os_linux.cpp revision 1929:2d4762ec74af
1/* 2 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25# define __STDC_FORMAT_MACROS 26 27// no precompiled headers 28#include "classfile/classLoader.hpp" 29#include "classfile/systemDictionary.hpp" 30#include "classfile/vmSymbols.hpp" 31#include "code/icBuffer.hpp" 32#include "code/vtableStubs.hpp" 33#include "compiler/compileBroker.hpp" 34#include "interpreter/interpreter.hpp" 35#include "jvm_linux.h" 36#include "memory/allocation.inline.hpp" 37#include "memory/filemap.hpp" 38#include "mutex_linux.inline.hpp" 39#include "oops/oop.inline.hpp" 40#include "os_share_linux.hpp" 41#include "prims/jniFastGetField.hpp" 42#include "prims/jvm.h" 43#include "prims/jvm_misc.hpp" 44#include "runtime/arguments.hpp" 45#include "runtime/extendedPC.hpp" 46#include "runtime/globals.hpp" 47#include "runtime/hpi.hpp" 48#include "runtime/interfaceSupport.hpp" 49#include "runtime/java.hpp" 50#include "runtime/javaCalls.hpp" 51#include "runtime/mutexLocker.hpp" 52#include "runtime/objectMonitor.hpp" 53#include "runtime/osThread.hpp" 54#include "runtime/perfMemory.hpp" 55#include "runtime/sharedRuntime.hpp" 56#include "runtime/statSampler.hpp" 57#include "runtime/stubRoutines.hpp" 58#include "runtime/threadCritical.hpp" 59#include "runtime/timer.hpp" 60#include "services/attachListener.hpp" 61#include "services/runtimeService.hpp" 62#include "thread_linux.inline.hpp" 63#include "utilities/decoder.hpp" 64#include "utilities/defaultStream.hpp" 65#include "utilities/events.hpp" 66#include "utilities/growableArray.hpp" 67#include "utilities/vmError.hpp" 68#ifdef TARGET_ARCH_x86 69# include "assembler_x86.inline.hpp" 70# include "nativeInst_x86.hpp" 71#endif 72#ifdef TARGET_ARCH_sparc 73# include "assembler_sparc.inline.hpp" 74# include "nativeInst_sparc.hpp" 75#endif 76#ifdef TARGET_ARCH_zero 77# include "assembler_zero.inline.hpp" 78# include "nativeInst_zero.hpp" 79#endif 80#ifdef COMPILER1 81#include "c1/c1_Runtime1.hpp" 82#endif 83#ifdef COMPILER2 84#include "opto/runtime.hpp" 85#endif 86 87// put OS-includes here 88# include <sys/types.h> 89# include <sys/mman.h> 90# include <sys/stat.h> 91# include <sys/select.h> 92# include <pthread.h> 93# include <signal.h> 94# include <errno.h> 95# include <dlfcn.h> 96# include <stdio.h> 97# include <unistd.h> 98# include <sys/resource.h> 99# include <pthread.h> 100# include <sys/stat.h> 101# include <sys/time.h> 102# include <sys/times.h> 103# include <sys/utsname.h> 104# include <sys/socket.h> 105# include <sys/wait.h> 106# include <pwd.h> 107# include <poll.h> 108# include <semaphore.h> 109# include <fcntl.h> 110# include <string.h> 111# include <syscall.h> 112# include <sys/sysinfo.h> 113# include <gnu/libc-version.h> 114# include <sys/ipc.h> 115# include <sys/shm.h> 116# include <link.h> 117# include <stdint.h> 118# include <inttypes.h> 119 120#define MAX_PATH (2 * K) 121 122// for timer info max values which include all bits 123#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 124#define SEC_IN_NANOSECS 1000000000LL 125 126//////////////////////////////////////////////////////////////////////////////// 127// global variables 128julong os::Linux::_physical_memory = 0; 129 130address os::Linux::_initial_thread_stack_bottom = NULL; 131uintptr_t os::Linux::_initial_thread_stack_size = 0; 132 133int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 134int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 135Mutex* os::Linux::_createThread_lock = NULL; 136pthread_t os::Linux::_main_thread; 137int os::Linux::_page_size = -1; 138bool os::Linux::_is_floating_stack = false; 139bool os::Linux::_is_NPTL = false; 140bool os::Linux::_supports_fast_thread_cpu_time = false; 141const char * os::Linux::_glibc_version = NULL; 142const char * os::Linux::_libpthread_version = NULL; 143 144static jlong initial_time_count=0; 145 146static int clock_tics_per_sec = 100; 147 148// For diagnostics to print a message once. see run_periodic_checks 149static sigset_t check_signal_done; 150static bool check_signals = true;; 151 152static pid_t _initial_pid = 0; 153 154/* Signal number used to suspend/resume a thread */ 155 156/* do not use any signal number less than SIGSEGV, see 4355769 */ 157static int SR_signum = SIGUSR2; 158sigset_t SR_sigset; 159 160/* Used to protect dlsym() calls */ 161static pthread_mutex_t dl_mutex; 162 163//////////////////////////////////////////////////////////////////////////////// 164// utility functions 165 166static int SR_initialize(); 167static int SR_finalize(); 168 169julong os::available_memory() { 170 return Linux::available_memory(); 171} 172 173julong os::Linux::available_memory() { 174 // values in struct sysinfo are "unsigned long" 175 struct sysinfo si; 176 sysinfo(&si); 177 178 return (julong)si.freeram * si.mem_unit; 179} 180 181julong os::physical_memory() { 182 return Linux::physical_memory(); 183} 184 185julong os::allocatable_physical_memory(julong size) { 186#ifdef _LP64 187 return size; 188#else 189 julong result = MIN2(size, (julong)3800*M); 190 if (!is_allocatable(result)) { 191 // See comments under solaris for alignment considerations 192 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); 193 result = MIN2(size, reasonable_size); 194 } 195 return result; 196#endif // _LP64 197} 198 199//////////////////////////////////////////////////////////////////////////////// 200// environment support 201 202bool os::getenv(const char* name, char* buf, int len) { 203 const char* val = ::getenv(name); 204 if (val != NULL && strlen(val) < (size_t)len) { 205 strcpy(buf, val); 206 return true; 207 } 208 if (len > 0) buf[0] = 0; // return a null string 209 return false; 210} 211 212 213// Return true if user is running as root. 214 215bool os::have_special_privileges() { 216 static bool init = false; 217 static bool privileges = false; 218 if (!init) { 219 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 220 init = true; 221 } 222 return privileges; 223} 224 225 226#ifndef SYS_gettid 227// i386: 224, ia64: 1105, amd64: 186, sparc 143 228#ifdef __ia64__ 229#define SYS_gettid 1105 230#elif __i386__ 231#define SYS_gettid 224 232#elif __amd64__ 233#define SYS_gettid 186 234#elif __sparc__ 235#define SYS_gettid 143 236#else 237#error define gettid for the arch 238#endif 239#endif 240 241// Cpu architecture string 242#if defined(ZERO) 243static char cpu_arch[] = ZERO_LIBARCH; 244#elif defined(IA64) 245static char cpu_arch[] = "ia64"; 246#elif defined(IA32) 247static char cpu_arch[] = "i386"; 248#elif defined(AMD64) 249static char cpu_arch[] = "amd64"; 250#elif defined(ARM) 251static char cpu_arch[] = "arm"; 252#elif defined(PPC) 253static char cpu_arch[] = "ppc"; 254#elif defined(SPARC) 255# ifdef _LP64 256static char cpu_arch[] = "sparcv9"; 257# else 258static char cpu_arch[] = "sparc"; 259# endif 260#else 261#error Add appropriate cpu_arch setting 262#endif 263 264 265// pid_t gettid() 266// 267// Returns the kernel thread id of the currently running thread. Kernel 268// thread id is used to access /proc. 269// 270// (Note that getpid() on LinuxThreads returns kernel thread id too; but 271// on NPTL, it returns the same pid for all threads, as required by POSIX.) 272// 273pid_t os::Linux::gettid() { 274 int rslt = syscall(SYS_gettid); 275 if (rslt == -1) { 276 // old kernel, no NPTL support 277 return getpid(); 278 } else { 279 return (pid_t)rslt; 280 } 281} 282 283// Most versions of linux have a bug where the number of processors are 284// determined by looking at the /proc file system. In a chroot environment, 285// the system call returns 1. This causes the VM to act as if it is 286// a single processor and elide locking (see is_MP() call). 287static bool unsafe_chroot_detected = false; 288static const char *unstable_chroot_error = "/proc file system not found.\n" 289 "Java may be unstable running multithreaded in a chroot " 290 "environment on Linux when /proc filesystem is not mounted."; 291 292void os::Linux::initialize_system_info() { 293 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 294 if (processor_count() == 1) { 295 pid_t pid = os::Linux::gettid(); 296 char fname[32]; 297 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 298 FILE *fp = fopen(fname, "r"); 299 if (fp == NULL) { 300 unsafe_chroot_detected = true; 301 } else { 302 fclose(fp); 303 } 304 } 305 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 306 assert(processor_count() > 0, "linux error"); 307} 308 309void os::init_system_properties_values() { 310// char arch[12]; 311// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 312 313 // The next steps are taken in the product version: 314 // 315 // Obtain the JAVA_HOME value from the location of libjvm[_g].so. 316 // This library should be located at: 317 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so. 318 // 319 // If "/jre/lib/" appears at the right place in the path, then we 320 // assume libjvm[_g].so is installed in a JDK and we use this path. 321 // 322 // Otherwise exit with message: "Could not create the Java virtual machine." 323 // 324 // The following extra steps are taken in the debugging version: 325 // 326 // If "/jre/lib/" does NOT appear at the right place in the path 327 // instead of exit check for $JAVA_HOME environment variable. 328 // 329 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 330 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so 331 // it looks like libjvm[_g].so is installed there 332 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so. 333 // 334 // Otherwise exit. 335 // 336 // Important note: if the location of libjvm.so changes this 337 // code needs to be changed accordingly. 338 339 // The next few definitions allow the code to be verbatim: 340#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n)) 341#define getenv(n) ::getenv(n) 342 343/* 344 * See ld(1): 345 * The linker uses the following search paths to locate required 346 * shared libraries: 347 * 1: ... 348 * ... 349 * 7: The default directories, normally /lib and /usr/lib. 350 */ 351#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 352#define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 353#else 354#define DEFAULT_LIBPATH "/lib:/usr/lib" 355#endif 356 357#define EXTENSIONS_DIR "/lib/ext" 358#define ENDORSED_DIR "/lib/endorsed" 359#define REG_DIR "/usr/java/packages" 360 361 { 362 /* sysclasspath, java_home, dll_dir */ 363 { 364 char *home_path; 365 char *dll_path; 366 char *pslash; 367 char buf[MAXPATHLEN]; 368 os::jvm_path(buf, sizeof(buf)); 369 370 // Found the full path to libjvm.so. 371 // Now cut the path to <java_home>/jre if we can. 372 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 373 pslash = strrchr(buf, '/'); 374 if (pslash != NULL) 375 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 376 dll_path = malloc(strlen(buf) + 1); 377 if (dll_path == NULL) 378 return; 379 strcpy(dll_path, buf); 380 Arguments::set_dll_dir(dll_path); 381 382 if (pslash != NULL) { 383 pslash = strrchr(buf, '/'); 384 if (pslash != NULL) { 385 *pslash = '\0'; /* get rid of /<arch> */ 386 pslash = strrchr(buf, '/'); 387 if (pslash != NULL) 388 *pslash = '\0'; /* get rid of /lib */ 389 } 390 } 391 392 home_path = malloc(strlen(buf) + 1); 393 if (home_path == NULL) 394 return; 395 strcpy(home_path, buf); 396 Arguments::set_java_home(home_path); 397 398 if (!set_boot_path('/', ':')) 399 return; 400 } 401 402 /* 403 * Where to look for native libraries 404 * 405 * Note: Due to a legacy implementation, most of the library path 406 * is set in the launcher. This was to accomodate linking restrictions 407 * on legacy Linux implementations (which are no longer supported). 408 * Eventually, all the library path setting will be done here. 409 * 410 * However, to prevent the proliferation of improperly built native 411 * libraries, the new path component /usr/java/packages is added here. 412 * Eventually, all the library path setting will be done here. 413 */ 414 { 415 char *ld_library_path; 416 417 /* 418 * Construct the invariant part of ld_library_path. Note that the 419 * space for the colon and the trailing null are provided by the 420 * nulls included by the sizeof operator (so actually we allocate 421 * a byte more than necessary). 422 */ 423 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 424 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 425 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 426 427 /* 428 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 429 * should always exist (until the legacy problem cited above is 430 * addressed). 431 */ 432 char *v = getenv("LD_LIBRARY_PATH"); 433 if (v != NULL) { 434 char *t = ld_library_path; 435 /* That's +1 for the colon and +1 for the trailing '\0' */ 436 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 437 sprintf(ld_library_path, "%s:%s", v, t); 438 } 439 Arguments::set_library_path(ld_library_path); 440 } 441 442 /* 443 * Extensions directories. 444 * 445 * Note that the space for the colon and the trailing null are provided 446 * by the nulls included by the sizeof operator (so actually one byte more 447 * than necessary is allocated). 448 */ 449 { 450 char *buf = malloc(strlen(Arguments::get_java_home()) + 451 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 452 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 453 Arguments::get_java_home()); 454 Arguments::set_ext_dirs(buf); 455 } 456 457 /* Endorsed standards default directory. */ 458 { 459 char * buf; 460 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 461 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 462 Arguments::set_endorsed_dirs(buf); 463 } 464 } 465 466#undef malloc 467#undef getenv 468#undef EXTENSIONS_DIR 469#undef ENDORSED_DIR 470 471 // Done 472 return; 473} 474 475//////////////////////////////////////////////////////////////////////////////// 476// breakpoint support 477 478void os::breakpoint() { 479 BREAKPOINT; 480} 481 482extern "C" void breakpoint() { 483 // use debugger to set breakpoint here 484} 485 486//////////////////////////////////////////////////////////////////////////////// 487// signal support 488 489debug_only(static bool signal_sets_initialized = false); 490static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 491 492bool os::Linux::is_sig_ignored(int sig) { 493 struct sigaction oact; 494 sigaction(sig, (struct sigaction*)NULL, &oact); 495 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 496 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 497 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 498 return true; 499 else 500 return false; 501} 502 503void os::Linux::signal_sets_init() { 504 // Should also have an assertion stating we are still single-threaded. 505 assert(!signal_sets_initialized, "Already initialized"); 506 // Fill in signals that are necessarily unblocked for all threads in 507 // the VM. Currently, we unblock the following signals: 508 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 509 // by -Xrs (=ReduceSignalUsage)); 510 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 511 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 512 // the dispositions or masks wrt these signals. 513 // Programs embedding the VM that want to use the above signals for their 514 // own purposes must, at this time, use the "-Xrs" option to prevent 515 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 516 // (See bug 4345157, and other related bugs). 517 // In reality, though, unblocking these signals is really a nop, since 518 // these signals are not blocked by default. 519 sigemptyset(&unblocked_sigs); 520 sigemptyset(&allowdebug_blocked_sigs); 521 sigaddset(&unblocked_sigs, SIGILL); 522 sigaddset(&unblocked_sigs, SIGSEGV); 523 sigaddset(&unblocked_sigs, SIGBUS); 524 sigaddset(&unblocked_sigs, SIGFPE); 525 sigaddset(&unblocked_sigs, SR_signum); 526 527 if (!ReduceSignalUsage) { 528 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 529 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 530 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 531 } 532 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 533 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 534 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 535 } 536 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 537 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 538 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 539 } 540 } 541 // Fill in signals that are blocked by all but the VM thread. 542 sigemptyset(&vm_sigs); 543 if (!ReduceSignalUsage) 544 sigaddset(&vm_sigs, BREAK_SIGNAL); 545 debug_only(signal_sets_initialized = true); 546 547} 548 549// These are signals that are unblocked while a thread is running Java. 550// (For some reason, they get blocked by default.) 551sigset_t* os::Linux::unblocked_signals() { 552 assert(signal_sets_initialized, "Not initialized"); 553 return &unblocked_sigs; 554} 555 556// These are the signals that are blocked while a (non-VM) thread is 557// running Java. Only the VM thread handles these signals. 558sigset_t* os::Linux::vm_signals() { 559 assert(signal_sets_initialized, "Not initialized"); 560 return &vm_sigs; 561} 562 563// These are signals that are blocked during cond_wait to allow debugger in 564sigset_t* os::Linux::allowdebug_blocked_signals() { 565 assert(signal_sets_initialized, "Not initialized"); 566 return &allowdebug_blocked_sigs; 567} 568 569void os::Linux::hotspot_sigmask(Thread* thread) { 570 571 //Save caller's signal mask before setting VM signal mask 572 sigset_t caller_sigmask; 573 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 574 575 OSThread* osthread = thread->osthread(); 576 osthread->set_caller_sigmask(caller_sigmask); 577 578 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 579 580 if (!ReduceSignalUsage) { 581 if (thread->is_VM_thread()) { 582 // Only the VM thread handles BREAK_SIGNAL ... 583 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 584 } else { 585 // ... all other threads block BREAK_SIGNAL 586 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 587 } 588 } 589} 590 591////////////////////////////////////////////////////////////////////////////// 592// detecting pthread library 593 594void os::Linux::libpthread_init() { 595 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 596 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 597 // generic name for earlier versions. 598 // Define macros here so we can build HotSpot on old systems. 599# ifndef _CS_GNU_LIBC_VERSION 600# define _CS_GNU_LIBC_VERSION 2 601# endif 602# ifndef _CS_GNU_LIBPTHREAD_VERSION 603# define _CS_GNU_LIBPTHREAD_VERSION 3 604# endif 605 606 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 607 if (n > 0) { 608 char *str = (char *)malloc(n); 609 confstr(_CS_GNU_LIBC_VERSION, str, n); 610 os::Linux::set_glibc_version(str); 611 } else { 612 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 613 static char _gnu_libc_version[32]; 614 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 615 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 616 os::Linux::set_glibc_version(_gnu_libc_version); 617 } 618 619 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 620 if (n > 0) { 621 char *str = (char *)malloc(n); 622 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 623 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 624 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 625 // is the case. LinuxThreads has a hard limit on max number of threads. 626 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 627 // On the other hand, NPTL does not have such a limit, sysconf() 628 // will return -1 and errno is not changed. Check if it is really NPTL. 629 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 630 strstr(str, "NPTL") && 631 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 632 free(str); 633 os::Linux::set_libpthread_version("linuxthreads"); 634 } else { 635 os::Linux::set_libpthread_version(str); 636 } 637 } else { 638 // glibc before 2.3.2 only has LinuxThreads. 639 os::Linux::set_libpthread_version("linuxthreads"); 640 } 641 642 if (strstr(libpthread_version(), "NPTL")) { 643 os::Linux::set_is_NPTL(); 644 } else { 645 os::Linux::set_is_LinuxThreads(); 646 } 647 648 // LinuxThreads have two flavors: floating-stack mode, which allows variable 649 // stack size; and fixed-stack mode. NPTL is always floating-stack. 650 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 651 os::Linux::set_is_floating_stack(); 652 } 653} 654 655///////////////////////////////////////////////////////////////////////////// 656// thread stack 657 658// Force Linux kernel to expand current thread stack. If "bottom" is close 659// to the stack guard, caller should block all signals. 660// 661// MAP_GROWSDOWN: 662// A special mmap() flag that is used to implement thread stacks. It tells 663// kernel that the memory region should extend downwards when needed. This 664// allows early versions of LinuxThreads to only mmap the first few pages 665// when creating a new thread. Linux kernel will automatically expand thread 666// stack as needed (on page faults). 667// 668// However, because the memory region of a MAP_GROWSDOWN stack can grow on 669// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 670// region, it's hard to tell if the fault is due to a legitimate stack 671// access or because of reading/writing non-exist memory (e.g. buffer 672// overrun). As a rule, if the fault happens below current stack pointer, 673// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 674// application (see Linux kernel fault.c). 675// 676// This Linux feature can cause SIGSEGV when VM bangs thread stack for 677// stack overflow detection. 