os_linux.cpp revision 691:956304450e80
1/* 2 * Copyright 1999-2009 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; 78const char * os::Linux::_glibc_version = NULL; 79const char * 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/* Used to protect dlsym() calls */ 98static pthread_mutex_t dl_mutex; 99 100//////////////////////////////////////////////////////////////////////////////// 101// utility functions 102 103static int SR_initialize(); 104static int SR_finalize(); 105 106julong os::available_memory() { 107 return Linux::available_memory(); 108} 109 110julong os::Linux::available_memory() { 111 // values in struct sysinfo are "unsigned long" 112 struct sysinfo si; 113 sysinfo(&si); 114 115 return (julong)si.freeram * si.mem_unit; 116} 117 118julong os::physical_memory() { 119 return Linux::physical_memory(); 120} 121 122julong os::allocatable_physical_memory(julong size) { 123#ifdef _LP64 124 return size; 125#else 126 julong result = MIN2(size, (julong)3800*M); 127 if (!is_allocatable(result)) { 128 // See comments under solaris for alignment considerations 129 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); 130 result = MIN2(size, reasonable_size); 131 } 132 return result; 133#endif // _LP64 134} 135 136//////////////////////////////////////////////////////////////////////////////// 137// environment support 138 139bool os::getenv(const char* name, char* buf, int len) { 140 const char* val = ::getenv(name); 141 if (val != NULL && strlen(val) < (size_t)len) { 142 strcpy(buf, val); 143 return true; 144 } 145 if (len > 0) buf[0] = 0; // return a null string 146 return false; 147} 148 149 150// Return true if user is running as root. 151 152bool os::have_special_privileges() { 153 static bool init = false; 154 static bool privileges = false; 155 if (!init) { 156 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 157 init = true; 158 } 159 return privileges; 160} 161 162 163#ifndef SYS_gettid 164// i386: 224, ia64: 1105, amd64: 186, sparc 143 165#ifdef __ia64__ 166#define SYS_gettid 1105 167#elif __i386__ 168#define SYS_gettid 224 169#elif __amd64__ 170#define SYS_gettid 186 171#elif __sparc__ 172#define SYS_gettid 143 173#else 174#error define gettid for the arch 175#endif 176#endif 177 178// Cpu architecture string 179#if defined(IA64) 180static char cpu_arch[] = "ia64"; 181#elif defined(IA32) 182static char cpu_arch[] = "i386"; 183#elif defined(AMD64) 184static char cpu_arch[] = "amd64"; 185#elif defined(SPARC) 186# ifdef _LP64 187static char cpu_arch[] = "sparcv9"; 188# else 189static char cpu_arch[] = "sparc"; 190# endif 191#else 192#error Add appropriate cpu_arch setting 193#endif 194 195 196// pid_t gettid() 197// 198// Returns the kernel thread id of the currently running thread. Kernel 199// thread id is used to access /proc. 200// 201// (Note that getpid() on LinuxThreads returns kernel thread id too; but 202// on NPTL, it returns the same pid for all threads, as required by POSIX.) 203// 204pid_t os::Linux::gettid() { 205 int rslt = syscall(SYS_gettid); 206 if (rslt == -1) { 207 // old kernel, no NPTL support 208 return getpid(); 209 } else { 210 return (pid_t)rslt; 211 } 212} 213 214// Most versions of linux have a bug where the number of processors are 215// determined by looking at the /proc file system. In a chroot environment, 216// the system call returns 1. This causes the VM to act as if it is 217// a single processor and elide locking (see is_MP() call). 218static bool unsafe_chroot_detected = false; 219static const char *unstable_chroot_error = "/proc file system not found.\n" 220 "Java may be unstable running multithreaded in a chroot " 221 "environment on Linux when /proc filesystem is not mounted."; 222 223void os::Linux::initialize_system_info() { 224 _processor_count = sysconf(_SC_NPROCESSORS_CONF); 225 if (_processor_count == 1) { 226 pid_t pid = os::Linux::gettid(); 227 char fname[32]; 228 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 229 FILE *fp = fopen(fname, "r"); 230 if (fp == NULL) { 231 unsafe_chroot_detected = true; 232 } else { 233 fclose(fp); 234 } 235 } 236 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 237 assert(_processor_count > 0, "linux error"); 238} 239 240void os::init_system_properties_values() { 241// char arch[12]; 242// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 243 244 // The next steps are taken in the product version: 245 // 246 // Obtain the JAVA_HOME value from the location of libjvm[_g].so. 247 // This library should be located at: 248 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so. 249 // 250 // If "/jre/lib/" appears at the right place in the path, then we 251 // assume libjvm[_g].so is installed in a JDK and we use this path. 252 // 253 // Otherwise exit with message: "Could not create the Java virtual machine." 254 // 255 // The following extra steps are taken in the debugging version: 256 // 257 // If "/jre/lib/" does NOT appear at the right place in the path 258 // instead of exit check for $JAVA_HOME environment variable. 259 // 260 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 261 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so 262 // it looks like libjvm[_g].so is installed there 263 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so. 264 // 265 // Otherwise exit. 266 // 267 // Important note: if the location of libjvm.so changes this 268 // code needs to be changed accordingly. 269 270 // The next few definitions allow the code to be verbatim: 271#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n)) 272#define getenv(n) ::getenv(n) 273 274/* 275 * See ld(1): 276 * The linker uses the following search paths to locate required 277 * shared libraries: 278 * 1: ... 279 * ... 280 * 7: The default directories, normally /lib and /usr/lib. 281 */ 282#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 283#define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 284#else 285#define DEFAULT_LIBPATH "/lib:/usr/lib" 286#endif 287 288#define EXTENSIONS_DIR "/lib/ext" 289#define ENDORSED_DIR "/lib/endorsed" 290#define REG_DIR "/usr/java/packages" 291 292 { 293 /* sysclasspath, java_home, dll_dir */ 294 { 295 char *home_path; 296 char *dll_path; 297 char *pslash; 298 char buf[MAXPATHLEN]; 299 os::jvm_path(buf, sizeof(buf)); 300 301 // Found the full path to libjvm.so. 302 // Now cut the path to <java_home>/jre if we can. 303 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 304 pslash = strrchr(buf, '/'); 305 if (pslash != NULL) 306 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 307 dll_path = malloc(strlen(buf) + 1); 308 if (dll_path == NULL) 309 return; 310 strcpy(dll_path, buf); 311 Arguments::set_dll_dir(dll_path); 312 313 if (pslash != NULL) { 314 pslash = strrchr(buf, '/'); 315 if (pslash != NULL) { 316 *pslash = '\0'; /* get rid of /<arch> */ 317 pslash = strrchr(buf, '/'); 318 if (pslash != NULL) 319 *pslash = '\0'; /* get rid of /lib */ 320 } 321 } 322 323 home_path = malloc(strlen(buf) + 1); 324 if (home_path == NULL) 325 return; 326 strcpy(home_path, buf); 327 Arguments::set_java_home(home_path); 328 329 if (!set_boot_path('/', ':')) 330 return; 331 } 332 333 /* 334 * Where to look for native libraries 335 * 336 * Note: Due to a legacy implementation, most of the library path 337 * is set in the launcher. This was to accomodate linking restrictions 338 * on legacy Linux implementations (which are no longer supported). 339 * Eventually, all the library path setting will be done here. 340 * 341 * However, to prevent the proliferation of improperly built native 342 * libraries, the new path component /usr/java/packages is added here. 343 * Eventually, all the library path setting will be done here. 344 */ 345 { 346 char *ld_library_path; 347 348 /* 349 * Construct the invariant part of ld_library_path. Note that the 350 * space for the colon and the trailing null are provided by the 351 * nulls included by the sizeof operator (so actually we allocate 352 * a byte more than necessary). 353 */ 354 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 355 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 356 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 357 358 /* 359 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 360 * should always exist (until the legacy problem cited above is 361 * addressed). 362 */ 363 char *v = getenv("LD_LIBRARY_PATH"); 364 if (v != NULL) { 365 char *t = ld_library_path; 366 /* That's +1 for the colon and +1 for the trailing '\0' */ 367 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 368 sprintf(ld_library_path, "%s:%s", v, t); 369 } 370 Arguments::set_library_path(ld_library_path); 371 } 372 373 /* 374 * Extensions directories. 375 * 376 * Note that the space for the colon and the trailing null are provided 377 * by the nulls included by the sizeof operator (so actually one byte more 378 * than necessary is allocated). 379 */ 380 { 381 char *buf = malloc(strlen(Arguments::get_java_home()) + 382 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 383 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 384 Arguments::get_java_home()); 385 Arguments::set_ext_dirs(buf); 386 } 387 388 /* Endorsed standards default directory. */ 389 { 390 char * buf; 391 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 392 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 393 Arguments::set_endorsed_dirs(buf); 394 } 395 } 396 397#undef malloc 398#undef getenv 399#undef EXTENSIONS_DIR 400#undef ENDORSED_DIR 401 402 // Done 403 return; 404} 405 406//////////////////////////////////////////////////////////////////////////////// 407// breakpoint support 408 409void os::breakpoint() { 410 BREAKPOINT; 411} 412 413extern "C" void breakpoint() { 414 // use debugger to set breakpoint here 415} 416 417//////////////////////////////////////////////////////////////////////////////// 418// signal support 419 420debug_only(static bool signal_sets_initialized = false); 421static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 422 423bool os::Linux::is_sig_ignored(int sig) { 424 struct sigaction oact; 425 sigaction(sig, (struct sigaction*)NULL, &oact); 426 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 427 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 428 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 429 return true; 430 else 431 return false; 432} 433 434void os::Linux::signal_sets_init() { 435 // Should also have an assertion stating we are still single-threaded. 