os_linux.cpp revision 20:e195fe4c40c7
1/* 2 * Copyright 1999-2007 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, 20 * CA 95054 USA or visit www.sun.com if you need additional information or 21 * have any questions. 22 * 23 */ 24 25// do not include precompiled header file 26# include "incls/_os_linux.cpp.incl" 27 28// put OS-includes here 29# include <sys/types.h> 30# include <sys/mman.h> 31# include <pthread.h> 32# include <signal.h> 33# include <errno.h> 34# include <dlfcn.h> 35# include <stdio.h> 36# include <unistd.h> 37# include <sys/resource.h> 38# include <pthread.h> 39# include <sys/stat.h> 40# include <sys/time.h> 41# include <sys/times.h> 42# include <sys/utsname.h> 43# include <sys/socket.h> 44# include <sys/wait.h> 45# include <pwd.h> 46# include <poll.h> 47# include <semaphore.h> 48# include <fcntl.h> 49# include <string.h> 50# include <syscall.h> 51# include <sys/sysinfo.h> 52# include <gnu/libc-version.h> 53# include <sys/ipc.h> 54# include <sys/shm.h> 55# include <link.h> 56 57#define MAX_PATH (2 * K) 58 59// for timer info max values which include all bits 60#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 61#define SEC_IN_NANOSECS 1000000000LL 62 63//////////////////////////////////////////////////////////////////////////////// 64// global variables 65julong os::Linux::_physical_memory = 0; 66 67address os::Linux::_initial_thread_stack_bottom = NULL; 68uintptr_t os::Linux::_initial_thread_stack_size = 0; 69 70int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 71int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 72Mutex* os::Linux::_createThread_lock = NULL; 73pthread_t os::Linux::_main_thread; 74int os::Linux::_page_size = -1; 75bool os::Linux::_is_floating_stack = false; 76bool os::Linux::_is_NPTL = false; 77bool os::Linux::_supports_fast_thread_cpu_time = false; 78char * os::Linux::_glibc_version = NULL; 79char * os::Linux::_libpthread_version = NULL; 80 81static jlong initial_time_count=0; 82 83static int clock_tics_per_sec = 100; 84 85// For diagnostics to print a message once. see run_periodic_checks 86static sigset_t check_signal_done; 87static bool check_signals = true;; 88 89static pid_t _initial_pid = 0; 90 91/* Signal number used to suspend/resume a thread */ 92 93/* do not use any signal number less than SIGSEGV, see 4355769 */ 94static int SR_signum = SIGUSR2; 95sigset_t SR_sigset; 96 97//////////////////////////////////////////////////////////////////////////////// 98// utility functions 99 100static int SR_initialize(); 101static int SR_finalize(); 102 103julong os::available_memory() { 104 return Linux::available_memory(); 105} 106 107julong os::Linux::available_memory() { 108 // values in struct sysinfo are "unsigned long" 109 struct sysinfo si; 110 sysinfo(&si); 111 112 return (julong)si.freeram * si.mem_unit; 113} 114 115julong os::physical_memory() { 116 return Linux::physical_memory(); 117} 118 119julong os::allocatable_physical_memory(julong size) { 120#ifdef _LP64 121 return size; 122#else 123 julong result = MIN2(size, (julong)3800*M); 124 if (!is_allocatable(result)) { 125 // See comments under solaris for alignment considerations 126 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); 127 result = MIN2(size, reasonable_size); 128 } 129 return result; 130#endif // _LP64 131} 132 133//////////////////////////////////////////////////////////////////////////////// 134// environment support 135 136bool os::getenv(const char* name, char* buf, int len) { 137 const char* val = ::getenv(name); 138 if (val != NULL && strlen(val) < (size_t)len) { 139 strcpy(buf, val); 140 return true; 141 } 142 if (len > 0) buf[0] = 0; // return a null string 143 return false; 144} 145 146 147// Return true if user is running as root. 148 149bool os::have_special_privileges() { 150 static bool init = false; 151 static bool privileges = false; 152 if (!init) { 153 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 154 init = true; 155 } 156 return privileges; 157} 158 159 160#ifndef SYS_gettid 161// i386: 224, ia64: 1105, amd64: 186, sparc 143 162#ifdef __ia64__ 163#define SYS_gettid 1105 164#elif __i386__ 165#define SYS_gettid 224 166#elif __amd64__ 167#define SYS_gettid 186 168#elif __sparc__ 169#define SYS_gettid 143 170#else 171#error define gettid for the arch 172#endif 173#endif 174 175// Cpu architecture string 176#if defined(IA64) 177static char cpu_arch[] = "ia64"; 178#elif defined(IA32) 179static char cpu_arch[] = "i386"; 180#elif defined(AMD64) 181static char cpu_arch[] = "amd64"; 182#elif defined(SPARC) 183# ifdef _LP64 184static char cpu_arch[] = "sparcv9"; 185# else 186static char cpu_arch[] = "sparc"; 187# endif 188#else 189#error Add appropriate cpu_arch setting 190#endif 191 192 193// pid_t gettid() 194// 195// Returns the kernel thread id of the currently running thread. Kernel 196// thread id is used to access /proc. 197// 198// (Note that getpid() on LinuxThreads returns kernel thread id too; but 199// on NPTL, it returns the same pid for all threads, as required by POSIX.) 200// 201pid_t os::Linux::gettid() { 202 int rslt = syscall(SYS_gettid); 203 if (rslt == -1) { 204 // old kernel, no NPTL support 205 return getpid(); 206 } else { 207 return (pid_t)rslt; 208 } 209} 210 211// Most versions of linux have a bug where the number of processors are 212// determined by looking at the /proc file system. In a chroot environment, 213// the system call returns 1. This causes the VM to act as if it is 214// a single processor and elide locking (see is_MP() call). 215static bool unsafe_chroot_detected = false; 216static char *unstable_chroot_error = "/proc file system not found.\n" 217 "Java may be unstable running multithreaded in a chroot " 218 "environment on Linux when /proc filesystem is not mounted."; 219 220void os::Linux::initialize_system_info() { 221 _processor_count = sysconf(_SC_NPROCESSORS_CONF); 222 if (_processor_count == 1) { 223 pid_t pid = os::Linux::gettid(); 224 char fname[32]; 225 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 226 FILE *fp = fopen(fname, "r"); 227 if (fp == NULL) { 228 unsafe_chroot_detected = true; 229 } else { 230 fclose(fp); 231 } 232 } 233 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 234 assert(_processor_count > 0, "linux error"); 235} 236 237void os::init_system_properties_values() { 238// char arch[12]; 239// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 240 241 // The next steps are taken in the product version: 242 // 243 // Obtain the JAVA_HOME value from the location of libjvm[_g].so. 244 // This library should be located at: 245 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so. 246 // 247 // If "/jre/lib/" appears at the right place in the path, then we 248 // assume libjvm[_g].so is installed in a JDK and we use this path. 249 // 250 // Otherwise exit with message: "Could not create the Java virtual machine." 251 // 252 // The following extra steps are taken in the debugging version: 253 // 254 // If "/jre/lib/" does NOT appear at the right place in the path 255 // instead of exit check for $JAVA_HOME environment variable. 256 // 257 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 258 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so 259 // it looks like libjvm[_g].so is installed there 260 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so. 261 // 262 // Otherwise exit. 263 // 264 // Important note: if the location of libjvm.so changes this 265 // code needs to be changed accordingly. 266 267 // The next few definitions allow the code to be verbatim: 268#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n)) 269#define getenv(n) ::getenv(n) 270 271/* 272 * See ld(1): 273 * The linker uses the following search paths to locate required 274 * shared libraries: 275 * 1: ... 276 * ... 277 * 7: The default directories, normally /lib and /usr/lib. 278 */ 279#define DEFAULT_LIBPATH "/lib:/usr/lib" 280 281#define EXTENSIONS_DIR "/lib/ext" 282#define ENDORSED_DIR "/lib/endorsed" 283#define REG_DIR "/usr/java/packages" 284 285 { 286 /* sysclasspath, java_home, dll_dir */ 287 { 288 char *home_path; 289 char *dll_path; 290 char *pslash; 291 char buf[MAXPATHLEN]; 292 os::jvm_path(buf, sizeof(buf)); 293 294 // Found the full path to libjvm.so. 295 // Now cut the path to <java_home>/jre if we can. 296 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 297 pslash = strrchr(buf, '/'); 298 if (pslash != NULL) 299 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 300 dll_path = malloc(strlen(buf) + 1); 301 if (dll_path == NULL) 302 return; 303 strcpy(dll_path, buf); 304 Arguments::set_dll_dir(dll_path); 305 306 if (pslash != NULL) { 307 pslash = strrchr(buf, '/'); 308 if (pslash != NULL) { 309 *pslash = '\0'; /* get rid of /<arch> */ 310 pslash = strrchr(buf, '/'); 311 if (pslash != NULL) 312 *pslash = '\0'; /* get rid of /lib */ 313 } 314 } 315 316 home_path = malloc(strlen(buf) + 1); 317 if (home_path == NULL) 318 return; 319 strcpy(home_path, buf); 320 Arguments::set_java_home(home_path); 321 322 if (!set_boot_path('/', ':')) 323 return; 324 } 325 326 /* 327 * Where to look for native libraries 328 * 329 * Note: Due to a legacy implementation, most of the library path 330 * is set in the launcher. This was to accomodate linking restrictions 331 * on legacy Linux implementations (which are no longer supported). 332 * Eventually, all the library path setting will be done here. 333 * 334 * However, to prevent the proliferation of improperly built native 335 * libraries, the new path component /usr/java/packages is added here. 336 * Eventually, all the library path setting will be done here. 337 */ 338 { 339 char *ld_library_path; 340 341 /* 342 * Construct the invariant part of ld_library_path. Note that the 343 * space for the colon and the trailing null are provided by the 344 * nulls included by the sizeof operator (so actually we allocate 345 * a byte more than necessary). 346 */ 347 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 348 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 349 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 350 351 /* 352 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 353 * should always exist (until the legacy problem cited above is 354 * addressed). 355 */ 356 char *v = getenv("LD_LIBRARY_PATH"); 357 if (v != NULL) { 358 char *t = ld_library_path; 359 /* That's +1 for the colon and +1 for the trailing '\0' */ 360 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 361 sprintf(ld_library_path, "%s:%s", v, t); 362 } 363 Arguments::set_library_path(ld_library_path); 364 } 365 366 /* 367 * Extensions directories. 368 * 369 * Note that the space for the colon and the trailing null are provided 370 * by the nulls included by the sizeof operator (so actually one byte more 371 * than necessary is allocated). 372 */ 373 { 374 char *buf = malloc(strlen(Arguments::get_java_home()) + 375 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 376 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 377 Arguments::get_java_home()); 378 Arguments::set_ext_dirs(buf); 379 } 380 381 /* Endorsed standards default directory. */ 382 { 383 char * buf; 384 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 385 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 386 Arguments::set_endorsed_dirs(buf); 387 } 388 } 389 390#undef malloc 391#undef getenv 392#undef EXTENSIONS_DIR 393#undef ENDORSED_DIR 394 395 // Done 396 return; 397} 398 399//////////////////////////////////////////////////////////////////////////////// 400// breakpoint support 401 402void os::breakpoint() { 403 BREAKPOINT; 404} 405 406extern "C" void breakpoint() { 407 // use debugger to set breakpoint here 408} 409 410//////////////////////////////////////////////////////////////////////////////// 411// signal support 412 413debug_only(static bool signal_sets_initialized = false); 414static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 415 416bool os::Linux::is_sig_ignored(int sig) { 417 struct sigaction oact; 418 sigaction(sig, (struct sigaction*)NULL, &oact); 419 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 420 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 421 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 422 return true; 423 else 424 return false; 425} 426 427void os::Linux::signal_sets_init() { 428 // Should also have an assertion stating we are still single-threaded. 429 assert(!signal_sets_initialized, "Already initialized"); 430 // Fill in signals that are necessarily unblocked for all threads in 431 // the VM. Currently, we unblock the following signals: 432 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 433 // by -Xrs (=ReduceSignalUsage)); 434 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 435 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 436 // the dispositions or masks wrt these signals. 437 // Programs embedding the VM that want to use the above signals for their 438 // own purposes must, at this time, use the "-Xrs" option to prevent 439 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 440 // (See bug 4345157, and other related bugs). 441 // In reality, though, unblocking these signals is really a nop, since 442 // these signals are not blocked by default. 443 sigemptyset(&unblocked_sigs); 444 sigemptyset(&allowdebug_blocked_sigs); 445 sigaddset(&unblocked_sigs, SIGILL); 446 sigaddset(&unblocked_sigs, SIGSEGV); 447 sigaddset(&unblocked_sigs, SIGBUS); 448 sigaddset(&unblocked_sigs, SIGFPE); 449 sigaddset(&unblocked_sigs, SR_signum); 450 451 if (!ReduceSignalUsage) { 452 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 453 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 454 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 455 } 456 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 457 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 458 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 459 } 460 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 461 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 462 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 463 } 464 } 465 // Fill in signals that are blocked by all but the VM thread. 466 sigemptyset(&vm_sigs); 467 if (!ReduceSignalUsage) 468 sigaddset(&vm_sigs, BREAK_SIGNAL); 469 debug_only(signal_sets_initialized = true); 470 471} 472 473// These are signals that are unblocked while a thread is running Java. 474// (For some reason, they get blocked by default.) 475sigset_t* os::Linux::unblocked_signals() { 476 assert(signal_sets_initialized, "Not initialized"); 477 return &unblocked_sigs; 478} 479 480// These are the signals that are blocked while a (non-VM) thread is 481// running Java. Only the VM thread handles these signals. 482sigset_t* os::Linux::vm_signals() { 483 assert(signal_sets_initialized, "Not initialized"); 484 return &vm_sigs; 485} 486 487// These are signals that are blocked during cond_wait to allow debugger in 488sigset_t* os::Linux::allowdebug_blocked_signals() { 489 assert(signal_sets_initialized, "Not initialized"); 490 return &allowdebug_blocked_sigs; 491} 492 493void os::Linux::hotspot_sigmask(Thread* thread) { 494 495 //Save caller's signal mask before setting VM signal mask 496 sigset_t caller_sigmask; 497 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 498 499 OSThread* osthread = thread->osthread(); 500 osthread->set_caller_sigmask(caller_sigmask); 501 502 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 503 504 if (!ReduceSignalUsage) { 505 if (thread->is_VM_thread()) { 506 // Only the VM thread handles BREAK_SIGNAL ... 507 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 508 } else { 509 // ... all other threads block BREAK_SIGNAL 510 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 511 } 512 } 513} 514 515////////////////////////////////////////////////////////////////////////////// 516// detecting pthread library 517 518void os::Linux::libpthread_init() { 519 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 520 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 521 // generic name for earlier versions. 522 // Define macros here so we can build HotSpot on old systems. 523# ifndef _CS_GNU_LIBC_VERSION 524# define _CS_GNU_LIBC_VERSION 2 525# endif 526# ifndef _CS_GNU_LIBPTHREAD_VERSION 527# define _CS_GNU_LIBPTHREAD_VERSION 3 528# endif 529 530 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 531 if (n > 0) { 532 char *str = (char *)malloc(n); 533 confstr(_CS_GNU_LIBC_VERSION, str, n); 534 os::Linux::set_glibc_version(str); 535 } else { 536 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 537 static char _gnu_libc_version[32]; 538 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 539 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 540 os::Linux::set_glibc_version(_gnu_libc_version); 541 } 542 543 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 544 if (n > 0) { 545 char *str = (char *)malloc(n); 546 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 547 548 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 549 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 550 // is the case: 551 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 552 strstr(str, "NPTL")) { 553 // LinuxThreads has a hard limit on max number of threads. So 554 // sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 555 // On the other hand, NPTL does not have such a limit, sysconf() 556 // will return -1 and errno is not changed. Check if it is really 557 // NPTL: 558 if (sysconf(_SC_THREAD_THREADS_MAX) > 0) { 559 free(str); 560 str = "linuxthreads"; 561 } 562 } 563 os::Linux::set_libpthread_version(str); 564 } else { 565 // glibc before 2.3.2 only has LinuxThreads. 566 os::Linux::set_libpthread_version("linuxthreads"); 567 } 568 569 if (strstr(libpthread_version(), "NPTL")) { 570 os::Linux::set_is_NPTL(); 571 } else { 572 os::Linux::set_is_LinuxThreads(); 573 } 574 575 // LinuxThreads have two flavors: floating-stack mode, which allows variable 576 // stack size; and fixed-stack mode. NPTL is always floating-stack. 577 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 578 os::Linux::set_is_floating_stack(); 579 } 580} 581 582///////////////////////////////////////////////////////////////////////////// 583// thread stack 584 585// Force Linux kernel to expand current thread stack. If "bottom" is close 586// to the stack guard, caller should block all signals. 587// 588// MAP_GROWSDOWN: 589// A special mmap() flag that is used to implement thread stacks. It tells 590// kernel that the memory region should extend downwards when needed. This 591// allows early versions of LinuxThreads to only mmap the first few pages 592// when creating a new thread. Linux kernel will automatically expand thread 593// stack as needed (on page faults). 594// 595// However, because the memory region of a MAP_GROWSDOWN stack can grow on 596// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 597// region, it's hard to tell if the fault is due to a legitimate stack 598// access or because of reading/writing non-exist memory (e.g. buffer 599// overrun). As a rule, if the fault happens below current stack pointer, 600// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 601// application (see Linux kernel fault.c). 602// 603// This Linux feature can cause SIGSEGV when VM bangs thread stack for 604// stack overflow detection. 