678// 679// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 680// not use this flag. However, the stack of initial thread is not created 681// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 682// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 683// and then attach the thread to JVM. 684// 685// To get around the problem and allow stack banging on Linux, we need to 686// manually expand thread stack after receiving the SIGSEGV. 687// 688// There are two ways to expand thread stack to address "bottom", we used 689// both of them in JVM before 1.5: 690// 1. adjust stack pointer first so that it is below "bottom", and then 691// touch "bottom" 692// 2. mmap() the page in question 693// 694// Now alternate signal stack is gone, it's harder to use 2. For instance, 695// if current sp is already near the lower end of page 101, and we need to 696// call mmap() to map page 100, it is possible that part of the mmap() frame 697// will be placed in page 100. When page 100 is mapped, it is zero-filled. 698// That will destroy the mmap() frame and cause VM to crash. 699// 700// The following code works by adjusting sp first, then accessing the "bottom" 701// page to force a page fault. Linux kernel will then automatically expand the 702// stack mapping. 703// 704// _expand_stack_to() assumes its frame size is less than page size, which 705// should always be true if the function is not inlined. 706 707#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 708#define NOINLINE 709#else 710#define NOINLINE __attribute__ ((noinline)) 711#endif 712 713static void _expand_stack_to(address bottom) NOINLINE; 714 715static void _expand_stack_to(address bottom) { 716 address sp; 717 size_t size; 718 volatile char *p; 719 720 // Adjust bottom to point to the largest address within the same page, it 721 // gives us a one-page buffer if alloca() allocates slightly more memory. 722 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 723 bottom += os::Linux::page_size() - 1; 724 725 // sp might be slightly above current stack pointer; if that's the case, we 726 // will alloca() a little more space than necessary, which is OK. Don't use 727 // os::current_stack_pointer(), as its result can be slightly below current 728 // stack pointer, causing us to not alloca enough to reach "bottom". 729 sp = (address)&sp; 730 731 if (sp > bottom) { 732 size = sp - bottom; 733 p = (volatile char *)alloca(size); 734 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 735 p[0] = '\0'; 736 } 737} 738 739bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 740 assert(t!=NULL, "just checking"); 741 assert(t->osthread()->expanding_stack(), "expand should be set"); 742 assert(t->stack_base() != NULL, "stack_base was not initialized"); 743 744 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 745 sigset_t mask_all, old_sigset; 746 sigfillset(&mask_all); 747 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 748 _expand_stack_to(addr); 749 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 750 return true; 751 } 752 return false; 753} 754 755////////////////////////////////////////////////////////////////////////////// 756// create new thread 757 758static address highest_vm_reserved_address(); 759 760// check if it's safe to start a new thread 761static bool _thread_safety_check(Thread* thread) { 762 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 763 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 764 // Heap is mmap'ed at lower end of memory space. Thread stacks are 765 // allocated (MAP_FIXED) from high address space. Every thread stack 766 // occupies a fixed size slot (usually 2Mbytes, but user can change 767 // it to other values if they rebuild LinuxThreads). 768 // 769 // Problem with MAP_FIXED is that mmap() can still succeed even part of 770 // the memory region has already been mmap'ed. That means if we have too 771 // many threads and/or very large heap, eventually thread stack will 772 // collide with heap. 773 // 774 // Here we try to prevent heap/stack collision by comparing current 775 // stack bottom with the highest address that has been mmap'ed by JVM 776 // plus a safety margin for memory maps created by native code. 777 // 778 // This feature can be disabled by setting ThreadSafetyMargin to 0 779 // 780 if (ThreadSafetyMargin > 0) { 781 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 782 783 // not safe if our stack extends below the safety margin 784 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 785 } else { 786 return true; 787 } 788 } else { 789 // Floating stack LinuxThreads or NPTL: 790 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 791 // there's not enough space left, pthread_create() will fail. If we come 792 // here, that means enough space has been reserved for stack. 793 return true; 794 } 795} 796 797// Thread start routine for all newly created threads 798static void *java_start(Thread *thread) { 799 // Try to randomize the cache line index of hot stack frames. 800 // This helps when threads of the same stack traces evict each other's 801 // cache lines. The threads can be either from the same JVM instance, or 802 // from different JVM instances. The benefit is especially true for 803 // processors with hyperthreading technology. 804 static int counter = 0; 805 int pid = os::current_process_id(); 806 alloca(((pid ^ counter++) & 7) * 128); 807 808 ThreadLocalStorage::set_thread(thread); 809 810 OSThread* osthread = thread->osthread(); 811 Monitor* sync = osthread->startThread_lock(); 812 813 // non floating stack LinuxThreads needs extra check, see above 814 if (!_thread_safety_check(thread)) { 815 // notify parent thread 816 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 817 osthread->set_state(ZOMBIE); 818 sync->notify_all(); 819 return NULL; 820 } 821 822 // thread_id is kernel thread id (similar to Solaris LWP id) 823 osthread->set_thread_id(os::Linux::gettid()); 824 825 if (UseNUMA) { 826 int lgrp_id = os::numa_get_group_id(); 827 if (lgrp_id != -1) { 828 thread->set_lgrp_id(lgrp_id); 829 } 830 } 831 // initialize signal mask for this thread 832 os::Linux::hotspot_sigmask(thread); 833 834 // initialize floating point control register 835 os::Linux::init_thread_fpu_state(); 836 837 // handshaking with parent thread 838 { 839 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 840 841 // notify parent thread 842 osthread->set_state(INITIALIZED); 843 sync->notify_all(); 844 845 // wait until os::start_thread() 846 while (osthread->get_state() == INITIALIZED) { 847 sync->wait(Mutex::_no_safepoint_check_flag); 848 } 849 } 850 851 // call one more level start routine 852 thread->run(); 853 854 return 0; 855} 856 857bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 858 assert(thread->osthread() == NULL, "caller responsible"); 859 860 // Allocate the OSThread object 861 OSThread* osthread = new OSThread(NULL, NULL); 862 if (osthread == NULL) { 863 return false; 864 } 865 866 // set the correct thread state 867 osthread->set_thread_type(thr_type); 868 869 // Initial state is ALLOCATED but not INITIALIZED 870 osthread->set_state(ALLOCATED); 871 872 thread->set_osthread(osthread); 873 874 // init thread attributes 875 pthread_attr_t attr; 876 pthread_attr_init(&attr); 877 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 878 879 // stack size 880 if (os::Linux::supports_variable_stack_size()) { 881 // calculate stack size if it's not specified by caller 882 if (stack_size == 0) { 883 stack_size = os::Linux::default_stack_size(thr_type); 884 885 switch (thr_type) { 886 case os::java_thread: 887 // Java threads use ThreadStackSize which default value can be 888 // changed with the flag -Xss 889 assert (JavaThread::stack_size_at_create() > 0, "this should be set"); 890 stack_size = JavaThread::stack_size_at_create(); 891 break; 892 case os::compiler_thread: 893 if (CompilerThreadStackSize > 0) { 894 stack_size = (size_t)(CompilerThreadStackSize * K); 895 break; 896 } // else fall through: 897 // use VMThreadStackSize if CompilerThreadStackSize is not defined 898 case os::vm_thread: 899 case os::pgc_thread: 900 case os::cgc_thread: 901 case os::watcher_thread: 902 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 903 break; 904 } 905 } 906 907 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 908 pthread_attr_setstacksize(&attr, stack_size); 909 } else { 910 // let pthread_create() pick the default value. 911 } 912 913 // glibc guard page 914 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 915 916 ThreadState state; 917 918 { 919 // Serialize thread creation if we are running with fixed stack LinuxThreads 920 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 921 if (lock) { 922 os::Linux::createThread_lock()->lock_without_safepoint_check(); 923 } 924 925 pthread_t tid; 926 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 927 928 pthread_attr_destroy(&attr); 929 930 if (ret != 0) { 931 if (PrintMiscellaneous && (Verbose || WizardMode)) { 932 perror("pthread_create()"); 933 } 934 // Need to clean up stuff we've allocated so far 935 thread->set_osthread(NULL); 936 delete osthread; 937 if (lock) os::Linux::createThread_lock()->unlock(); 938 return false; 939 } 940 941 // Store pthread info into the OSThread 942 osthread->set_pthread_id(tid); 943 944 // Wait until child thread is either initialized or aborted 945 { 946 Monitor* sync_with_child = osthread->startThread_lock(); 947 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 948 while ((state = osthread->get_state()) == ALLOCATED) { 949 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 950 } 951 } 952 953 if (lock) { 954 os::Linux::createThread_lock()->unlock(); 955 } 956 } 957 958 // Aborted due to thread limit being reached 959 if (state == ZOMBIE) { 960 thread->set_osthread(NULL); 961 delete osthread; 962 return false; 963 } 964 965 // The thread is returned suspended (in state INITIALIZED), 966 // and is started higher up in the call chain 967 assert(state == INITIALIZED, "race condition"); 968 return true; 969} 970 971///////////////////////////////////////////////////////////////////////////// 972// attach existing thread 973 974// bootstrap the main thread 975bool os::create_main_thread(JavaThread* thread) { 976 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 977 return create_attached_thread(thread); 978} 979 980bool os::create_attached_thread(JavaThread* thread) { 981#ifdef ASSERT 982 thread->verify_not_published(); 983#endif 984 985 // Allocate the OSThread object 986 OSThread* osthread = new OSThread(NULL, NULL); 987 988 if (osthread == NULL) { 989 return false; 990 } 991 992 // Store pthread info into the OSThread 993 osthread->set_thread_id(os::Linux::gettid()); 994 osthread->set_pthread_id(::pthread_self()); 995 996 // initialize floating point control register 997 os::Linux::init_thread_fpu_state(); 998 999 // Initial thread state is RUNNABLE 1000 osthread->set_state(RUNNABLE); 1001 1002 thread->set_osthread(osthread); 1003 1004 if (UseNUMA) { 1005 int lgrp_id = os::numa_get_group_id(); 1006 if (lgrp_id != -1) { 1007 thread->set_lgrp_id(lgrp_id); 1008 } 1009 } 1010 1011 if (os::Linux::is_initial_thread()) { 1012 // If current thread is initial thread, its stack is mapped on demand, 1013 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 1014 // the entire stack region to avoid SEGV in stack banging. 1015 // It is also useful to get around the heap-stack-gap problem on SuSE 1016 // kernel (see 4821821 for details). We first expand stack to the top 1017 // of yellow zone, then enable stack yellow zone (order is significant, 1018 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1019 // is no gap between the last two virtual memory regions. 1020 1021 JavaThread *jt = (JavaThread *)thread; 1022 address addr = jt->stack_yellow_zone_base(); 1023 assert(addr != NULL, "initialization problem?"); 1024 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1025 1026 osthread->set_expanding_stack(); 1027 os::Linux::manually_expand_stack(jt, addr); 1028 osthread->clear_expanding_stack(); 1029 } 1030 1031 // initialize signal mask for this thread 1032 // and save the caller's signal mask 1033 os::Linux::hotspot_sigmask(thread); 1034 1035 return true; 1036} 1037 1038void os::pd_start_thread(Thread* thread) { 1039 OSThread * osthread = thread->osthread(); 1040 assert(osthread->get_state() != INITIALIZED, "just checking"); 1041 Monitor* sync_with_child = osthread->startThread_lock(); 1042 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1043 sync_with_child->notify(); 1044} 1045 1046// Free Linux resources related to the OSThread 1047void os::free_thread(OSThread* osthread) { 1048 assert(osthread != NULL, "osthread not set"); 1049 1050 if (Thread::current()->osthread() == osthread) { 1051 // Restore caller's signal mask 1052 sigset_t sigmask = osthread->caller_sigmask(); 1053 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1054 } 1055 1056 delete osthread; 1057} 1058 1059////////////////////////////////////////////////////////////////////////////// 1060// thread local storage 1061 1062int os::allocate_thread_local_storage() { 1063 pthread_key_t key; 1064 int rslt = pthread_key_create(&key, NULL); 1065 assert(rslt == 0, "cannot allocate thread local storage"); 1066 return (int)key; 1067} 1068 1069// Note: This is currently not used by VM, as we don't destroy TLS key 1070// on VM exit. 1071void os::free_thread_local_storage(int index) { 1072 int rslt = pthread_key_delete((pthread_key_t)index); 1073 assert(rslt == 0, "invalid index"); 1074} 1075 1076void os::thread_local_storage_at_put(int index, void* value) { 1077 int rslt = pthread_setspecific((pthread_key_t)index, value); 1078 assert(rslt == 0, "pthread_setspecific failed"); 1079} 1080 1081extern "C" Thread* get_thread() { 1082 return ThreadLocalStorage::thread(); 1083} 1084 1085////////////////////////////////////////////////////////////////////////////// 1086// initial thread 1087 1088// Check if current thread is the initial thread, similar to Solaris thr_main. 1089bool os::Linux::is_initial_thread(void) { 1090 char dummy; 1091 // If called before init complete, thread stack bottom will be null. 1092 // Can be called if fatal error occurs before initialization. 1093 if (initial_thread_stack_bottom() == NULL) return false; 1094 assert(initial_thread_stack_bottom() != NULL && 1095 initial_thread_stack_size() != 0, 1096 "os::init did not locate initial thread's stack region"); 1097 if ((address)&dummy >= initial_thread_stack_bottom() && 1098 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1099 return true; 1100 else return false; 1101} 1102 1103// Find the virtual memory area that contains addr 1104static bool find_vma(address addr, address* vma_low, address* vma_high) { 1105 FILE *fp = fopen("/proc/self/maps", "r"); 1106 if (fp) { 1107 address low, high; 1108 while (!feof(fp)) { 1109 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1110 if (low <= addr && addr < high) { 1111 if (vma_low) *vma_low = low; 1112 if (vma_high) *vma_high = high; 1113 fclose (fp); 1114 return true; 1115 } 1116 } 1117 for (;;) { 1118 int ch = fgetc(fp); 1119 if (ch == EOF || ch == (int)'\n') break; 1120 } 1121 } 1122 fclose(fp); 1123 } 1124 return false; 1125} 1126 1127// Locate initial thread stack. This special handling of initial thread stack 1128// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1129// bogus value for initial thread. 1130void os::Linux::capture_initial_stack(size_t max_size) { 1131 // stack size is the easy part, get it from RLIMIT_STACK 1132 size_t stack_size; 1133 struct rlimit rlim; 1134 getrlimit(RLIMIT_STACK, &rlim); 1135 stack_size = rlim.rlim_cur; 1136 1137 // 6308388: a bug in ld.so will relocate its own .data section to the 1138 // lower end of primordial stack; reduce ulimit -s value a little bit 1139 // so we won't install guard page on ld.so's data section. 1140 stack_size -= 2 * page_size(); 1141 1142 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1143 // 7.1, in both cases we will get 2G in return value. 1144 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1145 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1146 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1147 // in case other parts in glibc still assumes 2M max stack size. 1148 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1149#ifndef IA64 1150 if (stack_size > 2 * K * K) stack_size = 2 * K * K; 1151#else 1152 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1153 if (stack_size > 4 * K * K) stack_size = 4 * K * K; 1154#endif 1155 1156 // Try to figure out where the stack base (top) is. This is harder. 1157 // 1158 // When an application is started, glibc saves the initial stack pointer in 1159 // a global variable "__libc_stack_end", which is then used by system 1160 // libraries. __libc_stack_end should be pretty close to stack top. The 1161 // variable is available since the very early days. However, because it is 1162 // a private interface, it could disappear in the future. 1163 // 1164 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1165 // to __libc_stack_end, it is very close to stack top, but isn't the real 1166 // stack top. Note that /proc may not exist if VM is running as a chroot 1167 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1168 // /proc/<pid>/stat could change in the future (though unlikely). 1169 // 1170 // We try __libc_stack_end first. If that doesn't work, look for 1171 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1172 // as a hint, which should work well in most cases. 1173 1174 uintptr_t stack_start; 1175 1176 // try __libc_stack_end first 1177 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1178 if (p && *p) { 1179 stack_start = *p; 1180 } else { 1181 // see if we can get the start_stack field from /proc/self/stat 1182 FILE *fp; 1183 int pid; 1184 char state; 1185 int ppid; 1186 int pgrp; 1187 int session; 1188 int nr; 1189 int tpgrp; 1190 unsigned long flags; 1191 unsigned long minflt; 1192 unsigned long cminflt; 1193 unsigned long majflt; 1194 unsigned long cmajflt; 1195 unsigned long utime; 1196 unsigned long stime; 1197 long cutime; 1198 long cstime; 1199 long prio; 1200 long nice; 1201 long junk; 1202 long it_real; 1203 uintptr_t start; 1204 uintptr_t vsize; 1205 intptr_t rss; 1206 uintptr_t rsslim; 1207 uintptr_t scodes; 1208 uintptr_t ecode; 1209 int i; 1210 1211 // Figure what the primordial thread stack base is. Code is inspired 1212 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1213 // followed by command name surrounded by parentheses, state, etc. 1214 char stat[2048]; 1215 int statlen; 1216 1217 fp = fopen("/proc/self/stat", "r"); 1218 if (fp) { 1219 statlen = fread(stat, 1, 2047, fp); 1220 stat[statlen] = '\0'; 1221 fclose(fp); 1222 1223 // Skip pid and the command string. Note that we could be dealing with 1224 // weird command names, e.g. user could decide to rename java launcher 1225 // to "java 1.4.2 :)", then the stat file would look like 1226 // 1234 (java 1.4.2 :)) R ... ... 1227 // We don't really need to know the command string, just find the last 1228 // occurrence of ")" and then start parsing from there. See bug 4726580. 1229 char * s = strrchr(stat, ')'); 1230 1231 i = 0; 1232 if (s) { 1233 // Skip blank chars 1234 do s++; while (isspace(*s)); 1235 1236#define _UFM UINTX_FORMAT 1237#define _DFM INTX_FORMAT 1238 1239 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1240 /* 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 */ 1241 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM, 1242 &state, /* 3 %c */ 1243 &ppid, /* 4 %d */ 1244 &pgrp, /* 5 %d */ 1245 &session, /* 6 %d */ 1246 &nr, /* 7 %d */ 1247 &tpgrp, /* 8 %d */ 1248 &flags, /* 9 %lu */ 1249 &minflt, /* 10 %lu */ 1250 &cminflt, /* 11 %lu */ 1251 &majflt, /* 12 %lu */ 1252 &cmajflt, /* 13 %lu */ 1253 &utime, /* 14 %lu */ 1254 &stime, /* 15 %lu */ 1255 &cutime, /* 16 %ld */ 1256 &cstime, /* 17 %ld */ 1257 &prio, /* 18 %ld */ 1258 &nice, /* 19 %ld */ 1259 &junk, /* 20 %ld */ 1260 &it_real, /* 21 %ld */ 1261 &start, /* 22 UINTX_FORMAT */ 1262 &vsize, /* 23 UINTX_FORMAT */ 1263 &rss, /* 24 INTX_FORMAT */ 1264 &rsslim, /* 25 UINTX_FORMAT */ 1265 &scodes, /* 26 UINTX_FORMAT */ 1266 &ecode, /* 27 UINTX_FORMAT */ 1267 &stack_start); /* 28 UINTX_FORMAT */ 1268 } 1269 1270#undef _UFM 1271#undef _DFM 1272 1273 if (i != 28 - 2) { 1274 assert(false, "Bad conversion from /proc/self/stat"); 1275 // product mode - assume we are the initial thread, good luck in the 1276 // embedded case. 1277 warning("Can't detect initial thread stack location - bad conversion"); 1278 stack_start = (uintptr_t) &rlim; 1279 } 1280 } else { 1281 // For some reason we can't open /proc/self/stat (for example, running on 1282 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1283 // most cases, so don't abort: 1284 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1285 stack_start = (uintptr_t) &rlim; 1286 } 1287 } 1288 1289 // Now we have a pointer (stack_start) very close to the stack top, the 1290 // next thing to do is to figure out the exact location of stack top. We 1291 // can find out the virtual memory area that contains stack_start by 1292 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1293 // and its upper limit is the real stack top. (again, this would fail if 1294 // running inside chroot, because /proc may not exist.) 