436 assert(!signal_sets_initialized, "Already initialized"); 437 // Fill in signals that are necessarily unblocked for all threads in 438 // the VM. Currently, we unblock the following signals: 439 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 440 // by -Xrs (=ReduceSignalUsage)); 441 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 442 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 443 // the dispositions or masks wrt these signals. 444 // Programs embedding the VM that want to use the above signals for their 445 // own purposes must, at this time, use the "-Xrs" option to prevent 446 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 447 // (See bug 4345157, and other related bugs). 448 // In reality, though, unblocking these signals is really a nop, since 449 // these signals are not blocked by default. 450 sigemptyset(&unblocked_sigs); 451 sigemptyset(&allowdebug_blocked_sigs); 452 sigaddset(&unblocked_sigs, SIGILL); 453 sigaddset(&unblocked_sigs, SIGSEGV); 454 sigaddset(&unblocked_sigs, SIGBUS); 455 sigaddset(&unblocked_sigs, SIGFPE); 456 sigaddset(&unblocked_sigs, SR_signum); 457 458 if (!ReduceSignalUsage) { 459 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 460 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 461 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 462 } 463 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 464 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 465 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 466 } 467 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 468 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 469 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 470 } 471 } 472 // Fill in signals that are blocked by all but the VM thread. 473 sigemptyset(&vm_sigs); 474 if (!ReduceSignalUsage) 475 sigaddset(&vm_sigs, BREAK_SIGNAL); 476 debug_only(signal_sets_initialized = true); 477 478} 479 480// These are signals that are unblocked while a thread is running Java. 481// (For some reason, they get blocked by default.) 482sigset_t* os::Linux::unblocked_signals() { 483 assert(signal_sets_initialized, "Not initialized"); 484 return &unblocked_sigs; 485} 486 487// These are the signals that are blocked while a (non-VM) thread is 488// running Java. Only the VM thread handles these signals. 489sigset_t* os::Linux::vm_signals() { 490 assert(signal_sets_initialized, "Not initialized"); 491 return &vm_sigs; 492} 493 494// These are signals that are blocked during cond_wait to allow debugger in 495sigset_t* os::Linux::allowdebug_blocked_signals() { 496 assert(signal_sets_initialized, "Not initialized"); 497 return &allowdebug_blocked_sigs; 498} 499 500void os::Linux::hotspot_sigmask(Thread* thread) { 501 502 //Save caller's signal mask before setting VM signal mask 503 sigset_t caller_sigmask; 504 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 505 506 OSThread* osthread = thread->osthread(); 507 osthread->set_caller_sigmask(caller_sigmask); 508 509 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 510 511 if (!ReduceSignalUsage) { 512 if (thread->is_VM_thread()) { 513 // Only the VM thread handles BREAK_SIGNAL ... 514 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 515 } else { 516 // ... all other threads block BREAK_SIGNAL 517 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 518 } 519 } 520} 521 522////////////////////////////////////////////////////////////////////////////// 523// detecting pthread library 524 525void os::Linux::libpthread_init() { 526 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 527 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 528 // generic name for earlier versions. 529 // Define macros here so we can build HotSpot on old systems. 530# ifndef _CS_GNU_LIBC_VERSION 531# define _CS_GNU_LIBC_VERSION 2 532# endif 533# ifndef _CS_GNU_LIBPTHREAD_VERSION 534# define _CS_GNU_LIBPTHREAD_VERSION 3 535# endif 536 537 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 538 if (n > 0) { 539 char *str = (char *)malloc(n); 540 confstr(_CS_GNU_LIBC_VERSION, str, n); 541 os::Linux::set_glibc_version(str); 542 } else { 543 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 544 static char _gnu_libc_version[32]; 545 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 546 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 547 os::Linux::set_glibc_version(_gnu_libc_version); 548 } 549 550 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 551 if (n > 0) { 552 char *str = (char *)malloc(n); 553 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 554 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 555 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 556 // is the case. LinuxThreads has a hard limit on max number of threads. 557 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 558 // On the other hand, NPTL does not have such a limit, sysconf() 559 // will return -1 and errno is not changed. Check if it is really NPTL. 560 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 561 strstr(str, "NPTL") && 562 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 563 free(str); 564 os::Linux::set_libpthread_version("linuxthreads"); 565 } else { 566 os::Linux::set_libpthread_version(str); 567 } 568 } else { 569 // glibc before 2.3.2 only has LinuxThreads. 570 os::Linux::set_libpthread_version("linuxthreads"); 571 } 572 573 if (strstr(libpthread_version(), "NPTL")) { 574 os::Linux::set_is_NPTL(); 575 } else { 576 os::Linux::set_is_LinuxThreads(); 577 } 578 579 // LinuxThreads have two flavors: floating-stack mode, which allows variable 580 // stack size; and fixed-stack mode. NPTL is always floating-stack. 581 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 582 os::Linux::set_is_floating_stack(); 583 } 584} 585 586///////////////////////////////////////////////////////////////////////////// 587// thread stack 588 589// Force Linux kernel to expand current thread stack. If "bottom" is close 590// to the stack guard, caller should block all signals. 591// 592// MAP_GROWSDOWN: 593// A special mmap() flag that is used to implement thread stacks. It tells 594// kernel that the memory region should extend downwards when needed. This 595// allows early versions of LinuxThreads to only mmap the first few pages 596// when creating a new thread. Linux kernel will automatically expand thread 597// stack as needed (on page faults). 598// 599// However, because the memory region of a MAP_GROWSDOWN stack can grow on 600// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 601// region, it's hard to tell if the fault is due to a legitimate stack 602// access or because of reading/writing non-exist memory (e.g. buffer 603// overrun). As a rule, if the fault happens below current stack pointer, 604// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 605// application (see Linux kernel fault.c). 606// 607// This Linux feature can cause SIGSEGV when VM bangs thread stack for 608// stack overflow detection. 609// 610// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 611// not use this flag. However, the stack of initial thread is not created 612// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 613// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 614// and then attach the thread to JVM. 615// 616// To get around the problem and allow stack banging on Linux, we need to 617// manually expand thread stack after receiving the SIGSEGV. 618// 619// There are two ways to expand thread stack to address "bottom", we used 620// both of them in JVM before 1.5: 621// 1. adjust stack pointer first so that it is below "bottom", and then 622// touch "bottom" 623// 2. mmap() the page in question 624// 625// Now alternate signal stack is gone, it's harder to use 2. For instance, 626// if current sp is already near the lower end of page 101, and we need to 627// call mmap() to map page 100, it is possible that part of the mmap() frame 628// will be placed in page 100. When page 100 is mapped, it is zero-filled. 629// That will destroy the mmap() frame and cause VM to crash. 630// 631// The following code works by adjusting sp first, then accessing the "bottom" 632// page to force a page fault. Linux kernel will then automatically expand the 633// stack mapping. 634// 635// _expand_stack_to() assumes its frame size is less than page size, which 636// should always be true if the function is not inlined. 637 638#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 639#define NOINLINE 640#else 641#define NOINLINE __attribute__ ((noinline)) 642#endif 643 644static void _expand_stack_to(address bottom) NOINLINE; 645 646static void _expand_stack_to(address bottom) { 647 address sp; 648 size_t size; 649 volatile char *p; 650 651 // Adjust bottom to point to the largest address within the same page, it 652 // gives us a one-page buffer if alloca() allocates slightly more memory. 653 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 654 bottom += os::Linux::page_size() - 1; 655 656 // sp might be slightly above current stack pointer; if that's the case, we 657 // will alloca() a little more space than necessary, which is OK. Don't use 658 // os::current_stack_pointer(), as its result can be slightly below current 659 // stack pointer, causing us to not alloca enough to reach "bottom". 660 sp = (address)&sp; 661 662 if (sp > bottom) { 663 size = sp - bottom; 664 p = (volatile char *)alloca(size); 665 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 666 p[0] = '\0'; 667 } 668} 669 670bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 671 assert(t!=NULL, "just checking"); 672 assert(t->osthread()->expanding_stack(), "expand should be set"); 673 assert(t->stack_base() != NULL, "stack_base was not initialized"); 674 675 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 676 sigset_t mask_all, old_sigset; 677 sigfillset(&mask_all); 678 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 679 _expand_stack_to(addr); 680 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 681 return true; 682 } 683 return false; 684} 685 686////////////////////////////////////////////////////////////////////////////// 687// create new thread 688 689static address highest_vm_reserved_address(); 690 691// check if it's safe to start a new thread 692static bool _thread_safety_check(Thread* thread) { 693 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 694 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 695 // Heap is mmap'ed at lower end of memory space. Thread stacks are 696 // allocated (MAP_FIXED) from high address space. Every thread stack 697 // occupies a fixed size slot (usually 2Mbytes, but user can change 698 // it to other values if they rebuild LinuxThreads). 