605// 606// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 607// not use this flag. However, the stack of initial thread is not created 608// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 609// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 610// and then attach the thread to JVM. 611// 612// To get around the problem and allow stack banging on Linux, we need to 613// manually expand thread stack after receiving the SIGSEGV. 614// 615// There are two ways to expand thread stack to address "bottom", we used 616// both of them in JVM before 1.5: 617// 1. adjust stack pointer first so that it is below "bottom", and then 618// touch "bottom" 619// 2. mmap() the page in question 620// 621// Now alternate signal stack is gone, it's harder to use 2. For instance, 622// if current sp is already near the lower end of page 101, and we need to 623// call mmap() to map page 100, it is possible that part of the mmap() frame 624// will be placed in page 100. When page 100 is mapped, it is zero-filled. 625// That will destroy the mmap() frame and cause VM to crash. 626// 627// The following code works by adjusting sp first, then accessing the "bottom" 628// page to force a page fault. Linux kernel will then automatically expand the 629// stack mapping. 630// 631// _expand_stack_to() assumes its frame size is less than page size, which 632// should always be true if the function is not inlined. 633 634#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 635#define NOINLINE 636#else 637#define NOINLINE __attribute__ ((noinline)) 638#endif 639 640static void _expand_stack_to(address bottom) NOINLINE; 641 642static void _expand_stack_to(address bottom) { 643 address sp; 644 size_t size; 645 volatile char *p; 646 647 // Adjust bottom to point to the largest address within the same page, it 648 // gives us a one-page buffer if alloca() allocates slightly more memory. 649 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 650 bottom += os::Linux::page_size() - 1; 651 652 // sp might be slightly above current stack pointer; if that's the case, we 653 // will alloca() a little more space than necessary, which is OK. Don't use 654 // os::current_stack_pointer(), as its result can be slightly below current 655 // stack pointer, causing us to not alloca enough to reach "bottom". 656 sp = (address)&sp; 657 658 if (sp > bottom) { 659 size = sp - bottom; 660 p = (volatile char *)alloca(size); 661 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 662 p[0] = '\0'; 663 } 664} 665 666bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 667 assert(t!=NULL, "just checking"); 668 assert(t->osthread()->expanding_stack(), "expand should be set"); 669 assert(t->stack_base() != NULL, "stack_base was not initialized"); 670 671 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 672 sigset_t mask_all, old_sigset; 673 sigfillset(&mask_all); 674 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 675 _expand_stack_to(addr); 676 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 677 return true; 678 } 679 return false; 680} 681 682////////////////////////////////////////////////////////////////////////////// 683// create new thread 684 685static address highest_vm_reserved_address(); 686 687// check if it's safe to start a new thread 688static bool _thread_safety_check(Thread* thread) { 689 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 690 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 691 // Heap is mmap'ed at lower end of memory space. Thread stacks are 692 // allocated (MAP_FIXED) from high address space. Every thread stack 693 // occupies a fixed size slot (usually 2Mbytes, but user can change 694 // it to other values if they rebuild LinuxThreads). 695 // 696 // Problem with MAP_FIXED is that mmap() can still succeed even part of 697 // the memory region has already been mmap'ed. That means if we have too 698 // many threads and/or very large heap, eventually thread stack will 699 // collide with heap. 700 // 701 // Here we try to prevent heap/stack collision by comparing current 702 // stack bottom with the highest address that has been mmap'ed by JVM 703 // plus a safety margin for memory maps created by native code. 704 // 705 // This feature can be disabled by setting ThreadSafetyMargin to 0 706 // 707 if (ThreadSafetyMargin > 0) { 708 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 709 710 // not safe if our stack extends below the safety margin 711 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 712 } else { 713 return true; 714 } 715 } else { 716 // Floating stack LinuxThreads or NPTL: 717 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 718 // there's not enough space left, pthread_create() will fail. If we come 719 // here, that means enough space has been reserved for stack. 720 return true; 721 } 722} 723 724// Thread start routine for all newly created threads 725static void *java_start(Thread *thread) { 726 // Try to randomize the cache line index of hot stack frames. 727 // This helps when threads of the same stack traces evict each other's 728 // cache lines. The threads can be either from the same JVM instance, or 729 // from different JVM instances. The benefit is especially true for 730 // processors with hyperthreading technology. 731 static int counter = 0; 732 int pid = os::current_process_id(); 733 alloca(((pid ^ counter++) & 7) * 128); 734 735 ThreadLocalStorage::set_thread(thread); 736 737 OSThread* osthread = thread->osthread(); 738 Monitor* sync = osthread->startThread_lock(); 739 740 // non floating stack LinuxThreads needs extra check, see above 741 if (!_thread_safety_check(thread)) { 742 // notify parent thread 743 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 744 osthread->set_state(ZOMBIE); 745 sync->notify_all(); 746 return NULL; 747 } 748 749 // thread_id is kernel thread id (similar to Solaris LWP id) 750 osthread->set_thread_id(os::Linux::gettid()); 751 752 if (UseNUMA) { 753 int lgrp_id = os::numa_get_group_id(); 754 if (lgrp_id != -1) { 755 thread->set_lgrp_id(lgrp_id); 756 } 757 } 758 // initialize signal mask for this thread 759 os::Linux::hotspot_sigmask(thread); 760 761 // initialize floating point control register 762 os::Linux::init_thread_fpu_state(); 763 764 // handshaking with parent thread 765 { 766 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 767 768 // notify parent thread 769 osthread->set_state(INITIALIZED); 770 sync->notify_all(); 771 772 // wait until os::start_thread() 773 while (osthread->get_state() == INITIALIZED) { 774 sync->wait(Mutex::_no_safepoint_check_flag); 775 } 776 } 777 778 // call one more level start routine 779 thread->run(); 780 781 return 0; 782} 783 784bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 785 assert(thread->osthread() == NULL, "caller responsible"); 786 787 // Allocate the OSThread object 788 OSThread* osthread = new OSThread(NULL, NULL); 789 if (osthread == NULL) { 790 return false; 791 } 792 793 // set the correct thread state 794 osthread->set_thread_type(thr_type); 795 796 // Initial state is ALLOCATED but not INITIALIZED 797 osthread->set_state(ALLOCATED); 798 799 thread->set_osthread(osthread); 800 801 // init thread attributes 802 pthread_attr_t attr; 803 pthread_attr_init(&attr); 804 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 805 806 // stack size 807 if (os::Linux::supports_variable_stack_size()) { 808 // calculate stack size if it's not specified by caller 809 if (stack_size == 0) { 810 stack_size = os::Linux::default_stack_size(thr_type); 811 812 switch (thr_type) { 813 case os::java_thread: 814 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss 815 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create(); 816 break; 817 case os::compiler_thread: 818 if (CompilerThreadStackSize > 0) { 819 stack_size = (size_t)(CompilerThreadStackSize * K); 820 break; 821 } // else fall through: 822 // use VMThreadStackSize if CompilerThreadStackSize is not defined 823 case os::vm_thread: 824 case os::pgc_thread: 825 case os::cgc_thread: 826 case os::watcher_thread: 827 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 828 break; 829 } 830 } 831 832 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 833 pthread_attr_setstacksize(&attr, stack_size); 834 } else { 835 // let pthread_create() pick the default value. 836 } 837 838 // glibc guard page 839 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 840 841 ThreadState state; 842 843 { 844 // Serialize thread creation if we are running with fixed stack LinuxThreads 845 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 846 if (lock) { 847 os::Linux::createThread_lock()->lock_without_safepoint_check(); 848 } 849 850 pthread_t tid; 851 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 852 853 pthread_attr_destroy(&attr); 854 855 if (ret != 0) { 856 if (PrintMiscellaneous && (Verbose || WizardMode)) { 857 perror("pthread_create()"); 858 } 859 // Need to clean up stuff we've allocated so far 860 thread->set_osthread(NULL); 861 delete osthread; 862 if (lock) os::Linux::createThread_lock()->unlock(); 863 return false; 864 } 865 866 // Store pthread info into the OSThread 867 osthread->set_pthread_id(tid); 868 869 // Wait until child thread is either initialized or aborted 870 { 871 Monitor* sync_with_child = osthread->startThread_lock(); 872 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 873 while ((state = osthread->get_state()) == ALLOCATED) { 874 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 875 } 876 } 877 878 if (lock) { 879 os::Linux::createThread_lock()->unlock(); 880 } 881 } 882 883 // Aborted due to thread limit being reached 884 if (state == ZOMBIE) { 885 thread->set_osthread(NULL); 886 delete osthread; 887 return false; 888 } 889 890 // The thread is returned suspended (in state INITIALIZED), 891 // and is started higher up in the call chain 892 assert(state == INITIALIZED, "race condition"); 893 return true; 894} 895 896///////////////////////////////////////////////////////////////////////////// 897// attach existing thread 898 899// bootstrap the main thread 900bool os::create_main_thread(JavaThread* thread) { 901 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 902 return create_attached_thread(thread); 903} 904 905bool os::create_attached_thread(JavaThread* thread) { 906#ifdef ASSERT 907 thread->verify_not_published(); 908#endif 909 910 // Allocate the OSThread object 911 OSThread* osthread = new OSThread(NULL, NULL); 912 913 if (osthread == NULL) { 914 return false; 915 } 916 917 // Store pthread info into the OSThread 918 osthread->set_thread_id(os::Linux::gettid()); 919 osthread->set_pthread_id(::pthread_self()); 920 921 // initialize floating point control register 922 os::Linux::init_thread_fpu_state(); 923 924 // Initial thread state is RUNNABLE 925 osthread->set_state(RUNNABLE); 926 927 thread->set_osthread(osthread); 928 929 if (UseNUMA) { 930 int lgrp_id = os::numa_get_group_id(); 931 if (lgrp_id != -1) { 932 thread->set_lgrp_id(lgrp_id); 933 } 934 } 935 936 if (os::Linux::is_initial_thread()) { 937 // If current thread is initial thread, its stack is mapped on demand, 938 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 939 // the entire stack region to avoid SEGV in stack banging. 940 // It is also useful to get around the heap-stack-gap problem on SuSE 941 // kernel (see 4821821 for details). We first expand stack to the top 942 // of yellow zone, then enable stack yellow zone (order is significant, 943 // enabling yellow zone first will crash JVM on SuSE Linux), so there 944 // is no gap between the last two virtual memory regions. 945 946 JavaThread *jt = (JavaThread *)thread; 947 address addr = jt->stack_yellow_zone_base(); 948 assert(addr != NULL, "initialization problem?"); 949 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 950 951 osthread->set_expanding_stack(); 952 os::Linux::manually_expand_stack(jt, addr); 953 osthread->clear_expanding_stack(); 954 } 955 956 // initialize signal mask for this thread 957 // and save the caller's signal mask 958 os::Linux::hotspot_sigmask(thread); 959 960 return true; 961} 962 963void os::pd_start_thread(Thread* thread) { 964 OSThread * osthread = thread->osthread(); 965 assert(osthread->get_state() != INITIALIZED, "just checking"); 966 Monitor* sync_with_child = osthread->startThread_lock(); 967 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 968 sync_with_child->notify(); 969} 970 971// Free Linux resources related to the OSThread 972void os::free_thread(OSThread* osthread) { 973 assert(osthread != NULL, "osthread not set"); 974 975 if (Thread::current()->osthread() == osthread) { 976 // Restore caller's signal mask 977 sigset_t sigmask = osthread->caller_sigmask(); 978 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 979 } 980 981 delete osthread; 982} 983 984////////////////////////////////////////////////////////////////////////////// 985// thread local storage 986 987int os::allocate_thread_local_storage() { 988 pthread_key_t key; 989 int rslt = pthread_key_create(&key, NULL); 990 assert(rslt == 0, "cannot allocate thread local storage"); 991 return (int)key; 992} 993 994// Note: This is currently not used by VM, as we don't destroy TLS key 995// on VM exit. 996void os::free_thread_local_storage(int index) { 997 int rslt = pthread_key_delete((pthread_key_t)index); 998 assert(rslt == 0, "invalid index"); 999} 1000 1001void os::thread_local_storage_at_put(int index, void* value) { 1002 int rslt = pthread_setspecific((pthread_key_t)index, value); 1003 assert(rslt == 0, "pthread_setspecific failed"); 1004} 1005 1006extern "C" Thread* get_thread() { 1007 return ThreadLocalStorage::thread(); 1008} 1009 1010////////////////////////////////////////////////////////////////////////////// 1011// initial thread 1012 1013// Check if current thread is the initial thread, similar to Solaris thr_main. 1014bool os::Linux::is_initial_thread(void) { 1015 char dummy; 1016 // If called before init complete, thread stack bottom will be null. 1017 // Can be called if fatal error occurs before initialization. 1018 if (initial_thread_stack_bottom() == NULL) return false; 1019 assert(initial_thread_stack_bottom() != NULL && 1020 initial_thread_stack_size() != 0, 1021 "os::init did not locate initial thread's stack region"); 1022 if ((address)&dummy >= initial_thread_stack_bottom() && 1023 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1024 return true; 1025 else return false; 1026} 1027 1028// Find the virtual memory area that contains addr 1029static bool find_vma(address addr, address* vma_low, address* vma_high) { 1030 FILE *fp = fopen("/proc/self/maps", "r"); 1031 if (fp) { 1032 address low, high; 1033 while (!feof(fp)) { 1034 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1035 if (low <= addr && addr < high) { 1036 if (vma_low) *vma_low = low; 1037 if (vma_high) *vma_high = high; 1038 fclose (fp); 1039 return true; 1040 } 1041 } 1042 for (;;) { 1043 int ch = fgetc(fp); 1044 if (ch == EOF || ch == (int)'\n') break; 1045 } 1046 } 1047 fclose(fp); 1048 } 1049 return false; 1050} 1051 1052// Locate initial thread stack. This special handling of initial thread stack 1053// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1054// bogus value for initial thread. 1055void os::Linux::capture_initial_stack(size_t max_size) { 1056 // stack size is the easy part, get it from RLIMIT_STACK 1057 size_t stack_size; 1058 struct rlimit rlim; 1059 getrlimit(RLIMIT_STACK, &rlim); 1060 stack_size = rlim.rlim_cur; 1061 1062 // 6308388: a bug in ld.so will relocate its own .data section to the 1063 // lower end of primordial stack; reduce ulimit -s value a little bit 1064 // so we won't install guard page on ld.so's data section. 1065 stack_size -= 2 * page_size(); 1066 1067 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1068 // 7.1, in both cases we will get 2G in return value. 1069 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1070 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1071 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1072 // in case other parts in glibc still assumes 2M max stack size. 1073 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1074#ifndef IA64 1075 if (stack_size > 2 * K * K) stack_size = 2 * K * K; 1076#else 1077 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1078 if (stack_size > 4 * K * K) stack_size = 4 * K * K; 1079#endif 1080 1081 // Try to figure out where the stack base (top) is. This is harder. 1082 // 1083 // When an application is started, glibc saves the initial stack pointer in 1084 // a global variable "__libc_stack_end", which is then used by system 1085 // libraries. __libc_stack_end should be pretty close to stack top. The 1086 // variable is available since the very early days. However, because it is 1087 // a private interface, it could disappear in the future. 1088 // 1089 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1090 // to __libc_stack_end, it is very close to stack top, but isn't the real 1091 // stack top. Note that /proc may not exist if VM is running as a chroot 1092 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1093 // /proc/<pid>/stat could change in the future (though unlikely). 1094 // 1095 // We try __libc_stack_end first. If that doesn't work, look for 1096 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1097 // as a hint, which should work well in most cases. 1098 1099 uintptr_t stack_start; 1100 1101 // try __libc_stack_end first 1102 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1103 if (p && *p) { 1104 stack_start = *p; 1105 } else { 1106 // see if we can get the start_stack field from /proc/self/stat 1107 FILE *fp; 1108 int pid; 1109 char state; 1110 int ppid; 1111 int pgrp; 1112 int session; 1113 int nr; 1114 int tpgrp; 1115 unsigned long flags; 1116 unsigned long minflt; 1117 unsigned long cminflt; 1118 unsigned long majflt; 1119 unsigned long cmajflt; 1120 unsigned long utime; 1121 unsigned long stime; 1122 long cutime; 1123 long cstime; 1124 long prio; 1125 long nice; 1126 long junk; 1127 long it_real; 1128 uintptr_t start; 1129 uintptr_t vsize; 1130 uintptr_t rss; 1131 unsigned long rsslim; 1132 uintptr_t scodes; 1133 uintptr_t ecode; 1134 int i; 1135 1136 // Figure what the primordial thread stack base is. Code is inspired 1137 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1138 // followed by command name surrounded by parentheses, state, etc. 1139 char stat[2048]; 1140 int statlen; 1141 1142 fp = fopen("/proc/self/stat", "r"); 1143 if (fp) { 1144 statlen = fread(stat, 1, 2047, fp); 1145 stat[statlen] = '\0'; 1146 fclose(fp); 1147 1148 // Skip pid and the command string. Note that we could be dealing with 1149 // weird command names, e.g. user could decide to rename java launcher 1150 // to "java 1.4.2 :)", then the stat file would look like 1151 // 1234 (java 1.4.2 :)) R ... ... 1152 // We don't really need to know the command string, just find the last 1153 // occurrence of ")" and then start parsing from there. See bug 4726580. 