1295 1296 uintptr_t stack_top; 1297 address low, high; 1298 if (find_vma((address)stack_start, &low, &high)) { 1299 // success, "high" is the true stack top. (ignore "low", because initial 1300 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1301 stack_top = (uintptr_t)high; 1302 } else { 1303 // failed, likely because /proc/self/maps does not exist 1304 warning("Can't detect initial thread stack location - find_vma failed"); 1305 // best effort: stack_start is normally within a few pages below the real 1306 // stack top, use it as stack top, and reduce stack size so we won't put 1307 // guard page outside stack. 1308 stack_top = stack_start; 1309 stack_size -= 16 * page_size(); 1310 } 1311 1312 // stack_top could be partially down the page so align it 1313 stack_top = align_size_up(stack_top, page_size()); 1314 1315 if (max_size && stack_size > max_size) { 1316 _initial_thread_stack_size = max_size; 1317 } else { 1318 _initial_thread_stack_size = stack_size; 1319 } 1320 1321 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1322 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1323} 1324 1325//////////////////////////////////////////////////////////////////////////////// 1326// time support 1327 1328// Time since start-up in seconds to a fine granularity. 1329// Used by VMSelfDestructTimer and the MemProfiler. 1330double os::elapsedTime() { 1331 1332 return (double)(os::elapsed_counter()) * 0.000001; 1333} 1334 1335jlong os::elapsed_counter() { 1336 timeval time; 1337 int status = gettimeofday(&time, NULL); 1338 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1339} 1340 1341jlong os::elapsed_frequency() { 1342 return (1000 * 1000); 1343} 1344 1345// For now, we say that linux does not support vtime. I have no idea 1346// whether it can actually be made to (DLD, 9/13/05). 1347 1348bool os::supports_vtime() { return false; } 1349bool os::enable_vtime() { return false; } 1350bool os::vtime_enabled() { return false; } 1351double os::elapsedVTime() { 1352 // better than nothing, but not much 1353 return elapsedTime(); 1354} 1355 1356jlong os::javaTimeMillis() { 1357 timeval time; 1358 int status = gettimeofday(&time, NULL); 1359 assert(status != -1, "linux error"); 1360 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1361} 1362 1363#ifndef CLOCK_MONOTONIC 1364#define CLOCK_MONOTONIC (1) 1365#endif 1366 1367void os::Linux::clock_init() { 1368 // we do dlopen's in this particular order due to bug in linux 1369 // dynamical loader (see 6348968) leading to crash on exit 1370 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1371 if (handle == NULL) { 1372 handle = dlopen("librt.so", RTLD_LAZY); 1373 } 1374 1375 if (handle) { 1376 int (*clock_getres_func)(clockid_t, struct timespec*) = 1377 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1378 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1379 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1380 if (clock_getres_func && clock_gettime_func) { 1381 // See if monotonic clock is supported by the kernel. Note that some 1382 // early implementations simply return kernel jiffies (updated every 1383 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1384 // for nano time (though the monotonic property is still nice to have). 1385 // It's fixed in newer kernels, however clock_getres() still returns 1386 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1387 // resolution for now. Hopefully as people move to new kernels, this 1388 // won't be a problem. 1389 struct timespec res; 1390 struct timespec tp; 1391 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1392 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1393 // yes, monotonic clock is supported 1394 _clock_gettime = clock_gettime_func; 1395 } else { 1396 // close librt if there is no monotonic clock 1397 dlclose(handle); 1398 } 1399 } 1400 } 1401} 1402 1403#ifndef SYS_clock_getres 1404 1405#if defined(IA32) || defined(AMD64) 1406#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1407#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1408#else 1409#warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1410#define sys_clock_getres(x,y) -1 1411#endif 1412 1413#else 1414#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1415#endif 1416 1417void os::Linux::fast_thread_clock_init() { 1418 if (!UseLinuxPosixThreadCPUClocks) { 1419 return; 1420 } 1421 clockid_t clockid; 1422 struct timespec tp; 1423 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1424 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1425 1426 // Switch to using fast clocks for thread cpu time if 1427 // the sys_clock_getres() returns 0 error code. 1428 // Note, that some kernels may support the current thread 1429 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1430 // returned by the pthread_getcpuclockid(). 1431 // If the fast Posix clocks are supported then the sys_clock_getres() 1432 // must return at least tp.tv_sec == 0 which means a resolution 1433 // better than 1 sec. This is extra check for reliability. 1434 1435 if(pthread_getcpuclockid_func && 1436 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1437 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1438 1439 _supports_fast_thread_cpu_time = true; 1440 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1441 } 1442} 1443 1444jlong os::javaTimeNanos() { 1445 if (Linux::supports_monotonic_clock()) { 1446 struct timespec tp; 1447 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1448 assert(status == 0, "gettime error"); 1449 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1450 return result; 1451 } else { 1452 timeval time; 1453 int status = gettimeofday(&time, NULL); 1454 assert(status != -1, "linux error"); 1455 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1456 return 1000 * usecs; 1457 } 1458} 1459 1460void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1461 if (Linux::supports_monotonic_clock()) { 1462 info_ptr->max_value = ALL_64_BITS; 1463 1464 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1465 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1466 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1467 } else { 1468 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1469 info_ptr->max_value = ALL_64_BITS; 1470 1471 // gettimeofday is a real time clock so it skips 1472 info_ptr->may_skip_backward = true; 1473 info_ptr->may_skip_forward = true; 1474 } 1475 1476 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1477} 1478 1479// Return the real, user, and system times in seconds from an 1480// arbitrary fixed point in the past. 1481bool os::getTimesSecs(double* process_real_time, 1482 double* process_user_time, 1483 double* process_system_time) { 1484 struct tms ticks; 1485 clock_t real_ticks = times(&ticks); 1486 1487 if (real_ticks == (clock_t) (-1)) { 1488 return false; 1489 } else { 1490 double ticks_per_second = (double) clock_tics_per_sec; 1491 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1492 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1493 *process_real_time = ((double) real_ticks) / ticks_per_second; 1494 1495 return true; 1496 } 1497} 1498 1499 1500char * os::local_time_string(char *buf, size_t buflen) { 1501 struct tm t; 1502 time_t long_time; 1503 time(&long_time); 1504 localtime_r(&long_time, &t); 1505 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1506 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1507 t.tm_hour, t.tm_min, t.tm_sec); 1508 return buf; 1509} 1510 1511struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1512 return localtime_r(clock, res); 1513} 1514 1515//////////////////////////////////////////////////////////////////////////////// 1516// runtime exit support 1517 1518// Note: os::shutdown() might be called very early during initialization, or 1519// called from signal handler. Before adding something to os::shutdown(), make 1520// sure it is async-safe and can handle partially initialized VM. 1521void os::shutdown() { 1522 1523 // allow PerfMemory to attempt cleanup of any persistent resources 1524 perfMemory_exit(); 1525 1526 // needs to remove object in file system 1527 AttachListener::abort(); 1528 1529 // flush buffered output, finish log files 1530 ostream_abort(); 1531 1532 // Check for abort hook 1533 abort_hook_t abort_hook = Arguments::abort_hook(); 1534 if (abort_hook != NULL) { 1535 abort_hook(); 1536 } 1537 1538} 1539 1540// Note: os::abort() might be called very early during initialization, or 1541// called from signal handler. Before adding something to os::abort(), make 1542// sure it is async-safe and can handle partially initialized VM. 1543void os::abort(bool dump_core) { 1544 os::shutdown(); 1545 if (dump_core) { 1546#ifndef PRODUCT 1547 fdStream out(defaultStream::output_fd()); 1548 out.print_raw("Current thread is "); 1549 char buf[16]; 1550 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1551 out.print_raw_cr(buf); 1552 out.print_raw_cr("Dumping core ..."); 1553#endif 1554 ::abort(); // dump core 1555 } 1556 1557 ::exit(1); 1558} 1559 1560// Die immediately, no exit hook, no abort hook, no cleanup. 1561void os::die() { 1562 // _exit() on LinuxThreads only kills current thread 1563 ::abort(); 1564} 1565 1566// unused on linux for now. 1567void os::set_error_file(const char *logfile) {} 1568 1569intx os::current_thread_id() { return (intx)pthread_self(); } 1570int os::current_process_id() { 1571 1572 // Under the old linux thread library, linux gives each thread 1573 // its own process id. Because of this each thread will return 1574 // a different pid if this method were to return the result 1575 // of getpid(2). Linux provides no api that returns the pid 1576 // of the launcher thread for the vm. This implementation 1577 // returns a unique pid, the pid of the launcher thread 1578 // that starts the vm 'process'. 1579 1580 // Under the NPTL, getpid() returns the same pid as the 1581 // launcher thread rather than a unique pid per thread. 1582 // Use gettid() if you want the old pre NPTL behaviour. 1583 1584 // if you are looking for the result of a call to getpid() that 1585 // returns a unique pid for the calling thread, then look at the 1586 // OSThread::thread_id() method in osThread_linux.hpp file 1587 1588 return (int)(_initial_pid ? _initial_pid : getpid()); 1589} 1590 1591// DLL functions 1592 1593const char* os::dll_file_extension() { return ".so"; } 1594 1595const char* os::get_temp_directory() { 1596 const char *prop = Arguments::get_property("java.io.tmpdir"); 1597 return prop == NULL ? "/tmp" : prop; 1598} 1599 1600static bool file_exists(const char* filename) { 1601 struct stat statbuf; 1602 if (filename == NULL || strlen(filename) == 0) { 1603 return false; 1604 } 1605 return os::stat(filename, &statbuf) == 0; 1606} 1607 1608void os::dll_build_name(char* buffer, size_t buflen, 1609 const char* pname, const char* fname) { 1610 // Copied from libhpi 1611 const size_t pnamelen = pname ? strlen(pname) : 0; 1612 1613 // Quietly truncate on buffer overflow. Should be an error. 1614 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1615 *buffer = '\0'; 1616 return; 1617 } 1618 1619 if (pnamelen == 0) { 1620 snprintf(buffer, buflen, "lib%s.so", fname); 1621 } else if (strchr(pname, *os::path_separator()) != NULL) { 1622 int n; 1623 char** pelements = split_path(pname, &n); 1624 for (int i = 0 ; i < n ; i++) { 1625 // Really shouldn't be NULL, but check can't hurt 1626 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1627 continue; // skip the empty path values 1628 } 1629 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1630 if (file_exists(buffer)) { 1631 break; 1632 } 1633 } 1634 // release the storage 1635 for (int i = 0 ; i < n ; i++) { 1636 if (pelements[i] != NULL) { 1637 FREE_C_HEAP_ARRAY(char, pelements[i]); 1638 } 1639 } 1640 if (pelements != NULL) { 1641 FREE_C_HEAP_ARRAY(char*, pelements); 1642 } 1643 } else { 1644 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1645 } 1646} 1647 1648const char* os::get_current_directory(char *buf, int buflen) { 1649 return getcwd(buf, buflen); 1650} 1651 1652// check if addr is inside libjvm[_g].so 1653bool os::address_is_in_vm(address addr) { 1654 static address libjvm_base_addr; 1655 Dl_info dlinfo; 1656 1657 if (libjvm_base_addr == NULL) { 1658 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); 1659 libjvm_base_addr = (address)dlinfo.dli_fbase; 1660 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1661 } 1662 1663 if (dladdr((void *)addr, &dlinfo)) { 1664 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1665 } 1666 1667 return false; 1668} 1669 1670bool os::dll_address_to_function_name(address addr, char *buf, 1671 int buflen, int *offset) { 1672 Dl_info dlinfo; 1673 1674 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { 1675 if (buf != NULL) { 1676 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1677 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1678 } 1679 } 1680 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1681 return true; 1682 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) { 1683 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1684 dlinfo.dli_fname, buf, buflen, offset) == Decoder::no_error) { 1685 return true; 1686 } 1687 } 1688 1689 if (buf != NULL) buf[0] = '\0'; 1690 if (offset != NULL) *offset = -1; 1691 return false; 1692} 1693 1694struct _address_to_library_name { 1695 address addr; // input : memory address 1696 size_t buflen; // size of fname 1697 char* fname; // output: library name 1698 address base; // library base addr 1699}; 1700 1701static int address_to_library_name_callback(struct dl_phdr_info *info, 1702 size_t size, void *data) { 1703 int i; 1704 bool found = false; 1705 address libbase = NULL; 1706 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1707 1708 // iterate through all loadable segments 1709 for (i = 0; i < info->dlpi_phnum; i++) { 1710 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1711 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1712 // base address of a library is the lowest address of its loaded 1713 // segments. 1714 if (libbase == NULL || libbase > segbase) { 1715 libbase = segbase; 1716 } 1717 // see if 'addr' is within current segment 1718 if (segbase <= d->addr && 1719 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1720 found = true; 1721 } 1722 } 1723 } 1724 1725 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1726 // so dll_address_to_library_name() can fall through to use dladdr() which 1727 // can figure out executable name from argv[0]. 1728 if (found && info->dlpi_name && info->dlpi_name[0]) { 1729 d->base = libbase; 1730 if (d->fname) { 1731 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1732 } 1733 return 1; 1734 } 1735 return 0; 1736} 1737 1738bool os::dll_address_to_library_name(address addr, char* buf, 1739 int buflen, int* offset) { 1740 Dl_info dlinfo; 1741 struct _address_to_library_name data; 1742 1743 // There is a bug in old glibc dladdr() implementation that it could resolve 1744 // to wrong library name if the .so file has a base address != NULL. Here 1745 // we iterate through the program headers of all loaded libraries to find 1746 // out which library 'addr' really belongs to. This workaround can be 1747 // removed once the minimum requirement for glibc is moved to 2.3.x. 1748 data.addr = addr; 1749 data.fname = buf; 1750 data.buflen = buflen; 1751 data.base = NULL; 1752 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1753 1754 if (rslt) { 1755 // buf already contains library name 1756 if (offset) *offset = addr - data.base; 1757 return true; 1758 } else if (dladdr((void*)addr, &dlinfo)){ 1759 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1760 if (offset) *offset = addr - (address)dlinfo.dli_fbase; 1761 return true; 1762 } else { 1763 if (buf) buf[0] = '\0'; 1764 if (offset) *offset = -1; 1765 return false; 1766 } 1767} 1768 1769 // Loads .dll/.so and 1770 // in case of error it checks if .dll/.so was built for the 1771 // same architecture as Hotspot is running on 1772 1773void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1774{ 1775 void * result= ::dlopen(filename, RTLD_LAZY); 1776 if (result != NULL) { 1777 // Successful loading 1778 return result; 1779 } 1780 1781 Elf32_Ehdr elf_head; 1782 1783 // Read system error message into ebuf 1784 // It may or may not be overwritten below 1785 ::strncpy(ebuf, ::dlerror(), ebuflen-1); 1786 ebuf[ebuflen-1]='\0'; 1787 int diag_msg_max_length=ebuflen-strlen(ebuf); 1788 char* diag_msg_buf=ebuf+strlen(ebuf); 1789 1790 if (diag_msg_max_length==0) { 1791 // No more space in ebuf for additional diagnostics message 1792 return NULL; 1793 } 1794 1795 1796 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1797 1798 if (file_descriptor < 0) { 1799 // Can't open library, report dlerror() message 1800 return NULL; 1801 } 1802 1803 bool failed_to_read_elf_head= 1804 (sizeof(elf_head)!= 1805 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1806 1807 ::close(file_descriptor); 1808 if (failed_to_read_elf_head) { 1809 // file i/o error - report dlerror() msg 1810 return NULL; 1811 } 1812 1813 typedef struct { 1814 Elf32_Half code; // Actual value as defined in elf.h 1815 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1816 char elf_class; // 32 or 64 bit 1817 char endianess; // MSB or LSB 1818 char* name; // String representation 1819 } arch_t; 1820 1821 #ifndef EM_486 1822 #define EM_486 6 /* Intel 80486 */ 1823 #endif 1824 1825 static const arch_t arch_array[]={ 1826 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1827 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1828 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1829 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1830 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1831 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1832 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1833 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1834 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1835 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1836 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1837 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1838 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1839 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1840 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1841 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1842 }; 1843 1844 #if (defined IA32) 1845 static Elf32_Half running_arch_code=EM_386; 1846 #elif (defined AMD64) 1847 static Elf32_Half running_arch_code=EM_X86_64; 1848 #elif (defined IA64) 1849 static Elf32_Half running_arch_code=EM_IA_64; 1850 #elif (defined __sparc) && (defined _LP64) 1851 static Elf32_Half running_arch_code=EM_SPARCV9; 1852 #elif (defined __sparc) && (!defined _LP64) 1853 static Elf32_Half running_arch_code=EM_SPARC; 1854 #elif (defined __powerpc64__) 1855 static Elf32_Half running_arch_code=EM_PPC64; 1856 #elif (defined __powerpc__) 1857 static Elf32_Half running_arch_code=EM_PPC; 1858 #elif (defined ARM) 1859 static Elf32_Half running_arch_code=EM_ARM; 1860 #elif (defined S390) 1861 static Elf32_Half running_arch_code=EM_S390; 1862 #elif (defined ALPHA) 1863 static Elf32_Half running_arch_code=EM_ALPHA; 1864 #elif (defined MIPSEL) 1865 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1866 #elif (defined PARISC) 1867 static Elf32_Half running_arch_code=EM_PARISC; 1868 #elif (defined MIPS) 1869 static Elf32_Half running_arch_code=EM_MIPS; 1870 #elif (defined M68K) 1871 static Elf32_Half running_arch_code=EM_68K; 1872 #else 1873 #error Method os::dll_load requires that one of following is defined:\ 1874 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 1875 #endif 1876 1877 // Identify compatability class for VM's architecture and library's architecture 1878 // Obtain string descriptions for architectures 1879 1880 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1881 int running_arch_index=-1; 1882 1883 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1884 if (running_arch_code == arch_array[i].code) { 1885 running_arch_index = i; 1886 } 1887 if (lib_arch.code == arch_array[i].code) { 1888 lib_arch.compat_class = arch_array[i].compat_class; 1889 lib_arch.name = arch_array[i].name; 1890 } 1891 } 1892 1893 assert(running_arch_index != -1, 1894 "Didn't find running architecture code (running_arch_code) in arch_array"); 1895 if (running_arch_index == -1) { 1896 // Even though running architecture detection failed 1897 // we may still continue with reporting dlerror() message 1898 return NULL; 1899 } 1900 1901 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1902 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1903 return NULL; 1904 } 1905 1906#ifndef S390 1907 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1908 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1909 return NULL; 1910 } 1911#endif // !S390 1912 1913 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1914 if ( lib_arch.name!=NULL ) { 1915 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1916 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1917 lib_arch.name, arch_array[running_arch_index].name); 1918 } else { 1919 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1920 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1921 lib_arch.code, 1922 arch_array[running_arch_index].name); 1923 } 1924 } 1925 1926 return NULL; 1927} 1928 1929/* 1930 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 1931 * chances are you might want to run the generated bits against glibc-2.0 1932 * libdl.so, so always use locking for any version of glibc. 