699 // 700 // Problem with MAP_FIXED is that mmap() can still succeed even part of 701 // the memory region has already been mmap'ed. That means if we have too 702 // many threads and/or very large heap, eventually thread stack will 703 // collide with heap. 704 // 705 // Here we try to prevent heap/stack collision by comparing current 706 // stack bottom with the highest address that has been mmap'ed by JVM 707 // plus a safety margin for memory maps created by native code. 708 // 709 // This feature can be disabled by setting ThreadSafetyMargin to 0 710 // 711 if (ThreadSafetyMargin > 0) { 712 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 713 714 // not safe if our stack extends below the safety margin 715 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 716 } else { 717 return true; 718 } 719 } else { 720 // Floating stack LinuxThreads or NPTL: 721 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 722 // there's not enough space left, pthread_create() will fail. If we come 723 // here, that means enough space has been reserved for stack. 724 return true; 725 } 726} 727 728// Thread start routine for all newly created threads 729static void *java_start(Thread *thread) { 730 // Try to randomize the cache line index of hot stack frames. 731 // This helps when threads of the same stack traces evict each other's 732 // cache lines. The threads can be either from the same JVM instance, or 733 // from different JVM instances. The benefit is especially true for 734 // processors with hyperthreading technology. 735 static int counter = 0; 736 int pid = os::current_process_id(); 737 alloca(((pid ^ counter++) & 7) * 128); 738 739 ThreadLocalStorage::set_thread(thread); 740 741 OSThread* osthread = thread->osthread(); 742 Monitor* sync = osthread->startThread_lock(); 743 744 // non floating stack LinuxThreads needs extra check, see above 745 if (!_thread_safety_check(thread)) { 746 // notify parent thread 747 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 748 osthread->set_state(ZOMBIE); 749 sync->notify_all(); 750 return NULL; 751 } 752 753 // thread_id is kernel thread id (similar to Solaris LWP id) 754 osthread->set_thread_id(os::Linux::gettid()); 755 756 if (UseNUMA) { 757 int lgrp_id = os::numa_get_group_id(); 758 if (lgrp_id != -1) { 759 thread->set_lgrp_id(lgrp_id); 760 } 761 } 762 // initialize signal mask for this thread 763 os::Linux::hotspot_sigmask(thread); 764 765 // initialize floating point control register 766 os::Linux::init_thread_fpu_state(); 767 768 // handshaking with parent thread 769 { 770 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 771 772 // notify parent thread 773 osthread->set_state(INITIALIZED); 774 sync->notify_all(); 775 776 // wait until os::start_thread() 777 while (osthread->get_state() == INITIALIZED) { 778 sync->wait(Mutex::_no_safepoint_check_flag); 779 } 780 } 781 782 // call one more level start routine 783 thread->run(); 784 785 return 0; 786} 787 788bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 789 assert(thread->osthread() == NULL, "caller responsible"); 790 791 // Allocate the OSThread object 792 OSThread* osthread = new OSThread(NULL, NULL); 793 if (osthread == NULL) { 794 return false; 795 } 796 797 // set the correct thread state 798 osthread->set_thread_type(thr_type); 799 800 // Initial state is ALLOCATED but not INITIALIZED 801 osthread->set_state(ALLOCATED); 802 803 thread->set_osthread(osthread); 804 805 // init thread attributes 806 pthread_attr_t attr; 807 pthread_attr_init(&attr); 808 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 809 810 // stack size 811 if (os::Linux::supports_variable_stack_size()) { 812 // calculate stack size if it's not specified by caller 813 if (stack_size == 0) { 814 stack_size = os::Linux::default_stack_size(thr_type); 815 816 switch (thr_type) { 817 case os::java_thread: 818 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss 819 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create(); 820 break; 821 case os::compiler_thread: 822 if (CompilerThreadStackSize > 0) { 823 stack_size = (size_t)(CompilerThreadStackSize * K); 824 break; 825 } // else fall through: 826 // use VMThreadStackSize if CompilerThreadStackSize is not defined 827 case os::vm_thread: 828 case os::pgc_thread: 829 case os::cgc_thread: 830 case os::watcher_thread: 831 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 832 break; 833 } 834 } 835 836 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 837 pthread_attr_setstacksize(&attr, stack_size); 838 } else { 839 // let pthread_create() pick the default value. 840 } 841 842 // glibc guard page 843 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 844 845 ThreadState state; 846 847 { 848 // Serialize thread creation if we are running with fixed stack LinuxThreads 849 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 850 if (lock) { 851 os::Linux::createThread_lock()->lock_without_safepoint_check(); 852 } 853 854 pthread_t tid; 855 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 856 857 pthread_attr_destroy(&attr); 858 859 if (ret != 0) { 860 if (PrintMiscellaneous && (Verbose || WizardMode)) { 861 perror("pthread_create()"); 862 } 863 // Need to clean up stuff we've allocated so far 864 thread->set_osthread(NULL); 865 delete osthread; 866 if (lock) os::Linux::createThread_lock()->unlock(); 867 return false; 868 } 869 870 // Store pthread info into the OSThread 871 osthread->set_pthread_id(tid); 872 873 // Wait until child thread is either initialized or aborted 874 { 875 Monitor* sync_with_child = osthread->startThread_lock(); 876 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 877 while ((state = osthread->get_state()) == ALLOCATED) { 878 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 879 } 880 } 881 882 if (lock) { 883 os::Linux::createThread_lock()->unlock(); 884 } 885 } 886 887 // Aborted due to thread limit being reached 888 if (state == ZOMBIE) { 889 thread->set_osthread(NULL); 890 delete osthread; 891 return false; 892 } 893 894 // The thread is returned suspended (in state INITIALIZED), 895 // and is started higher up in the call chain 896 assert(state == INITIALIZED, "race condition"); 897 return true; 898} 899 900///////////////////////////////////////////////////////////////////////////// 901// attach existing thread 902 903// bootstrap the main thread 904bool os::create_main_thread(JavaThread* thread) { 905 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 906 return create_attached_thread(thread); 907} 908 909bool os::create_attached_thread(JavaThread* thread) { 910#ifdef ASSERT 911 thread->verify_not_published(); 912#endif 913 914 // Allocate the OSThread object 915 OSThread* osthread = new OSThread(NULL, NULL); 916 917 if (osthread == NULL) { 918 return false; 919 } 920 921 // Store pthread info into the OSThread 922 osthread->set_thread_id(os::Linux::gettid()); 923 osthread->set_pthread_id(::pthread_self()); 924 925 // initialize floating point control register 926 os::Linux::init_thread_fpu_state(); 927 928 // Initial thread state is RUNNABLE 929 osthread->set_state(RUNNABLE); 930 931 thread->set_osthread(osthread); 932 933 if (UseNUMA) { 934 int lgrp_id = os::numa_get_group_id(); 935 if (lgrp_id != -1) { 936 thread->set_lgrp_id(lgrp_id); 937 } 938 } 939 940 if (os::Linux::is_initial_thread()) { 941 // If current thread is initial thread, its stack is mapped on demand, 942 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 943 // the entire stack region to avoid SEGV in stack banging. 944 // It is also useful to get around the heap-stack-gap problem on SuSE 945 // kernel (see 4821821 for details). We first expand stack to the top 946 // of yellow zone, then enable stack yellow zone (order is significant, 947 // enabling yellow zone first will crash JVM on SuSE Linux), so there 948 // is no gap between the last two virtual memory regions. 949 950 JavaThread *jt = (JavaThread *)thread; 951 address addr = jt->stack_yellow_zone_base(); 952 assert(addr != NULL, "initialization problem?"); 953 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 954 955 osthread->set_expanding_stack(); 956 os::Linux::manually_expand_stack(jt, addr); 957 osthread->clear_expanding_stack(); 958 } 959 960 // initialize signal mask for this thread 961 // and save the caller's signal mask 962 os::Linux::hotspot_sigmask(thread); 963 964 return true; 965} 966 967void os::pd_start_thread(Thread* thread) { 968 OSThread * osthread = thread->osthread(); 969 assert(osthread->get_state() != INITIALIZED, "just checking"); 970 Monitor* sync_with_child = osthread->startThread_lock(); 971 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 972 sync_with_child->notify(); 973} 974 975// Free Linux resources related to the OSThread 976void os::free_thread(OSThread* osthread) { 977 assert(osthread != NULL, "osthread not set"); 978 979 if (Thread::current()->osthread() == osthread) { 980 // Restore caller's signal mask 981 sigset_t sigmask = osthread->caller_sigmask(); 982 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 983 } 984 985 delete osthread; 986} 987 988////////////////////////////////////////////////////////////////////////////// 989// thread local storage 990 991int os::allocate_thread_local_storage() { 992 pthread_key_t key; 993 int rslt = pthread_key_create(&key, NULL); 994 assert(rslt == 0, "cannot allocate thread local storage"); 995 return (int)key; 996} 997 998// Note: This is currently not used by VM, as we don't destroy TLS key 999// on VM exit. 1000void os::free_thread_local_storage(int index) { 1001 int rslt = pthread_key_delete((pthread_key_t)index); 1002 assert(rslt == 0, "invalid index"); 1003} 1004 1005void os::thread_local_storage_at_put(int index, void* value) { 1006 int rslt = pthread_setspecific((pthread_key_t)index, value); 1007 assert(rslt == 0, "pthread_setspecific failed"); 1008} 1009 1010extern "C" Thread* get_thread() { 1011 return ThreadLocalStorage::thread(); 1012} 1013 1014////////////////////////////////////////////////////////////////////////////// 1015// initial thread 1016 1017// Check if current thread is the initial thread, similar to Solaris thr_main. 1018bool os::Linux::is_initial_thread(void) { 1019 char dummy; 1020 // If called before init complete, thread stack bottom will be null. 1021 // Can be called if fatal error occurs before initialization. 