1154 char * s = strrchr(stat, ')'); 1155 1156 i = 0; 1157 if (s) { 1158 // Skip blank chars 1159 do s++; while (isspace(*s)); 1160 1161 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1162 /* 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 */ 1163 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld %lu %lu %ld %lu %lu %lu %lu", 1164 &state, /* 3 %c */ 1165 &ppid, /* 4 %d */ 1166 &pgrp, /* 5 %d */ 1167 &session, /* 6 %d */ 1168 &nr, /* 7 %d */ 1169 &tpgrp, /* 8 %d */ 1170 &flags, /* 9 %lu */ 1171 &minflt, /* 10 %lu */ 1172 &cminflt, /* 11 %lu */ 1173 &majflt, /* 12 %lu */ 1174 &cmajflt, /* 13 %lu */ 1175 &utime, /* 14 %lu */ 1176 &stime, /* 15 %lu */ 1177 &cutime, /* 16 %ld */ 1178 &cstime, /* 17 %ld */ 1179 &prio, /* 18 %ld */ 1180 &nice, /* 19 %ld */ 1181 &junk, /* 20 %ld */ 1182 &it_real, /* 21 %ld */ 1183 &start, /* 22 %lu */ 1184 &vsize, /* 23 %lu */ 1185 &rss, /* 24 %ld */ 1186 &rsslim, /* 25 %lu */ 1187 &scodes, /* 26 %lu */ 1188 &ecode, /* 27 %lu */ 1189 &stack_start); /* 28 %lu */ 1190 } 1191 1192 if (i != 28 - 2) { 1193 assert(false, "Bad conversion from /proc/self/stat"); 1194 // product mode - assume we are the initial thread, good luck in the 1195 // embedded case. 1196 warning("Can't detect initial thread stack location - bad conversion"); 1197 stack_start = (uintptr_t) &rlim; 1198 } 1199 } else { 1200 // For some reason we can't open /proc/self/stat (for example, running on 1201 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1202 // most cases, so don't abort: 1203 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1204 stack_start = (uintptr_t) &rlim; 1205 } 1206 } 1207 1208 // Now we have a pointer (stack_start) very close to the stack top, the 1209 // next thing to do is to figure out the exact location of stack top. We 1210 // can find out the virtual memory area that contains stack_start by 1211 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1212 // and its upper limit is the real stack top. (again, this would fail if 1213 // running inside chroot, because /proc may not exist.) 1214 1215 uintptr_t stack_top; 1216 address low, high; 1217 if (find_vma((address)stack_start, &low, &high)) { 1218 // success, "high" is the true stack top. (ignore "low", because initial 1219 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1220 stack_top = (uintptr_t)high; 1221 } else { 1222 // failed, likely because /proc/self/maps does not exist 1223 warning("Can't detect initial thread stack location - find_vma failed"); 1224 // best effort: stack_start is normally within a few pages below the real 1225 // stack top, use it as stack top, and reduce stack size so we won't put 1226 // guard page outside stack. 1227 stack_top = stack_start; 1228 stack_size -= 16 * page_size(); 1229 } 1230 1231 // stack_top could be partially down the page so align it 1232 stack_top = align_size_up(stack_top, page_size()); 1233 1234 if (max_size && stack_size > max_size) { 1235 _initial_thread_stack_size = max_size; 1236 } else { 1237 _initial_thread_stack_size = stack_size; 1238 } 1239 1240 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1241 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1242} 1243 1244//////////////////////////////////////////////////////////////////////////////// 1245// time support 1246 1247// Time since start-up in seconds to a fine granularity. 1248// Used by VMSelfDestructTimer and the MemProfiler. 1249double os::elapsedTime() { 1250 1251 return (double)(os::elapsed_counter()) * 0.000001; 1252} 1253 1254jlong os::elapsed_counter() { 1255 timeval time; 1256 int status = gettimeofday(&time, NULL); 1257 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1258} 1259 1260jlong os::elapsed_frequency() { 1261 return (1000 * 1000); 1262} 1263 1264jlong os::timeofday() { 1265 timeval time; 1266 int status = gettimeofday(&time, NULL); 1267 assert(status != -1, "linux error"); 1268 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1269} 1270 1271// Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis 1272// _use_global_time is only set if CacheTimeMillis is true 1273jlong os::javaTimeMillis() { 1274 return (_use_global_time ? read_global_time() : timeofday()); 1275} 1276 1277#ifndef CLOCK_MONOTONIC 1278#define CLOCK_MONOTONIC (1) 1279#endif 1280 1281void os::Linux::clock_init() { 1282 // we do dlopen's in this particular order due to bug in linux 1283 // dynamical loader (see 6348968) leading to crash on exit 1284 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1285 if (handle == NULL) { 1286 handle = dlopen("librt.so", RTLD_LAZY); 1287 } 1288 1289 if (handle) { 1290 int (*clock_getres_func)(clockid_t, struct timespec*) = 1291 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1292 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1293 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1294 if (clock_getres_func && clock_gettime_func) { 1295 // See if monotonic clock is supported by the kernel. Note that some 1296 // early implementations simply return kernel jiffies (updated every 1297 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1298 // for nano time (though the monotonic property is still nice to have). 1299 // It's fixed in newer kernels, however clock_getres() still returns 1300 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1301 // resolution for now. Hopefully as people move to new kernels, this 1302 // won't be a problem. 1303 struct timespec res; 1304 struct timespec tp; 1305 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1306 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1307 // yes, monotonic clock is supported 1308 _clock_gettime = clock_gettime_func; 1309 } else { 1310 // close librt if there is no monotonic clock 1311 dlclose(handle); 1312 } 1313 } 1314 } 1315} 1316 1317#ifndef SYS_clock_getres 1318 1319#if defined(IA32) || defined(AMD64) 1320#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1321#else 1322#error Value of SYS_clock_getres not known on this platform 1323#endif 1324 1325#endif 1326 1327#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1328 1329void os::Linux::fast_thread_clock_init() { 1330 if (!UseLinuxPosixThreadCPUClocks) { 1331 return; 1332 } 1333 clockid_t clockid; 1334 struct timespec tp; 1335 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1336 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1337 1338 // Switch to using fast clocks for thread cpu time if 1339 // the sys_clock_getres() returns 0 error code. 1340 // Note, that some kernels may support the current thread 1341 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1342 // returned by the pthread_getcpuclockid(). 1343 // If the fast Posix clocks are supported then the sys_clock_getres() 1344 // must return at least tp.tv_sec == 0 which means a resolution 1345 // better than 1 sec. This is extra check for reliability. 1346 1347 if(pthread_getcpuclockid_func && 1348 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1349 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1350 1351 _supports_fast_thread_cpu_time = true; 1352 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1353 } 1354} 1355 1356jlong os::javaTimeNanos() { 1357 if (Linux::supports_monotonic_clock()) { 1358 struct timespec tp; 1359 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1360 assert(status == 0, "gettime error"); 1361 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1362 return result; 1363 } else { 1364 timeval time; 1365 int status = gettimeofday(&time, NULL); 1366 assert(status != -1, "linux error"); 1367 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1368 return 1000 * usecs; 1369 } 1370} 1371 1372void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1373 if (Linux::supports_monotonic_clock()) { 1374 info_ptr->max_value = ALL_64_BITS; 1375 1376 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1377 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1378 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1379 } else { 1380 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1381 info_ptr->max_value = ALL_64_BITS; 1382 1383 // gettimeofday is a real time clock so it skips 1384 info_ptr->may_skip_backward = true; 1385 info_ptr->may_skip_forward = true; 1386 } 1387 1388 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1389} 1390 1391// Return the real, user, and system times in seconds from an 1392// arbitrary fixed point in the past. 1393bool os::getTimesSecs(double* process_real_time, 1394 double* process_user_time, 1395 double* process_system_time) { 1396 struct tms ticks; 1397 clock_t real_ticks = times(&ticks); 1398 1399 if (real_ticks == (clock_t) (-1)) { 1400 return false; 1401 } else { 1402 double ticks_per_second = (double) clock_tics_per_sec; 1403 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1404 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1405 *process_real_time = ((double) real_ticks) / ticks_per_second; 1406 1407 return true; 1408 } 1409} 1410 1411 1412char * os::local_time_string(char *buf, size_t buflen) { 1413 struct tm t; 1414 time_t long_time; 1415 time(&long_time); 1416 localtime_r(&long_time, &t); 1417 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1418 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1419 t.tm_hour, t.tm_min, t.tm_sec); 1420 return buf; 1421} 1422 1423//////////////////////////////////////////////////////////////////////////////// 1424// runtime exit support 1425 1426// Note: os::shutdown() might be called very early during initialization, or 1427// called from signal handler. Before adding something to os::shutdown(), make 1428// sure it is async-safe and can handle partially initialized VM. 1429void os::shutdown() { 1430 1431 // allow PerfMemory to attempt cleanup of any persistent resources 1432 perfMemory_exit(); 1433 1434 // needs to remove object in file system 1435 AttachListener::abort(); 1436 1437 // flush buffered output, finish log files 1438 ostream_abort(); 1439 1440 // Check for abort hook 1441 abort_hook_t abort_hook = Arguments::abort_hook(); 1442 if (abort_hook != NULL) { 1443 abort_hook(); 1444 } 1445 1446} 1447 1448// Note: os::abort() might be called very early during initialization, or 1449// called from signal handler. Before adding something to os::abort(), make 1450// sure it is async-safe and can handle partially initialized VM. 1451void os::abort(bool dump_core) { 1452 os::shutdown(); 1453 if (dump_core) { 1454#ifndef PRODUCT 1455 fdStream out(defaultStream::output_fd()); 1456 out.print_raw("Current thread is "); 1457 char buf[16]; 1458 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1459 out.print_raw_cr(buf); 1460 out.print_raw_cr("Dumping core ..."); 1461#endif 1462 ::abort(); // dump core 1463 } 1464 1465 ::exit(1); 1466} 1467 1468// Die immediately, no exit hook, no abort hook, no cleanup. 1469void os::die() { 1470 // _exit() on LinuxThreads only kills current thread 1471 ::abort(); 1472} 1473 1474// unused on linux for now. 1475void os::set_error_file(const char *logfile) {} 1476 1477intx os::current_thread_id() { return (intx)pthread_self(); } 1478int os::current_process_id() { 1479 1480 // Under the old linux thread library, linux gives each thread 1481 // its own process id. Because of this each thread will return 1482 // a different pid if this method were to return the result 1483 // of getpid(2). Linux provides no api that returns the pid 1484 // of the launcher thread for the vm. This implementation 1485 // returns a unique pid, the pid of the launcher thread 1486 // that starts the vm 'process'. 1487 1488 // Under the NPTL, getpid() returns the same pid as the 1489 // launcher thread rather than a unique pid per thread. 1490 // Use gettid() if you want the old pre NPTL behaviour. 1491 1492 // if you are looking for the result of a call to getpid() that 1493 // returns a unique pid for the calling thread, then look at the 1494 // OSThread::thread_id() method in osThread_linux.hpp file 1495 1496 return (int)(_initial_pid ? _initial_pid : getpid()); 1497} 1498 1499// DLL functions 1500 1501const char* os::dll_file_extension() { return ".so"; } 1502 1503const char* os::get_temp_directory() { return "/tmp/"; } 1504 1505const char* os::get_current_directory(char *buf, int buflen) { 1506 return getcwd(buf, buflen); 1507} 1508 1509// check if addr is inside libjvm[_g].so 1510bool os::address_is_in_vm(address addr) { 1511 static address libjvm_base_addr; 1512 Dl_info dlinfo; 1513 1514 if (libjvm_base_addr == NULL) { 1515 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); 1516 libjvm_base_addr = (address)dlinfo.dli_fbase; 1517 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1518 } 1519 1520 if (dladdr((void *)addr, &dlinfo)) { 1521 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1522 } 1523 1524 return false; 1525} 1526 1527bool os::dll_address_to_function_name(address addr, char *buf, 1528 int buflen, int *offset) { 1529 Dl_info dlinfo; 1530 1531 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { 1532 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1533 if (offset) *offset = addr - (address)dlinfo.dli_saddr; 1534 return true; 1535 } else { 1536 if (buf) buf[0] = '\0'; 1537 if (offset) *offset = -1; 1538 return false; 1539 } 1540} 1541 1542struct _address_to_library_name { 1543 address addr; // input : memory address 1544 size_t buflen; // size of fname 1545 char* fname; // output: library name 1546 address base; // library base addr 1547}; 1548 1549static int address_to_library_name_callback(struct dl_phdr_info *info, 1550 size_t size, void *data) { 1551 int i; 1552 bool found = false; 1553 address libbase = NULL; 1554 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1555 1556 // iterate through all loadable segments 1557 for (i = 0; i < info->dlpi_phnum; i++) { 1558 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1559 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1560 // base address of a library is the lowest address of its loaded 1561 // segments. 1562 if (libbase == NULL || libbase > segbase) { 1563 libbase = segbase; 1564 } 1565 // see if 'addr' is within current segment 1566 if (segbase <= d->addr && 1567 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1568 found = true; 1569 } 1570 } 1571 } 1572 1573 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1574 // so dll_address_to_library_name() can fall through to use dladdr() which 1575 // can figure out executable name from argv[0]. 1576 if (found && info->dlpi_name && info->dlpi_name[0]) { 1577 d->base = libbase; 1578 if (d->fname) { 1579 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1580 } 1581 return 1; 1582 } 1583 return 0; 1584} 1585 1586bool os::dll_address_to_library_name(address addr, char* buf, 1587 int buflen, int* offset) { 1588 Dl_info dlinfo; 1589 struct _address_to_library_name data; 1590 1591 // There is a bug in old glibc dladdr() implementation that it could resolve 1592 // to wrong library name if the .so file has a base address != NULL. Here 1593 // we iterate through the program headers of all loaded libraries to find 1594 // out which library 'addr' really belongs to. This workaround can be 1595 // removed once the minimum requirement for glibc is moved to 2.3.x. 1596 data.addr = addr; 1597 data.fname = buf; 1598 data.buflen = buflen; 1599 data.base = NULL; 1600 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1601 1602 if (rslt) { 1603 // buf already contains library name 1604 if (offset) *offset = addr - data.base; 1605 return true; 1606 } else if (dladdr((void*)addr, &dlinfo)){ 1607 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1608 if (offset) *offset = addr - (address)dlinfo.dli_fbase; 1609 return true; 1610 } else { 1611 if (buf) buf[0] = '\0'; 1612 if (offset) *offset = -1; 1613 return false; 1614 } 1615} 1616 1617 // Loads .dll/.so and 1618 // in case of error it checks if .dll/.so was built for the 1619 // same architecture as Hotspot is running on 1620 1621void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1622{ 1623 void * result= ::dlopen(filename, RTLD_LAZY); 1624 if (result != NULL) { 1625 // Successful loading 1626 return result; 1627 } 1628 1629 Elf32_Ehdr elf_head; 1630 1631 // Read system error message into ebuf 1632 // It may or may not be overwritten below 1633 ::strncpy(ebuf, ::dlerror(), ebuflen-1); 1634 ebuf[ebuflen-1]='\0'; 1635 int diag_msg_max_length=ebuflen-strlen(ebuf); 1636 char* diag_msg_buf=ebuf+strlen(ebuf); 1637 1638 if (diag_msg_max_length==0) { 1639 // No more space in ebuf for additional diagnostics message 1640 return NULL; 1641 } 1642 1643 1644 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1645 1646 if (file_descriptor < 0) { 1647 // Can't open library, report dlerror() message 1648 return NULL; 1649 } 1650 1651 bool failed_to_read_elf_head= 1652 (sizeof(elf_head)!= 1653 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1654 1655 ::close(file_descriptor); 1656 if (failed_to_read_elf_head) { 1657 // file i/o error - report dlerror() msg 1658 return NULL; 1659 } 1660 1661 typedef struct { 1662 Elf32_Half code; // Actual value as defined in elf.h 1663 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1664 char elf_class; // 32 or 64 bit 1665 char endianess; // MSB or LSB 1666 char* name; // String representation 1667 } arch_t; 1668 1669 #ifndef EM_486 1670 #define EM_486 6 /* Intel 80486 */ 1671 #endif 1672 1673 static const arch_t arch_array[]={ 1674 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1675 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1676 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1677 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1678 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1679 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1680 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1681 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1682 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"} 1683 }; 1684 1685 #if (defined IA32) 1686 static Elf32_Half running_arch_code=EM_386; 1687 #elif (defined AMD64) 1688 static Elf32_Half running_arch_code=EM_X86_64; 1689 #elif (defined IA64) 1690 static Elf32_Half running_arch_code=EM_IA_64; 1691 #elif (defined __sparc) && (defined _LP64) 1692 static Elf32_Half running_arch_code=EM_SPARCV9; 1693 #elif (defined __sparc) && (!defined _LP64) 1694 static Elf32_Half running_arch_code=EM_SPARC; 1695 #elif (defined __powerpc64__) 1696 static Elf32_Half running_arch_code=EM_PPC64; 1697 #elif (defined __powerpc__) 1698 static Elf32_Half running_arch_code=EM_PPC; 1699 #else 1700 #error Method os::dll_load requires that one of following is defined:\ 1701 IA32, AMD64, IA64, __sparc, __powerpc__ 1702 #endif 1703 1704 // Identify compatability class for VM's architecture and library's architecture 1705 // Obtain string descriptions for architectures 1706 1707 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1708 int running_arch_index=-1; 1709 1710 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1711 if (running_arch_code == arch_array[i].code) { 1712 running_arch_index = i; 1713 } 1714 if (lib_arch.code == arch_array[i].code) { 1715 lib_arch.compat_class = arch_array[i].compat_class; 1716 lib_arch.name = arch_array[i].name; 1717 } 1718 } 1719 1720 assert(running_arch_index != -1, 1721 "Didn't find running architecture code (running_arch_code) in arch_array"); 1722 if (running_arch_index == -1) { 1723 // Even though running architecture detection failed 1724 // we may still continue with reporting dlerror() message 1725 return NULL; 1726 } 1727 1728 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1729 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1730 return NULL; 1731 } 1732 1733 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1734 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1735 return NULL; 1736 } 1737 1738 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1739 if ( lib_arch.name!