1933 */ 1934void* os::dll_lookup(void* handle, const char* name) { 1935 pthread_mutex_lock(&dl_mutex); 1936 void* res = dlsym(handle, name); 1937 pthread_mutex_unlock(&dl_mutex); 1938 return res; 1939} 1940 1941 1942bool _print_ascii_file(const char* filename, outputStream* st) { 1943 int fd = open(filename, O_RDONLY); 1944 if (fd == -1) { 1945 return false; 1946 } 1947 1948 char buf[32]; 1949 int bytes; 1950 while ((bytes = read(fd, buf, sizeof(buf))) > 0) { 1951 st->print_raw(buf, bytes); 1952 } 1953 1954 close(fd); 1955 1956 return true; 1957} 1958 1959void os::print_dll_info(outputStream *st) { 1960 st->print_cr("Dynamic libraries:"); 1961 1962 char fname[32]; 1963 pid_t pid = os::Linux::gettid(); 1964 1965 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1966 1967 if (!_print_ascii_file(fname, st)) { 1968 st->print("Can not get library information for pid = %d\n", pid); 1969 } 1970} 1971 1972 1973void os::print_os_info(outputStream* st) { 1974 st->print("OS:"); 1975 1976 // Try to identify popular distros. 1977 // Most Linux distributions have /etc/XXX-release file, which contains 1978 // the OS version string. Some have more than one /etc/XXX-release file 1979 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 1980 // so the order is important. 1981 if (!_print_ascii_file("/etc/mandrake-release", st) && 1982 !_print_ascii_file("/etc/sun-release", st) && 1983 !_print_ascii_file("/etc/redhat-release", st) && 1984 !_print_ascii_file("/etc/SuSE-release", st) && 1985 !_print_ascii_file("/etc/turbolinux-release", st) && 1986 !_print_ascii_file("/etc/gentoo-release", st) && 1987 !_print_ascii_file("/etc/debian_version", st) && 1988 !_print_ascii_file("/etc/ltib-release", st) && 1989 !_print_ascii_file("/etc/angstrom-version", st)) { 1990 st->print("Linux"); 1991 } 1992 st->cr(); 1993 1994 // kernel 1995 st->print("uname:"); 1996 struct utsname name; 1997 uname(&name); 1998 st->print(name.sysname); st->print(" "); 1999 st->print(name.release); st->print(" "); 2000 st->print(name.version); st->print(" "); 2001 st->print(name.machine); 2002 st->cr(); 2003 2004 // Print warning if unsafe chroot environment detected 2005 if (unsafe_chroot_detected) { 2006 st->print("WARNING!! "); 2007 st->print_cr(unstable_chroot_error); 2008 } 2009 2010 // libc, pthread 2011 st->print("libc:"); 2012 st->print(os::Linux::glibc_version()); st->print(" "); 2013 st->print(os::Linux::libpthread_version()); st->print(" "); 2014 if (os::Linux::is_LinuxThreads()) { 2015 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2016 } 2017 st->cr(); 2018 2019 // rlimit 2020 st->print("rlimit:"); 2021 struct rlimit rlim; 2022 2023 st->print(" STACK "); 2024 getrlimit(RLIMIT_STACK, &rlim); 2025 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 2026 else st->print("%uk", rlim.rlim_cur >> 10); 2027 2028 st->print(", CORE "); 2029 getrlimit(RLIMIT_CORE, &rlim); 2030 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 2031 else st->print("%uk", rlim.rlim_cur >> 10); 2032 2033 st->print(", NPROC "); 2034 getrlimit(RLIMIT_NPROC, &rlim); 2035 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 2036 else st->print("%d", rlim.rlim_cur); 2037 2038 st->print(", NOFILE "); 2039 getrlimit(RLIMIT_NOFILE, &rlim); 2040 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 2041 else st->print("%d", rlim.rlim_cur); 2042 2043 st->print(", AS "); 2044 getrlimit(RLIMIT_AS, &rlim); 2045 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 2046 else st->print("%uk", rlim.rlim_cur >> 10); 2047 st->cr(); 2048 2049 // load average 2050 st->print("load average:"); 2051 double loadavg[3]; 2052 os::loadavg(loadavg, 3); 2053 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]); 2054 st->cr(); 2055 2056 // meminfo 2057 st->print("\n/proc/meminfo:\n"); 2058 _print_ascii_file("/proc/meminfo", st); 2059 st->cr(); 2060} 2061 2062void os::print_memory_info(outputStream* st) { 2063 2064 st->print("Memory:"); 2065 st->print(" %dk page", os::vm_page_size()>>10); 2066 2067 // values in struct sysinfo are "unsigned long" 2068 struct sysinfo si; 2069 sysinfo(&si); 2070 2071 st->print(", physical " UINT64_FORMAT "k", 2072 os::physical_memory() >> 10); 2073 st->print("(" UINT64_FORMAT "k free)", 2074 os::available_memory() >> 10); 2075 st->print(", swap " UINT64_FORMAT "k", 2076 ((jlong)si.totalswap * si.mem_unit) >> 10); 2077 st->print("(" UINT64_FORMAT "k free)", 2078 ((jlong)si.freeswap * si.mem_unit) >> 10); 2079 st->cr(); 2080} 2081 2082// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 2083// but they're the same for all the linux arch that we support 2084// and they're the same for solaris but there's no common place to put this. 2085const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 2086 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 2087 "ILL_COPROC", "ILL_BADSTK" }; 2088 2089const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 2090 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 2091 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 2092 2093const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 2094 2095const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 2096 2097void os::print_siginfo(outputStream* st, void* siginfo) { 2098 st->print("siginfo:"); 2099 2100 const int buflen = 100; 2101 char buf[buflen]; 2102 siginfo_t *si = (siginfo_t*)siginfo; 2103 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 2104 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 2105 st->print("si_errno=%s", buf); 2106 } else { 2107 st->print("si_errno=%d", si->si_errno); 2108 } 2109 const int c = si->si_code; 2110 assert(c > 0, "unexpected si_code"); 2111 switch (si->si_signo) { 2112 case SIGILL: 2113 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 2114 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2115 break; 2116 case SIGFPE: 2117 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 2118 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2119 break; 2120 case SIGSEGV: 2121 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 2122 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2123 break; 2124 case SIGBUS: 2125 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 2126 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2127 break; 2128 default: 2129 st->print(", si_code=%d", si->si_code); 2130 // no si_addr 2131 } 2132 2133 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2134 UseSharedSpaces) { 2135 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2136 if (mapinfo->is_in_shared_space(si->si_addr)) { 2137 st->print("\n\nError accessing class data sharing archive." \ 2138 " Mapped file inaccessible during execution, " \ 2139 " possible disk/network problem."); 2140 } 2141 } 2142 st->cr(); 2143} 2144 2145 2146static void print_signal_handler(outputStream* st, int sig, 2147 char* buf, size_t buflen); 2148 2149void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2150 st->print_cr("Signal Handlers:"); 2151 print_signal_handler(st, SIGSEGV, buf, buflen); 2152 print_signal_handler(st, SIGBUS , buf, buflen); 2153 print_signal_handler(st, SIGFPE , buf, buflen); 2154 print_signal_handler(st, SIGPIPE, buf, buflen); 2155 print_signal_handler(st, SIGXFSZ, buf, buflen); 2156 print_signal_handler(st, SIGILL , buf, buflen); 2157 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2158 print_signal_handler(st, SR_signum, buf, buflen); 2159 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2160 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2161 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2162 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2163} 2164 2165static char saved_jvm_path[MAXPATHLEN] = {0}; 2166 2167// Find the full path to the current module, libjvm.so or libjvm_g.so 2168void os::jvm_path(char *buf, jint buflen) { 2169 // Error checking. 2170 if (buflen < MAXPATHLEN) { 2171 assert(false, "must use a large-enough buffer"); 2172 buf[0] = '\0'; 2173 return; 2174 } 2175 // Lazy resolve the path to current module. 2176 if (saved_jvm_path[0] != 0) { 2177 strcpy(buf, saved_jvm_path); 2178 return; 2179 } 2180 2181 char dli_fname[MAXPATHLEN]; 2182 bool ret = dll_address_to_library_name( 2183 CAST_FROM_FN_PTR(address, os::jvm_path), 2184 dli_fname, sizeof(dli_fname), NULL); 2185 assert(ret != 0, "cannot locate libjvm"); 2186 char *rp = realpath(dli_fname, buf); 2187 if (rp == NULL) 2188 return; 2189 2190 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) { 2191 // Support for the gamma launcher. Typical value for buf is 2192 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 2193 // the right place in the string, then assume we are installed in a JDK and 2194 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2195 // up the path so it looks like libjvm.so is installed there (append a 2196 // fake suffix hotspot/libjvm.so). 2197 const char *p = buf + strlen(buf) - 1; 2198 for (int count = 0; p > buf && count < 5; ++count) { 2199 for (--p; p > buf && *p != '/'; --p) 2200 /* empty */ ; 2201 } 2202 2203 if (strncmp(p, "/jre/lib/", 9) != 0) { 2204 // Look for JAVA_HOME in the environment. 2205 char* java_home_var = ::getenv("JAVA_HOME"); 2206 if (java_home_var != NULL && java_home_var[0] != 0) { 2207 char* jrelib_p; 2208 int len; 2209 2210 // Check the current module name "libjvm.so" or "libjvm_g.so". 2211 p = strrchr(buf, '/'); 2212 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2213 p = strstr(p, "_g") ? "_g" : ""; 2214 2215 rp = realpath(java_home_var, buf); 2216 if (rp == NULL) 2217 return; 2218 2219 // determine if this is a legacy image or modules image 2220 // modules image doesn't have "jre" subdirectory 2221 len = strlen(buf); 2222 jrelib_p = buf + len; 2223 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2224 if (0 != access(buf, F_OK)) { 2225 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2226 } 2227 2228 if (0 == access(buf, F_OK)) { 2229 // Use current module name "libjvm[_g].so" instead of 2230 // "libjvm"debug_only("_g")".so" since for fastdebug version 2231 // we should have "libjvm.so" but debug_only("_g") adds "_g"! 2232 // It is used when we are choosing the HPI library's name 2233 // "libhpi[_g].so" in hpi::initialize_get_interface(). 2234 len = strlen(buf); 2235 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p); 2236 } else { 2237 // Go back to path of .so 2238 rp = realpath(dli_fname, buf); 2239 if (rp == NULL) 2240 return; 2241 } 2242 } 2243 } 2244 } 2245 2246 strcpy(saved_jvm_path, buf); 2247} 2248 2249void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2250 // no prefix required, not even "_" 2251} 2252 2253void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2254 // no suffix required 2255} 2256 2257//////////////////////////////////////////////////////////////////////////////// 2258// sun.misc.Signal support 2259 2260static volatile jint sigint_count = 0; 2261 2262static void 2263UserHandler(int sig, void *siginfo, void *context) { 2264 // 4511530 - sem_post is serialized and handled by the manager thread. When 2265 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2266 // don't want to flood the manager thread with sem_post requests. 2267 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2268 return; 2269 2270 // Ctrl-C is pressed during error reporting, likely because the error 2271 // handler fails to abort. Let VM die immediately. 2272 if (sig == SIGINT && is_error_reported()) { 2273 os::die(); 2274 } 2275 2276 os::signal_notify(sig); 2277} 2278 2279void* os::user_handler() { 2280 return CAST_FROM_FN_PTR(void*, UserHandler); 2281} 2282 2283extern "C" { 2284 typedef void (*sa_handler_t)(int); 2285 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2286} 2287 2288void* os::signal(int signal_number, void* handler) { 2289 struct sigaction sigAct, oldSigAct; 2290 2291 sigfillset(&(sigAct.sa_mask)); 2292 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2293 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2294 2295 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2296 // -1 means registration failed 2297 return (void *)-1; 2298 } 2299 2300 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2301} 2302 2303void os::signal_raise(int signal_number) { 2304 ::raise(signal_number); 2305} 2306 2307/* 2308 * The following code is moved from os.cpp for making this 2309 * code platform specific, which it is by its very nature. 2310 */ 2311 2312// Will be modified when max signal is changed to be dynamic 2313int os::sigexitnum_pd() { 2314 return NSIG; 2315} 2316 2317// a counter for each possible signal value 2318static volatile jint pending_signals[NSIG+1] = { 0 }; 2319 2320// Linux(POSIX) specific hand shaking semaphore. 2321static sem_t sig_sem; 2322 2323void os::signal_init_pd() { 2324 // Initialize signal structures 2325 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2326 2327 // Initialize signal semaphore 2328 ::sem_init(&sig_sem, 0, 0); 2329} 2330 2331void os::signal_notify(int sig) { 2332 Atomic::inc(&pending_signals[sig]); 2333 ::sem_post(&sig_sem); 2334} 2335 2336static int check_pending_signals(bool wait) { 2337 Atomic::store(0, &sigint_count); 2338 for (;;) { 2339 for (int i = 0; i < NSIG + 1; i++) { 2340 jint n = pending_signals[i]; 2341 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2342 return i; 2343 } 2344 } 2345 if (!wait) { 2346 return -1; 2347 } 2348 JavaThread *thread = JavaThread::current(); 2349 ThreadBlockInVM tbivm(thread); 2350 2351 bool threadIsSuspended; 2352 do { 2353 thread->set_suspend_equivalent(); 2354 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2355 ::sem_wait(&sig_sem); 2356 2357 // were we externally suspended while we were waiting? 2358 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2359 if (threadIsSuspended) { 2360 // 2361 // The semaphore has been incremented, but while we were waiting 2362 // another thread suspended us. We don't want to continue running 2363 // while suspended because that would surprise the thread that 2364 // suspended us. 2365 // 2366 ::sem_post(&sig_sem); 2367 2368 thread->java_suspend_self(); 2369 } 2370 } while (threadIsSuspended); 2371 } 2372} 2373 2374int os::signal_lookup() { 2375 return check_pending_signals(false); 2376} 2377 2378int os::signal_wait() { 2379 return check_pending_signals(true); 2380} 2381 2382//////////////////////////////////////////////////////////////////////////////// 2383// Virtual Memory 2384 2385int os::vm_page_size() { 2386 // Seems redundant as all get out 2387 assert(os::Linux::page_size() != -1, "must call os::init"); 2388 return os::Linux::page_size(); 2389} 2390 2391// Solaris allocates memory by pages. 2392int os::vm_allocation_granularity() { 2393 assert(os::Linux::page_size() != -1, "must call os::init"); 2394 return os::Linux::page_size(); 2395} 2396 2397// Rationale behind this function: 2398// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2399// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2400// samples for JITted code. Here we create private executable mapping over the code cache 2401// and then we can use standard (well, almost, as mapping can change) way to provide 2402// info for the reporting script by storing timestamp and location of symbol 2403void linux_wrap_code(char* base, size_t size) { 2404 static volatile jint cnt = 0; 2405 2406 if (!UseOprofile) { 2407 return; 2408 } 2409 2410 char buf[PATH_MAX+1]; 2411 int num = Atomic::add(1, &cnt); 2412 2413 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2414 os::get_temp_directory(), os::current_process_id(), num); 2415 unlink(buf); 2416 2417 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU); 2418 2419 if (fd != -1) { 2420 off_t rv = lseek(fd, size-2, SEEK_SET); 2421 if (rv != (off_t)-1) { 2422 if (write(fd, "", 1) == 1) { 2423 mmap(base, size, 2424 PROT_READ|PROT_WRITE|PROT_EXEC, 2425 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2426 } 2427 } 2428 close(fd); 2429 unlink(buf); 2430 } 2431} 2432 2433// NOTE: Linux kernel does not really reserve the pages for us. 2434// All it does is to check if there are enough free pages 2435// left at the time of mmap(). This could be a potential 2436// problem. 2437bool os::commit_memory(char* addr, size_t size, bool exec) { 2438 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2439 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2440 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2441 return res != (uintptr_t) MAP_FAILED; 2442} 2443 2444bool os::commit_memory(char* addr, size_t size, size_t alignment_hint, 2445 bool exec) { 2446 return commit_memory(addr, size, exec); 2447} 2448 2449void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { } 2450 2451void os::free_memory(char *addr, size_t bytes) { 2452 ::mmap(addr, bytes, PROT_READ | PROT_WRITE, 2453 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2454} 2455 2456void os::numa_make_global(char *addr, size_t bytes) { 2457 Linux::numa_interleave_memory(addr, bytes); 2458} 2459 2460void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2461 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2462} 2463 2464bool os::numa_topology_changed() { return false; } 2465 2466size_t os::numa_get_groups_num() { 2467 int max_node = Linux::numa_max_node(); 2468 return max_node > 0 ? max_node + 1 : 1; 2469} 2470 2471int os::numa_get_group_id() { 2472 int cpu_id = Linux::sched_getcpu(); 2473 if (cpu_id != -1) { 2474 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2475 if (lgrp_id != -1) { 2476 return lgrp_id; 2477 } 2478 } 2479 return 0; 2480} 2481 2482size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2483 for (size_t i = 0; i < size; i++) { 2484 ids[i] = i; 2485 } 2486 return size; 2487} 2488 2489bool os::get_page_info(char *start, page_info* info) { 2490 return false; 2491} 2492 2493char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2494 return end; 2495} 2496 2497extern "C" void numa_warn(int number, char *where, ...) { } 2498extern "C" void numa_error(char *where) { } 2499 2500 2501// If we are running with libnuma version > 2, then we should 2502// be trying to use symbols with versions 1.1 2503// If we are running with earlier version, which did not have symbol versions, 2504// we should use the base version. 2505void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2506 void *f = dlvsym(handle, name, "libnuma_1.1"); 2507 if (f == NULL) { 2508 f = dlsym(handle, name); 2509 } 2510 return f; 2511} 2512 2513bool os::Linux::libnuma_init() { 2514 // sched_getcpu() should be in libc. 2515 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2516 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2517 2518 if (sched_getcpu() != -1) { // Does it work? 2519 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2520 if (handle != NULL) { 2521 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2522 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2523 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2524 libnuma_dlsym(handle, "numa_max_node"))); 2525 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2526 libnuma_dlsym(handle, "numa_available"))); 2527 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2528 libnuma_dlsym(handle, "numa_tonode_memory"))); 2529 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2530 libnuma_dlsym(handle, "numa_interleave_memory"))); 2531 2532 2533 if (numa_available() != -1) { 2534 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2535 // Create a cpu -> node mapping 2536 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true); 2537 rebuild_cpu_to_node_map(); 2538 return true; 2539 } 2540 } 2541 } 2542 return false; 2543} 2544 2545// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2546// The table is later used in get_node_by_cpu(). 2547void os::Linux::rebuild_cpu_to_node_map() { 2548 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2549 // in libnuma (possible values are starting from 16, 2550 // and continuing up with every other power of 2, but less 2551 // than the maximum number of CPUs supported by kernel), and 2552 // is a subject to change (in libnuma version 2 the requirements 2553 // are more reasonable) we'll just hardcode the number they use 2554 // in the library. 2555 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2556 2557 size_t cpu_num = os::active_processor_count(); 2558 size_t cpu_map_size = NCPUS / BitsPerCLong; 2559 size_t cpu_map_valid_size = 2560 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2561 2562 cpu_to_node()->clear(); 2563 cpu_to_node()->at_grow(cpu_num - 1); 2564 size_t node_num = numa_get_groups_num(); 2565 2566 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size); 2567 for (size_t i = 0; i < node_num; i++) { 2568 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2569 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2570 if (cpu_map[j] != 0) { 2571 for (size_t k = 0; k < BitsPerCLong; k++) { 2572 if (cpu_map[j] & (1UL << k)) { 2573 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2574 } 2575 } 2576 } 2577 } 2578 } 2579 } 2580 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2581} 2582 2583int os::Linux::get_node_by_cpu(int cpu_id) { 2584 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2585 return cpu_to_node()->at(cpu_id); 2586 } 2587 return -1; 2588} 2589 2590GrowableArray<int>* os::Linux::_cpu_to_node; 2591os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2592os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2593os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2594os::Linux::numa_available_func_t os::Linux::_numa_available; 2595os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2596os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2597unsigned long* os::Linux::_numa_all_nodes; 2598 2599bool os::uncommit_memory(char* addr, size_t size) { 2600 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2601 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2602 return res != (uintptr_t) MAP_FAILED; 2603} 2604 2605// Linux uses a growable mapping for the stack, and if the mapping for 2606// the stack guard pages is not removed when we detach a thread the 2607// stack cannot grow beyond the pages where the stack guard was 2608// mapped. If at some point later in the process the stack expands to 2609// that point, the Linux kernel cannot expand the stack any further 2610// because the guard pages are in the way, and a segfault occurs. 