1022 if (initial_thread_stack_bottom() == NULL) return false; 1023 assert(initial_thread_stack_bottom() != NULL && 1024 initial_thread_stack_size() != 0, 1025 "os::init did not locate initial thread's stack region"); 1026 if ((address)&dummy >= initial_thread_stack_bottom() && 1027 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1028 return true; 1029 else return false; 1030} 1031 1032// Find the virtual memory area that contains addr 1033static bool find_vma(address addr, address* vma_low, address* vma_high) { 1034 FILE *fp = fopen("/proc/self/maps", "r"); 1035 if (fp) { 1036 address low, high; 1037 while (!feof(fp)) { 1038 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1039 if (low <= addr && addr < high) { 1040 if (vma_low) *vma_low = low; 1041 if (vma_high) *vma_high = high; 1042 fclose (fp); 1043 return true; 1044 } 1045 } 1046 for (;;) { 1047 int ch = fgetc(fp); 1048 if (ch == EOF || ch == (int)'\n') break; 1049 } 1050 } 1051 fclose(fp); 1052 } 1053 return false; 1054} 1055 1056// Locate initial thread stack. This special handling of initial thread stack 1057// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1058// bogus value for initial thread. 1059void os::Linux::capture_initial_stack(size_t max_size) { 1060 // stack size is the easy part, get it from RLIMIT_STACK 1061 size_t stack_size; 1062 struct rlimit rlim; 1063 getrlimit(RLIMIT_STACK, &rlim); 1064 stack_size = rlim.rlim_cur; 1065 1066 // 6308388: a bug in ld.so will relocate its own .data section to the 1067 // lower end of primordial stack; reduce ulimit -s value a little bit 1068 // so we won't install guard page on ld.so's data section. 1069 stack_size -= 2 * page_size(); 1070 1071 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1072 // 7.1, in both cases we will get 2G in return value. 1073 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1074 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1075 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1076 // in case other parts in glibc still assumes 2M max stack size. 1077 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1078#ifndef IA64 1079 if (stack_size > 2 * K * K) stack_size = 2 * K * K; 1080#else 1081 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1082 if (stack_size > 4 * K * K) stack_size = 4 * K * K; 1083#endif 1084 1085 // Try to figure out where the stack base (top) is. This is harder. 1086 // 1087 // When an application is started, glibc saves the initial stack pointer in 1088 // a global variable "__libc_stack_end", which is then used by system 1089 // libraries. __libc_stack_end should be pretty close to stack top. The 1090 // variable is available since the very early days. However, because it is 1091 // a private interface, it could disappear in the future. 1092 // 1093 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1094 // to __libc_stack_end, it is very close to stack top, but isn't the real 1095 // stack top. Note that /proc may not exist if VM is running as a chroot 1096 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1097 // /proc/<pid>/stat could change in the future (though unlikely). 1098 // 1099 // We try __libc_stack_end first. If that doesn't work, look for 1100 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1101 // as a hint, which should work well in most cases. 1102 1103 uintptr_t stack_start; 1104 1105 // try __libc_stack_end first 1106 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1107 if (p && *p) { 1108 stack_start = *p; 1109 } else { 1110 // see if we can get the start_stack field from /proc/self/stat 1111 FILE *fp; 1112 int pid; 1113 char state; 1114 int ppid; 1115 int pgrp; 1116 int session; 1117 int nr; 1118 int tpgrp; 1119 unsigned long flags; 1120 unsigned long minflt; 1121 unsigned long cminflt; 1122 unsigned long majflt; 1123 unsigned long cmajflt; 1124 unsigned long utime; 1125 unsigned long stime; 1126 long cutime; 1127 long cstime; 1128 long prio; 1129 long nice; 1130 long junk; 1131 long it_real; 1132 uintptr_t start; 1133 uintptr_t vsize; 1134 uintptr_t rss; 1135 unsigned long rsslim; 1136 uintptr_t scodes; 1137 uintptr_t ecode; 1138 int i; 1139 1140 // Figure what the primordial thread stack base is. Code is inspired 1141 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1142 // followed by command name surrounded by parentheses, state, etc. 1143 char stat[2048]; 1144 int statlen; 1145 1146 fp = fopen("/proc/self/stat", "r"); 1147 if (fp) { 1148 statlen = fread(stat, 1, 2047, fp); 1149 stat[statlen] = '\0'; 1150 fclose(fp); 1151 1152 // Skip pid and the command string. Note that we could be dealing with 1153 // weird command names, e.g. user could decide to rename java launcher 1154 // to "java 1.4.2 :)", then the stat file would look like 1155 // 1234 (java 1.4.2 :)) R ... ... 1156 // We don't really need to know the command string, just find the last 1157 // occurrence of ")" and then start parsing from there. See bug 4726580. 1158 char * s = strrchr(stat, ')'); 1159 1160 i = 0; 1161 if (s) { 1162 // Skip blank chars 1163 do s++; while (isspace(*s)); 1164 1165 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1166 /* 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 */ 1167 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " 1168 UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT 1169 " %lu " 1170 UINTX_FORMAT UINTX_FORMAT UINTX_FORMAT, 1171 &state, /* 3 %c */ 1172 &ppid, /* 4 %d */ 1173 &pgrp, /* 5 %d */ 1174 &session, /* 6 %d */ 1175 &nr, /* 7 %d */ 1176 &tpgrp, /* 8 %d */ 1177 &flags, /* 9 %lu */ 1178 &minflt, /* 10 %lu */ 1179 &cminflt, /* 11 %lu */ 1180 &majflt, /* 12 %lu */ 1181 &cmajflt, /* 13 %lu */ 1182 &utime, /* 14 %lu */ 1183 &stime, /* 15 %lu */ 1184 &cutime, /* 16 %ld */ 1185 &cstime, /* 17 %ld */ 1186 &prio, /* 18 %ld */ 1187 &nice, /* 19 %ld */ 1188 &junk, /* 20 %ld */ 1189 &it_real, /* 21 %ld */ 1190 &start, /* 22 UINTX_FORMAT */ 1191 &vsize, /* 23 UINTX_FORMAT */ 1192 &rss, /* 24 UINTX_FORMAT */ 1193 &rsslim, /* 25 %lu */ 1194 &scodes, /* 26 UINTX_FORMAT */ 1195 &ecode, /* 27 UINTX_FORMAT */ 1196 &stack_start); /* 28 UINTX_FORMAT */ 1197 } 1198 1199 if (i != 28 - 2) { 1200 assert(false, "Bad conversion from /proc/self/stat"); 1201 // product mode - assume we are the initial thread, good luck in the 1202 // embedded case. 1203 warning("Can't detect initial thread stack location - bad conversion"); 1204 stack_start = (uintptr_t) &rlim; 1205 } 1206 } else { 1207 // For some reason we can't open /proc/self/stat (for example, running on 1208 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1209 // most cases, so don't abort: 1210 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1211 stack_start = (uintptr_t) &rlim; 1212 } 1213 } 1214 1215 // Now we have a pointer (stack_start) very close to the stack top, the 1216 // next thing to do is to figure out the exact location of stack top. We 1217 // can find out the virtual memory area that contains stack_start by 1218 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1219 // and its upper limit is the real stack top. (again, this would fail if 1220 // running inside chroot, because /proc may not exist.) 1221 1222 uintptr_t stack_top; 1223 address low, high; 1224 if (find_vma((address)stack_start, &low, &high)) { 1225 // success, "high" is the true stack top. (ignore "low", because initial 1226 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1227 stack_top = (uintptr_t)high; 1228 } else { 1229 // failed, likely because /proc/self/maps does not exist 1230 warning("Can't detect initial thread stack location - find_vma failed"); 1231 // best effort: stack_start is normally within a few pages below the real 1232 // stack top, use it as stack top, and reduce stack size so we won't put 1233 // guard page outside stack. 1234 stack_top = stack_start; 1235 stack_size -= 16 * page_size(); 1236 } 1237 1238 // stack_top could be partially down the page so align it 1239 stack_top = align_size_up(stack_top, page_size()); 1240 1241 if (max_size && stack_size > max_size) { 1242 _initial_thread_stack_size = max_size; 1243 } else { 1244 _initial_thread_stack_size = stack_size; 1245 } 1246 1247 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1248 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1249} 1250 1251//////////////////////////////////////////////////////////////////////////////// 1252// time support 1253 1254// Time since start-up in seconds to a fine granularity. 1255// Used by VMSelfDestructTimer and the MemProfiler. 1256double os::elapsedTime() { 1257 1258 return (double)(os::elapsed_counter()) * 0.000001; 1259} 1260 1261jlong os::elapsed_counter() { 1262 timeval time; 1263 int status = gettimeofday(&time, NULL); 1264 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1265} 1266 1267jlong os::elapsed_frequency() { 1268 return (1000 * 1000); 1269} 1270 1271// For now, we say that linux does not support vtime. I have no idea 1272// whether it can actually be made to (DLD, 9/13/05). 1273 1274bool os::supports_vtime() { return false; } 1275bool os::enable_vtime() { return false; } 1276bool os::vtime_enabled() { return false; } 1277double os::elapsedVTime() { 1278 // better than nothing, but not much 1279 return elapsedTime(); 1280} 1281 1282jlong os::javaTimeMillis() { 1283 timeval time; 1284 int status = gettimeofday(&time, NULL); 1285 assert(status != -1, "linux error"); 1286 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1287} 1288 1289#ifndef CLOCK_MONOTONIC 1290#define CLOCK_MONOTONIC (1) 1291#endif 1292 1293void os::Linux::clock_init() { 1294 // we do dlopen's in this particular order due to bug in linux 1295 // dynamical loader (see 6348968) leading to crash on exit 1296 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1297 if (handle == NULL) { 1298 handle = dlopen("librt.so", RTLD_LAZY); 1299 } 1300 1301 if (handle) { 1302 int (*clock_getres_func)(clockid_t, struct timespec*) = 1303 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1304 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1305 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1306 if (clock_getres_func && clock_gettime_func) { 1307 // See if monotonic clock is supported by the kernel. Note that some 1308 // early implementations simply return kernel jiffies (updated every 1309 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1310 // for nano time (though the monotonic property is still nice to have). 1311 // It's fixed in newer kernels, however clock_getres() still returns 1312 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1313 // resolution for now. Hopefully as people move to new kernels, this 1314 // won't be a problem. 