=NULL ) { 1740 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1741 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1742 lib_arch.name, arch_array[running_arch_index].name); 1743 } else { 1744 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1745 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1746 lib_arch.code, 1747 arch_array[running_arch_index].name); 1748 } 1749 } 1750 1751 return NULL; 1752} 1753 1754 1755 1756 1757bool _print_ascii_file(const char* filename, outputStream* st) { 1758 int fd = open(filename, O_RDONLY); 1759 if (fd == -1) { 1760 return false; 1761 } 1762 1763 char buf[32]; 1764 int bytes; 1765 while ((bytes = read(fd, buf, sizeof(buf))) > 0) { 1766 st->print_raw(buf, bytes); 1767 } 1768 1769 close(fd); 1770 1771 return true; 1772} 1773 1774void os::print_dll_info(outputStream *st) { 1775 st->print_cr("Dynamic libraries:"); 1776 1777 char fname[32]; 1778 pid_t pid = os::Linux::gettid(); 1779 1780 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1781 1782 if (!_print_ascii_file(fname, st)) { 1783 st->print("Can not get library information for pid = %d\n", pid); 1784 } 1785} 1786 1787 1788void os::print_os_info(outputStream* st) { 1789 st->print("OS:"); 1790 1791 // Try to identify popular distros. 1792 // Most Linux distributions have /etc/XXX-release file, which contains 1793 // the OS version string. Some have more than one /etc/XXX-release file 1794 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 1795 // so the order is important. 1796 if (!_print_ascii_file("/etc/mandrake-release", st) && 1797 !_print_ascii_file("/etc/sun-release", st) && 1798 !_print_ascii_file("/etc/redhat-release", st) && 1799 !_print_ascii_file("/etc/SuSE-release", st) && 1800 !_print_ascii_file("/etc/turbolinux-release", st) && 1801 !_print_ascii_file("/etc/gentoo-release", st) && 1802 !_print_ascii_file("/etc/debian_version", st)) { 1803 st->print("Linux"); 1804 } 1805 st->cr(); 1806 1807 // kernel 1808 st->print("uname:"); 1809 struct utsname name; 1810 uname(&name); 1811 st->print(name.sysname); st->print(" "); 1812 st->print(name.release); st->print(" "); 1813 st->print(name.version); st->print(" "); 1814 st->print(name.machine); 1815 st->cr(); 1816 1817 // Print warning if unsafe chroot environment detected 1818 if (unsafe_chroot_detected) { 1819 st->print("WARNING!! "); 1820 st->print_cr(unstable_chroot_error); 1821 } 1822 1823 // libc, pthread 1824 st->print("libc:"); 1825 st->print(os::Linux::glibc_version()); st->print(" "); 1826 st->print(os::Linux::libpthread_version()); st->print(" "); 1827 if (os::Linux::is_LinuxThreads()) { 1828 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 1829 } 1830 st->cr(); 1831 1832 // rlimit 1833 st->print("rlimit:"); 1834 struct rlimit rlim; 1835 1836 st->print(" STACK "); 1837 getrlimit(RLIMIT_STACK, &rlim); 1838 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1839 else st->print("%uk", rlim.rlim_cur >> 10); 1840 1841 st->print(", CORE "); 1842 getrlimit(RLIMIT_CORE, &rlim); 1843 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1844 else st->print("%uk", rlim.rlim_cur >> 10); 1845 1846 st->print(", NPROC "); 1847 getrlimit(RLIMIT_NPROC, &rlim); 1848 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1849 else st->print("%d", rlim.rlim_cur); 1850 1851 st->print(", NOFILE "); 1852 getrlimit(RLIMIT_NOFILE, &rlim); 1853 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1854 else st->print("%d", rlim.rlim_cur); 1855 1856 st->print(", AS "); 1857 getrlimit(RLIMIT_AS, &rlim); 1858 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1859 else st->print("%uk", rlim.rlim_cur >> 10); 1860 st->cr(); 1861 1862 // load average 1863 st->print("load average:"); 1864 double loadavg[3]; 1865 os::loadavg(loadavg, 3); 1866 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]); 1867 st->cr(); 1868} 1869 1870void os::print_memory_info(outputStream* st) { 1871 1872 st->print("Memory:"); 1873 st->print(" %dk page", os::vm_page_size()>>10); 1874 1875 // values in struct sysinfo are "unsigned long" 1876 struct sysinfo si; 1877 sysinfo(&si); 1878 1879 st->print(", physical " UINT64_FORMAT "k", 1880 os::physical_memory() >> 10); 1881 st->print("(" UINT64_FORMAT "k free)", 1882 os::available_memory() >> 10); 1883 st->print(", swap " UINT64_FORMAT "k", 1884 ((jlong)si.totalswap * si.mem_unit) >> 10); 1885 st->print("(" UINT64_FORMAT "k free)", 1886 ((jlong)si.freeswap * si.mem_unit) >> 10); 1887 st->cr(); 1888} 1889 1890// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 1891// but they're the same for all the linux arch that we support 1892// and they're the same for solaris but there's no common place to put this. 1893const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 1894 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 1895 "ILL_COPROC", "ILL_BADSTK" }; 1896 1897const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 1898 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 1899 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 1900 1901const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 1902 1903const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 1904 1905void os::print_siginfo(outputStream* st, void* siginfo) { 1906 st->print("siginfo:"); 1907 1908 const int buflen = 100; 1909 char buf[buflen]; 1910 siginfo_t *si = (siginfo_t*)siginfo; 1911 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 1912 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 1913 st->print("si_errno=%s", buf); 1914 } else { 1915 st->print("si_errno=%d", si->si_errno); 1916 } 1917 const int c = si->si_code; 1918 assert(c > 0, "unexpected si_code"); 1919 switch (si->si_signo) { 1920 case SIGILL: 1921 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 1922 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1923 break; 1924 case SIGFPE: 1925 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 1926 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1927 break; 1928 case SIGSEGV: 1929 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 1930 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1931 break; 1932 case SIGBUS: 1933 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 1934 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1935 break; 1936 default: 1937 st->print(", si_code=%d", si->si_code); 1938 // no si_addr 1939 } 1940 1941 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 1942 UseSharedSpaces) { 1943 FileMapInfo* mapinfo = FileMapInfo::current_info(); 1944 if (mapinfo->is_in_shared_space(si->si_addr)) { 1945 st->print("\n\nError accessing class data sharing archive." \ 1946 " Mapped file inaccessible during execution, " \ 1947 " possible disk/network problem."); 1948 } 1949 } 1950 st->cr(); 1951} 1952 1953 1954static void print_signal_handler(outputStream* st, int sig, 1955 char* buf, size_t buflen); 1956 1957void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 1958 st->print_cr("Signal Handlers:"); 1959 print_signal_handler(st, SIGSEGV, buf, buflen); 1960 print_signal_handler(st, SIGBUS , buf, buflen); 1961 print_signal_handler(st, SIGFPE , buf, buflen); 1962 print_signal_handler(st, SIGPIPE, buf, buflen); 1963 print_signal_handler(st, SIGXFSZ, buf, buflen); 1964 print_signal_handler(st, SIGILL , buf, buflen); 1965 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 1966 print_signal_handler(st, SR_signum, buf, buflen); 1967 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 1968 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 1969 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 1970 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 1971} 1972 1973static char saved_jvm_path[MAXPATHLEN] = {0}; 1974 1975// Find the full path to the current module, libjvm.so or libjvm_g.so 1976void os::jvm_path(char *buf, jint len) { 1977 // Error checking. 1978 if (len < MAXPATHLEN) { 1979 assert(false, "must use a large-enough buffer"); 1980 buf[0] = '\0'; 1981 return; 1982 } 1983 // Lazy resolve the path to current module. 1984 if (saved_jvm_path[0] != 0) { 1985 strcpy(buf, saved_jvm_path); 1986 return; 1987 } 1988 1989 char dli_fname[MAXPATHLEN]; 1990 bool ret = dll_address_to_library_name( 1991 CAST_FROM_FN_PTR(address, os::jvm_path), 1992 dli_fname, sizeof(dli_fname), NULL); 1993 assert(ret != 0, "cannot locate libjvm"); 1994 realpath(dli_fname, buf); 1995 1996 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) { 1997 // Support for the gamma launcher. Typical value for buf is 1998 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 1999 // the right place in the string, then assume we are installed in a JDK and 2000 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2001 // up the path so it looks like libjvm.so is installed there (append a 2002 // fake suffix hotspot/libjvm.so). 2003 const char *p = buf + strlen(buf) - 1; 2004 for (int count = 0; p > buf && count < 5; ++count) { 2005 for (--p; p > buf && *p != '/'; --p) 2006 /* empty */ ; 2007 } 2008 2009 if (strncmp(p, "/jre/lib/", 9) != 0) { 2010 // Look for JAVA_HOME in the environment. 2011 char* java_home_var = ::getenv("JAVA_HOME"); 2012 if (java_home_var != NULL && java_home_var[0] != 0) { 2013 // Check the current module name "libjvm.so" or "libjvm_g.so". 2014 p = strrchr(buf, '/'); 2015 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2016 p = strstr(p, "_g") ? "_g" : ""; 2017 2018 realpath(java_home_var, buf); 2019 sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch); 2020 if (0 == access(buf, F_OK)) { 2021 // Use current module name "libjvm[_g].so" instead of 2022 // "libjvm"debug_only("_g")".so" since for fastdebug version 2023 // we should have "libjvm.so" but debug_only("_g") adds "_g"! 2024 // It is used when we are choosing the HPI library's name 2025 // "libhpi[_g].so" in hpi::initialize_get_interface(). 2026 sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p); 2027 } else { 2028 // Go back to path of .so 2029 realpath(dli_fname, buf); 2030 } 2031 } 2032 } 2033 } 2034 2035 strcpy(saved_jvm_path, buf); 2036} 2037 2038void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2039 // no prefix required, not even "_" 2040} 2041 2042void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2043 // no suffix required 2044} 2045 2046//////////////////////////////////////////////////////////////////////////////// 2047// sun.misc.Signal support 2048 2049static volatile jint sigint_count = 0; 2050 2051static void 2052UserHandler(int sig, void *siginfo, void *context) { 2053 // 4511530 - sem_post is serialized and handled by the manager thread. When 2054 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2055 // don't want to flood the manager thread with sem_post requests. 2056 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2057 return; 2058 2059 // Ctrl-C is pressed during error reporting, likely because the error 2060 // handler fails to abort. Let VM die immediately. 2061 if (sig == SIGINT && is_error_reported()) { 2062 os::die(); 2063 } 2064 2065 os::signal_notify(sig); 2066} 2067 2068void* os::user_handler() { 2069 return CAST_FROM_FN_PTR(void*, UserHandler); 2070} 2071 2072extern "C" { 2073 typedef void (*sa_handler_t)(int); 2074 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2075} 2076 2077void* os::signal(int signal_number, void* handler) { 2078 struct sigaction sigAct, oldSigAct; 2079 2080 sigfillset(&(sigAct.sa_mask)); 2081 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2082 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2083 2084 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2085 // -1 means registration failed 2086 return (void *)-1; 2087 } 2088 2089 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2090} 2091 2092void os::signal_raise(int signal_number) { 2093 ::raise(signal_number); 2094} 2095 2096/* 2097 * The following code is moved from os.cpp for making this 2098 * code platform specific, which it is by its very nature. 2099 */ 2100 2101// Will be modified when max signal is changed to be dynamic 2102int os::sigexitnum_pd() { 2103 return NSIG; 2104} 2105 2106// a counter for each possible signal value 2107static volatile jint pending_signals[NSIG+1] = { 0 }; 2108 2109// Linux(POSIX) specific hand shaking semaphore. 2110static sem_t sig_sem; 2111 2112void os::signal_init_pd() { 2113 // Initialize signal structures 2114 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2115 2116 // Initialize signal semaphore 2117 ::sem_init(&sig_sem, 0, 0); 2118} 2119 2120void os::signal_notify(int sig) { 2121 Atomic::inc(&pending_signals[sig]); 2122 ::sem_post(&sig_sem); 2123} 2124 2125static int check_pending_signals(bool wait) { 2126 Atomic::store(0, &sigint_count); 2127 for (;;) { 2128 for (int i = 0; i < NSIG + 1; i++) { 2129 jint n = pending_signals[i]; 2130 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2131 return i; 2132 } 2133 } 2134 if (!wait) { 2135 return -1; 2136 } 2137 JavaThread *thread = JavaThread::current(); 2138 ThreadBlockInVM tbivm(thread); 2139 2140 bool threadIsSuspended; 2141 do { 2142 thread->set_suspend_equivalent(); 2143 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2144 ::sem_wait(&sig_sem); 2145 2146 // were we externally suspended while we were waiting? 2147 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2148 if (threadIsSuspended) { 2149 // 2150 // The semaphore has been incremented, but while we were waiting 2151 // another thread suspended us. We don't want to continue running 2152 // while suspended because that would surprise the thread that 2153 // suspended us. 2154 // 2155 ::sem_post(&sig_sem); 2156 2157 thread->java_suspend_self(); 2158 } 2159 } while (threadIsSuspended); 2160 } 2161} 2162 2163int os::signal_lookup() { 2164 return check_pending_signals(false); 2165} 2166 2167int os::signal_wait() { 2168 return check_pending_signals(true); 2169} 2170 2171//////////////////////////////////////////////////////////////////////////////// 2172// Virtual Memory 2173 2174int os::vm_page_size() { 2175 // Seems redundant as all get out 2176 assert(os::Linux::page_size() != -1, "must call os::init"); 2177 return os::Linux::page_size(); 2178} 2179 2180// Solaris allocates memory by pages. 2181int os::vm_allocation_granularity() { 2182 assert(os::Linux::page_size() != -1, "must call os::init"); 2183 return os::Linux::page_size(); 2184} 2185 2186// Rationale behind this function: 2187// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2188// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2189// samples for JITted code. Here we create private executable mapping over the code cache 2190// and then we can use standard (well, almost, as mapping can change) way to provide 2191// info for the reporting script by storing timestamp and location of symbol 2192void linux_wrap_code(char* base, size_t size) { 2193 static volatile jint cnt = 0; 2194 2195 if (!UseOprofile) { 2196 return; 2197 } 2198 2199 char buf[40]; 2200 int num = Atomic::add(1, &cnt); 2201 2202 sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num); 2203 unlink(buf); 2204 2205 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU); 2206 2207 if (fd != -1) { 2208 off_t rv = lseek(fd, size-2, SEEK_SET); 2209 if (rv != (off_t)-1) { 2210 if (write(fd, "", 1) == 1) { 2211 mmap(base, size, 2212 PROT_READ|PROT_WRITE|PROT_EXEC, 2213 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2214 } 2215 } 2216 close(fd); 2217 unlink(buf); 2218 } 2219} 2220 2221// NOTE: Linux kernel does not really reserve the pages for us. 2222// All it does is to check if there are enough free pages 2223// left at the time of mmap(). This could be a potential 2224// problem. 2225bool os::commit_memory(char* addr, size_t size) { 2226 uintptr_t res = (uintptr_t) ::mmap(addr, size, 2227 PROT_READ|PROT_WRITE|PROT_EXEC, 2228 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2229 return res != (uintptr_t) MAP_FAILED; 2230} 2231 2232bool os::commit_memory(char* addr, size_t size, size_t alignment_hint) { 2233 return commit_memory(addr, size); 2234} 2235 2236void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { } 2237void os::free_memory(char *addr, size_t bytes) { } 2238void os::numa_make_global(char *addr, size_t bytes) { } 2239void os::numa_make_local(char *addr, size_t bytes) { } 2240bool os::numa_topology_changed() { return false; } 2241size_t os::numa_get_groups_num() { return 1; } 2242int os::numa_get_group_id() { return 0; } 2243size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2244 if (size > 0) { 2245 ids[0] = 0; 2246 return 1; 2247 } 2248 return 0; 2249} 2250 2251bool os::get_page_info(char *start, page_info* info) { 2252 return false; 2253} 2254 2255char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2256 return end; 2257} 2258 2259bool os::uncommit_memory(char* addr, size_t size) { 2260 return ::mmap(addr, size, 2261 PROT_READ|PROT_WRITE|PROT_EXEC, 2262 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0) 2263 != MAP_FAILED; 2264} 2265 2266static address _highest_vm_reserved_address = NULL; 2267 2268// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2269// at 'requested_addr'. If there are existing memory mappings at the same 2270// location, however, they will be overwritten. If 'fixed' is false, 2271// 'requested_addr' is only treated as a hint, the return value may or 2272// may not start from the requested address. Unlike Linux mmap(), this 2273// function returns NULL to indicate failure. 2274static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2275 char * addr; 2276 int flags; 2277 2278 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2279 if (fixed) { 2280 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2281 flags |= MAP_FIXED; 2282 } 2283 2284 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE|PROT_EXEC, 2285 flags, -1, 0); 2286 2287 if (addr != MAP_FAILED) { 2288 // anon_mmap() should only get called during VM initialization, 2289 // don't need lock (actually we can skip locking even it can be called 2290 // from multiple threads, because _highest_vm_reserved_address is just a 2291 // hint about the upper limit of non-stack memory regions.) 2292 if ((address)addr + bytes > _highest_vm_reserved_address) { 2293 _highest_vm_reserved_address = (address)addr + bytes; 2294 } 2295 } 2296 2297 return addr == MAP_FAILED ? NULL : addr; 2298} 2299 2300// Don't update _highest_vm_reserved_address, because there might be memory 2301// regions above addr + size. If so, releasing a memory region only creates 2302// a hole in the address space, it doesn't help prevent heap-stack collision. 2303// 2304static int anon_munmap(char * addr, size_t size) { 2305 return ::munmap(addr, size) == 0; 2306} 2307 2308char* os::reserve_memory(size_t bytes, char* requested_addr, 2309 size_t alignment_hint) { 2310 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 2311} 2312 2313bool os::release_memory(char* addr, size_t size) { 2314 return anon_munmap(addr, size); 2315} 2316 2317static address highest_vm_reserved_address() { 2318 return _highest_vm_reserved_address; 2319} 2320 2321static bool linux_mprotect(char* addr, size_t size, int prot) { 2322 // Linux wants the mprotect address argument to be page aligned. 2323 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 2324 2325 // According to SUSv3, mprotect() should only be used with mappings 2326 // established by mmap(), and mmap() always maps whole pages. Unaligned 2327 // 'addr' likely indicates problem in the VM (e.g. trying to change 2328 // protection of malloc'ed or statically allocated memory). Check the 2329 // caller if you hit this assert. 