2611// 2612// However, it's essential not to split the stack region by unmapping 2613// a region (leaving a hole) that's already part of the stack mapping, 2614// so if the stack mapping has already grown beyond the guard pages at 2615// the time we create them, we have to truncate the stack mapping. 2616// So, we need to know the extent of the stack mapping when 2617// create_stack_guard_pages() is called. 2618 2619// Find the bounds of the stack mapping. Return true for success. 2620// 2621// We only need this for stacks that are growable: at the time of 2622// writing thread stacks don't use growable mappings (i.e. those 2623// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 2624// only applies to the main thread. 2625static bool 2626get_stack_bounds(uintptr_t *bottom, uintptr_t *top) 2627{ 2628 FILE *f = fopen("/proc/self/maps", "r"); 2629 if (f == NULL) 2630 return false; 2631 2632 while (!feof(f)) { 2633 size_t dummy; 2634 char *str = NULL; 2635 ssize_t len = getline(&str, &dummy, f); 2636 if (len == -1) { 2637 fclose(f); 2638 return false; 2639 } 2640 2641 if (len > 0 && str[len-1] == '\n') { 2642 str[len-1] = 0; 2643 len--; 2644 } 2645 2646 static const char *stack_str = "[stack]"; 2647 if (len > (ssize_t)strlen(stack_str) 2648 && (strcmp(str + len - strlen(stack_str), stack_str) == 0)) { 2649 if (sscanf(str, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) { 2650 uintptr_t sp = (uintptr_t)__builtin_frame_address(0); 2651 if (sp >= *bottom && sp <= *top) { 2652 free(str); 2653 fclose(f); 2654 return true; 2655 } 2656 } 2657 } 2658 free(str); 2659 } 2660 fclose(f); 2661 return false; 2662} 2663 2664// If the (growable) stack mapping already extends beyond the point 2665// where we're going to put our guard pages, truncate the mapping at 2666// that point by munmap()ping it. This ensures that when we later 2667// munmap() the guard pages we don't leave a hole in the stack 2668// mapping. This only affects the main/initial thread, but guard 2669// against future OS changes 2670bool os::create_stack_guard_pages(char* addr, size_t size) { 2671 uintptr_t stack_extent, stack_base; 2672 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2673 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2674 assert(os::Linux::is_initial_thread(), 2675 "growable stack in non-initial thread"); 2676 if (stack_extent < (uintptr_t)addr) 2677 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent); 2678 } 2679 2680 return os::commit_memory(addr, size); 2681} 2682 2683// If this is a growable mapping, remove the guard pages entirely by 2684// munmap()ping them. If not, just call uncommit_memory(). This only 2685// affects the main/initial thread, but guard against future OS changes 2686bool os::remove_stack_guard_pages(char* addr, size_t size) { 2687 uintptr_t stack_extent, stack_base; 2688 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2689 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2690 assert(os::Linux::is_initial_thread(), 2691 "growable stack in non-initial thread"); 2692 2693 return ::munmap(addr, size) == 0; 2694 } 2695 2696 return os::uncommit_memory(addr, size); 2697} 2698 2699static address _highest_vm_reserved_address = NULL; 2700 2701// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2702// at 'requested_addr'. If there are existing memory mappings at the same 2703// location, however, they will be overwritten. If 'fixed' is false, 2704// 'requested_addr' is only treated as a hint, the return value may or 2705// may not start from the requested address. Unlike Linux mmap(), this 2706// function returns NULL to indicate failure. 2707static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2708 char * addr; 2709 int flags; 2710 2711 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2712 if (fixed) { 2713 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2714 flags |= MAP_FIXED; 2715 } 2716 2717 // Map uncommitted pages PROT_READ and PROT_WRITE, change access 2718 // to PROT_EXEC if executable when we commit the page. 2719 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE, 2720 flags, -1, 0); 2721 2722 if (addr != MAP_FAILED) { 2723 // anon_mmap() should only get called during VM initialization, 2724 // don't need lock (actually we can skip locking even it can be called 2725 // from multiple threads, because _highest_vm_reserved_address is just a 2726 // hint about the upper limit of non-stack memory regions.) 2727 if ((address)addr + bytes > _highest_vm_reserved_address) { 2728 _highest_vm_reserved_address = (address)addr + bytes; 2729 } 2730 } 2731 2732 return addr == MAP_FAILED ? NULL : addr; 2733} 2734 2735// Don't update _highest_vm_reserved_address, because there might be memory 2736// regions above addr + size. If so, releasing a memory region only creates 2737// a hole in the address space, it doesn't help prevent heap-stack collision. 2738// 2739static int anon_munmap(char * addr, size_t size) { 2740 return ::munmap(addr, size) == 0; 2741} 2742 2743char* os::reserve_memory(size_t bytes, char* requested_addr, 2744 size_t alignment_hint) { 2745 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 2746} 2747 2748bool os::release_memory(char* addr, size_t size) { 2749 return anon_munmap(addr, size); 2750} 2751 2752static address highest_vm_reserved_address() { 2753 return _highest_vm_reserved_address; 2754} 2755 2756static bool linux_mprotect(char* addr, size_t size, int prot) { 2757 // Linux wants the mprotect address argument to be page aligned. 2758 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 2759 2760 // According to SUSv3, mprotect() should only be used with mappings 2761 // established by mmap(), and mmap() always maps whole pages. Unaligned 2762 // 'addr' likely indicates problem in the VM (e.g. trying to change 2763 // protection of malloc'ed or statically allocated memory). Check the 2764 // caller if you hit this assert. 2765 assert(addr == bottom, "sanity check"); 2766 2767 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 2768 return ::mprotect(bottom, size, prot) == 0; 2769} 2770 2771// Set protections specified 2772bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 2773 bool is_committed) { 2774 unsigned int p = 0; 2775 switch (prot) { 2776 case MEM_PROT_NONE: p = PROT_NONE; break; 2777 case MEM_PROT_READ: p = PROT_READ; break; 2778 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 2779 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 2780 default: 2781 ShouldNotReachHere(); 2782 } 2783 // is_committed is unused. 2784 return linux_mprotect(addr, bytes, p); 2785} 2786 2787bool os::guard_memory(char* addr, size_t size) { 2788 return linux_mprotect(addr, size, PROT_NONE); 2789} 2790 2791bool os::unguard_memory(char* addr, size_t size) { 2792 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 2793} 2794 2795// Large page support 2796 2797static size_t _large_page_size = 0; 2798 2799bool os::large_page_init() { 2800 if (!UseLargePages) return false; 2801 2802 if (LargePageSizeInBytes) { 2803 _large_page_size = LargePageSizeInBytes; 2804 } else { 2805 // large_page_size on Linux is used to round up heap size. x86 uses either 2806 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 2807 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 2808 // page as large as 256M. 2809 // 2810 // Here we try to figure out page size by parsing /proc/meminfo and looking 2811 // for a line with the following format: 2812 // Hugepagesize: 2048 kB 2813 // 2814 // If we can't determine the value (e.g. /proc is not mounted, or the text 2815 // format has been changed), we'll use the largest page size supported by 2816 // the processor. 2817 2818#ifndef ZERO 2819 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 2820 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 2821#endif // ZERO 2822 2823 FILE *fp = fopen("/proc/meminfo", "r"); 2824 if (fp) { 2825 while (!feof(fp)) { 2826 int x = 0; 2827 char buf[16]; 2828 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 2829 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 2830 _large_page_size = x * K; 2831 break; 2832 } 2833 } else { 2834 // skip to next line 2835 for (;;) { 2836 int ch = fgetc(fp); 2837 if (ch == EOF || ch == (int)'\n') break; 2838 } 2839 } 2840 } 2841 fclose(fp); 2842 } 2843 } 2844 2845 const size_t default_page_size = (size_t)Linux::page_size(); 2846 if (_large_page_size > default_page_size) { 2847 _page_sizes[0] = _large_page_size; 2848 _page_sizes[1] = default_page_size; 2849 _page_sizes[2] = 0; 2850 } 2851 2852 // Large page support is available on 2.6 or newer kernel, some vendors 2853 // (e.g. Redhat) have backported it to their 2.4 based distributions. 2854 // We optimistically assume the support is available. If later it turns out 2855 // not true, VM will automatically switch to use regular page size. 2856 return true; 2857} 2858 2859#ifndef SHM_HUGETLB 2860#define SHM_HUGETLB 04000 2861#endif 2862 2863char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) { 2864 // "exec" is passed in but not used. Creating the shared image for 2865 // the code cache doesn't have an SHM_X executable permission to check. 2866 assert(UseLargePages, "only for large pages"); 2867 2868 key_t key = IPC_PRIVATE; 2869 char *addr; 2870 2871 bool warn_on_failure = UseLargePages && 2872 (!FLAG_IS_DEFAULT(UseLargePages) || 2873 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 2874 ); 2875 char msg[128]; 2876 2877 // Create a large shared memory region to attach to based on size. 2878 // Currently, size is the total size of the heap 2879 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 2880 if (shmid == -1) { 2881 // Possible reasons for shmget failure: 2882 // 1. shmmax is too small for Java heap. 2883 // > check shmmax value: cat /proc/sys/kernel/shmmax 2884 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 2885 // 2. not enough large page memory. 2886 // > check available large pages: cat /proc/meminfo 2887 // > increase amount of large pages: 2888 // echo new_value > /proc/sys/vm/nr_hugepages 2889 // Note 1: different Linux may use different name for this property, 2890 // e.g. on Redhat AS-3 it is "hugetlb_pool". 2891 // Note 2: it's possible there's enough physical memory available but 2892 // they are so fragmented after a long run that they can't 2893 // coalesce into large pages. Try to reserve large pages when 2894 // the system is still "fresh". 2895 if (warn_on_failure) { 2896 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 2897 warning(msg); 2898 } 2899 return NULL; 2900 } 2901 2902 // attach to the region 2903 addr = (char*)shmat(shmid, req_addr, 0); 2904 int err = errno; 2905 2906 // Remove shmid. If shmat() is successful, the actual shared memory segment 2907 // will be deleted when it's detached by shmdt() or when the process 2908 // terminates. If shmat() is not successful this will remove the shared 2909 // segment immediately. 2910 shmctl(shmid, IPC_RMID, NULL); 2911 2912 if ((intptr_t)addr == -1) { 2913 if (warn_on_failure) { 2914 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 2915 warning(msg); 2916 } 2917 return NULL; 2918 } 2919 2920 return addr; 2921} 2922 2923bool os::release_memory_special(char* base, size_t bytes) { 2924 // detaching the SHM segment will also delete it, see reserve_memory_special() 2925 int rslt = shmdt(base); 2926 return rslt == 0; 2927} 2928 2929size_t os::large_page_size() { 2930 return _large_page_size; 2931} 2932 2933// Linux does not support anonymous mmap with large page memory. The only way 2934// to reserve large page memory without file backing is through SysV shared 2935// memory API. The entire memory region is committed and pinned upfront. 2936// Hopefully this will change in the future... 2937bool os::can_commit_large_page_memory() { 2938 return false; 2939} 2940 2941bool os::can_execute_large_page_memory() { 2942 return false; 2943} 2944 2945// Reserve memory at an arbitrary address, only if that area is 2946// available (and not reserved for something else). 2947 2948char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 2949 const int max_tries = 10; 2950 char* base[max_tries]; 2951 size_t size[max_tries]; 2952 const size_t gap = 0x000000; 2953 2954 // Assert only that the size is a multiple of the page size, since 2955 // that's all that mmap requires, and since that's all we really know 2956 // about at this low abstraction level. If we need higher alignment, 2957 // we can either pass an alignment to this method or verify alignment 2958 // in one of the methods further up the call chain. See bug 5044738. 2959 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 2960 2961 // Repeatedly allocate blocks until the block is allocated at the 2962 // right spot. Give up after max_tries. Note that reserve_memory() will 2963 // automatically update _highest_vm_reserved_address if the call is 2964 // successful. The variable tracks the highest memory address every reserved 2965 // by JVM. It is used to detect heap-stack collision if running with 2966 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 2967 // space than needed, it could confuse the collision detecting code. To 2968 // solve the problem, save current _highest_vm_reserved_address and 2969 // calculate the correct value before return. 2970 address old_highest = _highest_vm_reserved_address; 2971 2972 // Linux mmap allows caller to pass an address as hint; give it a try first, 2973 // if kernel honors the hint then we can return immediately. 2974 char * addr = anon_mmap(requested_addr, bytes, false); 2975 if (addr == requested_addr) { 2976 return requested_addr; 2977 } 2978 2979 if (addr != NULL) { 2980 // mmap() is successful but it fails to reserve at the requested address 2981 anon_munmap(addr, bytes); 2982 } 2983 2984 int i; 2985 for (i = 0; i < max_tries; ++i) { 2986 base[i] = reserve_memory(bytes); 2987 2988 if (base[i] != NULL) { 2989 // Is this the block we wanted? 2990 if (base[i] == requested_addr) { 2991 size[i] = bytes; 2992 break; 2993 } 2994 2995 // Does this overlap the block we wanted? Give back the overlapped 2996 // parts and try again. 2997 2998 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 2999 if (top_overlap >= 0 && top_overlap < bytes) { 3000 unmap_memory(base[i], top_overlap); 3001 base[i] += top_overlap; 3002 size[i] = bytes - top_overlap; 3003 } else { 3004 size_t bottom_overlap = base[i] + bytes - requested_addr; 3005 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3006 unmap_memory(requested_addr, bottom_overlap); 3007 size[i] = bytes - bottom_overlap; 3008 } else { 3009 size[i] = bytes; 3010 } 3011 } 3012 } 3013 } 3014 3015 // Give back the unused reserved pieces. 3016 3017 for (int j = 0; j < i; ++j) { 3018 if (base[j] != NULL) { 3019 unmap_memory(base[j], size[j]); 3020 } 3021 } 3022 3023 if (i < max_tries) { 3024 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3025 return requested_addr; 3026 } else { 3027 _highest_vm_reserved_address = old_highest; 3028 return NULL; 3029 } 3030} 3031 3032size_t os::read(int fd, void *buf, unsigned int nBytes) { 3033 return ::read(fd, buf, nBytes); 3034} 3035 3036// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 3037// Solaris uses poll(), linux uses park(). 3038// Poll() is likely a better choice, assuming that Thread.interrupt() 3039// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 3040// SIGSEGV, see 4355769. 3041 3042const int NANOSECS_PER_MILLISECS = 1000000; 3043 3044int os::sleep(Thread* thread, jlong millis, bool interruptible) { 3045 assert(thread == Thread::current(), "thread consistency check"); 3046 3047 ParkEvent * const slp = thread->_SleepEvent ; 3048 slp->reset() ; 3049 OrderAccess::fence() ; 3050 3051 if (interruptible) { 3052 jlong prevtime = javaTimeNanos(); 3053 3054 for (;;) { 3055 if (os::is_interrupted(thread, true)) { 3056 return OS_INTRPT; 3057 } 3058 3059 jlong newtime = javaTimeNanos(); 3060 3061 if (newtime - prevtime < 0) { 3062 // time moving backwards, should only happen if no monotonic clock 3063 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3064 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3065 } else { 3066 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 3067 } 3068 3069 if(millis <= 0) { 3070 return OS_OK; 3071 } 3072 3073 prevtime = newtime; 3074 3075 { 3076 assert(thread->is_Java_thread(), "sanity check"); 3077 JavaThread *jt = (JavaThread *) thread; 3078 ThreadBlockInVM tbivm(jt); 3079 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 3080 3081 jt->set_suspend_equivalent(); 3082 // cleared by handle_special_suspend_equivalent_condition() or 3083 // java_suspend_self() via check_and_wait_while_suspended() 3084 3085 slp->park(millis); 3086 3087 // were we externally suspended while we were waiting? 3088 jt->check_and_wait_while_suspended(); 3089 } 3090 } 3091 } else { 3092 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 3093 jlong prevtime = javaTimeNanos(); 3094 3095 for (;;) { 3096 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 3097 // the 1st iteration ... 3098 jlong newtime = javaTimeNanos(); 3099 3100 if (newtime - prevtime < 0) { 3101 // time moving backwards, should only happen if no monotonic clock 3102 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3103 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3104 } else { 3105 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 3106 } 3107 3108 if(millis <= 0) break ; 3109 3110 prevtime = newtime; 3111 slp->park(millis); 3112 } 3113 return OS_OK ; 3114 } 3115} 3116 3117int os::naked_sleep() { 3118 // %% make the sleep time an integer flag. for now use 1 millisec. 3119 return os::sleep(Thread::current(), 1, false); 3120} 3121 3122// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3123void os::infinite_sleep() { 3124 while (true) { // sleep forever ... 3125 ::sleep(100); // ... 100 seconds at a time 3126 } 3127} 3128 3129// Used to convert frequent JVM_Yield() to nops 3130bool os::dont_yield() { 3131 return DontYieldALot; 3132} 3133 3134void os::yield() { 3135 sched_yield(); 3136} 3137 3138os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3139 3140void os::yield_all(int attempts) { 3141 // Yields to all threads, including threads with lower priorities 3142 // Threads on Linux are all with same priority. The Solaris style 3143 // os::yield_all() with nanosleep(1ms) is not necessary. 3144 sched_yield(); 3145} 3146 3147// Called from the tight loops to possibly influence time-sharing heuristics 3148void os::loop_breaker(int attempts) { 3149 os::yield_all(attempts); 3150} 3151 3152//////////////////////////////////////////////////////////////////////////////// 3153// thread priority support 3154 3155// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3156// only supports dynamic priority, static priority must be zero. For real-time 3157// applications, Linux supports SCHED_RR which allows static priority (1-99). 3158// However, for large multi-threaded applications, SCHED_RR is not only slower 3159// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3160// of 5 runs - Sep 2005). 3161// 3162// The following code actually changes the niceness of kernel-thread/LWP. It 3163// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3164// not the entire user process, and user level threads are 1:1 mapped to kernel 3165// threads. It has always been the case, but could change in the future. For 3166// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3167// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3168 3169int os::java_to_os_priority[MaxPriority + 1] = { 3170 19, // 0 Entry should never be used 3171 3172 4, // 1 MinPriority 3173 3, // 2 3174 2, // 3 3175 3176 1, // 4 3177 0, // 5 NormPriority 3178 -1, // 6 3179 3180 -2, // 7 3181 -3, // 8 3182 -4, // 9 NearMaxPriority 3183 3184 -5 // 10 MaxPriority 3185}; 3186 3187static int prio_init() { 3188 if (ThreadPriorityPolicy == 1) { 3189 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3190 // if effective uid is not root. Perhaps, a more elegant way of doing 3191 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3192 if (geteuid() != 0) { 3193 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3194 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3195 } 3196 ThreadPriorityPolicy = 0; 3197 } 3198 } 3199 return 0; 3200} 3201 3202OSReturn os::set_native_priority(Thread* thread, int newpri) { 3203 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3204 3205 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3206 return (ret == 0) ? OS_OK : OS_ERR; 3207} 3208 3209OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3210 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3211 *priority_ptr = java_to_os_priority[NormPriority]; 3212 return OS_OK; 3213 } 3214 3215 errno = 0; 3216 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3217 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3218} 3219 3220// Hint to the underlying OS that a task switch would not be good. 3221// Void return because it's a hint and can fail. 3222void os::hint_no_preempt() {} 3223 3224//////////////////////////////////////////////////////////////////////////////// 3225// suspend/resume support 3226 3227// the low-level signal-based suspend/resume support is a remnant from the 3228// old VM-suspension that used to be for java-suspension, safepoints etc, 3229// within hotspot. Now there is a single use-case for this: 3230// - calling get_thread_pc() on the VMThread by the flat-profiler task 3231// that runs in the watcher thread. 