1315 struct timespec res; 1316 struct timespec tp; 1317 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1318 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1319 // yes, monotonic clock is supported 1320 _clock_gettime = clock_gettime_func; 1321 } else { 1322 // close librt if there is no monotonic clock 1323 dlclose(handle); 1324 } 1325 } 1326 } 1327} 1328 1329#ifndef SYS_clock_getres 1330 1331#if defined(IA32) || defined(AMD64) 1332#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1333#else 1334#error Value of SYS_clock_getres not known on this platform 1335#endif 1336 1337#endif 1338 1339#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1340 1341void os::Linux::fast_thread_clock_init() { 1342 if (!UseLinuxPosixThreadCPUClocks) { 1343 return; 1344 } 1345 clockid_t clockid; 1346 struct timespec tp; 1347 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1348 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1349 1350 // Switch to using fast clocks for thread cpu time if 1351 // the sys_clock_getres() returns 0 error code. 1352 // Note, that some kernels may support the current thread 1353 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1354 // returned by the pthread_getcpuclockid(). 1355 // If the fast Posix clocks are supported then the sys_clock_getres() 1356 // must return at least tp.tv_sec == 0 which means a resolution 1357 // better than 1 sec. This is extra check for reliability. 1358 1359 if(pthread_getcpuclockid_func && 1360 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1361 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1362 1363 _supports_fast_thread_cpu_time = true; 1364 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1365 } 1366} 1367 1368jlong os::javaTimeNanos() { 1369 if (Linux::supports_monotonic_clock()) { 1370 struct timespec tp; 1371 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1372 assert(status == 0, "gettime error"); 1373 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1374 return result; 1375 } else { 1376 timeval time; 1377 int status = gettimeofday(&time, NULL); 1378 assert(status != -1, "linux error"); 1379 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1380 return 1000 * usecs; 1381 } 1382} 1383 1384void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1385 if (Linux::supports_monotonic_clock()) { 1386 info_ptr->max_value = ALL_64_BITS; 1387 1388 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1389 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1390 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1391 } else { 1392 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1393 info_ptr->max_value = ALL_64_BITS; 1394 1395 // gettimeofday is a real time clock so it skips 1396 info_ptr->may_skip_backward = true; 1397 info_ptr->may_skip_forward = true; 1398 } 1399 1400 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1401} 1402 1403// Return the real, user, and system times in seconds from an 1404// arbitrary fixed point in the past. 1405bool os::getTimesSecs(double* process_real_time, 1406 double* process_user_time, 1407 double* process_system_time) { 1408 struct tms ticks; 1409 clock_t real_ticks = times(&ticks); 1410 1411 if (real_ticks == (clock_t) (-1)) { 1412 return false; 1413 } else { 1414 double ticks_per_second = (double) clock_tics_per_sec; 1415 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1416 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1417 *process_real_time = ((double) real_ticks) / ticks_per_second; 1418 1419 return true; 1420 } 1421} 1422 1423 1424char * os::local_time_string(char *buf, size_t buflen) { 1425 struct tm t; 1426 time_t long_time; 1427 time(&long_time); 1428 localtime_r(&long_time, &t); 1429 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1430 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1431 t.tm_hour, t.tm_min, t.tm_sec); 1432 return buf; 1433} 1434 1435struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1436 return localtime_r(clock, res); 1437} 1438 1439//////////////////////////////////////////////////////////////////////////////// 1440// runtime exit support 1441 1442// Note: os::shutdown() might be called very early during initialization, or 1443// called from signal handler. Before adding something to os::shutdown(), make 1444// sure it is async-safe and can handle partially initialized VM. 1445void os::shutdown() { 1446 1447 // allow PerfMemory to attempt cleanup of any persistent resources 1448 perfMemory_exit(); 1449 1450 // needs to remove object in file system 1451 AttachListener::abort(); 1452 1453 // flush buffered output, finish log files 1454 ostream_abort(); 1455 1456 // Check for abort hook 1457 abort_hook_t abort_hook = Arguments::abort_hook(); 1458 if (abort_hook != NULL) { 1459 abort_hook(); 1460 } 1461 1462} 1463 1464// Note: os::abort() might be called very early during initialization, or 1465// called from signal handler. Before adding something to os::abort(), make 1466// sure it is async-safe and can handle partially initialized VM. 1467void os::abort(bool dump_core) { 1468 os::shutdown(); 1469 if (dump_core) { 1470#ifndef PRODUCT 1471 fdStream out(defaultStream::output_fd()); 1472 out.print_raw("Current thread is "); 1473 char buf[16]; 1474 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1475 out.print_raw_cr(buf); 1476 out.print_raw_cr("Dumping core ..."); 1477#endif 1478 ::abort(); // dump core 1479 } 1480 1481 ::exit(1); 1482} 1483 1484// Die immediately, no exit hook, no abort hook, no cleanup. 1485void os::die() { 1486 // _exit() on LinuxThreads only kills current thread 1487 ::abort(); 1488} 1489 1490// unused on linux for now. 1491void os::set_error_file(const char *logfile) {} 1492 1493intx os::current_thread_id() { return (intx)pthread_self(); } 1494int os::current_process_id() { 1495 1496 // Under the old linux thread library, linux gives each thread 1497 // its own process id. Because of this each thread will return 1498 // a different pid if this method were to return the result 1499 // of getpid(2). Linux provides no api that returns the pid 1500 // of the launcher thread for the vm. This implementation 1501 // returns a unique pid, the pid of the launcher thread 1502 // that starts the vm 'process'. 1503 1504 // Under the NPTL, getpid() returns the same pid as the 1505 // launcher thread rather than a unique pid per thread. 1506 // Use gettid() if you want the old pre NPTL behaviour. 1507 1508 // if you are looking for the result of a call to getpid() that 1509 // returns a unique pid for the calling thread, then look at the 1510 // OSThread::thread_id() method in osThread_linux.hpp file 1511 1512 return (int)(_initial_pid ? _initial_pid : getpid()); 1513} 1514 1515// DLL functions 1516 1517const char* os::dll_file_extension() { return ".so"; } 1518 1519const char* os::get_temp_directory() { return "/tmp/"; } 1520 1521static bool file_exists(const char* filename) { 1522 struct stat statbuf; 1523 if (filename == NULL || strlen(filename) == 0) { 1524 return false; 1525 } 1526 return os::stat(filename, &statbuf) == 0; 1527} 1528 1529void os::dll_build_name(char* buffer, size_t buflen, 1530 const char* pname, const char* fname) { 1531 // Copied from libhpi 1532 const size_t pnamelen = pname ? strlen(pname) : 0; 1533 1534 // Quietly truncate on buffer overflow. Should be an error. 1535 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1536 *buffer = '\0'; 1537 return; 1538 } 1539 1540 if (pnamelen == 0) { 1541 snprintf(buffer, buflen, "lib%s.so", fname); 1542 } else if (strchr(pname, *os::path_separator()) != NULL) { 1543 int n; 1544 char** pelements = split_path(pname, &n); 1545 for (int i = 0 ; i < n ; i++) { 1546 // Really shouldn't be NULL, but check can't hurt 1547 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1548 continue; // skip the empty path values 1549 } 1550 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1551 if (file_exists(buffer)) { 1552 break; 1553 } 1554 } 1555 // release the storage 1556 for (int i = 0 ; i < n ; i++) { 1557 if (pelements[i] != NULL) { 1558 FREE_C_HEAP_ARRAY(char, pelements[i]); 1559 } 1560 } 1561 if (pelements != NULL) { 1562 FREE_C_HEAP_ARRAY(char*, pelements); 1563 } 1564 } else { 1565 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1566 } 1567} 1568 1569const char* os::get_current_directory(char *buf, int buflen) { 1570 return getcwd(buf, buflen); 1571} 1572 1573// check if addr is inside libjvm[_g].so 1574bool os::address_is_in_vm(address addr) { 1575 static address libjvm_base_addr; 1576 Dl_info dlinfo; 1577 1578 if (libjvm_base_addr == NULL) { 1579 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); 1580 libjvm_base_addr = (address)dlinfo.dli_fbase; 1581 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1582 } 1583 1584 if (dladdr((void *)addr, &dlinfo)) { 1585 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1586 } 1587 1588 return false; 1589} 1590 1591bool os::dll_address_to_function_name(address addr, char *buf, 1592 int buflen, int *offset) { 1593 Dl_info dlinfo; 1594 1595 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { 1596 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1597 if (offset) *offset = addr - (address)dlinfo.dli_saddr; 1598 return true; 1599 } else { 1600 if (buf) buf[0] = '\0'; 1601 if (offset) *offset = -1; 1602 return false; 1603 } 1604} 1605 1606struct _address_to_library_name { 1607 address addr; // input : memory address 1608 size_t buflen; // size of fname 1609 char* fname; // output: library name 1610 address base; // library base addr 1611}; 1612 1613static int address_to_library_name_callback(struct dl_phdr_info *info, 1614 size_t size, void *data) { 1615 int i; 1616 bool found = false; 1617 address libbase = NULL; 1618 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1619 1620 // iterate through all loadable segments 1621 for (i = 0; i < info->dlpi_phnum; i++) { 1622 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1623 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1624 // base address of a library is the lowest address of its loaded 1625 // segments. 1626 if (libbase == NULL || libbase > segbase) { 1627 libbase = segbase; 1628 } 1629 // see if 'addr' is within current segment 1630 if (segbase <= d->addr && 1631 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1632 found = true; 1633 } 1634 } 1635 } 1636 1637 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1638 // so dll_address_to_library_name() can fall through to use dladdr() which 1639 // can figure out executable name from argv[0]. 