2330 assert(addr == bottom, "sanity check"); 2331 2332 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 2333 return ::mprotect(bottom, size, prot) == 0; 2334} 2335 2336bool os::protect_memory(char* addr, size_t size) { 2337 return linux_mprotect(addr, size, PROT_READ); 2338} 2339 2340bool os::guard_memory(char* addr, size_t size) { 2341 return linux_mprotect(addr, size, PROT_NONE); 2342} 2343 2344bool os::unguard_memory(char* addr, size_t size) { 2345 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE|PROT_EXEC); 2346} 2347 2348// Large page support 2349 2350static size_t _large_page_size = 0; 2351 2352bool os::large_page_init() { 2353 if (!UseLargePages) return false; 2354 2355 if (LargePageSizeInBytes) { 2356 _large_page_size = LargePageSizeInBytes; 2357 } else { 2358 // large_page_size on Linux is used to round up heap size. x86 uses either 2359 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 2360 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 2361 // page as large as 256M. 2362 // 2363 // Here we try to figure out page size by parsing /proc/meminfo and looking 2364 // for a line with the following format: 2365 // Hugepagesize: 2048 kB 2366 // 2367 // If we can't determine the value (e.g. /proc is not mounted, or the text 2368 // format has been changed), we'll use the largest page size supported by 2369 // the processor. 2370 2371 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M); 2372 2373 FILE *fp = fopen("/proc/meminfo", "r"); 2374 if (fp) { 2375 while (!feof(fp)) { 2376 int x = 0; 2377 char buf[16]; 2378 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 2379 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 2380 _large_page_size = x * K; 2381 break; 2382 } 2383 } else { 2384 // skip to next line 2385 for (;;) { 2386 int ch = fgetc(fp); 2387 if (ch == EOF || ch == (int)'\n') break; 2388 } 2389 } 2390 } 2391 fclose(fp); 2392 } 2393 } 2394 2395 const size_t default_page_size = (size_t)Linux::page_size(); 2396 if (_large_page_size > default_page_size) { 2397 _page_sizes[0] = _large_page_size; 2398 _page_sizes[1] = default_page_size; 2399 _page_sizes[2] = 0; 2400 } 2401 2402 // Large page support is available on 2.6 or newer kernel, some vendors 2403 // (e.g. Redhat) have backported it to their 2.4 based distributions. 2404 // We optimistically assume the support is available. If later it turns out 2405 // not true, VM will automatically switch to use regular page size. 2406 return true; 2407} 2408 2409#ifndef SHM_HUGETLB 2410#define SHM_HUGETLB 04000 2411#endif 2412 2413char* os::reserve_memory_special(size_t bytes) { 2414 assert(UseLargePages, "only for large pages"); 2415 2416 key_t key = IPC_PRIVATE; 2417 char *addr; 2418 2419 bool warn_on_failure = UseLargePages && 2420 (!FLAG_IS_DEFAULT(UseLargePages) || 2421 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 2422 ); 2423 char msg[128]; 2424 2425 // Create a large shared memory region to attach to based on size. 2426 // Currently, size is the total size of the heap 2427 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 2428 if (shmid == -1) { 2429 // Possible reasons for shmget failure: 2430 // 1. shmmax is too small for Java heap. 2431 // > check shmmax value: cat /proc/sys/kernel/shmmax 2432 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 2433 // 2. not enough large page memory. 2434 // > check available large pages: cat /proc/meminfo 2435 // > increase amount of large pages: 2436 // echo new_value > /proc/sys/vm/nr_hugepages 2437 // Note 1: different Linux may use different name for this property, 2438 // e.g. on Redhat AS-3 it is "hugetlb_pool". 2439 // Note 2: it's possible there's enough physical memory available but 2440 // they are so fragmented after a long run that they can't 2441 // coalesce into large pages. Try to reserve large pages when 2442 // the system is still "fresh". 2443 if (warn_on_failure) { 2444 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 2445 warning(msg); 2446 } 2447 return NULL; 2448 } 2449 2450 // attach to the region 2451 addr = (char*)shmat(shmid, NULL, 0); 2452 int err = errno; 2453 2454 // Remove shmid. If shmat() is successful, the actual shared memory segment 2455 // will be deleted when it's detached by shmdt() or when the process 2456 // terminates. If shmat() is not successful this will remove the shared 2457 // segment immediately. 2458 shmctl(shmid, IPC_RMID, NULL); 2459 2460 if ((intptr_t)addr == -1) { 2461 if (warn_on_failure) { 2462 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 2463 warning(msg); 2464 } 2465 return NULL; 2466 } 2467 2468 return addr; 2469} 2470 2471bool os::release_memory_special(char* base, size_t bytes) { 2472 // detaching the SHM segment will also delete it, see reserve_memory_special() 2473 int rslt = shmdt(base); 2474 return rslt == 0; 2475} 2476 2477size_t os::large_page_size() { 2478 return _large_page_size; 2479} 2480 2481// Linux does not support anonymous mmap with large page memory. The only way 2482// to reserve large page memory without file backing is through SysV shared 2483// memory API. The entire memory region is committed and pinned upfront. 2484// Hopefully this will change in the future... 2485bool os::can_commit_large_page_memory() { 2486 return false; 2487} 2488 2489// Reserve memory at an arbitrary address, only if that area is 2490// available (and not reserved for something else). 2491 2492char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 2493 const int max_tries = 10; 2494 char* base[max_tries]; 2495 size_t size[max_tries]; 2496 const size_t gap = 0x000000; 2497 2498 // Assert only that the size is a multiple of the page size, since 2499 // that's all that mmap requires, and since that's all we really know 2500 // about at this low abstraction level. If we need higher alignment, 2501 // we can either pass an alignment to this method or verify alignment 2502 // in one of the methods further up the call chain. See bug 5044738. 2503 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 2504 2505 // Repeatedly allocate blocks until the block is allocated at the 2506 // right spot. Give up after max_tries. Note that reserve_memory() will 2507 // automatically update _highest_vm_reserved_address if the call is 2508 // successful. The variable tracks the highest memory address every reserved 2509 // by JVM. It is used to detect heap-stack collision if running with 2510 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 2511 // space than needed, it could confuse the collision detecting code. To 2512 // solve the problem, save current _highest_vm_reserved_address and 2513 // calculate the correct value before return. 2514 address old_highest = _highest_vm_reserved_address; 2515 2516 // Linux mmap allows caller to pass an address as hint; give it a try first, 2517 // if kernel honors the hint then we can return immediately. 2518 char * addr = anon_mmap(requested_addr, bytes, false); 2519 if (addr == requested_addr) { 2520 return requested_addr; 2521 } 2522 2523 if (addr != NULL) { 2524 // mmap() is successful but it fails to reserve at the requested address 2525 anon_munmap(addr, bytes); 2526 } 2527 2528 int i; 2529 for (i = 0; i < max_tries; ++i) { 2530 base[i] = reserve_memory(bytes); 2531 2532 if (base[i] != NULL) { 2533 // Is this the block we wanted? 2534 if (base[i] == requested_addr) { 2535 size[i] = bytes; 2536 break; 2537 } 2538 2539 // Does this overlap the block we wanted? Give back the overlapped 2540 // parts and try again. 2541 2542 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 2543 if (top_overlap >= 0 && top_overlap < bytes) { 2544 unmap_memory(base[i], top_overlap); 2545 base[i] += top_overlap; 2546 size[i] = bytes - top_overlap; 2547 } else { 2548 size_t bottom_overlap = base[i] + bytes - requested_addr; 2549 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 2550 unmap_memory(requested_addr, bottom_overlap); 2551 size[i] = bytes - bottom_overlap; 2552 } else { 2553 size[i] = bytes; 2554 } 2555 } 2556 } 2557 } 2558 2559 // Give back the unused reserved pieces. 2560 2561 for (int j = 0; j < i; ++j) { 2562 if (base[j] != NULL) { 2563 unmap_memory(base[j], size[j]); 2564 } 2565 } 2566 2567 if (i < max_tries) { 2568 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 2569 return requested_addr; 2570 } else { 2571 _highest_vm_reserved_address = old_highest; 2572 return NULL; 2573 } 2574} 2575 2576size_t os::read(int fd, void *buf, unsigned int nBytes) { 2577 return ::read(fd, buf, nBytes); 2578} 2579 2580// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 2581// Solaris uses poll(), linux uses park(). 2582// Poll() is likely a better choice, assuming that Thread.interrupt() 2583// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 2584// SIGSEGV, see 4355769. 2585 2586const int NANOSECS_PER_MILLISECS = 1000000; 2587 2588int os::sleep(Thread* thread, jlong millis, bool interruptible) { 2589 assert(thread == Thread::current(), "thread consistency check"); 2590 2591 ParkEvent * const slp = thread->_SleepEvent ; 2592 slp->reset() ; 2593 OrderAccess::fence() ; 2594 2595 if (interruptible) { 2596 jlong prevtime = javaTimeNanos(); 2597 2598 for (;;) { 2599 if (os::is_interrupted(thread, true)) { 2600 return OS_INTRPT; 2601 } 2602 2603 jlong newtime = javaTimeNanos(); 2604 2605 if (newtime - prevtime < 0) { 2606 // time moving backwards, should only happen if no monotonic clock 2607 // not a guarantee() because JVM should not abort on kernel/glibc bugs 2608 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 2609 } else { 2610 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 2611 } 2612 2613 if(millis <= 0) { 2614 return OS_OK; 2615 } 2616 2617 prevtime = newtime; 2618 2619 { 2620 assert(thread->is_Java_thread(), "sanity check"); 2621 JavaThread *jt = (JavaThread *) thread; 2622 ThreadBlockInVM tbivm(jt); 2623 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 2624 2625 jt->set_suspend_equivalent(); 2626 // cleared by handle_special_suspend_equivalent_condition() or 2627 // java_suspend_self() via check_and_wait_while_suspended() 2628 2629 slp->park(millis); 2630 2631 // were we externally suspended while we were waiting? 2632 jt->check_and_wait_while_suspended(); 2633 } 2634 } 2635 } else { 2636 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 2637 jlong prevtime = javaTimeNanos(); 2638 2639 for (;;) { 2640 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 2641 // the 1st iteration ... 2642 jlong newtime = javaTimeNanos(); 2643 2644 if (newtime - prevtime < 0) { 2645 // time moving backwards, should only happen if no monotonic clock 2646 // not a guarantee() because JVM should not abort on kernel/glibc bugs 2647 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 2648 } else { 2649 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 2650 } 2651 2652 if(millis <= 0) break ; 2653 2654 prevtime = newtime; 2655 slp->park(millis); 2656 } 2657 return OS_OK ; 2658 } 2659} 2660 2661int os::naked_sleep() { 2662 // %% make the sleep time an integer flag. for now use 1 millisec. 2663 return os::sleep(Thread::current(), 1, false); 2664} 2665 2666// Sleep forever; naked call to OS-specific sleep; use with CAUTION 2667void os::infinite_sleep() { 2668 while (true) { // sleep forever ... 2669 ::sleep(100); // ... 100 seconds at a time 2670 } 2671} 2672 2673// Used to convert frequent JVM_Yield() to nops 2674bool os::dont_yield() { 2675 return DontYieldALot; 2676} 2677 2678void os::yield() { 2679 sched_yield(); 2680} 2681 2682os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 2683 2684void os::yield_all(int attempts) { 2685 // Yields to all threads, including threads with lower priorities 2686 // Threads on Linux are all with same priority. The Solaris style 2687 // os::yield_all() with nanosleep(1ms) is not necessary. 2688 sched_yield(); 2689} 2690 2691// Called from the tight loops to possibly influence time-sharing heuristics 2692void os::loop_breaker(int attempts) { 2693 os::yield_all(attempts); 2694} 2695 2696//////////////////////////////////////////////////////////////////////////////// 2697// thread priority support 2698 2699// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 2700// only supports dynamic priority, static priority must be zero. For real-time 2701// applications, Linux supports SCHED_RR which allows static priority (1-99). 2702// However, for large multi-threaded applications, SCHED_RR is not only slower 2703// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 2704// of 5 runs - Sep 2005). 2705// 2706// The following code actually changes the niceness of kernel-thread/LWP. It 2707// has an assumption that setpriority() only modifies one kernel-thread/LWP, 2708// not the entire user process, and user level threads are 1:1 mapped to kernel 2709// threads. It has always been the case, but could change in the future. For 2710// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 2711// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 2712 2713int os::java_to_os_priority[MaxPriority + 1] = { 2714 19, // 0 Entry should never be used 2715 2716 4, // 1 MinPriority 2717 3, // 2 2718 2, // 3 2719 2720 1, // 4 2721 0, // 5 NormPriority 2722 -1, // 6 2723 2724 -2, // 7 2725 -3, // 8 2726 -4, // 9 NearMaxPriority 2727 2728 -5 // 10 MaxPriority 2729}; 2730 2731static int prio_init() { 2732 if (ThreadPriorityPolicy == 1) { 2733 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 2734 // if effective uid is not root. Perhaps, a more elegant way of doing 2735 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 2736 if (geteuid() != 0) { 2737 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 2738 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 2739 } 2740 ThreadPriorityPolicy = 0; 2741 } 2742 } 2743 return 0; 2744} 2745 2746OSReturn os::set_native_priority(Thread* thread, int newpri) { 2747 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 2748 2749 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 2750 return (ret == 0) ? OS_OK : OS_ERR; 2751} 2752 2753OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 2754 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 2755 *priority_ptr = java_to_os_priority[NormPriority]; 2756 return OS_OK; 2757 } 2758 2759 errno = 0; 2760 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 2761 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 2762} 2763 2764// Hint to the underlying OS that a task switch would not be good. 2765// Void return because it's a hint and can fail. 2766void os::hint_no_preempt() {} 2767 2768//////////////////////////////////////////////////////////////////////////////// 2769// suspend/resume support 2770 2771// the low-level signal-based suspend/resume support is a remnant from the 2772// old VM-suspension that used to be for java-suspension, safepoints etc, 2773// within hotspot. Now there is a single use-case for this: 2774// - calling get_thread_pc() on the VMThread by the flat-profiler task 2775// that runs in the watcher thread. 2776// The remaining code is greatly simplified from the more general suspension 2777// code that used to be used. 2778// 2779// The protocol is quite simple: 2780// - suspend: 2781// - sends a signal to the target thread 2782// - polls the suspend state of the osthread using a yield loop 2783// - target thread signal handler (SR_handler) sets suspend state 2784// and blocks in sigsuspend until continued 2785// - resume: 2786// - sets target osthread state to continue 2787// - sends signal to end the sigsuspend loop in the SR_handler 2788// 2789// Note that the SR_lock plays no role in this suspend/resume protocol. 2790// 2791 2792static void resume_clear_context(OSThread *osthread) { 2793 osthread->set_ucontext(NULL); 2794 osthread->set_siginfo(NULL); 2795 2796 // notify the suspend action is completed, we have now resumed 2797 osthread->sr.clear_suspended(); 2798} 2799 2800static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 2801 osthread->set_ucontext(context); 2802 osthread->set_siginfo(siginfo); 2803} 2804 2805// 2806// Handler function invoked when a thread's execution is suspended or 2807// resumed. We have to be careful that only async-safe functions are 2808// called here (Note: most pthread functions are not async safe and 2809// should be avoided.) 2810// 2811// Note: sigwait() is a more natural fit than sigsuspend() from an 2812// interface point of view, but sigwait() prevents the signal hander 2813// from being run. libpthread would get very confused by not having 2814// its signal handlers run and prevents sigwait()'s use with the 2815// mutex granting granting signal. 2816// 2817// Currently only ever called on the VMThread 2818// 2819static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 2820 // Save and restore errno to avoid confusing native code with EINTR 2821 // after sigsuspend. 2822 int old_errno = errno; 2823 2824 Thread* thread = Thread::current(); 2825 OSThread* osthread = thread->osthread(); 2826 assert(thread->is_VM_thread(), "Must be VMThread"); 2827 // read current suspend action 2828 int action = osthread->sr.suspend_action(); 2829 if (action == SR_SUSPEND) { 2830 suspend_save_context(osthread, siginfo, context); 2831 2832 // Notify the suspend action is about to be completed. do_suspend() 2833 // waits until SR_SUSPENDED is set and then returns. We will wait 2834 // here for a resume signal and that completes the suspend-other 2835 // action. do_suspend/do_resume is always called as a pair from 2836 // the same thread - so there are no races 2837 2838 // notify the caller 2839 osthread->sr.set_suspended(); 2840 2841 sigset_t suspend_set; // signals for sigsuspend() 2842 2843 // get current set of blocked signals and unblock resume signal 2844 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 2845 sigdelset(&suspend_set, SR_signum); 2846 2847 // wait here until we are resumed 2848 do { 2849 sigsuspend(&suspend_set); 2850 // ignore all returns until we get a resume signal 2851 } while (osthread->sr.suspend_action() != SR_CONTINUE); 2852 2853 resume_clear_context(osthread); 2854 2855 } else { 2856 assert(action == SR_CONTINUE, "unexpected sr action"); 2857 // nothing special to do - just leave the handler 2858 } 2859 2860 errno = old_errno; 2861} 2862 2863 2864static int SR_initialize() { 2865 struct sigaction act; 2866 char *s; 2867 /* Get signal number to use for suspend/resume */ 2868 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 2869 int sig = ::strtol(s, 0, 10); 2870 if (sig > 0 || sig < _NSIG) { 2871 SR_signum = sig; 2872 } 2873 } 2874 2875 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 2876 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 2877 2878 sigemptyset(&SR_sigset); 2879 sigaddset(&SR_sigset, SR_signum); 2880 2881 /* Set up signal handler for suspend/resume */ 2882 act.sa_flags = SA_RESTART|SA_SIGINFO; 2883 act.sa_handler = (void (*)(int)) SR_handler; 2884 2885 // SR_signum is blocked by default. 