3232// The remaining code is greatly simplified from the more general suspension 3233// code that used to be used. 3234// 3235// The protocol is quite simple: 3236// - suspend: 3237// - sends a signal to the target thread 3238// - polls the suspend state of the osthread using a yield loop 3239// - target thread signal handler (SR_handler) sets suspend state 3240// and blocks in sigsuspend until continued 3241// - resume: 3242// - sets target osthread state to continue 3243// - sends signal to end the sigsuspend loop in the SR_handler 3244// 3245// Note that the SR_lock plays no role in this suspend/resume protocol. 3246// 3247 3248static void resume_clear_context(OSThread *osthread) { 3249 osthread->set_ucontext(NULL); 3250 osthread->set_siginfo(NULL); 3251 3252 // notify the suspend action is completed, we have now resumed 3253 osthread->sr.clear_suspended(); 3254} 3255 3256static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3257 osthread->set_ucontext(context); 3258 osthread->set_siginfo(siginfo); 3259} 3260 3261// 3262// Handler function invoked when a thread's execution is suspended or 3263// resumed. We have to be careful that only async-safe functions are 3264// called here (Note: most pthread functions are not async safe and 3265// should be avoided.) 3266// 3267// Note: sigwait() is a more natural fit than sigsuspend() from an 3268// interface point of view, but sigwait() prevents the signal hander 3269// from being run. libpthread would get very confused by not having 3270// its signal handlers run and prevents sigwait()'s use with the 3271// mutex granting granting signal. 3272// 3273// Currently only ever called on the VMThread 3274// 3275static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3276 // Save and restore errno to avoid confusing native code with EINTR 3277 // after sigsuspend. 3278 int old_errno = errno; 3279 3280 Thread* thread = Thread::current(); 3281 OSThread* osthread = thread->osthread(); 3282 assert(thread->is_VM_thread(), "Must be VMThread"); 3283 // read current suspend action 3284 int action = osthread->sr.suspend_action(); 3285 if (action == SR_SUSPEND) { 3286 suspend_save_context(osthread, siginfo, context); 3287 3288 // Notify the suspend action is about to be completed. do_suspend() 3289 // waits until SR_SUSPENDED is set and then returns. We will wait 3290 // here for a resume signal and that completes the suspend-other 3291 // action. do_suspend/do_resume is always called as a pair from 3292 // the same thread - so there are no races 3293 3294 // notify the caller 3295 osthread->sr.set_suspended(); 3296 3297 sigset_t suspend_set; // signals for sigsuspend() 3298 3299 // get current set of blocked signals and unblock resume signal 3300 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3301 sigdelset(&suspend_set, SR_signum); 3302 3303 // wait here until we are resumed 3304 do { 3305 sigsuspend(&suspend_set); 3306 // ignore all returns until we get a resume signal 3307 } while (osthread->sr.suspend_action() != SR_CONTINUE); 3308 3309 resume_clear_context(osthread); 3310 3311 } else { 3312 assert(action == SR_CONTINUE, "unexpected sr action"); 3313 // nothing special to do - just leave the handler 3314 } 3315 3316 errno = old_errno; 3317} 3318 3319 3320static int SR_initialize() { 3321 struct sigaction act; 3322 char *s; 3323 /* Get signal number to use for suspend/resume */ 3324 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3325 int sig = ::strtol(s, 0, 10); 3326 if (sig > 0 || sig < _NSIG) { 3327 SR_signum = sig; 3328 } 3329 } 3330 3331 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3332 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3333 3334 sigemptyset(&SR_sigset); 3335 sigaddset(&SR_sigset, SR_signum); 3336 3337 /* Set up signal handler for suspend/resume */ 3338 act.sa_flags = SA_RESTART|SA_SIGINFO; 3339 act.sa_handler = (void (*)(int)) SR_handler; 3340 3341 // SR_signum is blocked by default. 3342 // 4528190 - We also need to block pthread restart signal (32 on all 3343 // supported Linux platforms). Note that LinuxThreads need to block 3344 // this signal for all threads to work properly. So we don't have 3345 // to use hard-coded signal number when setting up the mask. 3346 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 3347 3348 if (sigaction(SR_signum, &act, 0) == -1) { 3349 return -1; 3350 } 3351 3352 // Save signal flag 3353 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 3354 return 0; 3355} 3356 3357static int SR_finalize() { 3358 return 0; 3359} 3360 3361 3362// returns true on success and false on error - really an error is fatal 3363// but this seems the normal response to library errors 3364static bool do_suspend(OSThread* osthread) { 3365 // mark as suspended and send signal 3366 osthread->sr.set_suspend_action(SR_SUSPEND); 3367 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3368 assert_status(status == 0, status, "pthread_kill"); 3369 3370 // check status and wait until notified of suspension 3371 if (status == 0) { 3372 for (int i = 0; !osthread->sr.is_suspended(); i++) { 3373 os::yield_all(i); 3374 } 3375 osthread->sr.set_suspend_action(SR_NONE); 3376 return true; 3377 } 3378 else { 3379 osthread->sr.set_suspend_action(SR_NONE); 3380 return false; 3381 } 3382} 3383 3384static void do_resume(OSThread* osthread) { 3385 assert(osthread->sr.is_suspended(), "thread should be suspended"); 3386 osthread->sr.set_suspend_action(SR_CONTINUE); 3387 3388 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3389 assert_status(status == 0, status, "pthread_kill"); 3390 // check status and wait unit notified of resumption 3391 if (status == 0) { 3392 for (int i = 0; osthread->sr.is_suspended(); i++) { 3393 os::yield_all(i); 3394 } 3395 } 3396 osthread->sr.set_suspend_action(SR_NONE); 3397} 3398 3399//////////////////////////////////////////////////////////////////////////////// 3400// interrupt support 3401 3402void os::interrupt(Thread* thread) { 3403 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3404 "possibility of dangling Thread pointer"); 3405 3406 OSThread* osthread = thread->osthread(); 3407 3408 if (!osthread->interrupted()) { 3409 osthread->set_interrupted(true); 3410 // More than one thread can get here with the same value of osthread, 3411 // resulting in multiple notifications. We do, however, want the store 3412 // to interrupted() to be visible to other threads before we execute unpark(). 3413 OrderAccess::fence(); 3414 ParkEvent * const slp = thread->_SleepEvent ; 3415 if (slp != NULL) slp->unpark() ; 3416 } 3417 3418 // For JSR166. Unpark even if interrupt status already was set 3419 if (thread->is_Java_thread()) 3420 ((JavaThread*)thread)->parker()->unpark(); 3421 3422 ParkEvent * ev = thread->_ParkEvent ; 3423 if (ev != NULL) ev->unpark() ; 3424 3425} 3426 3427bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 3428 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3429 "possibility of dangling Thread pointer"); 3430 3431 OSThread* osthread = thread->osthread(); 3432 3433 bool interrupted = osthread->interrupted(); 3434 3435 if (interrupted && clear_interrupted) { 3436 osthread->set_interrupted(false); 3437 // consider thread->_SleepEvent->reset() ... optional optimization 3438 } 3439 3440 return interrupted; 3441} 3442 3443/////////////////////////////////////////////////////////////////////////////////// 3444// signal handling (except suspend/resume) 3445 3446// This routine may be used by user applications as a "hook" to catch signals. 3447// The user-defined signal handler must pass unrecognized signals to this 3448// routine, and if it returns true (non-zero), then the signal handler must 3449// return immediately. If the flag "abort_if_unrecognized" is true, then this 3450// routine will never retun false (zero), but instead will execute a VM panic 3451// routine kill the process. 3452// 3453// If this routine returns false, it is OK to call it again. This allows 3454// the user-defined signal handler to perform checks either before or after 3455// the VM performs its own checks. Naturally, the user code would be making 3456// a serious error if it tried to handle an exception (such as a null check 3457// or breakpoint) that the VM was generating for its own correct operation. 3458// 3459// This routine may recognize any of the following kinds of signals: 3460// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 3461// It should be consulted by handlers for any of those signals. 3462// 3463// The caller of this routine must pass in the three arguments supplied 3464// to the function referred to in the "sa_sigaction" (not the "sa_handler") 3465// field of the structure passed to sigaction(). This routine assumes that 3466// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 3467// 3468// Note that the VM will print warnings if it detects conflicting signal 3469// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 3470// 3471extern "C" int 3472JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 3473 void* ucontext, int abort_if_unrecognized); 3474 3475void signalHandler(int sig, siginfo_t* info, void* uc) { 3476 assert(info != NULL && uc != NULL, "it must be old kernel"); 3477 JVM_handle_linux_signal(sig, info, uc, true); 3478} 3479 3480 3481// This boolean allows users to forward their own non-matching signals 3482// to JVM_handle_linux_signal, harmlessly. 3483bool os::Linux::signal_handlers_are_installed = false; 3484 3485// For signal-chaining 3486struct sigaction os::Linux::sigact[MAXSIGNUM]; 3487unsigned int os::Linux::sigs = 0; 3488bool os::Linux::libjsig_is_loaded = false; 3489typedef struct sigaction *(*get_signal_t)(int); 3490get_signal_t os::Linux::get_signal_action = NULL; 3491 3492struct sigaction* os::Linux::get_chained_signal_action(int sig) { 3493 struct sigaction *actp = NULL; 3494 3495 if (libjsig_is_loaded) { 3496 // Retrieve the old signal handler from libjsig 3497 actp = (*get_signal_action)(sig); 3498 } 3499 if (actp == NULL) { 3500 // Retrieve the preinstalled signal handler from jvm 3501 actp = get_preinstalled_handler(sig); 3502 } 3503 3504 return actp; 3505} 3506 3507static bool call_chained_handler(struct sigaction *actp, int sig, 3508 siginfo_t *siginfo, void *context) { 3509 // Call the old signal handler 3510 if (actp->sa_handler == SIG_DFL) { 3511 // It's more reasonable to let jvm treat it as an unexpected exception 3512 // instead of taking the default action. 3513 return false; 3514 } else if (actp->sa_handler != SIG_IGN) { 3515 if ((actp->sa_flags & SA_NODEFER) == 0) { 3516 // automaticlly block the signal 3517 sigaddset(&(actp->sa_mask), sig); 3518 } 3519 3520 sa_handler_t hand; 3521 sa_sigaction_t sa; 3522 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 3523 // retrieve the chained handler 3524 if (siginfo_flag_set) { 3525 sa = actp->sa_sigaction; 3526 } else { 3527 hand = actp->sa_handler; 3528 } 3529 3530 if ((actp->sa_flags & SA_RESETHAND) != 0) { 3531 actp->sa_handler = SIG_DFL; 3532 } 3533 3534 // try to honor the signal mask 3535 sigset_t oset; 3536 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 3537 3538 // call into the chained handler 3539 if (siginfo_flag_set) { 3540 (*sa)(sig, siginfo, context); 3541 } else { 3542 (*hand)(sig); 3543 } 3544 3545 // restore the signal mask 3546 pthread_sigmask(SIG_SETMASK, &oset, 0); 3547 } 3548 // Tell jvm's signal handler the signal is taken care of. 3549 return true; 3550} 3551 3552bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 3553 bool chained = false; 3554 // signal-chaining 3555 if (UseSignalChaining) { 3556 struct sigaction *actp = get_chained_signal_action(sig); 3557 if (actp != NULL) { 3558 chained = call_chained_handler(actp, sig, siginfo, context); 3559 } 3560 } 3561 return chained; 3562} 3563 3564struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 3565 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 3566 return &sigact[sig]; 3567 } 3568 return NULL; 3569} 3570 3571void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 3572 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3573 sigact[sig] = oldAct; 3574 sigs |= (unsigned int)1 << sig; 3575} 3576 3577// for diagnostic 3578int os::Linux::sigflags[MAXSIGNUM]; 3579 3580int os::Linux::get_our_sigflags(int sig) { 3581 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3582 return sigflags[sig]; 3583} 3584 3585void os::Linux::set_our_sigflags(int sig, int flags) { 3586 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3587 sigflags[sig] = flags; 3588} 3589 3590void os::Linux::set_signal_handler(int sig, bool set_installed) { 3591 // Check for overwrite. 3592 struct sigaction oldAct; 3593 sigaction(sig, (struct sigaction*)NULL, &oldAct); 3594 3595 void* oldhand = oldAct.sa_sigaction 3596 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3597 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3598 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 3599 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 3600 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 3601 if (AllowUserSignalHandlers || !set_installed) { 3602 // Do not overwrite; user takes responsibility to forward to us. 3603 return; 3604 } else if (UseSignalChaining) { 3605 // save the old handler in jvm 3606 save_preinstalled_handler(sig, oldAct); 3607 // libjsig also interposes the sigaction() call below and saves the 3608 // old sigaction on it own. 3609 } else { 3610 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 3611 "%#lx for signal %d.", (long)oldhand, sig)); 3612 } 3613 } 3614 3615 struct sigaction sigAct; 3616 sigfillset(&(sigAct.sa_mask)); 3617 sigAct.sa_handler = SIG_DFL; 3618 if (!set_installed) { 3619 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3620 } else { 3621 sigAct.sa_sigaction = signalHandler; 3622 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3623 } 3624 // Save flags, which are set by ours 3625 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3626 sigflags[sig] = sigAct.sa_flags; 3627 3628 int ret = sigaction(sig, &sigAct, &oldAct); 3629 assert(ret == 0, "check"); 3630 3631 void* oldhand2 = oldAct.sa_sigaction 3632 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3633 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3634 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 3635} 3636 3637// install signal handlers for signals that HotSpot needs to 3638// handle in order to support Java-level exception handling. 3639 3640void os::Linux::install_signal_handlers() { 3641 if (!signal_handlers_are_installed) { 3642 signal_handlers_are_installed = true; 3643 3644 // signal-chaining 3645 typedef void (*signal_setting_t)(); 3646 signal_setting_t begin_signal_setting = NULL; 3647 signal_setting_t end_signal_setting = NULL; 3648 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3649 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 3650 if (begin_signal_setting != NULL) { 3651 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3652 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 3653 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 3654 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 3655 libjsig_is_loaded = true; 3656 assert(UseSignalChaining, "should enable signal-chaining"); 3657 } 3658 if (libjsig_is_loaded) { 3659 // Tell libjsig jvm is setting signal handlers 3660 (*begin_signal_setting)(); 3661 } 3662 3663 set_signal_handler(SIGSEGV, true); 3664 set_signal_handler(SIGPIPE, true); 3665 set_signal_handler(SIGBUS, true); 3666 set_signal_handler(SIGILL, true); 3667 set_signal_handler(SIGFPE, true); 3668 set_signal_handler(SIGXFSZ, true); 3669 3670 if (libjsig_is_loaded) { 3671 // Tell libjsig jvm finishes setting signal handlers 3672 (*end_signal_setting)(); 3673 } 3674 3675 // We don't activate signal checker if libjsig is in place, we trust ourselves 3676 // and if UserSignalHandler is installed all bets are off 3677 if (CheckJNICalls) { 3678 if (libjsig_is_loaded) { 3679 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 3680 check_signals = false; 3681 } 3682 if (AllowUserSignalHandlers) { 3683 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 3684 check_signals = false; 3685 } 3686 } 3687 } 3688} 3689 3690// This is the fastest way to get thread cpu time on Linux. 3691// Returns cpu time (user+sys) for any thread, not only for current. 3692// POSIX compliant clocks are implemented in the kernels 2.6.16+. 3693// It might work on 2.6.10+ with a special kernel/glibc patch. 3694// For reference, please, see IEEE Std 1003.1-2004: 3695// http://www.unix.org/single_unix_specification 3696 3697jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 3698 struct timespec tp; 3699 int rc = os::Linux::clock_gettime(clockid, &tp); 3700 assert(rc == 0, "clock_gettime is expected to return 0 code"); 3701 3702 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec; 3703} 3704 3705///// 3706// glibc on Linux platform uses non-documented flag 3707// to indicate, that some special sort of signal 3708// trampoline is used. 3709// We will never set this flag, and we should 3710// ignore this flag in our diagnostic 3711#ifdef SIGNIFICANT_SIGNAL_MASK 3712#undef SIGNIFICANT_SIGNAL_MASK 3713#endif 3714#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 3715 3716static const char* get_signal_handler_name(address handler, 3717 char* buf, int buflen) { 3718 int offset; 3719 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 3720 if (found) { 3721 // skip directory names 3722 const char *p1, *p2; 3723 p1 = buf; 3724 size_t len = strlen(os::file_separator()); 3725 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 3726 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 3727 } else { 3728 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 3729 } 3730 return buf; 3731} 3732 3733static void print_signal_handler(outputStream* st, int sig, 3734 char* buf, size_t buflen) { 3735 struct sigaction sa; 3736 3737 sigaction(sig, NULL, &sa); 3738 3739 // See comment for SIGNIFICANT_SIGNAL_MASK define 3740 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3741 3742 st->print("%s: ", os::exception_name(sig, buf, buflen)); 3743 3744 address handler = (sa.sa_flags & SA_SIGINFO) 3745 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 3746 : CAST_FROM_FN_PTR(address, sa.sa_handler); 3747 3748 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 3749 st->print("SIG_DFL"); 3750 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 3751 st->print("SIG_IGN"); 3752 } else { 3753 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 3754 } 3755 3756 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 3757 3758 address rh = VMError::get_resetted_sighandler(sig); 3759 // May be, handler was resetted by VMError? 3760 if(rh != NULL) { 3761 handler = rh; 3762 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 3763 } 3764 3765 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 3766 3767 // Check: is it our handler? 3768 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 3769 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 3770 // It is our signal handler 3771 // check for flags, reset system-used one! 3772 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 3773 st->print( 3774 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 3775 os::Linux::get_our_sigflags(sig)); 3776 } 3777 } 3778 st->cr(); 3779} 3780 3781 3782#define DO_SIGNAL_CHECK(sig) \ 3783 if (!sigismember(&check_signal_done, sig)) \ 3784 os::Linux::check_signal_handler(sig) 3785 3786// This method is a periodic task to check for misbehaving JNI applications 3787// under CheckJNI, we can add any periodic checks here 3788 3789void os::run_periodic_checks() { 3790 3791 if (check_signals == false) return; 3792 3793 // SEGV and BUS if overridden could potentially prevent 3794 // generation of hs*.log in the event of a crash, debugging 3795 // such a case can be very challenging, so we absolutely 3796 // check the following for a good measure: 3797 DO_SIGNAL_CHECK(SIGSEGV); 3798 DO_SIGNAL_CHECK(SIGILL); 3799 DO_SIGNAL_CHECK(SIGFPE); 3800 DO_SIGNAL_CHECK(SIGBUS); 3801 DO_SIGNAL_CHECK(SIGPIPE); 3802 DO_SIGNAL_CHECK(SIGXFSZ); 3803 3804 3805 // ReduceSignalUsage allows the user to override these handlers 3806 // see comments at the very top and jvm_solaris.h 3807 if (!ReduceSignalUsage) { 3808 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 3809 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 3810 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 3811 DO_SIGNAL_CHECK(BREAK_SIGNAL); 3812 } 3813 3814 DO_SIGNAL_CHECK(SR_signum); 3815 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 3816} 3817 3818typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 3819 3820static os_sigaction_t os_sigaction = NULL; 3821 3822void os::Linux::check_signal_handler(int sig) { 3823 char buf[O_BUFLEN]; 3824 address jvmHandler = NULL; 3825 3826 3827 struct sigaction act; 3828 if (os_sigaction == NULL) { 3829 // only trust the default sigaction, in case it has been interposed 3830 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 3831 if (os_sigaction == NULL) return; 3832 } 3833 3834 os_sigaction(sig, (struct sigaction*)NULL, &act); 3835 3836 3837 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3838 3839 address thisHandler = (act.sa_flags & SA_SIGINFO) 3840 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 3841 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 3842 3843 3844 switch(sig) { 3845 case SIGSEGV: 3846 case SIGBUS: 3847 case SIGFPE: 3848 case SIGPIPE: 3849 case SIGILL: 3850 case SIGXFSZ: 3851 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 3852 break; 3853 3854 case SHUTDOWN1_SIGNAL: 3855 case SHUTDOWN2_SIGNAL: 3856 case SHUTDOWN3_SIGNAL: 3857 case BREAK_SIGNAL: 3858 jvmHandler = (address)user_handler(); 3859 break; 3860 3861 case INTERRUPT_SIGNAL: 3862 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 3863 break; 3864 3865 default: 3866 if (sig == SR_signum) { 3867 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 3868 } else { 3869 return; 3870 } 3871 break; 3872 } 3873 3874 if (thisHandler != jvmHandler) { 3875 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 3876 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 3877 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 3878 // No need to check this sig any longer 3879 sigaddset(&check_signal_done, sig); 3880 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 3881 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 3882 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 3883 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 3884 // No need to check this sig any longer 3885 sigaddset(&check_signal_done, sig); 3886 } 3887 3888 // Dump all the signal 3889 if (sigismember(&check_signal_done, sig)) { 3890 print_signal_handlers(tty, buf, O_BUFLEN); 3891 } 3892} 3893 3894extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 3895 3896extern bool signal_name(int signo, char* buf, size_t len); 3897 3898const char* os::exception_name(int exception_code, char* buf, size_t size) { 3899 if (0 < exception_code && exception_code <= SIGRTMAX) { 3900 // signal 3901 if (!signal_name(exception_code, buf, size)) { 3902 jio_snprintf(buf, size, "SIG%d", exception_code); 3903 } 3904 return buf; 3905 } else { 3906 return NULL; 3907 } 3908} 3909 3910// this is called _before_ the most of global arguments have been parsed 3911void os::init(void) { 3912 char dummy; /* used to get a guess on initial stack address */ 3913// first_hrtime = gethrtime(); 3914 3915 // With LinuxThreads the JavaMain thread pid (primordial thread) 3916 // is different than the pid of the java launcher thread. 