1640 if (found && info->dlpi_name && info->dlpi_name[0]) { 1641 d->base = libbase; 1642 if (d->fname) { 1643 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1644 } 1645 return 1; 1646 } 1647 return 0; 1648} 1649 1650bool os::dll_address_to_library_name(address addr, char* buf, 1651 int buflen, int* offset) { 1652 Dl_info dlinfo; 1653 struct _address_to_library_name data; 1654 1655 // There is a bug in old glibc dladdr() implementation that it could resolve 1656 // to wrong library name if the .so file has a base address != NULL. Here 1657 // we iterate through the program headers of all loaded libraries to find 1658 // out which library 'addr' really belongs to. This workaround can be 1659 // removed once the minimum requirement for glibc is moved to 2.3.x. 1660 data.addr = addr; 1661 data.fname = buf; 1662 data.buflen = buflen; 1663 data.base = NULL; 1664 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1665 1666 if (rslt) { 1667 // buf already contains library name 1668 if (offset) *offset = addr - data.base; 1669 return true; 1670 } else if (dladdr((void*)addr, &dlinfo)){ 1671 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1672 if (offset) *offset = addr - (address)dlinfo.dli_fbase; 1673 return true; 1674 } else { 1675 if (buf) buf[0] = '\0'; 1676 if (offset) *offset = -1; 1677 return false; 1678 } 1679} 1680 1681 // Loads .dll/.so and 1682 // in case of error it checks if .dll/.so was built for the 1683 // same architecture as Hotspot is running on 1684 1685void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1686{ 1687 void * result= ::dlopen(filename, RTLD_LAZY); 1688 if (result != NULL) { 1689 // Successful loading 1690 return result; 1691 } 1692 1693 Elf32_Ehdr elf_head; 1694 1695 // Read system error message into ebuf 1696 // It may or may not be overwritten below 1697 ::strncpy(ebuf, ::dlerror(), ebuflen-1); 1698 ebuf[ebuflen-1]='\0'; 1699 int diag_msg_max_length=ebuflen-strlen(ebuf); 1700 char* diag_msg_buf=ebuf+strlen(ebuf); 1701 1702 if (diag_msg_max_length==0) { 1703 // No more space in ebuf for additional diagnostics message 1704 return NULL; 1705 } 1706 1707 1708 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1709 1710 if (file_descriptor < 0) { 1711 // Can't open library, report dlerror() message 1712 return NULL; 1713 } 1714 1715 bool failed_to_read_elf_head= 1716 (sizeof(elf_head)!= 1717 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1718 1719 ::close(file_descriptor); 1720 if (failed_to_read_elf_head) { 1721 // file i/o error - report dlerror() msg 1722 return NULL; 1723 } 1724 1725 typedef struct { 1726 Elf32_Half code; // Actual value as defined in elf.h 1727 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1728 char elf_class; // 32 or 64 bit 1729 char endianess; // MSB or LSB 1730 char* name; // String representation 1731 } arch_t; 1732 1733 #ifndef EM_486 1734 #define EM_486 6 /* Intel 80486 */ 1735 #endif 1736 1737 static const arch_t arch_array[]={ 1738 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1739 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1740 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1741 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1742 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1743 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1744 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1745 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1746 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"} 1747 }; 1748 1749 #if (defined IA32) 1750 static Elf32_Half running_arch_code=EM_386; 1751 #elif (defined AMD64) 1752 static Elf32_Half running_arch_code=EM_X86_64; 1753 #elif (defined IA64) 1754 static Elf32_Half running_arch_code=EM_IA_64; 1755 #elif (defined __sparc) && (defined _LP64) 1756 static Elf32_Half running_arch_code=EM_SPARCV9; 1757 #elif (defined __sparc) && (!defined _LP64) 1758 static Elf32_Half running_arch_code=EM_SPARC; 1759 #elif (defined __powerpc64__) 1760 static Elf32_Half running_arch_code=EM_PPC64; 1761 #elif (defined __powerpc__) 1762 static Elf32_Half running_arch_code=EM_PPC; 1763 #else 1764 #error Method os::dll_load requires that one of following is defined:\ 1765 IA32, AMD64, IA64, __sparc, __powerpc__ 1766 #endif 1767 1768 // Identify compatability class for VM's architecture and library's architecture 1769 // Obtain string descriptions for architectures 1770 1771 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1772 int running_arch_index=-1; 1773 1774 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1775 if (running_arch_code == arch_array[i].code) { 1776 running_arch_index = i; 1777 } 1778 if (lib_arch.code == arch_array[i].code) { 1779 lib_arch.compat_class = arch_array[i].compat_class; 1780 lib_arch.name = arch_array[i].name; 1781 } 1782 } 1783 1784 assert(running_arch_index != -1, 1785 "Didn't find running architecture code (running_arch_code) in arch_array"); 1786 if (running_arch_index == -1) { 1787 // Even though running architecture detection failed 1788 // we may still continue with reporting dlerror() message 1789 return NULL; 1790 } 1791 1792 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1793 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1794 return NULL; 1795 } 1796 1797 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1798 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1799 return NULL; 1800 } 1801 1802 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1803 if ( lib_arch.name!=NULL ) { 1804 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1805 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1806 lib_arch.name, arch_array[running_arch_index].name); 1807 } else { 1808 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1809 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1810 lib_arch.code, 1811 arch_array[running_arch_index].name); 1812 } 1813 } 1814 1815 return NULL; 1816} 1817 1818/* 1819 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 1820 * chances are you might want to run the generated bits against glibc-2.0 1821 * libdl.so, so always use locking for any version of glibc. 1822 */ 1823void* os::dll_lookup(void* handle, const char* name) { 1824 pthread_mutex_lock(&dl_mutex); 1825 void* res = dlsym(handle, name); 1826 pthread_mutex_unlock(&dl_mutex); 1827 return res; 1828} 1829 1830 1831bool _print_ascii_file(const char* filename, outputStream* st) { 1832 int fd = open(filename, O_RDONLY); 1833 if (fd == -1) { 1834 return false; 1835 } 1836 1837 char buf[32]; 1838 int bytes; 1839 while ((bytes = read(fd, buf, sizeof(buf))) > 0) { 1840 st->print_raw(buf, bytes); 1841 } 1842 1843 close(fd); 1844 1845 return true; 1846} 1847 1848void os::print_dll_info(outputStream *st) { 1849 st->print_cr("Dynamic libraries:"); 1850 1851 char fname[32]; 1852 pid_t pid = os::Linux::gettid(); 1853 1854 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1855 1856 if (!_print_ascii_file(fname, st)) { 1857 st->print("Can not get library information for pid = %d\n", pid); 1858 } 1859} 1860 1861 1862void os::print_os_info(outputStream* st) { 1863 st->print("OS:"); 1864 1865 // Try to identify popular distros. 1866 // Most Linux distributions have /etc/XXX-release file, which contains 1867 // the OS version string. Some have more than one /etc/XXX-release file 1868 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 1869 // so the order is important. 1870 if (!_print_ascii_file("/etc/mandrake-release", st) && 1871 !_print_ascii_file("/etc/sun-release", st) && 1872 !_print_ascii_file("/etc/redhat-release", st) && 1873 !_print_ascii_file("/etc/SuSE-release", st) && 1874 !_print_ascii_file("/etc/turbolinux-release", st) && 1875 !_print_ascii_file("/etc/gentoo-release", st) && 1876 !_print_ascii_file("/etc/debian_version", st)) { 1877 st->print("Linux"); 1878 } 1879 st->cr(); 1880 1881 // kernel 1882 st->print("uname:"); 1883 struct utsname name; 1884 uname(&name); 1885 st->print(name.sysname); st->print(" "); 1886 st->print(name.release); st->print(" "); 1887 st->print(name.version); st->print(" "); 1888 st->print(name.machine); 1889 st->cr(); 1890 1891 // Print warning if unsafe chroot environment detected 1892 if (unsafe_chroot_detected) { 1893 st->print("WARNING!! "); 1894 st->print_cr(unstable_chroot_error); 1895 } 1896 1897 // libc, pthread 1898 st->print("libc:"); 1899 st->print(os::Linux::glibc_version()); st->print(" "); 1900 st->print(os::Linux::libpthread_version()); st->print(" "); 1901 if (os::Linux::is_LinuxThreads()) { 1902 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 1903 } 1904 st->cr(); 1905 1906 // rlimit 1907 st->print("rlimit:"); 1908 struct rlimit rlim; 1909 1910 st->print(" STACK "); 1911 getrlimit(RLIMIT_STACK, &rlim); 1912 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1913 else st->print("%uk", rlim.rlim_cur >> 10); 1914 1915 st->print(", CORE "); 1916 getrlimit(RLIMIT_CORE, &rlim); 1917 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1918 else st->print("%uk", rlim.rlim_cur >> 10); 1919 1920 st->print(", NPROC "); 1921 getrlimit(RLIMIT_NPROC, &rlim); 1922 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1923 else st->print("%d", rlim.rlim_cur); 1924 1925 st->print(", NOFILE "); 1926 getrlimit(RLIMIT_NOFILE, &rlim); 1927 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1928 else st->print("%d", rlim.rlim_cur); 1929 1930 st->print(", AS "); 1931 getrlimit(RLIMIT_AS, &rlim); 1932 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1933 else st->print("%uk", rlim.rlim_cur >> 10); 1934 st->cr(); 1935 1936 // load average 1937 st->print("load average:"); 1938 double loadavg[3]; 1939 os::loadavg(loadavg, 3); 1940 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]); 1941 st->cr(); 1942} 1943 1944void os::print_memory_info(outputStream* st) { 1945 1946 st->print("Memory:"); 1947 st->print(" %dk page", os::vm_page_size()>>10); 1948 1949 // values in struct sysinfo are "unsigned long" 1950 struct sysinfo si; 1951 sysinfo(&si); 1952 1953 st->print(", physical " UINT64_FORMAT "k", 1954 os::physical_memory() >> 10); 1955 st->print("(" UINT64_FORMAT "k free)", 1956 os::available_memory() >> 10); 1957 st->print(", swap " UINT64_FORMAT "k", 1958 ((jlong)si.totalswap * si.mem_unit) >> 10); 1959 st->print("(" UINT64_FORMAT "k free)", 1960 ((jlong)si.freeswap * si.mem_unit) >> 10); 1961 st->cr(); 1962} 1963 1964// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 1965// but they're the same for all the linux arch that we support 1966// and they're the same for solaris but there's no common place to put this. 1967const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 