2886 // 4528190 - We also need to block pthread restart signal (32 on all 2887 // supported Linux platforms). Note that LinuxThreads need to block 2888 // this signal for all threads to work properly. So we don't have 2889 // to use hard-coded signal number when setting up the mask. 2890 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 2891 2892 if (sigaction(SR_signum, &act, 0) == -1) { 2893 return -1; 2894 } 2895 2896 // Save signal flag 2897 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 2898 return 0; 2899} 2900 2901static int SR_finalize() { 2902 return 0; 2903} 2904 2905 2906// returns true on success and false on error - really an error is fatal 2907// but this seems the normal response to library errors 2908static bool do_suspend(OSThread* osthread) { 2909 // mark as suspended and send signal 2910 osthread->sr.set_suspend_action(SR_SUSPEND); 2911 int status = pthread_kill(osthread->pthread_id(), SR_signum); 2912 assert_status(status == 0, status, "pthread_kill"); 2913 2914 // check status and wait until notified of suspension 2915 if (status == 0) { 2916 for (int i = 0; !osthread->sr.is_suspended(); i++) { 2917 os::yield_all(i); 2918 } 2919 osthread->sr.set_suspend_action(SR_NONE); 2920 return true; 2921 } 2922 else { 2923 osthread->sr.set_suspend_action(SR_NONE); 2924 return false; 2925 } 2926} 2927 2928static void do_resume(OSThread* osthread) { 2929 assert(osthread->sr.is_suspended(), "thread should be suspended"); 2930 osthread->sr.set_suspend_action(SR_CONTINUE); 2931 2932 int status = pthread_kill(osthread->pthread_id(), SR_signum); 2933 assert_status(status == 0, status, "pthread_kill"); 2934 // check status and wait unit notified of resumption 2935 if (status == 0) { 2936 for (int i = 0; osthread->sr.is_suspended(); i++) { 2937 os::yield_all(i); 2938 } 2939 } 2940 osthread->sr.set_suspend_action(SR_NONE); 2941} 2942 2943//////////////////////////////////////////////////////////////////////////////// 2944// interrupt support 2945 2946void os::interrupt(Thread* thread) { 2947 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 2948 "possibility of dangling Thread pointer"); 2949 2950 OSThread* osthread = thread->osthread(); 2951 2952 if (!osthread->interrupted()) { 2953 osthread->set_interrupted(true); 2954 // More than one thread can get here with the same value of osthread, 2955 // resulting in multiple notifications. We do, however, want the store 2956 // to interrupted() to be visible to other threads before we execute unpark(). 2957 OrderAccess::fence(); 2958 ParkEvent * const slp = thread->_SleepEvent ; 2959 if (slp != NULL) slp->unpark() ; 2960 } 2961 2962 // For JSR166. Unpark even if interrupt status already was set 2963 if (thread->is_Java_thread()) 2964 ((JavaThread*)thread)->parker()->unpark(); 2965 2966 ParkEvent * ev = thread->_ParkEvent ; 2967 if (ev != NULL) ev->unpark() ; 2968 2969} 2970 2971bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 2972 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 2973 "possibility of dangling Thread pointer"); 2974 2975 OSThread* osthread = thread->osthread(); 2976 2977 bool interrupted = osthread->interrupted(); 2978 2979 if (interrupted && clear_interrupted) { 2980 osthread->set_interrupted(false); 2981 // consider thread->_SleepEvent->reset() ... optional optimization 2982 } 2983 2984 return interrupted; 2985} 2986 2987/////////////////////////////////////////////////////////////////////////////////// 2988// signal handling (except suspend/resume) 2989 2990// This routine may be used by user applications as a "hook" to catch signals. 2991// The user-defined signal handler must pass unrecognized signals to this 2992// routine, and if it returns true (non-zero), then the signal handler must 2993// return immediately. If the flag "abort_if_unrecognized" is true, then this 2994// routine will never retun false (zero), but instead will execute a VM panic 2995// routine kill the process. 2996// 2997// If this routine returns false, it is OK to call it again. This allows 2998// the user-defined signal handler to perform checks either before or after 2999// the VM performs its own checks. Naturally, the user code would be making 3000// a serious error if it tried to handle an exception (such as a null check 3001// or breakpoint) that the VM was generating for its own correct operation. 3002// 3003// This routine may recognize any of the following kinds of signals: 3004// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 3005// It should be consulted by handlers for any of those signals. 3006// 3007// The caller of this routine must pass in the three arguments supplied 3008// to the function referred to in the "sa_sigaction" (not the "sa_handler") 3009// field of the structure passed to sigaction(). This routine assumes that 3010// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 3011// 3012// Note that the VM will print warnings if it detects conflicting signal 3013// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 3014// 3015extern "C" int 3016JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 3017 void* ucontext, int abort_if_unrecognized); 3018 3019void signalHandler(int sig, siginfo_t* info, void* uc) { 3020 assert(info != NULL && uc != NULL, "it must be old kernel"); 3021 JVM_handle_linux_signal(sig, info, uc, true); 3022} 3023 3024 3025// This boolean allows users to forward their own non-matching signals 3026// to JVM_handle_linux_signal, harmlessly. 3027bool os::Linux::signal_handlers_are_installed = false; 3028 3029// For signal-chaining 3030struct sigaction os::Linux::sigact[MAXSIGNUM]; 3031unsigned int os::Linux::sigs = 0; 3032bool os::Linux::libjsig_is_loaded = false; 3033typedef struct sigaction *(*get_signal_t)(int); 3034get_signal_t os::Linux::get_signal_action = NULL; 3035 3036struct sigaction* os::Linux::get_chained_signal_action(int sig) { 3037 struct sigaction *actp = NULL; 3038 3039 if (libjsig_is_loaded) { 3040 // Retrieve the old signal handler from libjsig 3041 actp = (*get_signal_action)(sig); 3042 } 3043 if (actp == NULL) { 3044 // Retrieve the preinstalled signal handler from jvm 3045 actp = get_preinstalled_handler(sig); 3046 } 3047 3048 return actp; 3049} 3050 3051static bool call_chained_handler(struct sigaction *actp, int sig, 3052 siginfo_t *siginfo, void *context) { 3053 // Call the old signal handler 3054 if (actp->sa_handler == SIG_DFL) { 3055 // It's more reasonable to let jvm treat it as an unexpected exception 3056 // instead of taking the default action. 3057 return false; 3058 } else if (actp->sa_handler != SIG_IGN) { 3059 if ((actp->sa_flags & SA_NODEFER) == 0) { 3060 // automaticlly block the signal 3061 sigaddset(&(actp->sa_mask), sig); 3062 } 3063 3064 sa_handler_t hand; 3065 sa_sigaction_t sa; 3066 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 3067 // retrieve the chained handler 3068 if (siginfo_flag_set) { 3069 sa = actp->sa_sigaction; 3070 } else { 3071 hand = actp->sa_handler; 3072 } 3073 3074 if ((actp->sa_flags & SA_RESETHAND) != 0) { 3075 actp->sa_handler = SIG_DFL; 3076 } 3077 3078 // try to honor the signal mask 3079 sigset_t oset; 3080 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 3081 3082 // call into the chained handler 3083 if (siginfo_flag_set) { 3084 (*sa)(sig, siginfo, context); 3085 } else { 3086 (*hand)(sig); 3087 } 3088 3089 // restore the signal mask 3090 pthread_sigmask(SIG_SETMASK, &oset, 0); 3091 } 3092 // Tell jvm's signal handler the signal is taken care of. 3093 return true; 3094} 3095 3096bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 3097 bool chained = false; 3098 // signal-chaining 3099 if (UseSignalChaining) { 3100 struct sigaction *actp = get_chained_signal_action(sig); 3101 if (actp != NULL) { 3102 chained = call_chained_handler(actp, sig, siginfo, context); 3103 } 3104 } 3105 return chained; 3106} 3107 3108struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 3109 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 3110 return &sigact[sig]; 3111 } 3112 return NULL; 3113} 3114 3115void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 3116 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3117 sigact[sig] = oldAct; 3118 sigs |= (unsigned int)1 << sig; 3119} 3120 3121// for diagnostic 3122int os::Linux::sigflags[MAXSIGNUM]; 3123 3124int os::Linux::get_our_sigflags(int sig) { 3125 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3126 return sigflags[sig]; 3127} 3128 3129void os::Linux::set_our_sigflags(int sig, int flags) { 3130 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3131 sigflags[sig] = flags; 3132} 3133 3134void os::Linux::set_signal_handler(int sig, bool set_installed) { 3135 // Check for overwrite. 3136 struct sigaction oldAct; 3137 sigaction(sig, (struct sigaction*)NULL, &oldAct); 3138 3139 void* oldhand = oldAct.sa_sigaction 3140 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3141 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3142 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 3143 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 3144 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 3145 if (AllowUserSignalHandlers || !set_installed) { 3146 // Do not overwrite; user takes responsibility to forward to us. 3147 return; 3148 } else if (UseSignalChaining) { 3149 // save the old handler in jvm 3150 save_preinstalled_handler(sig, oldAct); 3151 // libjsig also interposes the sigaction() call below and saves the 3152 // old sigaction on it own. 3153 } else { 3154 fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig); 3155 } 3156 } 3157 3158 struct sigaction sigAct; 3159 sigfillset(&(sigAct.sa_mask)); 3160 sigAct.sa_handler = SIG_DFL; 3161 if (!set_installed) { 3162 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3163 } else { 3164 sigAct.sa_sigaction = signalHandler; 3165 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3166 } 3167 // Save flags, which are set by ours 3168 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3169 sigflags[sig] = sigAct.sa_flags; 3170 3171 int ret = sigaction(sig, &sigAct, &oldAct); 3172 assert(ret == 0, "check"); 3173 3174 void* oldhand2 = oldAct.sa_sigaction 3175 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3176 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3177 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 3178} 3179 3180// install signal handlers for signals that HotSpot needs to 3181// handle in order to support Java-level exception handling. 3182 3183void os::Linux::install_signal_handlers() { 3184 if (!signal_handlers_are_installed) { 3185 signal_handlers_are_installed = true; 3186 3187 // signal-chaining 3188 typedef void (*signal_setting_t)(); 3189 signal_setting_t begin_signal_setting = NULL; 3190 signal_setting_t end_signal_setting = NULL; 3191 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3192 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 3193 if (begin_signal_setting != NULL) { 3194 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3195 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 3196 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 3197 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 3198 libjsig_is_loaded = true; 3199 assert(UseSignalChaining, "should enable signal-chaining"); 3200 } 3201 if (libjsig_is_loaded) { 3202 // Tell libjsig jvm is setting signal handlers 3203 (*begin_signal_setting)(); 3204 } 3205 3206 set_signal_handler(SIGSEGV, true); 3207 set_signal_handler(SIGPIPE, true); 3208 set_signal_handler(SIGBUS, true); 3209 set_signal_handler(SIGILL, true); 3210 set_signal_handler(SIGFPE, true); 3211 set_signal_handler(SIGXFSZ, true); 3212 3213 if (libjsig_is_loaded) { 3214 // Tell libjsig jvm finishes setting signal handlers 3215 (*end_signal_setting)(); 3216 } 3217 3218 // We don't activate signal checker if libjsig is in place, we trust ourselves 3219 // and if UserSignalHandler is installed all bets are off 3220 if (CheckJNICalls) { 3221 if (libjsig_is_loaded) { 3222 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 3223 check_signals = false; 3224 } 3225 if (AllowUserSignalHandlers) { 3226 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 3227 check_signals = false; 3228 } 3229 } 3230 } 3231} 3232 3233// This is the fastest way to get thread cpu time on Linux. 3234// Returns cpu time (user+sys) for any thread, not only for current. 3235// POSIX compliant clocks are implemented in the kernels 2.6.16+. 3236// It might work on 2.6.10+ with a special kernel/glibc patch. 3237// For reference, please, see IEEE Std 1003.1-2004: 3238// http://www.unix.org/single_unix_specification 3239 3240jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 3241 struct timespec tp; 3242 int rc = os::Linux::clock_gettime(clockid, &tp); 3243 assert(rc == 0, "clock_gettime is expected to return 0 code"); 3244 3245 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec; 3246} 3247 3248///// 3249// glibc on Linux platform uses non-documented flag 3250// to indicate, that some special sort of signal 3251// trampoline is used. 3252// We will never set this flag, and we should 3253// ignore this flag in our diagnostic 3254#ifdef SIGNIFICANT_SIGNAL_MASK 3255#undef SIGNIFICANT_SIGNAL_MASK 3256#endif 3257#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 3258 3259static const char* get_signal_handler_name(address handler, 3260 char* buf, int buflen) { 3261 int offset; 3262 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 3263 if (found) { 3264 // skip directory names 3265 const char *p1, *p2; 3266 p1 = buf; 3267 size_t len = strlen(os::file_separator()); 3268 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 3269 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 3270 } else { 3271 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 3272 } 3273 return buf; 3274} 3275 3276static void print_signal_handler(outputStream* st, int sig, 3277 char* buf, size_t buflen) { 3278 struct sigaction sa; 3279 3280 sigaction(sig, NULL, &sa); 3281 3282 // See comment for SIGNIFICANT_SIGNAL_MASK define 3283 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3284 3285 st->print("%s: ", os::exception_name(sig, buf, buflen)); 3286 3287 address handler = (sa.sa_flags & SA_SIGINFO) 3288 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 3289 : CAST_FROM_FN_PTR(address, sa.sa_handler); 3290 3291 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 3292 st->print("SIG_DFL"); 3293 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 3294 st->print("SIG_IGN"); 3295 } else { 3296 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 3297 } 3298 3299 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 3300 3301 address rh = VMError::get_resetted_sighandler(sig); 3302 // May be, handler was resetted by VMError? 3303 if(rh != NULL) { 3304 handler = rh; 3305 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 3306 } 3307 3308 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 3309 3310 // Check: is it our handler? 3311 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 3312 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 3313 // It is our signal handler 3314 // check for flags, reset system-used one! 3315 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 3316 st->print( 3317 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 3318 os::Linux::get_our_sigflags(sig)); 3319 } 3320 } 3321 st->cr(); 3322} 3323 3324 3325#define DO_SIGNAL_CHECK(sig) \ 3326 if (!sigismember(&check_signal_done, sig)) \ 3327 os::Linux::check_signal_handler(sig) 3328 3329// This method is a periodic task to check for misbehaving JNI applications 3330// under CheckJNI, we can add any periodic checks here 3331 3332void os::run_periodic_checks() { 3333 3334 if (check_signals == false) return; 3335 3336 // SEGV and BUS if overridden could potentially prevent 3337 // generation of hs*.log in the event of a crash, debugging 3338 // such a case can be very challenging, so we absolutely 3339 // check the following for a good measure: 3340 DO_SIGNAL_CHECK(SIGSEGV); 3341 DO_SIGNAL_CHECK(SIGILL); 3342 DO_SIGNAL_CHECK(SIGFPE); 3343 DO_SIGNAL_CHECK(SIGBUS); 3344 DO_SIGNAL_CHECK(SIGPIPE); 3345 DO_SIGNAL_CHECK(SIGXFSZ); 3346 3347 3348 // ReduceSignalUsage allows the user to override these handlers 3349 // see comments at the very top and jvm_solaris.h 3350 if (!ReduceSignalUsage) { 3351 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 3352 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 3353 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 3354 DO_SIGNAL_CHECK(BREAK_SIGNAL); 3355 } 3356 3357 DO_SIGNAL_CHECK(SR_signum); 3358 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 3359} 3360 3361typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 3362 3363static os_sigaction_t os_sigaction = NULL; 3364 3365void os::Linux::check_signal_handler(int sig) { 3366 char buf[O_BUFLEN]; 3367 address jvmHandler = NULL; 3368 3369 3370 struct sigaction act; 3371 if (os_sigaction == NULL) { 3372 // only trust the default sigaction, in case it has been interposed 3373 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 3374 if (os_sigaction == NULL) return; 3375 } 3376 3377 os_sigaction(sig, (struct sigaction*)NULL, &act); 3378 3379 3380 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3381 3382 address thisHandler = (act.sa_flags & SA_SIGINFO) 3383 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 3384 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 3385 3386 3387 switch(sig) { 3388 case SIGSEGV: 3389 case SIGBUS: 3390 case SIGFPE: 3391 case SIGPIPE: 3392 case SIGILL: 3393 case SIGXFSZ: 3394 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 3395 break; 3396 3397 case SHUTDOWN1_SIGNAL: 3398 case SHUTDOWN2_SIGNAL: 3399 case SHUTDOWN3_SIGNAL: 3400 case BREAK_SIGNAL: 3401 jvmHandler = (address)user_handler(); 3402 break; 3403 3404 case INTERRUPT_SIGNAL: 3405 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 3406 break; 3407 3408 default: 3409 if (sig == SR_signum) { 3410 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 3411 } else { 3412 return; 3413 } 3414 break; 3415 } 3416 3417 if (thisHandler != jvmHandler) { 3418 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 3419 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 3420 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 3421 // No need to check this sig any longer 3422 sigaddset(&check_signal_done, sig); 3423 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 3424 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 3425 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 3426 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 3427 // No need to check this sig any longer 3428 sigaddset(&check_signal_done, sig); 3429 } 3430 3431 // Dump all the signal 3432 if (sigismember(&check_signal_done, sig)) { 3433 print_signal_handlers(tty, buf, O_BUFLEN); 3434 } 3435} 3436 3437extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 3438 3439extern bool signal_name(int signo, char* buf, size_t len); 3440 3441const char* os::exception_name(int exception_code, char* buf, size_t size) { 3442 if (0 < exception_code && exception_code <= SIGRTMAX) { 3443 // signal 3444 if (!signal_name(exception_code, buf, size)) { 3445 jio_snprintf(buf, size, "SIG%d", exception_code); 3446 } 3447 return buf; 3448 } else { 3449 return NULL; 3450 } 3451} 3452 3453// this is called _before_ the most of global arguments have been parsed 3454void os::init(void) { 3455 char dummy; /* used to get a guess on initial stack address */ 3456// first_hrtime = gethrtime(); 3457 3458 // With LinuxThreads the JavaMain thread pid (primordial thread) 3459 // is different than the pid of the java launcher thread. 3460 // So, on Linux, the launcher thread pid is passed to the VM 3461 // via the sun.java.launcher.pid property. 3462 // Use this property instead of getpid() if it was correctly passed. 3463 // See bug 6351349. 