3917 // So, on Linux, the launcher thread pid is passed to the VM 3918 // via the sun.java.launcher.pid property. 3919 // Use this property instead of getpid() if it was correctly passed. 3920 // See bug 6351349. 3921 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 3922 3923 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 3924 3925 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 3926 3927 init_random(1234567); 3928 3929 ThreadCritical::initialize(); 3930 3931 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 3932 if (Linux::page_size() == -1) { 3933 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 3934 strerror(errno))); 3935 } 3936 init_page_sizes((size_t) Linux::page_size()); 3937 3938 Linux::initialize_system_info(); 3939 3940 // main_thread points to the aboriginal thread 3941 Linux::_main_thread = pthread_self(); 3942 3943 Linux::clock_init(); 3944 initial_time_count = os::elapsed_counter(); 3945 pthread_mutex_init(&dl_mutex, NULL); 3946} 3947 3948// To install functions for atexit system call 3949extern "C" { 3950 static void perfMemory_exit_helper() { 3951 perfMemory_exit(); 3952 } 3953} 3954 3955// this is called _after_ the global arguments have been parsed 3956jint os::init_2(void) 3957{ 3958 Linux::fast_thread_clock_init(); 3959 3960 // Allocate a single page and mark it as readable for safepoint polling 3961 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3962 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 3963 3964 os::set_polling_page( polling_page ); 3965 3966#ifndef PRODUCT 3967 if(Verbose && PrintMiscellaneous) 3968 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 3969#endif 3970 3971 if (!UseMembar) { 3972 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3973 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); 3974 os::set_memory_serialize_page( mem_serialize_page ); 3975 3976#ifndef PRODUCT 3977 if(Verbose && PrintMiscellaneous) 3978 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 3979#endif 3980 } 3981 3982 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init()); 3983 3984 // initialize suspend/resume support - must do this before signal_sets_init() 3985 if (SR_initialize() != 0) { 3986 perror("SR_initialize failed"); 3987 return JNI_ERR; 3988 } 3989 3990 Linux::signal_sets_init(); 3991 Linux::install_signal_handlers(); 3992 3993 // Check minimum allowable stack size for thread creation and to initialize 3994 // the java system classes, including StackOverflowError - depends on page 3995 // size. Add a page for compiler2 recursion in main thread. 3996 // Add in 2*BytesPerWord times page size to account for VM stack during 3997 // class initialization depending on 32 or 64 bit VM. 3998 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 3999 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+ 4000 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size()); 4001 4002 size_t threadStackSizeInBytes = ThreadStackSize * K; 4003 if (threadStackSizeInBytes != 0 && 4004 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4005 tty->print_cr("\nThe stack size specified is too small, " 4006 "Specify at least %dk", 4007 os::Linux::min_stack_allowed/ K); 4008 return JNI_ERR; 4009 } 4010 4011 // Make the stack size a multiple of the page size so that 4012 // the yellow/red zones can be guarded. 4013 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4014 vm_page_size())); 4015 4016 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4017 4018 Linux::libpthread_init(); 4019 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4020 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4021 Linux::glibc_version(), Linux::libpthread_version(), 4022 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4023 } 4024 4025 if (UseNUMA) { 4026 if (!Linux::libnuma_init()) { 4027 UseNUMA = false; 4028 } else { 4029 if ((Linux::numa_max_node() < 1)) { 4030 // There's only one node(they start from 0), disable NUMA. 4031 UseNUMA = false; 4032 } 4033 } 4034 if (!UseNUMA && ForceNUMA) { 4035 UseNUMA = true; 4036 } 4037 } 4038 4039 if (MaxFDLimit) { 4040 // set the number of file descriptors to max. print out error 4041 // if getrlimit/setrlimit fails but continue regardless. 4042 struct rlimit nbr_files; 4043 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4044 if (status != 0) { 4045 if (PrintMiscellaneous && (Verbose || WizardMode)) 4046 perror("os::init_2 getrlimit failed"); 4047 } else { 4048 nbr_files.rlim_cur = nbr_files.rlim_max; 4049 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4050 if (status != 0) { 4051 if (PrintMiscellaneous && (Verbose || WizardMode)) 4052 perror("os::init_2 setrlimit failed"); 4053 } 4054 } 4055 } 4056 4057 // Initialize lock used to serialize thread creation (see os::create_thread) 4058 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4059 4060 // Initialize HPI. 4061 jint hpi_result = hpi::initialize(); 4062 if (hpi_result != JNI_OK) { 4063 tty->print_cr("There was an error trying to initialize the HPI library."); 4064 return hpi_result; 4065 } 4066 4067 // at-exit methods are called in the reverse order of their registration. 4068 // atexit functions are called on return from main or as a result of a 4069 // call to exit(3C). There can be only 32 of these functions registered 4070 // and atexit() does not set errno. 4071 4072 if (PerfAllowAtExitRegistration) { 4073 // only register atexit functions if PerfAllowAtExitRegistration is set. 4074 // atexit functions can be delayed until process exit time, which 4075 // can be problematic for embedded VM situations. Embedded VMs should 4076 // call DestroyJavaVM() to assure that VM resources are released. 4077 4078 // note: perfMemory_exit_helper atexit function may be removed in 4079 // the future if the appropriate cleanup code can be added to the 4080 // VM_Exit VMOperation's doit method. 4081 if (atexit(perfMemory_exit_helper) != 0) { 4082 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 4083 } 4084 } 4085 4086 // initialize thread priority policy 4087 prio_init(); 4088 4089 return JNI_OK; 4090} 4091 4092// this is called at the end of vm_initialization 4093void os::init_3(void) { } 4094 4095// Mark the polling page as unreadable 4096void os::make_polling_page_unreadable(void) { 4097 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4098 fatal("Could not disable polling page"); 4099}; 4100 4101// Mark the polling page as readable 4102void os::make_polling_page_readable(void) { 4103 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4104 fatal("Could not enable polling page"); 4105 } 4106}; 4107 4108int os::active_processor_count() { 4109 // Linux doesn't yet have a (official) notion of processor sets, 4110 // so just return the number of online processors. 4111 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4112 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4113 return online_cpus; 4114} 4115 4116bool os::distribute_processes(uint length, uint* distribution) { 4117 // Not yet implemented. 4118 return false; 4119} 4120 4121bool os::bind_to_processor(uint processor_id) { 4122 // Not yet implemented. 4123 return false; 4124} 4125 4126/// 4127 4128// Suspends the target using the signal mechanism and then grabs the PC before 4129// resuming the target. Used by the flat-profiler only 4130ExtendedPC os::get_thread_pc(Thread* thread) { 4131 // Make sure that it is called by the watcher for the VMThread 4132 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4133 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4134 4135 ExtendedPC epc; 4136 4137 OSThread* osthread = thread->osthread(); 4138 if (do_suspend(osthread)) { 4139 if (osthread->ucontext() != NULL) { 4140 epc = os::Linux::ucontext_get_pc(osthread->ucontext()); 4141 } else { 4142 // NULL context is unexpected, double-check this is the VMThread 4143 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4144 } 4145 do_resume(osthread); 4146 } 4147 // failure means pthread_kill failed for some reason - arguably this is 4148 // a fatal problem, but such problems are ignored elsewhere 4149 4150 return epc; 4151} 4152 4153int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4154{ 4155 if (is_NPTL()) { 4156 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4157 } else { 4158#ifndef IA64 4159 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4160 // word back to default 64bit precision if condvar is signaled. Java 4161 // wants 53bit precision. Save and restore current value. 4162 int fpu = get_fpu_control_word(); 4163#endif // IA64 4164 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4165#ifndef IA64 4166 set_fpu_control_word(fpu); 4167#endif // IA64 4168 return status; 4169 } 4170} 4171 4172//////////////////////////////////////////////////////////////////////////////// 4173// debug support 4174 4175static address same_page(address x, address y) { 4176 int page_bits = -os::vm_page_size(); 4177 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) 4178 return x; 4179 else if (x > y) 4180 return (address)(intptr_t(y) | ~page_bits) + 1; 4181 else 4182 return (address)(intptr_t(y) & page_bits); 4183} 4184 4185bool os::find(address addr, outputStream* st) { 4186 Dl_info dlinfo; 4187 memset(&dlinfo, 0, sizeof(dlinfo)); 4188 if (dladdr(addr, &dlinfo)) { 4189 st->print(PTR_FORMAT ": ", addr); 4190 if (dlinfo.dli_sname != NULL) { 4191 st->print("%s+%#x", dlinfo.dli_sname, 4192 addr - (intptr_t)dlinfo.dli_saddr); 4193 } else if (dlinfo.dli_fname) { 4194 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4195 } else { 4196 st->print("<absolute address>"); 4197 } 4198 if (dlinfo.dli_fname) { 4199 st->print(" in %s", dlinfo.dli_fname); 4200 } 4201 if (dlinfo.dli_fbase) { 4202 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4203 } 4204 st->cr(); 4205 4206 if (Verbose) { 4207 // decode some bytes around the PC 4208 address begin = same_page(addr-40, addr); 4209 address end = same_page(addr+40, addr); 4210 address lowest = (address) dlinfo.dli_sname; 4211 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4212 if (begin < lowest) begin = lowest; 4213 Dl_info dlinfo2; 4214 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr 4215 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 4216 end = (address) dlinfo2.dli_saddr; 4217 Disassembler::decode(begin, end, st); 4218 } 4219 return true; 4220 } 4221 return false; 4222} 4223 4224//////////////////////////////////////////////////////////////////////////////// 4225// misc 4226 4227// This does not do anything on Linux. This is basically a hook for being 4228// able to use structured exception handling (thread-local exception filters) 4229// on, e.g., Win32. 4230void 4231os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4232 JavaCallArguments* args, Thread* thread) { 4233 f(value, method, args, thread); 4234} 4235 4236void os::print_statistics() { 4237} 4238 4239int os::message_box(const char* title, const char* message) { 4240 int i; 4241 fdStream err(defaultStream::error_fd()); 4242 for (i = 0; i < 78; i++) err.print_raw("="); 4243 err.cr(); 4244 err.print_raw_cr(title); 4245 for (i = 0; i < 78; i++) err.print_raw("-"); 4246 err.cr(); 4247 err.print_raw_cr(message); 4248 for (i = 0; i < 78; i++) err.print_raw("="); 4249 err.cr(); 4250 4251 char buf[16]; 4252 // Prevent process from exiting upon "read error" without consuming all CPU 4253 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 4254 4255 return buf[0] == 'y' || buf[0] == 'Y'; 4256} 4257 4258int os::stat(const char *path, struct stat *sbuf) { 4259 char pathbuf[MAX_PATH]; 4260 if (strlen(path) > MAX_PATH - 1) { 4261 errno = ENAMETOOLONG; 4262 return -1; 4263 } 4264 hpi::native_path(strcpy(pathbuf, path)); 4265 return ::stat(pathbuf, sbuf); 4266} 4267 4268bool os::check_heap(bool force) { 4269 return true; 4270} 4271 4272int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 4273 return ::vsnprintf(buf, count, format, args); 4274} 4275 4276// Is a (classpath) directory empty? 4277bool os::dir_is_empty(const char* path) { 4278 DIR *dir = NULL; 4279 struct dirent *ptr; 4280 4281 dir = opendir(path); 4282 if (dir == NULL) return true; 4283 4284 /* Scan the directory */ 4285 bool result = true; 4286 char buf[sizeof(struct dirent) + MAX_PATH]; 4287 while (result && (ptr = ::readdir(dir)) != NULL) { 4288 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 4289 result = false; 4290 } 4291 } 4292 closedir(dir); 4293 return result; 4294} 4295 4296// create binary file, rewriting existing file if required 4297int os::create_binary_file(const char* path, bool rewrite_existing) { 4298 int oflags = O_WRONLY | O_CREAT; 4299 if (!rewrite_existing) { 4300 oflags |= O_EXCL; 4301 } 4302 return ::open64(path, oflags, S_IREAD | S_IWRITE); 4303} 4304 4305// return current position of file pointer 4306jlong os::current_file_offset(int fd) { 4307 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 4308} 4309 4310// move file pointer to the specified offset 4311jlong os::seek_to_file_offset(int fd, jlong offset) { 4312 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 4313} 4314 4315// Map a block of memory. 4316char* os::map_memory(int fd, const char* file_name, size_t file_offset, 4317 char *addr, size_t bytes, bool read_only, 4318 bool allow_exec) { 4319 int prot; 4320 int flags; 4321 4322 if (read_only) { 4323 prot = PROT_READ; 4324 flags = MAP_SHARED; 4325 } else { 4326 prot = PROT_READ | PROT_WRITE; 4327 flags = MAP_PRIVATE; 4328 } 4329 4330 if (allow_exec) { 4331 prot |= PROT_EXEC; 4332 } 4333 4334 if (addr != NULL) { 4335 flags |= MAP_FIXED; 4336 } 4337 4338 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 4339 fd, file_offset); 4340 if (mapped_address == MAP_FAILED) { 4341 return NULL; 4342 } 4343 return mapped_address; 4344} 4345 4346 4347// Remap a block of memory. 4348char* os::remap_memory(int fd, const char* file_name, size_t file_offset, 4349 char *addr, size_t bytes, bool read_only, 4350 bool allow_exec) { 4351 // same as map_memory() on this OS 4352 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 4353 allow_exec); 4354} 4355 4356 4357// Unmap a block of memory. 4358bool os::unmap_memory(char* addr, size_t bytes) { 4359 return munmap(addr, bytes) == 0; 4360} 4361 4362static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 4363 4364static clockid_t thread_cpu_clockid(Thread* thread) { 4365 pthread_t tid = thread->osthread()->pthread_id(); 4366 clockid_t clockid; 4367 4368 // Get thread clockid 4369 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 4370 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 4371 return clockid; 4372} 4373 4374// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 4375// are used by JVM M&M and JVMTI to get user+sys or user CPU time 4376// of a thread. 4377// 4378// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 4379// the fast estimate available on the platform. 4380 4381jlong os::current_thread_cpu_time() { 4382 if (os::Linux::supports_fast_thread_cpu_time()) { 4383 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4384 } else { 4385 // return user + sys since the cost is the same 4386 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 4387 } 4388} 4389 4390jlong os::thread_cpu_time(Thread* thread) { 4391 // consistent with what current_thread_cpu_time() returns 4392 if (os::Linux::supports_fast_thread_cpu_time()) { 4393 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4394 } else { 4395 return slow_thread_cpu_time(thread, true /* user + sys */); 4396 } 4397} 4398 4399jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 4400 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4401 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4402 } else { 4403 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 4404 } 4405} 4406 4407jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4408 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4409 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4410 } else { 4411 return slow_thread_cpu_time(thread, user_sys_cpu_time); 4412 } 4413} 4414 4415// 4416// -1 on error. 4417// 4418 4419static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4420 static bool proc_pid_cpu_avail = true; 4421 static bool proc_task_unchecked = true; 4422 static const char *proc_stat_path = "/proc/%d/stat"; 4423 pid_t tid = thread->osthread()->thread_id(); 4424 int i; 4425 char *s; 4426 char stat[2048]; 4427 int statlen; 4428 char proc_name[64]; 4429 int count; 4430 long sys_time, user_time; 4431 char string[64]; 4432 char cdummy; 4433 int idummy; 4434 long ldummy; 4435 FILE *fp; 4436 4437 // We first try accessing /proc/<pid>/cpu since this is faster to 4438 // process. If this file is not present (linux kernels 2.5 and above) 4439 // then we open /proc/<pid>/stat. 4440 if ( proc_pid_cpu_avail ) { 4441 sprintf(proc_name, "/proc/%d/cpu", tid); 4442 fp = fopen(proc_name, "r"); 4443 if ( fp != NULL ) { 4444 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time); 4445 fclose(fp); 4446 if ( count != 3 ) return -1; 4447 4448 if (user_sys_cpu_time) { 4449 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4450 } else { 4451 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4452 } 4453 } 4454 else proc_pid_cpu_avail = false; 4455 } 4456 4457 // The /proc/<tid>/stat aggregates per-process usage on 4458 // new Linux kernels 2.6+ where NPTL is supported. 4459 // The /proc/self/task/<tid>/stat still has the per-thread usage. 4460 // See bug 6328462. 4461 // There can be no directory /proc/self/task on kernels 2.4 with NPTL 4462 // and possibly in some other cases, so we check its availability. 4463 if (proc_task_unchecked && os::Linux::is_NPTL()) { 4464 // This is executed only once 4465 proc_task_unchecked = false; 4466 fp = fopen("/proc/self/task", "r"); 4467 if (fp != NULL) { 4468 proc_stat_path = "/proc/self/task/%d/stat"; 4469 fclose(fp); 4470 } 4471 } 4472 4473 sprintf(proc_name, proc_stat_path, tid); 4474 fp = fopen(proc_name, "r"); 4475 if ( fp == NULL ) return -1; 4476 statlen = fread(stat, 1, 2047, fp); 4477 stat[statlen] = '\0'; 4478 fclose(fp); 4479 4480 // Skip pid and the command string. Note that we could be dealing with 4481 // weird command names, e.g. user could decide to rename java launcher 4482 // to "java 1.4.2 :)", then the stat file would look like 4483 // 1234 (java 1.4.2 :)) R ... ... 4484 // We don't really need to know the command string, just find the last 4485 // occurrence of ")" and then start parsing from there. See bug 4726580. 4486 s = strrchr(stat, ')'); 4487 i = 0; 4488 if (s == NULL ) return -1; 4489 4490 // Skip blank chars 4491 do s++; while (isspace(*s)); 4492 4493 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 4494 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 4495 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 4496 &user_time, &sys_time); 4497 if ( count != 13 ) return -1; 4498 if (user_sys_cpu_time) { 4499 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4500 } else { 4501 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4502 } 4503} 4504 4505void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4506 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4507 info_ptr->may_skip_backward = false; // elapsed time not wall time 4508 info_ptr->may_skip_forward = false; // elapsed time not wall time 4509 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4510} 4511 4512void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4513 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4514 info_ptr->may_skip_backward = false; // elapsed time not wall time 4515 info_ptr->may_skip_forward = false; // elapsed time not wall time 4516 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4517} 4518 4519bool os::is_thread_cpu_time_supported() { 4520 return true; 4521} 4522 4523// System loadavg support. Returns -1 if load average cannot be obtained. 