1968 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 1969 "ILL_COPROC", "ILL_BADSTK" }; 1970 1971const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 1972 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 1973 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 1974 1975const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 1976 1977const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 1978 1979void os::print_siginfo(outputStream* st, void* siginfo) { 1980 st->print("siginfo:"); 1981 1982 const int buflen = 100; 1983 char buf[buflen]; 1984 siginfo_t *si = (siginfo_t*)siginfo; 1985 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 1986 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 1987 st->print("si_errno=%s", buf); 1988 } else { 1989 st->print("si_errno=%d", si->si_errno); 1990 } 1991 const int c = si->si_code; 1992 assert(c > 0, "unexpected si_code"); 1993 switch (si->si_signo) { 1994 case SIGILL: 1995 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 1996 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1997 break; 1998 case SIGFPE: 1999 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 2000 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2001 break; 2002 case SIGSEGV: 2003 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 2004 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2005 break; 2006 case SIGBUS: 2007 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 2008 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2009 break; 2010 default: 2011 st->print(", si_code=%d", si->si_code); 2012 // no si_addr 2013 } 2014 2015 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2016 UseSharedSpaces) { 2017 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2018 if (mapinfo->is_in_shared_space(si->si_addr)) { 2019 st->print("\n\nError accessing class data sharing archive." \ 2020 " Mapped file inaccessible during execution, " \ 2021 " possible disk/network problem."); 2022 } 2023 } 2024 st->cr(); 2025} 2026 2027 2028static void print_signal_handler(outputStream* st, int sig, 2029 char* buf, size_t buflen); 2030 2031void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2032 st->print_cr("Signal Handlers:"); 2033 print_signal_handler(st, SIGSEGV, buf, buflen); 2034 print_signal_handler(st, SIGBUS , buf, buflen); 2035 print_signal_handler(st, SIGFPE , buf, buflen); 2036 print_signal_handler(st, SIGPIPE, buf, buflen); 2037 print_signal_handler(st, SIGXFSZ, buf, buflen); 2038 print_signal_handler(st, SIGILL , buf, buflen); 2039 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2040 print_signal_handler(st, SR_signum, buf, buflen); 2041 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2042 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2043 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2044 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2045} 2046 2047static char saved_jvm_path[MAXPATHLEN] = {0}; 2048 2049// Find the full path to the current module, libjvm.so or libjvm_g.so 2050void os::jvm_path(char *buf, jint len) { 2051 // Error checking. 2052 if (len < MAXPATHLEN) { 2053 assert(false, "must use a large-enough buffer"); 2054 buf[0] = '\0'; 2055 return; 2056 } 2057 // Lazy resolve the path to current module. 2058 if (saved_jvm_path[0] != 0) { 2059 strcpy(buf, saved_jvm_path); 2060 return; 2061 } 2062 2063 char dli_fname[MAXPATHLEN]; 2064 bool ret = dll_address_to_library_name( 2065 CAST_FROM_FN_PTR(address, os::jvm_path), 2066 dli_fname, sizeof(dli_fname), NULL); 2067 assert(ret != 0, "cannot locate libjvm"); 2068 if (realpath(dli_fname, buf) == NULL) 2069 return; 2070 2071 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) { 2072 // Support for the gamma launcher. Typical value for buf is 2073 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 2074 // the right place in the string, then assume we are installed in a JDK and 2075 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2076 // up the path so it looks like libjvm.so is installed there (append a 2077 // fake suffix hotspot/libjvm.so). 2078 const char *p = buf + strlen(buf) - 1; 2079 for (int count = 0; p > buf && count < 5; ++count) { 2080 for (--p; p > buf && *p != '/'; --p) 2081 /* empty */ ; 2082 } 2083 2084 if (strncmp(p, "/jre/lib/", 9) != 0) { 2085 // Look for JAVA_HOME in the environment. 2086 char* java_home_var = ::getenv("JAVA_HOME"); 2087 if (java_home_var != NULL && java_home_var[0] != 0) { 2088 // Check the current module name "libjvm.so" or "libjvm_g.so". 2089 p = strrchr(buf, '/'); 2090 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2091 p = strstr(p, "_g") ? "_g" : ""; 2092 2093 if (realpath(java_home_var, buf) == NULL) 2094 return; 2095 sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch); 2096 if (0 == access(buf, F_OK)) { 2097 // Use current module name "libjvm[_g].so" instead of 2098 // "libjvm"debug_only("_g")".so" since for fastdebug version 2099 // we should have "libjvm.so" but debug_only("_g") adds "_g"! 2100 // It is used when we are choosing the HPI library's name 2101 // "libhpi[_g].so" in hpi::initialize_get_interface(). 2102 sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p); 2103 } else { 2104 // Go back to path of .so 2105 if (realpath(dli_fname, buf) == NULL) 2106 return; 2107 } 2108 } 2109 } 2110 } 2111 2112 strcpy(saved_jvm_path, buf); 2113} 2114 2115void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2116 // no prefix required, not even "_" 2117} 2118 2119void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2120 // no suffix required 2121} 2122 2123//////////////////////////////////////////////////////////////////////////////// 2124// sun.misc.Signal support 2125 2126static volatile jint sigint_count = 0; 2127 2128static void 2129UserHandler(int sig, void *siginfo, void *context) { 2130 // 4511530 - sem_post is serialized and handled by the manager thread. When 2131 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2132 // don't want to flood the manager thread with sem_post requests. 2133 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2134 return; 2135 2136 // Ctrl-C is pressed during error reporting, likely because the error 2137 // handler fails to abort. Let VM die immediately. 2138 if (sig == SIGINT && is_error_reported()) { 2139 os::die(); 2140 } 2141 2142 os::signal_notify(sig); 2143} 2144 2145void* os::user_handler() { 2146 return CAST_FROM_FN_PTR(void*, UserHandler); 2147} 2148 2149extern "C" { 2150 typedef void (*sa_handler_t)(int); 2151 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2152} 2153 2154void* os::signal(int signal_number, void* handler) { 2155 struct sigaction sigAct, oldSigAct; 2156 2157 sigfillset(&(sigAct.sa_mask)); 2158 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2159 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2160 2161 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2162 // -1 means registration failed 2163 return (void *)-1; 2164 } 2165 2166 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2167} 2168 2169void os::signal_raise(int signal_number) { 2170 ::raise(signal_number); 2171} 2172 2173/* 2174 * The following code is moved from os.cpp for making this 2175 * code platform specific, which it is by its very nature. 2176 */ 2177 2178// Will be modified when max signal is changed to be dynamic 2179int os::sigexitnum_pd() { 2180 return NSIG; 2181} 2182 2183// a counter for each possible signal value 2184static volatile jint pending_signals[NSIG+1] = { 0 }; 2185 2186// Linux(POSIX) specific hand shaking semaphore. 2187static sem_t sig_sem; 2188 2189void os::signal_init_pd() { 2190 // Initialize signal structures 2191 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2192 2193 // Initialize signal semaphore 2194 ::sem_init(&sig_sem, 0, 0); 2195} 2196 2197void os::signal_notify(int sig) { 2198 Atomic::inc(&pending_signals[sig]); 2199 ::sem_post(&sig_sem); 2200} 2201 2202static int check_pending_signals(bool wait) { 2203 Atomic::store(0, &sigint_count); 2204 for (;;) { 2205 for (int i = 0; i < NSIG + 1; i++) { 2206 jint n = pending_signals[i]; 2207 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2208 return i; 2209 } 2210 } 2211 if (!wait) { 2212 return -1; 2213 } 2214 JavaThread *thread = JavaThread::current(); 2215 ThreadBlockInVM tbivm(thread); 2216 2217 bool threadIsSuspended; 2218 do { 2219 thread->set_suspend_equivalent(); 2220 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2221 ::sem_wait(&sig_sem); 2222 2223 // were we externally suspended while we were waiting? 2224 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2225 if (threadIsSuspended) { 2226 // 2227 // The semaphore has been incremented, but while we were waiting 2228 // another thread suspended us. We don't want to continue running 2229 // while suspended because that would surprise the thread that 2230 // suspended us. 2231 // 2232 ::sem_post(&sig_sem); 2233 2234 thread->java_suspend_self(); 2235 } 2236 } while (threadIsSuspended); 2237 } 2238} 2239 2240int os::signal_lookup() { 2241 return check_pending_signals(false); 2242} 2243 2244int os::signal_wait() { 2245 return check_pending_signals(true); 2246} 2247 2248//////////////////////////////////////////////////////////////////////////////// 2249// Virtual Memory 2250 2251int os::vm_page_size() { 2252 // Seems redundant as all get out 2253 assert(os::Linux::page_size() != -1, "must call os::init"); 2254 return os::Linux::page_size(); 2255} 2256 2257// Solaris allocates memory by pages. 2258int os::vm_allocation_granularity() { 2259 assert(os::Linux::page_size() != -1, "must call os::init"); 2260 return os::Linux::page_size(); 2261} 2262 2263// Rationale behind this function: 2264// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2265// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2266// samples for JITted code. Here we create private executable mapping over the code cache 2267// and then we can use standard (well, almost, as mapping can change) way to provide 2268// info for the reporting script by storing timestamp and location of symbol 2269void linux_wrap_code(char* base, size_t size) { 2270 static volatile jint cnt = 0; 2271 2272 if (!UseOprofile) { 2273 return; 2274 } 2275 2276 char buf[40]; 2277 int num = Atomic::add(1, &cnt); 2278 2279 sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num); 2280 unlink(buf); 2281 2282 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU); 2283 2284 if (fd != -1) { 2285 off_t rv = lseek(fd, size-2, SEEK_SET); 2286 if (rv != (off_t)-1) { 2287 if (write(fd, "", 1) == 1) { 2288 mmap(base, size, 2289 PROT_READ|PROT_WRITE|PROT_EXEC, 2290 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2291 } 2292 } 2293 close(fd); 2294 unlink(buf); 2295 } 2296} 2297 2298// NOTE: Linux kernel does not really reserve the pages for us. 2299// All it does is to check if there are enough free pages 2300// left at the time of mmap(). This could be a potential 2301// problem. 