3464 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 3465 3466 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 3467 3468 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 3469 3470 init_random(1234567); 3471 3472 ThreadCritical::initialize(); 3473 3474 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 3475 if (Linux::page_size() == -1) { 3476 fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno)); 3477 } 3478 init_page_sizes((size_t) Linux::page_size()); 3479 3480 Linux::initialize_system_info(); 3481 3482 // main_thread points to the aboriginal thread 3483 Linux::_main_thread = pthread_self(); 3484 3485 Linux::clock_init(); 3486 initial_time_count = os::elapsed_counter(); 3487} 3488 3489// To install functions for atexit system call 3490extern "C" { 3491 static void perfMemory_exit_helper() { 3492 perfMemory_exit(); 3493 } 3494} 3495 3496// this is called _after_ the global arguments have been parsed 3497jint os::init_2(void) 3498{ 3499 Linux::fast_thread_clock_init(); 3500 3501 // Allocate a single page and mark it as readable for safepoint polling 3502 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3503 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 3504 3505 os::set_polling_page( polling_page ); 3506 3507#ifndef PRODUCT 3508 if(Verbose && PrintMiscellaneous) 3509 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 3510#endif 3511 3512 if (!UseMembar) { 3513 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3514 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); 3515 os::set_memory_serialize_page( mem_serialize_page ); 3516 3517#ifndef PRODUCT 3518 if(Verbose && PrintMiscellaneous) 3519 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 3520#endif 3521 } 3522 3523 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init()); 3524 3525 // initialize suspend/resume support - must do this before signal_sets_init() 3526 if (SR_initialize() != 0) { 3527 perror("SR_initialize failed"); 3528 return JNI_ERR; 3529 } 3530 3531 Linux::signal_sets_init(); 3532 Linux::install_signal_handlers(); 3533 3534 size_t threadStackSizeInBytes = ThreadStackSize * K; 3535 if (threadStackSizeInBytes != 0 && 3536 threadStackSizeInBytes < Linux::min_stack_allowed) { 3537 tty->print_cr("\nThe stack size specified is too small, " 3538 "Specify at least %dk", 3539 Linux::min_stack_allowed / K); 3540 return JNI_ERR; 3541 } 3542 3543 // Make the stack size a multiple of the page size so that 3544 // the yellow/red zones can be guarded. 3545 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 3546 vm_page_size())); 3547 3548 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 3549 3550 Linux::libpthread_init(); 3551 if (PrintMiscellaneous && (Verbose || WizardMode)) { 3552 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 3553 Linux::glibc_version(), Linux::libpthread_version(), 3554 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 3555 } 3556 3557 if (MaxFDLimit) { 3558 // set the number of file descriptors to max. print out error 3559 // if getrlimit/setrlimit fails but continue regardless. 3560 struct rlimit nbr_files; 3561 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 3562 if (status != 0) { 3563 if (PrintMiscellaneous && (Verbose || WizardMode)) 3564 perror("os::init_2 getrlimit failed"); 3565 } else { 3566 nbr_files.rlim_cur = nbr_files.rlim_max; 3567 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 3568 if (status != 0) { 3569 if (PrintMiscellaneous && (Verbose || WizardMode)) 3570 perror("os::init_2 setrlimit failed"); 3571 } 3572 } 3573 } 3574 3575 // Initialize lock used to serialize thread creation (see os::create_thread) 3576 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 3577 3578 // Initialize HPI. 3579 jint hpi_result = hpi::initialize(); 3580 if (hpi_result != JNI_OK) { 3581 tty->print_cr("There was an error trying to initialize the HPI library."); 3582 return hpi_result; 3583 } 3584 3585 // at-exit methods are called in the reverse order of their registration. 3586 // atexit functions are called on return from main or as a result of a 3587 // call to exit(3C). There can be only 32 of these functions registered 3588 // and atexit() does not set errno. 3589 3590 if (PerfAllowAtExitRegistration) { 3591 // only register atexit functions if PerfAllowAtExitRegistration is set. 3592 // atexit functions can be delayed until process exit time, which 3593 // can be problematic for embedded VM situations. Embedded VMs should 3594 // call DestroyJavaVM() to assure that VM resources are released. 3595 3596 // note: perfMemory_exit_helper atexit function may be removed in 3597 // the future if the appropriate cleanup code can be added to the 3598 // VM_Exit VMOperation's doit method. 3599 if (atexit(perfMemory_exit_helper) != 0) { 3600 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 3601 } 3602 } 3603 3604 // initialize thread priority policy 3605 prio_init(); 3606 3607 return JNI_OK; 3608} 3609 3610// Mark the polling page as unreadable 3611void os::make_polling_page_unreadable(void) { 3612 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 3613 fatal("Could not disable polling page"); 3614}; 3615 3616// Mark the polling page as readable 3617void os::make_polling_page_readable(void) { 3618 if( !protect_memory((char *)_polling_page, Linux::page_size()) ) 3619 fatal("Could not enable polling page"); 3620}; 3621 3622int os::active_processor_count() { 3623 // Linux doesn't yet have a (official) notion of processor sets, 3624 // so just return the number of online processors. 3625 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 3626 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 3627 return online_cpus; 3628} 3629 3630bool os::distribute_processes(uint length, uint* distribution) { 3631 // Not yet implemented. 3632 return false; 3633} 3634 3635bool os::bind_to_processor(uint processor_id) { 3636 // Not yet implemented. 3637 return false; 3638} 3639 3640/// 3641 3642// Suspends the target using the signal mechanism and then grabs the PC before 3643// resuming the target. Used by the flat-profiler only 3644ExtendedPC os::get_thread_pc(Thread* thread) { 3645 // Make sure that it is called by the watcher for the VMThread 3646 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 3647 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 3648 3649 ExtendedPC epc; 3650 3651 OSThread* osthread = thread->osthread(); 3652 if (do_suspend(osthread)) { 3653 if (osthread->ucontext() != NULL) { 3654 epc = os::Linux::ucontext_get_pc(osthread->ucontext()); 3655 } else { 3656 // NULL context is unexpected, double-check this is the VMThread 3657 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 3658 } 3659 do_resume(osthread); 3660 } 3661 // failure means pthread_kill failed for some reason - arguably this is 3662 // a fatal problem, but such problems are ignored elsewhere 3663 3664 return epc; 3665} 3666 3667int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 3668{ 3669 if (is_NPTL()) { 3670 return pthread_cond_timedwait(_cond, _mutex, _abstime); 3671 } else { 3672#ifndef IA64 3673 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 3674 // word back to default 64bit precision if condvar is signaled. Java 3675 // wants 53bit precision. Save and restore current value. 3676 int fpu = get_fpu_control_word(); 3677#endif // IA64 3678 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 3679#ifndef IA64 3680 set_fpu_control_word(fpu); 3681#endif // IA64 3682 return status; 3683 } 3684} 3685 3686//////////////////////////////////////////////////////////////////////////////// 3687// debug support 3688 3689#ifndef PRODUCT 3690static address same_page(address x, address y) { 3691 int page_bits = -os::vm_page_size(); 3692 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) 3693 return x; 3694 else if (x > y) 3695 return (address)(intptr_t(y) | ~page_bits) + 1; 3696 else 3697 return (address)(intptr_t(y) & page_bits); 3698} 3699 3700bool os::find(address addr) { 3701 Dl_info dlinfo; 3702 memset(&dlinfo, 0, sizeof(dlinfo)); 3703 if (dladdr(addr, &dlinfo)) { 3704 tty->print(PTR_FORMAT ": ", addr); 3705 if (dlinfo.dli_sname != NULL) { 3706 tty->print("%s+%#x", dlinfo.dli_sname, 3707 addr - (intptr_t)dlinfo.dli_saddr); 3708 } else if (dlinfo.dli_fname) { 3709 tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 3710 } else { 3711 tty->print("<absolute address>"); 3712 } 3713 if (dlinfo.dli_fname) { 3714 tty->print(" in %s", dlinfo.dli_fname); 3715 } 3716 if (dlinfo.dli_fbase) { 3717 tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 3718 } 3719 tty->cr(); 3720 3721 if (Verbose) { 3722 // decode some bytes around the PC 3723 address begin = same_page(addr-40, addr); 3724 address end = same_page(addr+40, addr); 3725 address lowest = (address) dlinfo.dli_sname; 3726 if (!lowest) lowest = (address) dlinfo.dli_fbase; 3727 if (begin < lowest) begin = lowest; 3728 Dl_info dlinfo2; 3729 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr 3730 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 3731 end = (address) dlinfo2.dli_saddr; 3732 Disassembler::decode(begin, end); 3733 } 3734 return true; 3735 } 3736 return false; 3737} 3738 3739#endif 3740 3741//////////////////////////////////////////////////////////////////////////////// 3742// misc 3743 3744// This does not do anything on Linux. This is basically a hook for being 3745// able to use structured exception handling (thread-local exception filters) 3746// on, e.g., Win32. 3747void 3748os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 3749 JavaCallArguments* args, Thread* thread) { 3750 f(value, method, args, thread); 3751} 3752 3753void os::print_statistics() { 3754} 3755 3756int os::message_box(const char* title, const char* message) { 3757 int i; 3758 fdStream err(defaultStream::error_fd()); 3759 for (i = 0; i < 78; i++) err.print_raw("="); 3760 err.cr(); 3761 err.print_raw_cr(title); 3762 for (i = 0; i < 78; i++) err.print_raw("-"); 3763 err.cr(); 3764 err.print_raw_cr(message); 3765 for (i = 0; i < 78; i++) err.print_raw("="); 3766 err.cr(); 3767 3768 char buf[16]; 3769 // Prevent process from exiting upon "read error" without consuming all CPU 3770 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 3771 3772 return buf[0] == 'y' || buf[0] == 'Y'; 3773} 3774 3775int os::stat(const char *path, struct stat *sbuf) { 3776 char pathbuf[MAX_PATH]; 3777 if (strlen(path) > MAX_PATH - 1) { 3778 errno = ENAMETOOLONG; 3779 return -1; 3780 } 3781 hpi::native_path(strcpy(pathbuf, path)); 3782 return ::stat(pathbuf, sbuf); 3783} 3784 3785bool os::check_heap(bool force) { 3786 return true; 3787} 3788 3789int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 3790 return ::vsnprintf(buf, count, format, args); 3791} 3792 3793// Is a (classpath) directory empty? 3794bool os::dir_is_empty(const char* path) { 3795 DIR *dir = NULL; 3796 struct dirent *ptr; 3797 3798 dir = opendir(path); 3799 if (dir == NULL) return true; 3800 3801 /* Scan the directory */ 3802 bool result = true; 3803 char buf[sizeof(struct dirent) + MAX_PATH]; 3804 while (result && (ptr = ::readdir(dir)) != NULL) { 3805 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 3806 result = false; 3807 } 3808 } 3809 closedir(dir); 3810 return result; 3811} 3812 3813// create binary file, rewriting existing file if required 3814int os::create_binary_file(const char* path, bool rewrite_existing) { 3815 int oflags = O_WRONLY | O_CREAT; 3816 if (!rewrite_existing) { 3817 oflags |= O_EXCL; 3818 } 3819 return ::open64(path, oflags, S_IREAD | S_IWRITE); 3820} 3821 3822// return current position of file pointer 3823jlong os::current_file_offset(int fd) { 3824 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 3825} 3826 3827// move file pointer to the specified offset 3828jlong os::seek_to_file_offset(int fd, jlong offset) { 3829 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 3830} 3831 3832// Map a block of memory. 3833char* os::map_memory(int fd, const char* file_name, size_t file_offset, 3834 char *addr, size_t bytes, bool read_only, 3835 bool allow_exec) { 3836 int prot; 3837 int flags; 3838 3839 if (read_only) { 3840 prot = PROT_READ; 3841 flags = MAP_SHARED; 3842 } else { 3843 prot = PROT_READ | PROT_WRITE; 3844 flags = MAP_PRIVATE; 3845 } 3846 3847 if (allow_exec) { 3848 prot |= PROT_EXEC; 3849 } 3850 3851 if (addr != NULL) { 3852 flags |= MAP_FIXED; 3853 } 3854 3855 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 3856 fd, file_offset); 3857 if (mapped_address == MAP_FAILED) { 3858 return NULL; 3859 } 3860 return mapped_address; 3861} 3862 3863 3864// Remap a block of memory. 3865char* os::remap_memory(int fd, const char* file_name, size_t file_offset, 3866 char *addr, size_t bytes, bool read_only, 3867 bool allow_exec) { 3868 // same as map_memory() on this OS 3869 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 3870 allow_exec); 3871} 3872 3873 3874// Unmap a block of memory. 3875bool os::unmap_memory(char* addr, size_t bytes) { 3876 return munmap(addr, bytes) == 0; 3877} 3878 3879static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 3880 3881static clockid_t thread_cpu_clockid(Thread* thread) { 3882 pthread_t tid = thread->osthread()->pthread_id(); 3883 clockid_t clockid; 3884 3885 // Get thread clockid 3886 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 3887 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 3888 return clockid; 3889} 3890 3891// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 3892// are used by JVM M&M and JVMTI to get user+sys or user CPU time 3893// of a thread. 3894// 3895// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 3896// the fast estimate available on the platform. 3897 3898jlong os::current_thread_cpu_time() { 3899 if (os::Linux::supports_fast_thread_cpu_time()) { 3900 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 3901 } else { 3902 // return user + sys since the cost is the same 3903 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 3904 } 3905} 3906 3907jlong os::thread_cpu_time(Thread* thread) { 3908 // consistent with what current_thread_cpu_time() returns 3909 if (os::Linux::supports_fast_thread_cpu_time()) { 3910 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 3911 } else { 3912 return slow_thread_cpu_time(thread, true /* user + sys */); 3913 } 3914} 3915 3916jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 3917 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 3918 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 3919 } else { 3920 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 3921 } 3922} 3923 3924jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 3925 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 3926 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 3927 } else { 3928 return slow_thread_cpu_time(thread, user_sys_cpu_time); 3929 } 3930} 3931 3932// 3933// -1 on error. 3934// 3935 3936static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 3937 static bool proc_pid_cpu_avail = true; 3938 static bool proc_task_unchecked = true; 3939 static const char *proc_stat_path = "/proc/%d/stat"; 3940 pid_t tid = thread->osthread()->thread_id(); 3941 int i; 3942 char *s; 3943 char stat[2048]; 3944 int statlen; 3945 char proc_name[64]; 3946 int count; 3947 long sys_time, user_time; 3948 char string[64]; 3949 int idummy; 3950 long ldummy; 3951 FILE *fp; 3952 3953 // We first try accessing /proc/<pid>/cpu since this is faster to 3954 // process. If this file is not present (linux kernels 2.5 and above) 3955 // then we open /proc/<pid>/stat. 3956 if ( proc_pid_cpu_avail ) { 3957 sprintf(proc_name, "/proc/%d/cpu", tid); 3958 fp = fopen(proc_name, "r"); 3959 if ( fp != NULL ) { 3960 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time); 3961 fclose(fp); 3962 if ( count != 3 ) return -1; 3963 3964 if (user_sys_cpu_time) { 3965 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 3966 } else { 3967 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 3968 } 3969 } 3970 else proc_pid_cpu_avail = false; 3971 } 3972 3973 // The /proc/<tid>/stat aggregates per-process usage on 3974 // new Linux kernels 2.6+ where NPTL is supported. 3975 // The /proc/self/task/<tid>/stat still has the per-thread usage. 3976 // See bug 6328462. 3977 // There can be no directory /proc/self/task on kernels 2.4 with NPTL 3978 // and possibly in some other cases, so we check its availability. 3979 if (proc_task_unchecked && os::Linux::is_NPTL()) { 3980 // This is executed only once 3981 proc_task_unchecked = false; 3982 fp = fopen("/proc/self/task", "r"); 3983 if (fp != NULL) { 3984 proc_stat_path = "/proc/self/task/%d/stat"; 3985 fclose(fp); 3986 } 3987 } 3988 3989 sprintf(proc_name, proc_stat_path, tid); 3990 fp = fopen(proc_name, "r"); 3991 if ( fp == NULL ) return -1; 3992 statlen = fread(stat, 1, 2047, fp); 3993 stat[statlen] = '\0'; 3994 fclose(fp); 3995 3996 // Skip pid and the command string. Note that we could be dealing with 3997 // weird command names, e.g. user could decide to rename java launcher 3998 // to "java 1.4.2 :)", then the stat file would look like 3999 // 1234 (java 1.4.2 :)) R ... ... 4000 // We don't really need to know the command string, just find the last 4001 // occurrence of ")" and then start parsing from there. See bug 4726580. 4002 s = strrchr(stat, ')'); 4003 i = 0; 4004 if (s == NULL ) return -1; 4005 4006 // Skip blank chars 4007 do s++; while (isspace(*s)); 4008 4009 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 4010 &idummy, &idummy, &idummy, &idummy, &idummy, &idummy, 4011 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 4012 &user_time, &sys_time); 4013 if ( count != 13 ) return -1; 4014 if (user_sys_cpu_time) { 4015 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4016 } else { 4017 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4018 } 4019} 4020 4021void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4022 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4023 info_ptr->may_skip_backward = false; // elapsed time not wall time 4024 info_ptr->may_skip_forward = false; // elapsed time not wall time 4025 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4026} 4027 4028void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4029 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4030 info_ptr->may_skip_backward = false; // elapsed time not wall time 4031 info_ptr->may_skip_forward = false; // elapsed time not wall time 4032 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4033} 4034 4035bool os::is_thread_cpu_time_supported() { 4036 return true; 4037} 4038 4039// System loadavg support. Returns -1 if load average cannot be obtained. 4040// Linux doesn't yet have a (official) notion of processor sets, 4041// so just return the system wide load average. 4042int os::loadavg(double loadavg[], int nelem) { 4043 return ::getloadavg(loadavg, nelem); 4044} 4045 4046void os::pause() { 4047 char filename[MAX_PATH]; 4048 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 4049 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 4050 } else { 4051 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 4052 } 4053 4054 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 4055 if (fd != -1) { 4056 struct stat buf; 4057 close(fd); 4058 while (::stat(filename, &buf) == 0) { 4059 (void)::poll(NULL, 0, 100); 4060 } 4061 } else { 4062 jio_fprintf(stderr, 4063 "Could not open pause file '%s', continuing immediately.