; duke@435: assert (status == 0, "invariant") ; duke@435: status = pthread_mutex_unlock(_mutex); duke@435: assert (status == 0, "invariant") ; duke@435: } else { duke@435: status = pthread_mutex_unlock(_mutex); duke@435: assert (status == 0, "invariant") ; duke@435: status = pthread_cond_signal (_cond) ; duke@435: assert (status == 0, "invariant") ; duke@435: } duke@435: } else { duke@435: pthread_mutex_unlock(_mutex); duke@435: assert (status == 0, "invariant") ; duke@435: } duke@435: } duke@435: duke@435: duke@435: extern char** environ; duke@435: duke@435: #ifndef __NR_fork duke@435: #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) duke@435: #endif duke@435: duke@435: #ifndef __NR_execve duke@435: #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) duke@435: #endif duke@435: duke@435: // Run the specified command in a separate process. Return its exit value, duke@435: // or -1 on failure (e.g. can't fork a new process). duke@435: // Unlike system(), this function can be called from signal handler. It duke@435: // doesn't block SIGINT et al. duke@435: int os::fork_and_exec(char* cmd) { duke@435: char * argv[4]; duke@435: argv[0] = "sh"; duke@435: argv[1] = "-c"; duke@435: argv[2] = cmd; duke@435: argv[3] = NULL; duke@435: duke@435: // fork() in LinuxThreads/NPTL is not async-safe. It needs to run duke@435: // pthread_atfork handlers and reset pthread library. All we need is a duke@435: // separate process to execve. Make a direct syscall to fork process. duke@435: // On IA64 there's no fork syscall, we have to use fork() and hope for duke@435: // the best... duke@435: pid_t pid = NOT_IA64(syscall(__NR_fork);) duke@435: IA64_ONLY(fork();) duke@435: duke@435: if (pid < 0) { duke@435: // fork failed duke@435: return -1; duke@435: duke@435: } else if (pid == 0) { duke@435: // child process duke@435: duke@435: // execve() in LinuxThreads will call pthread_kill_other_threads_np() duke@435: // first to kill every thread on the thread list. Because this list is duke@435: // not reset by fork() (see notes above), execve() will instead kill duke@435: // every thread in the parent process. We know this is the only thread duke@435: // in the new process, so make a system call directly. duke@435: // IA64 should use normal execve() from glibc to match the glibc fork() duke@435: // above. duke@435: NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) duke@435: IA64_ONLY(execve("/bin/sh", argv, environ);) duke@435: duke@435: // execve failed duke@435: _exit(-1); duke@435: duke@435: } else { duke@435: // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't duke@435: // care about the actual exit code, for now. duke@435: duke@435: int status; duke@435: duke@435: // Wait for the child process to exit. This returns immediately if duke@435: // the child has already exited. */ duke@435: while (waitpid(pid, &status, 0) < 0) { duke@435: switch (errno) { duke@435: case ECHILD: return 0; duke@435: case EINTR: break; duke@435: default: return -1; duke@435: } duke@435: } duke@435: duke@435: if (WIFEXITED(status)) { duke@435: // The child exited normally; get its exit code. duke@435: return WEXITSTATUS(status); duke@435: } else if (WIFSIGNALED(status)) { duke@435: // The child exited because of a signal duke@435: // The best value to return is 0x80 + signal number, duke@435: // because that is what all Unix shells do, and because duke@435: // it allows callers to distinguish between process exit and duke@435: // process death by signal. duke@435: return 0x80 + WTERMSIG(status); duke@435: } else { duke@435: // Unknown exit code; pass it through duke@435: return status; duke@435: } duke@435: } duke@435: }