4524// Linux doesn't yet have a (official) notion of processor sets, 4525// so just return the system wide load average. 4526int os::loadavg(double loadavg[], int nelem) { 4527 return ::getloadavg(loadavg, nelem); 4528} 4529 4530void os::pause() { 4531 char filename[MAX_PATH]; 4532 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 4533 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 4534 } else { 4535 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 4536 } 4537 4538 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 4539 if (fd != -1) { 4540 struct stat buf; 4541 close(fd); 4542 while (::stat(filename, &buf) == 0) { 4543 (void)::poll(NULL, 0, 100); 4544 } 4545 } else { 4546 jio_fprintf(stderr, 4547 "Could not open pause file '%s', continuing immediately.\n", filename); 4548 } 4549} 4550 4551extern "C" { 4552 4553/** 4554 * NOTE: the following code is to keep the green threads code 4555 * in the libjava.so happy. Once the green threads is removed, 4556 * these code will no longer be needed. 4557 */ 4558int 4559jdk_waitpid(pid_t pid, int* status, int options) { 4560 return waitpid(pid, status, options); 4561} 4562 4563int 4564fork1() { 4565 return fork(); 4566} 4567 4568int 4569jdk_sem_init(sem_t *sem, int pshared, unsigned int value) { 4570 return sem_init(sem, pshared, value); 4571} 4572 4573int 4574jdk_sem_post(sem_t *sem) { 4575 return sem_post(sem); 4576} 4577 4578int 4579jdk_sem_wait(sem_t *sem) { 4580 return sem_wait(sem); 4581} 4582 4583int 4584jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) { 4585 return pthread_sigmask(how , newmask, oldmask); 4586} 4587 4588} 4589 4590// Refer to the comments in os_solaris.cpp park-unpark. 4591// 4592// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 4593// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 4594// For specifics regarding the bug see GLIBC BUGID 261237 : 4595// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 4596// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 4597// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 4598// is used. (The simple C test-case provided in the GLIBC bug report manifests the 4599// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 4600// and monitorenter when we're using 1-0 locking. All those operations may result in 4601// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 4602// of libpthread avoids the problem, but isn't practical. 4603// 4604// Possible remedies: 4605// 4606// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 4607// This is palliative and probabilistic, however. If the thread is preempted 4608// between the call to compute_abstime() and pthread_cond_timedwait(), more 4609// than the minimum period may have passed, and the abstime may be stale (in the 4610// past) resultin in a hang. Using this technique reduces the odds of a hang 4611// but the JVM is still vulnerable, particularly on heavily loaded systems. 4612// 4613// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 4614// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 4615// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 4616// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 4617// thread. 4618// 4619// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 4620// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 4621// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 4622// This also works well. In fact it avoids kernel-level scalability impediments 4623// on certain platforms that don't handle lots of active pthread_cond_timedwait() 4624// timers in a graceful fashion. 4625// 4626// 4. When the abstime value is in the past it appears that control returns 4627// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 4628// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 4629// can avoid the problem by reinitializing the condvar -- by cond_destroy() 4630// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 4631// It may be possible to avoid reinitialization by checking the return 4632// value from pthread_cond_timedwait(). In addition to reinitializing the 4633// condvar we must establish the invariant that cond_signal() is only called 4634// within critical sections protected by the adjunct mutex. This prevents 4635// cond_signal() from "seeing" a condvar that's in the midst of being 4636// reinitialized or that is corrupt. Sadly, this invariant obviates the 4637// desirable signal-after-unlock optimization that avoids futile context switching. 4638// 4639// I'm also concerned that some versions of NTPL might allocate an auxilliary 4640// structure when a condvar is used or initialized. cond_destroy() would 4641// release the helper structure. Our reinitialize-after-timedwait fix 4642// put excessive stress on malloc/free and locks protecting the c-heap. 4643// 4644// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 4645// It may be possible to refine (4) by checking the kernel and NTPL verisons 4646// and only enabling the work-around for vulnerable environments. 4647 4648// utility to compute the abstime argument to timedwait: 4649// millis is the relative timeout time 4650// abstime will be the absolute timeout time 4651// TODO: replace compute_abstime() with unpackTime() 4652 4653static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 4654 if (millis < 0) millis = 0; 4655 struct timeval now; 4656 int status = gettimeofday(&now, NULL); 4657 assert(status == 0, "gettimeofday"); 4658 jlong seconds = millis / 1000; 4659 millis %= 1000; 4660 if (seconds > 50000000) { // see man cond_timedwait(3T) 4661 seconds = 50000000; 4662 } 4663 abstime->tv_sec = now.tv_sec + seconds; 4664 long usec = now.tv_usec + millis * 1000; 4665 if (usec >= 1000000) { 4666 abstime->tv_sec += 1; 4667 usec -= 1000000; 4668 } 4669 abstime->tv_nsec = usec * 1000; 4670 return abstime; 4671} 4672 4673 4674// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 4675// Conceptually TryPark() should be equivalent to park(0). 4676 4677int os::PlatformEvent::TryPark() { 4678 for (;;) { 4679 const int v = _Event ; 4680 guarantee ((v == 0) || (v == 1), "invariant") ; 4681 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 4682 } 4683} 4684 4685void os::PlatformEvent::park() { // AKA "down()" 4686 // Invariant: Only the thread associated with the Event/PlatformEvent 4687 // may call park(). 4688 // TODO: assert that _Assoc != NULL or _Assoc == Self 4689 int v ; 4690 for (;;) { 4691 v = _Event ; 4692 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4693 } 4694 guarantee (v >= 0, "invariant") ; 4695 if (v == 0) { 4696 // Do this the hard way by blocking ... 4697 int status = pthread_mutex_lock(_mutex); 4698 assert_status(status == 0, status, "mutex_lock"); 4699 guarantee (_nParked == 0, "invariant") ; 4700 ++ _nParked ; 4701 while (_Event < 0) { 4702 status = pthread_cond_wait(_cond, _mutex); 4703 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 4704 // Treat this the same as if the wait was interrupted 4705 if (status == ETIME) { status = EINTR; } 4706 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 4707 } 4708 -- _nParked ; 4709 4710 // In theory we could move the ST of 0 into _Event past the unlock(), 4711 // but then we'd need a MEMBAR after the ST. 4712 _Event = 0 ; 4713 status = pthread_mutex_unlock(_mutex); 4714 assert_status(status == 0, status, "mutex_unlock"); 4715 } 4716 guarantee (_Event >= 0, "invariant") ; 4717} 4718 4719int os::PlatformEvent::park(jlong millis) { 4720 guarantee (_nParked == 0, "invariant") ; 4721 4722 int v ; 4723 for (;;) { 4724 v = _Event ; 4725 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4726 } 4727 guarantee (v >= 0, "invariant") ; 4728 if (v != 0) return OS_OK ; 4729 4730 // We do this the hard way, by blocking the thread. 4731 // Consider enforcing a minimum timeout value. 4732 struct timespec abst; 4733 compute_abstime(&abst, millis); 4734 4735 int ret = OS_TIMEOUT; 4736 int status = pthread_mutex_lock(_mutex); 4737 assert_status(status == 0, status, "mutex_lock"); 4738 guarantee (_nParked == 0, "invariant") ; 4739 ++_nParked ; 4740 4741 // Object.wait(timo) will return because of 4742 // (a) notification 4743 // (b) timeout 4744 // (c) thread.interrupt 4745 // 4746 // Thread.interrupt and object.notify{All} both call Event::set. 4747 // That is, we treat thread.interrupt as a special case of notification. 4748 // The underlying Solaris implementation, cond_timedwait, admits 4749 // spurious/premature wakeups, but the JLS/JVM spec prevents the 4750 // JVM from making those visible to Java code. As such, we must 4751 // filter out spurious wakeups. We assume all ETIME returns are valid. 4752 // 4753 // TODO: properly differentiate simultaneous notify+interrupt. 4754 // In that case, we should propagate the notify to another waiter. 4755 4756 while (_Event < 0) { 4757 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 4758 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4759 pthread_cond_destroy (_cond); 4760 pthread_cond_init (_cond, NULL) ; 4761 } 4762 assert_status(status == 0 || status == EINTR || 4763 status == ETIME || status == ETIMEDOUT, 4764 status, "cond_timedwait"); 4765 if (!FilterSpuriousWakeups) break ; // previous semantics 4766 if (status == ETIME || status == ETIMEDOUT) break ; 4767 // We consume and ignore EINTR and spurious wakeups. 4768 } 4769 --_nParked ; 4770 if (_Event >= 0) { 4771 ret = OS_OK; 4772 } 4773 _Event = 0 ; 4774 status = pthread_mutex_unlock(_mutex); 4775 assert_status(status == 0, status, "mutex_unlock"); 4776 assert (_nParked == 0, "invariant") ; 4777 return ret; 4778} 4779 4780void os::PlatformEvent::unpark() { 4781 int v, AnyWaiters ; 4782 for (;;) { 4783 v = _Event ; 4784 if (v > 0) { 4785 // The LD of _Event could have reordered or be satisfied 4786 // by a read-aside from this processor's write buffer. 4787 // To avoid problems execute a barrier and then 4788 // ratify the value. 4789 OrderAccess::fence() ; 4790 if (_Event == v) return ; 4791 continue ; 4792 } 4793 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ; 4794 } 4795 if (v < 0) { 4796 // Wait for the thread associated with the event to vacate 4797 int status = pthread_mutex_lock(_mutex); 4798 assert_status(status == 0, status, "mutex_lock"); 4799 AnyWaiters = _nParked ; 4800 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ; 4801 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 4802 AnyWaiters = 0 ; 4803 pthread_cond_signal (_cond); 4804 } 4805 status = pthread_mutex_unlock(_mutex); 4806 assert_status(status == 0, status, "mutex_unlock"); 4807 if (AnyWaiters != 0) { 4808 status = pthread_cond_signal(_cond); 4809 assert_status(status == 0, status, "cond_signal"); 4810 } 4811 } 4812 4813 // Note that we signal() _after dropping the lock for "immortal" Events. 4814 // This is safe and avoids a common class of futile wakeups. In rare 4815 // circumstances this can cause a thread to return prematurely from 4816 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 4817 // simply re-test the condition and re-park itself. 4818} 4819 4820 4821// JSR166 4822// ------------------------------------------------------- 4823 4824/* 4825 * The solaris and linux implementations of park/unpark are fairly 4826 * conservative for now, but can be improved. They currently use a 4827 * mutex/condvar pair, plus a a count. 4828 * Park decrements count if > 0, else does a condvar wait. Unpark 4829 * sets count to 1 and signals condvar. Only one thread ever waits 4830 * on the condvar. Contention seen when trying to park implies that someone 4831 * is unparking you, so don't wait. And spurious returns are fine, so there 4832 * is no need to track notifications. 4833 */ 4834 4835 4836#define NANOSECS_PER_SEC 1000000000 4837#define NANOSECS_PER_MILLISEC 1000000 4838#define MAX_SECS 100000000 4839/* 4840 * This code is common to linux and solaris and will be moved to a 4841 * common place in dolphin. 4842 * 4843 * The passed in time value is either a relative time in nanoseconds 4844 * or an absolute time in milliseconds. Either way it has to be unpacked 4845 * into suitable seconds and nanoseconds components and stored in the 4846 * given timespec structure. 4847 * Given time is a 64-bit value and the time_t used in the timespec is only 4848 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 4849 * overflow if times way in the future are given. Further on Solaris versions 4850 * prior to 10 there is a restriction (see cond_timedwait) that the specified 4851 * number of seconds, in abstime, is less than current_time + 100,000,000. 4852 * As it will be 28 years before "now + 100000000" will overflow we can 4853 * ignore overflow and just impose a hard-limit on seconds using the value 4854 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 4855 * years from "now". 4856 */ 4857 4858static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 4859 assert (time > 0, "convertTime"); 4860 4861 struct timeval now; 4862 int status = gettimeofday(&now, NULL); 4863 assert(status == 0, "gettimeofday"); 4864 4865 time_t max_secs = now.tv_sec + MAX_SECS; 4866 4867 if (isAbsolute) { 4868 jlong secs = time / 1000; 4869 if (secs > max_secs) { 4870 absTime->tv_sec = max_secs; 4871 } 4872 else { 4873 absTime->tv_sec = secs; 4874 } 4875 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 4876 } 4877 else { 4878 jlong secs = time / NANOSECS_PER_SEC; 4879 if (secs >= MAX_SECS) { 4880 absTime->tv_sec = max_secs; 4881 absTime->tv_nsec = 0; 4882 } 4883 else { 4884 absTime->tv_sec = now.tv_sec + secs; 4885 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 4886 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 4887 absTime->tv_nsec -= NANOSECS_PER_SEC; 4888 ++absTime->tv_sec; // note: this must be <= max_secs 4889 } 4890 } 4891 } 4892 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 4893 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 4894 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 4895 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 4896} 4897 4898void Parker::park(bool isAbsolute, jlong time) { 4899 // Optional fast-path check: 4900 // Return immediately if a permit is available. 4901 if (_counter > 0) { 4902 _counter = 0 ; 4903 OrderAccess::fence(); 4904 return ; 4905 } 4906 4907 Thread* thread = Thread::current(); 4908 assert(thread->is_Java_thread(), "Must be JavaThread"); 4909 JavaThread *jt = (JavaThread *)thread; 4910 4911 // Optional optimization -- avoid state transitions if there's an interrupt pending. 4912 // Check interrupt before trying to wait 4913 if (Thread::is_interrupted(thread, false)) { 4914 return; 4915 } 4916 4917 // Next, demultiplex/decode time arguments 4918 timespec absTime; 4919 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 4920 return; 4921 } 4922 if (time > 0) { 4923 unpackTime(&absTime, isAbsolute, time); 4924 } 4925 4926 4927 // Enter safepoint region 4928 // Beware of deadlocks such as 6317397. 4929 // The per-thread Parker:: mutex is a classic leaf-lock. 4930 // In particular a thread must never block on the Threads_lock while 4931 // holding the Parker:: mutex. If safepoints are pending both the 4932 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 4933 ThreadBlockInVM tbivm(jt); 4934 4935 // Don't wait if cannot get lock since interference arises from 4936 // unblocking. Also. check interrupt before trying wait 4937 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 4938 return; 4939 } 4940 4941 int status ; 4942 if (_counter > 0) { // no wait needed 4943 _counter = 0; 4944 status = pthread_mutex_unlock(_mutex); 4945 assert (status == 0, "invariant") ; 4946 OrderAccess::fence(); 4947 return; 4948 } 4949 4950#ifdef ASSERT 4951 // Don't catch signals while blocked; let the running threads have the signals. 4952 // (This allows a debugger to break into the running thread.) 4953 sigset_t oldsigs; 4954 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 4955 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 4956#endif 4957 4958 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 4959 jt->set_suspend_equivalent(); 4960 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 4961 4962 if (time == 0) { 4963 status = pthread_cond_wait (_cond, _mutex) ; 4964 } else { 4965 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 4966 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4967 pthread_cond_destroy (_cond) ; 4968 pthread_cond_init (_cond, NULL); 4969 } 4970 } 4971 assert_status(status == 0 || status == EINTR || 4972 status == ETIME || status == ETIMEDOUT, 4973 status, "cond_timedwait"); 4974 4975#ifdef ASSERT 4976 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 4977#endif 4978 4979 _counter = 0 ; 4980 status = pthread_mutex_unlock(_mutex) ; 4981 assert_status(status == 0, status, "invariant") ; 4982 // If externally suspended while waiting, re-suspend 4983 if (jt->handle_special_suspend_equivalent_condition()) { 4984 jt->java_suspend_self(); 4985 } 4986 4987 OrderAccess::fence(); 4988} 4989 4990void Parker::unpark() { 4991 int s, status ; 4992 status = pthread_mutex_lock(_mutex); 4993 assert (status == 0, "invariant") ; 4994 s = _counter; 4995 _counter = 1; 4996 if (s < 1) { 4997 if (WorkAroundNPTLTimedWaitHang) { 4998 status = pthread_cond_signal (_cond) ; 4999 assert (status == 0, "invariant") ; 5000 status = pthread_mutex_unlock(_mutex); 5001 assert (status == 0, "invariant") ; 5002 } else { 5003 status = pthread_mutex_unlock(_mutex); 5004 assert (status == 0, "invariant") ; 5005 status = pthread_cond_signal (_cond) ; 5006 assert (status == 0, "invariant") ; 5007 } 5008 } else { 5009 pthread_mutex_unlock(_mutex); 5010 assert (status == 0, "invariant") ; 5011 } 5012} 5013 5014 5015extern char** environ; 5016 5017#ifndef __NR_fork 5018#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5019#endif 5020 5021#ifndef __NR_execve 5022#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5023#endif 5024 5025// Run the specified command in a separate process. Return its exit value, 5026// or -1 on failure (e.g. can't fork a new process). 5027// Unlike system(), this function can be called from signal handler. It 5028// doesn't block SIGINT et al. 5029int os::fork_and_exec(char* cmd) { 5030 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5031 5032 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5033 // pthread_atfork handlers and reset pthread library. All we need is a 5034 // separate process to execve. Make a direct syscall to fork process. 5035 // On IA64 there's no fork syscall, we have to use fork() and hope for 5036 // the best... 5037 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5038 IA64_ONLY(fork();) 5039 5040 if (pid < 0) { 5041 // fork failed 5042 return -1; 5043 5044 } else if (pid == 0) { 5045 // child process 5046 5047 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5048 // first to kill every thread on the thread list. Because this list is 5049 // not reset by fork() (see notes above), execve() will instead kill 5050 // every thread in the parent process. We know this is the only thread 5051 // in the new process, so make a system call directly. 5052 // IA64 should use normal execve() from glibc to match the glibc fork() 5053 // above. 5054 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5055 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5056 5057 // execve failed 5058 _exit(-1); 5059 5060 } else { 5061 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5062 // care about the actual exit code, for now. 5063 5064 int status; 5065 5066 // Wait for the child process to exit. This returns immediately if 5067 // the child has already exited. */ 5068 while (waitpid(pid, &status, 0) < 0) { 5069 switch (errno) { 5070 case ECHILD: return 0; 5071 case EINTR: break; 5072 default: return -1; 5073 } 5074 } 5075 5076 if (WIFEXITED(status)) { 5077 // The child exited normally; get its exit code. 5078 return WEXITSTATUS(status); 5079 } else if (WIFSIGNALED(status)) { 5080 // The child exited because of a signal 5081 // The best value to return is 0x80 + signal number, 5082 // because that is what all Unix shells do, and because 5083 // it allows callers to distinguish between process exit and 5084 // process death by signal. 5085 return 0x80 + WTERMSIG(status); 5086 } else { 5087 // Unknown exit code; pass it through 5088 return status; 5089 } 5090 } 5091} 5092 5093// is_headless_jre() 5094// 5095// Test for the existence of libmawt in motif21 or xawt directories 5096// in order to report if we are running in a headless jre 5097// 5098bool os::is_headless_jre() { 5099 struct stat statbuf; 5100 char buf[MAXPATHLEN]; 5101 char libmawtpath[MAXPATHLEN]; 5102 const char *xawtstr = "/xawt/libmawt.so"; 5103 const char *motifstr = "/motif21/libmawt.so"; 5104 char *p; 5105 5106 // Get path to libjvm.so 5107 os::jvm_path(buf, sizeof(buf)); 5108 5109 // Get rid of libjvm.so 5110 p = strrchr(buf, '/'); 5111 if (p == NULL) return false; 5112 else *p = '\0'; 5113 5114 // Get rid of client or server 5115 p = strrchr(buf, '/'); 5116 if (p == NULL) return false; 5117 else *p = '\0'; 5118 5119 // check xawt/libmawt.so 5120 strcpy(libmawtpath, buf); 5121 strcat(libmawtpath, xawtstr); 5122 if (::stat(libmawtpath, &statbuf) == 0) return false; 5123 5124 // check motif21/libmawt.so 5125 strcpy(libmawtpath, buf); 5126 strcat(libmawtpath, motifstr); 5127 if (::stat(libmawtpath, &statbuf) == 0) return false; 5128 5129 return true; 5130} 5131 5132