\n", filename); 4064 } 4065} 4066 4067extern "C" { 4068 4069/** 4070 * NOTE: the following code is to keep the green threads code 4071 * in the libjava.so happy. Once the green threads is removed, 4072 * these code will no longer be needed. 4073 */ 4074int 4075jdk_waitpid(pid_t pid, int* status, int options) { 4076 return waitpid(pid, status, options); 4077} 4078 4079int 4080fork1() { 4081 return fork(); 4082} 4083 4084int 4085jdk_sem_init(sem_t *sem, int pshared, unsigned int value) { 4086 return sem_init(sem, pshared, value); 4087} 4088 4089int 4090jdk_sem_post(sem_t *sem) { 4091 return sem_post(sem); 4092} 4093 4094int 4095jdk_sem_wait(sem_t *sem) { 4096 return sem_wait(sem); 4097} 4098 4099int 4100jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) { 4101 return pthread_sigmask(how , newmask, oldmask); 4102} 4103 4104} 4105 4106// Refer to the comments in os_solaris.cpp park-unpark. 4107// 4108// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 4109// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 4110// For specifics regarding the bug see GLIBC BUGID 261237 : 4111// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 4112// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 4113// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 4114// is used. (The simple C test-case provided in the GLIBC bug report manifests the 4115// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 4116// and monitorenter when we're using 1-0 locking. All those operations may result in 4117// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 4118// of libpthread avoids the problem, but isn't practical. 4119// 4120// Possible remedies: 4121// 4122// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 4123// This is palliative and probabilistic, however. If the thread is preempted 4124// between the call to compute_abstime() and pthread_cond_timedwait(), more 4125// than the minimum period may have passed, and the abstime may be stale (in the 4126// past) resultin in a hang. Using this technique reduces the odds of a hang 4127// but the JVM is still vulnerable, particularly on heavily loaded systems. 4128// 4129// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 4130// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 4131// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 4132// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 4133// thread. 4134// 4135// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 4136// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 4137// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 4138// This also works well. In fact it avoids kernel-level scalability impediments 4139// on certain platforms that don't handle lots of active pthread_cond_timedwait() 4140// timers in a graceful fashion. 4141// 4142// 4. When the abstime value is in the past it appears that control returns 4143// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 4144// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 4145// can avoid the problem by reinitializing the condvar -- by cond_destroy() 4146// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 4147// It may be possible to avoid reinitialization by checking the return 4148// value from pthread_cond_timedwait(). In addition to reinitializing the 4149// condvar we must establish the invariant that cond_signal() is only called 4150// within critical sections protected by the adjunct mutex. This prevents 4151// cond_signal() from "seeing" a condvar that's in the midst of being 4152// reinitialized or that is corrupt. Sadly, this invariant obviates the 4153// desirable signal-after-unlock optimization that avoids futile context switching. 4154// 4155// I'm also concerned that some versions of NTPL might allocate an auxilliary 4156// structure when a condvar is used or initialized. cond_destroy() would 4157// release the helper structure. Our reinitialize-after-timedwait fix 4158// put excessive stress on malloc/free and locks protecting the c-heap. 4159// 4160// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 4161// It may be possible to refine (4) by checking the kernel and NTPL verisons 4162// and only enabling the work-around for vulnerable environments. 4163 4164// utility to compute the abstime argument to timedwait: 4165// millis is the relative timeout time 4166// abstime will be the absolute timeout time 4167// TODO: replace compute_abstime() with unpackTime() 4168 4169static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 4170 if (millis < 0) millis = 0; 4171 struct timeval now; 4172 int status = gettimeofday(&now, NULL); 4173 assert(status == 0, "gettimeofday"); 4174 jlong seconds = millis / 1000; 4175 millis %= 1000; 4176 if (seconds > 50000000) { // see man cond_timedwait(3T) 4177 seconds = 50000000; 4178 } 4179 abstime->tv_sec = now.tv_sec + seconds; 4180 long usec = now.tv_usec + millis * 1000; 4181 if (usec >= 1000000) { 4182 abstime->tv_sec += 1; 4183 usec -= 1000000; 4184 } 4185 abstime->tv_nsec = usec * 1000; 4186 return abstime; 4187} 4188 4189 4190// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 4191// Conceptually TryPark() should be equivalent to park(0). 4192 4193int os::PlatformEvent::TryPark() { 4194 for (;;) { 4195 const int v = _Event ; 4196 guarantee ((v == 0) || (v == 1), "invariant") ; 4197 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 4198 } 4199} 4200 4201void os::PlatformEvent::park() { // AKA "down()" 4202 // Invariant: Only the thread associated with the Event/PlatformEvent 4203 // may call park(). 4204 // TODO: assert that _Assoc != NULL or _Assoc == Self 4205 int v ; 4206 for (;;) { 4207 v = _Event ; 4208 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4209 } 4210 guarantee (v >= 0, "invariant") ; 4211 if (v == 0) { 4212 // Do this the hard way by blocking ... 4213 int status = pthread_mutex_lock(_mutex); 4214 assert_status(status == 0, status, "mutex_lock"); 4215 guarantee (_nParked == 0, "invariant") ; 4216 ++ _nParked ; 4217 while (_Event < 0) { 4218 status = pthread_cond_wait(_cond, _mutex); 4219 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 4220 // Treat this the same as if the wait was interrupted 4221 if (status == ETIME) { status = EINTR; } 4222 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 4223 } 4224 -- _nParked ; 4225 4226 // In theory we could move the ST of 0 into _Event past the unlock(), 4227 // but then we'd need a MEMBAR after the ST. 4228 _Event = 0 ; 4229 status = pthread_mutex_unlock(_mutex); 4230 assert_status(status == 0, status, "mutex_unlock"); 4231 } 4232 guarantee (_Event >= 0, "invariant") ; 4233} 4234 4235int os::PlatformEvent::park(jlong millis) { 4236 guarantee (_nParked == 0, "invariant") ; 4237 4238 int v ; 4239 for (;;) { 4240 v = _Event ; 4241 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4242 } 4243 guarantee (v >= 0, "invariant") ; 4244 if (v != 0) return OS_OK ; 4245 4246 // We do this the hard way, by blocking the thread. 4247 // Consider enforcing a minimum timeout value. 4248 struct timespec abst; 4249 compute_abstime(&abst, millis); 4250 4251 int ret = OS_TIMEOUT; 4252 int status = pthread_mutex_lock(_mutex); 4253 assert_status(status == 0, status, "mutex_lock"); 4254 guarantee (_nParked == 0, "invariant") ; 4255 ++_nParked ; 4256 4257 // Object.wait(timo) will return because of 4258 // (a) notification 4259 // (b) timeout 4260 // (c) thread.interrupt 4261 // 4262 // Thread.interrupt and object.notify{All} both call Event::set. 4263 // That is, we treat thread.interrupt as a special case of notification. 4264 // The underlying Solaris implementation, cond_timedwait, admits 4265 // spurious/premature wakeups, but the JLS/JVM spec prevents the 4266 // JVM from making those visible to Java code. As such, we must 4267 // filter out spurious wakeups. We assume all ETIME returns are valid. 4268 // 4269 // TODO: properly differentiate simultaneous notify+interrupt. 4270 // In that case, we should propagate the notify to another waiter. 4271 4272 while (_Event < 0) { 4273 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 4274 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4275 pthread_cond_destroy (_cond); 4276 pthread_cond_init (_cond, NULL) ; 4277 } 4278 assert_status(status == 0 || status == EINTR || 4279 status == ETIME || status == ETIMEDOUT, 4280 status, "cond_timedwait"); 4281 if (!FilterSpuriousWakeups) break ; // previous semantics 4282 if (status == ETIME || status == ETIMEDOUT) break ; 4283 // We consume and ignore EINTR and spurious wakeups. 4284 } 4285 --_nParked ; 4286 if (_Event >= 0) { 4287 ret = OS_OK; 4288 } 4289 _Event = 0 ; 4290 status = pthread_mutex_unlock(_mutex); 4291 assert_status(status == 0, status, "mutex_unlock"); 4292 assert (_nParked == 0, "invariant") ; 4293 return ret; 4294} 4295 4296void os::PlatformEvent::unpark() { 4297 int v, AnyWaiters ; 4298 for (;;) { 4299 v = _Event ; 4300 if (v > 0) { 4301 // The LD of _Event could have reordered or be satisfied 4302 // by a read-aside from this processor's write buffer. 4303 // To avoid problems execute a barrier and then 4304 // ratify the value. 4305 OrderAccess::fence() ; 4306 if (_Event == v) return ; 4307 continue ; 4308 } 4309 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ; 4310 } 4311 if (v < 0) { 4312 // Wait for the thread associated with the event to vacate 4313 int status = pthread_mutex_lock(_mutex); 4314 assert_status(status == 0, status, "mutex_lock"); 4315 AnyWaiters = _nParked ; 4316 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ; 4317 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 4318 AnyWaiters = 0 ; 4319 pthread_cond_signal (_cond); 4320 } 4321 status = pthread_mutex_unlock(_mutex); 4322 assert_status(status == 0, status, "mutex_unlock"); 4323 if (AnyWaiters != 0) { 4324 status = pthread_cond_signal(_cond); 4325 assert_status(status == 0, status, "cond_signal"); 4326 } 4327 } 4328 4329 // Note that we signal() _after dropping the lock for "immortal" Events. 4330 // This is safe and avoids a common class of futile wakeups. In rare 4331 // circumstances this can cause a thread to return prematurely from 4332 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 4333 // simply re-test the condition and re-park itself. 4334} 4335 4336 4337// JSR166 4338// ------------------------------------------------------- 4339 4340/* 4341 * The solaris and linux implementations of park/unpark are fairly 4342 * conservative for now, but can be improved. They currently use a 4343 * mutex/condvar pair, plus a a count. 4344 * Park decrements count if > 0, else does a condvar wait. Unpark 4345 * sets count to 1 and signals condvar. Only one thread ever waits 4346 * on the condvar. Contention seen when trying to park implies that someone 4347 * is unparking you, so don't wait. And spurious returns are fine, so there 4348 * is no need to track notifications. 4349 */ 4350 4351 4352#define NANOSECS_PER_SEC 1000000000 4353#define NANOSECS_PER_MILLISEC 1000000 4354#define MAX_SECS 100000000 4355/* 4356 * This code is common to linux and solaris and will be moved to a 4357 * common place in dolphin. 4358 * 4359 * The passed in time value is either a relative time in nanoseconds 4360 * or an absolute time in milliseconds. Either way it has to be unpacked 4361 * into suitable seconds and nanoseconds components and stored in the 4362 * given timespec structure. 4363 * Given time is a 64-bit value and the time_t used in the timespec is only 4364 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 4365 * overflow if times way in the future are given. Further on Solaris versions 4366 * prior to 10 there is a restriction (see cond_timedwait) that the specified 4367 * number of seconds, in abstime, is less than current_time + 100,000,000. 4368 * As it will be 28 years before "now + 100000000" will overflow we can 4369 * ignore overflow and just impose a hard-limit on seconds using the value 4370 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 4371 * years from "now". 4372 */ 4373 4374static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 4375 assert (time > 0, "convertTime"); 4376 4377 struct timeval now; 4378 int status = gettimeofday(&now, NULL); 4379 assert(status == 0, "gettimeofday"); 4380 4381 time_t max_secs = now.tv_sec + MAX_SECS; 4382 4383 if (isAbsolute) { 4384 jlong secs = time / 1000; 4385 if (secs > max_secs) { 4386 absTime->tv_sec = max_secs; 4387 } 4388 else { 4389 absTime->tv_sec = secs; 4390 } 4391 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 4392 } 4393 else { 4394 jlong secs = time / NANOSECS_PER_SEC; 4395 if (secs >= MAX_SECS) { 4396 absTime->tv_sec = max_secs; 4397 absTime->tv_nsec = 0; 4398 } 4399 else { 4400 absTime->tv_sec = now.tv_sec + secs; 4401 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 4402 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 4403 absTime->tv_nsec -= NANOSECS_PER_SEC; 4404 ++absTime->tv_sec; // note: this must be <= max_secs 4405 } 4406 } 4407 } 4408 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 4409 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 4410 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 4411 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 4412} 4413 4414void Parker::park(bool isAbsolute, jlong time) { 4415 // Optional fast-path check: 4416 // Return immediately if a permit is available. 4417 if (_counter > 0) { 4418 _counter = 0 ; 4419 return ; 4420 } 4421 4422 Thread* thread = Thread::current(); 4423 assert(thread->is_Java_thread(), "Must be JavaThread"); 4424 JavaThread *jt = (JavaThread *)thread; 4425 4426 // Optional optimization -- avoid state transitions if there's an interrupt pending. 4427 // Check interrupt before trying to wait 4428 if (Thread::is_interrupted(thread, false)) { 4429 return; 4430 } 4431 4432 // Next, demultiplex/decode time arguments 4433 timespec absTime; 4434 if (time < 0) { // don't wait at all 4435 return; 4436 } 4437 if (time > 0) { 4438 unpackTime(&absTime, isAbsolute, time); 4439 } 4440 4441 4442 // Enter safepoint region 4443 // Beware of deadlocks such as 6317397. 4444 // The per-thread Parker:: mutex is a classic leaf-lock. 4445 // In particular a thread must never block on the Threads_lock while 4446 // holding the Parker:: mutex. If safepoints are pending both the 4447 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 4448 ThreadBlockInVM tbivm(jt); 4449 4450 // Don't wait if cannot get lock since interference arises from 4451 // unblocking. Also. check interrupt before trying wait 4452 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 4453 return; 4454 } 4455 4456 int status ; 4457 if (_counter > 0) { // no wait needed 4458 _counter = 0; 4459 status = pthread_mutex_unlock(_mutex); 4460 assert (status == 0, "invariant") ; 4461 return; 4462 } 4463 4464#ifdef ASSERT 4465 // Don't catch signals while blocked; let the running threads have the signals. 4466 // (This allows a debugger to break into the running thread.) 4467 sigset_t oldsigs; 4468 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 4469 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 4470#endif 4471 4472 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 4473 jt->set_suspend_equivalent(); 4474 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 4475 4476 if (time == 0) { 4477 status = pthread_cond_wait (_cond, _mutex) ; 4478 } else { 4479 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 4480 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4481 pthread_cond_destroy (_cond) ; 4482 pthread_cond_init (_cond, NULL); 4483 } 4484 } 4485 assert_status(status == 0 || status == EINTR || 4486 status == ETIME || status == ETIMEDOUT, 4487 status, "cond_timedwait"); 4488 4489#ifdef ASSERT 4490 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 4491#endif 4492 4493 _counter = 0 ; 4494 status = pthread_mutex_unlock(_mutex) ; 4495 assert_status(status == 0, status, "invariant") ; 4496 // If externally suspended while waiting, re-suspend 4497 if (jt->handle_special_suspend_equivalent_condition()) { 4498 jt->java_suspend_self(); 4499 } 4500 4501} 4502 4503void Parker::unpark() { 4504 int s, status ; 4505 status = pthread_mutex_lock(_mutex); 4506 assert (status == 0, "invariant") ; 4507 s = _counter; 4508 _counter = 1; 4509 if (s < 1) { 4510 if (WorkAroundNPTLTimedWaitHang) { 4511 status = pthread_cond_signal (_cond) ; 4512 assert (status == 0, "invariant") ; 4513 status = pthread_mutex_unlock(_mutex); 4514 assert (status == 0, "invariant") ; 4515 } else { 4516 status = pthread_mutex_unlock(_mutex); 4517 assert (status == 0, "invariant") ; 4518 status = pthread_cond_signal (_cond) ; 4519 assert (status == 0, "invariant") ; 4520 } 4521 } else { 4522 pthread_mutex_unlock(_mutex); 4523 assert (status == 0, "invariant") ; 4524 } 4525} 4526 4527 4528extern char** environ; 4529 4530#ifndef __NR_fork 4531#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 4532#endif 4533 4534#ifndef __NR_execve 4535#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 4536#endif 4537 4538// Run the specified command in a separate process. Return its exit value, 4539// or -1 on failure (e.g. can't fork a new process). 4540// Unlike system(), this function can be called from signal handler. It 4541// doesn't block SIGINT et al. 4542int os::fork_and_exec(char* cmd) { 4543 char * argv[4]; 4544 argv[0] = "sh"; 4545 argv[1] = "-c"; 4546 argv[2] = cmd; 4547 argv[3] = NULL; 4548 4549 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 4550 // pthread_atfork handlers and reset pthread library. All we need is a 4551 // separate process to execve. Make a direct syscall to fork process. 4552 // On IA64 there's no fork syscall, we have to use fork() and hope for 4553 // the best... 4554 pid_t pid = NOT_IA64(syscall(__NR_fork);) 4555 IA64_ONLY(fork();) 4556 4557 if (pid < 0) { 4558 // fork failed 4559 return -1; 4560 4561 } else if (pid == 0) { 4562 // child process 4563 4564 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 4565 // first to kill every thread on the thread list. Because this list is 4566 // not reset by fork() (see notes above), execve() will instead kill 4567 // every thread in the parent process. We know this is the only thread 4568 // in the new process, so make a system call directly. 4569 // IA64 should use normal execve() from glibc to match the glibc fork() 4570 // above. 4571 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 4572 IA64_ONLY(execve("/bin/sh", argv, environ);) 4573 4574 // execve failed 4575 _exit(-1); 4576 4577 } else { 4578 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 4579 // care about the actual exit code, for now. 4580 4581 int status; 4582 4583 // Wait for the child process to exit. This returns immediately if 4584 // the child has already exited. */ 4585 while (waitpid(pid, &status, 0) < 0) { 4586 switch (errno) { 4587 case ECHILD: return 0; 4588 case EINTR: break; 4589 default: return -1; 4590 } 4591 } 4592 4593 if (WIFEXITED(status)) { 4594 // The child exited normally; get its exit code. 4595 return WEXITSTATUS(status); 4596 } else if (WIFSIGNALED(status)) { 4597 // The child exited because of a signal 4598 // The best value to return is 0x80 + signal number, 4599 // because that is what all Unix shells do, and because 4600 // it allows callers to distinguish between process exit and 4601 // process death by signal. 4602 return 0x80 + WTERMSIG(status); 4603 } else { 4604 // Unknown exit code; pass it through 4605 return status; 4606 } 4607 } 4608} 4609