2302bool os::commit_memory(char* addr, size_t size, bool exec) { 2303 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2304 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2305 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2306 return res != (uintptr_t) MAP_FAILED; 2307} 2308 2309bool os::commit_memory(char* addr, size_t size, size_t alignment_hint, 2310 bool exec) { 2311 return commit_memory(addr, size, exec); 2312} 2313 2314void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { } 2315 2316void os::free_memory(char *addr, size_t bytes) { 2317 uncommit_memory(addr, bytes); 2318} 2319 2320void os::numa_make_global(char *addr, size_t bytes) { 2321 Linux::numa_interleave_memory(addr, bytes); 2322} 2323 2324void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2325 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2326} 2327 2328bool os::numa_topology_changed() { return false; } 2329 2330size_t os::numa_get_groups_num() { 2331 int max_node = Linux::numa_max_node(); 2332 return max_node > 0 ? max_node + 1 : 1; 2333} 2334 2335int os::numa_get_group_id() { 2336 int cpu_id = Linux::sched_getcpu(); 2337 if (cpu_id != -1) { 2338 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2339 if (lgrp_id != -1) { 2340 return lgrp_id; 2341 } 2342 } 2343 return 0; 2344} 2345 2346size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2347 for (size_t i = 0; i < size; i++) { 2348 ids[i] = i; 2349 } 2350 return size; 2351} 2352 2353bool os::get_page_info(char *start, page_info* info) { 2354 return false; 2355} 2356 2357char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2358 return end; 2359} 2360 2361extern "C" void numa_warn(int number, char *where, ...) { } 2362extern "C" void numa_error(char *where) { } 2363 2364bool os::Linux::libnuma_init() { 2365 // sched_getcpu() should be in libc. 2366 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2367 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2368 2369 if (sched_getcpu() != -1) { // Does it work? 2370 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2371 if (handle != NULL) { 2372 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2373 dlsym(handle, "numa_node_to_cpus"))); 2374 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2375 dlsym(handle, "numa_max_node"))); 2376 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2377 dlsym(handle, "numa_available"))); 2378 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2379 dlsym(handle, "numa_tonode_memory"))); 2380 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2381 dlsym(handle, "numa_interleave_memory"))); 2382 2383 2384 if (numa_available() != -1) { 2385 set_numa_all_nodes((unsigned long*)dlsym(handle, "numa_all_nodes")); 2386 // Create a cpu -> node mapping 2387 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true); 2388 rebuild_cpu_to_node_map(); 2389 return true; 2390 } 2391 } 2392 } 2393 return false; 2394} 2395 2396// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2397// The table is later used in get_node_by_cpu(). 2398void os::Linux::rebuild_cpu_to_node_map() { 2399 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2400 // in libnuma (possible values are starting from 16, 2401 // and continuing up with every other power of 2, but less 2402 // than the maximum number of CPUs supported by kernel), and 2403 // is a subject to change (in libnuma version 2 the requirements 2404 // are more reasonable) we'll just hardcode the number they use 2405 // in the library. 2406 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2407 2408 size_t cpu_num = os::active_processor_count(); 2409 size_t cpu_map_size = NCPUS / BitsPerCLong; 2410 size_t cpu_map_valid_size = 2411 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2412 2413 cpu_to_node()->clear(); 2414 cpu_to_node()->at_grow(cpu_num - 1); 2415 size_t node_num = numa_get_groups_num(); 2416 2417 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size); 2418 for (size_t i = 0; i < node_num; i++) { 2419 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2420 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2421 if (cpu_map[j] != 0) { 2422 for (size_t k = 0; k < BitsPerCLong; k++) { 2423 if (cpu_map[j] & (1UL << k)) { 2424 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2425 } 2426 } 2427 } 2428 } 2429 } 2430 } 2431 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2432} 2433 2434int os::Linux::get_node_by_cpu(int cpu_id) { 2435 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2436 return cpu_to_node()->at(cpu_id); 2437 } 2438 return -1; 2439} 2440 2441GrowableArray<int>* os::Linux::_cpu_to_node; 2442os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2443os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2444os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2445os::Linux::numa_available_func_t os::Linux::_numa_available; 2446os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2447os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2448unsigned long* os::Linux::_numa_all_nodes; 2449 2450bool os::uncommit_memory(char* addr, size_t size) { 2451 return ::mmap(addr, size, PROT_NONE, 2452 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0) 2453 != MAP_FAILED; 2454} 2455 2456static address _highest_vm_reserved_address = NULL; 2457 2458// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2459// at 'requested_addr'. If there are existing memory mappings at the same 2460// location, however, they will be overwritten. If 'fixed' is false, 2461// 'requested_addr' is only treated as a hint, the return value may or 2462// may not start from the requested address. Unlike Linux mmap(), this 2463// function returns NULL to indicate failure. 2464static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2465 char * addr; 2466 int flags; 2467 2468 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2469 if (fixed) { 2470 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2471 flags |= MAP_FIXED; 2472 } 2473 2474 // Map uncommitted pages PROT_READ and PROT_WRITE, change access 2475 // to PROT_EXEC if executable when we commit the page. 2476 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE, 2477 flags, -1, 0); 2478 2479 if (addr != MAP_FAILED) { 2480 // anon_mmap() should only get called during VM initialization, 2481 // don't need lock (actually we can skip locking even it can be called 2482 // from multiple threads, because _highest_vm_reserved_address is just a 2483 // hint about the upper limit of non-stack memory regions.) 2484 if ((address)addr + bytes > _highest_vm_reserved_address) { 2485 _highest_vm_reserved_address = (address)addr + bytes; 2486 } 2487 } 2488 2489 return addr == MAP_FAILED ? NULL : addr; 2490} 2491 2492// Don't update _highest_vm_reserved_address, because there might be memory 2493// regions above addr + size. If so, releasing a memory region only creates 2494// a hole in the address space, it doesn't help prevent heap-stack collision. 2495// 2496static int anon_munmap(char * addr, size_t size) { 2497 return ::munmap(addr, size) == 0; 2498} 2499 2500char* os::reserve_memory(size_t bytes, char* requested_addr, 2501 size_t alignment_hint) { 2502 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 2503} 2504 2505bool os::release_memory(char* addr, size_t size) { 2506 return anon_munmap(addr, size); 2507} 2508 2509static address highest_vm_reserved_address() { 2510 return _highest_vm_reserved_address; 2511} 2512 2513static bool linux_mprotect(char* addr, size_t size, int prot) { 2514 // Linux wants the mprotect address argument to be page aligned. 2515 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 2516 2517 // According to SUSv3, mprotect() should only be used with mappings 2518 // established by mmap(), and mmap() always maps whole pages. Unaligned 2519 // 'addr' likely indicates problem in the VM (e.g. trying to change 2520 // protection of malloc'ed or statically allocated memory). Check the 2521 // caller if you hit this assert. 2522 assert(addr == bottom, "sanity check"); 2523 2524 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 2525 return ::mprotect(bottom, size, prot) == 0; 2526} 2527 2528// Set protections specified 2529bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 2530 bool is_committed) { 2531 unsigned int p = 0; 2532 switch (prot) { 2533 case MEM_PROT_NONE: p = PROT_NONE; break; 2534 case MEM_PROT_READ: p = PROT_READ; break; 2535 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 2536 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 2537 default: 2538 ShouldNotReachHere(); 2539 } 2540 // is_committed is unused. 2541 return linux_mprotect(addr, bytes, p); 2542} 2543 2544bool os::guard_memory(char* addr, size_t size) { 2545 return linux_mprotect(addr, size, PROT_NONE); 2546} 2547 2548bool os::unguard_memory(char* addr, size_t size) { 2549 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 2550} 2551 2552// Large page support 2553 2554static size_t _large_page_size = 0; 2555 2556bool os::large_page_init() { 2557 if (!UseLargePages) return false; 2558 2559 if (LargePageSizeInBytes) { 2560 _large_page_size = LargePageSizeInBytes; 2561 } else { 2562 // large_page_size on Linux is used to round up heap size. x86 uses either 2563 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 2564 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 2565 // page as large as 256M. 2566 // 2567 // Here we try to figure out page size by parsing /proc/meminfo and looking 2568 // for a line with the following format: 2569 // Hugepagesize: 2048 kB 2570 // 2571 // If we can't determine the value (e.g. /proc is not mounted, or the text 2572 // format has been changed), we'll use the largest page size supported by 2573 // the processor. 2574 2575 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M); 2576 2577 FILE *fp = fopen("/proc/meminfo", "r"); 2578 if (fp) { 2579 while (!feof(fp)) { 2580 int x = 0; 2581 char buf[16]; 2582 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 2583 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 2584 _large_page_size = x * K; 2585 break; 2586 } 2587 } else { 2588 // skip to next line 2589 for (;;) { 2590 int ch = fgetc(fp); 2591 if (ch == EOF || ch == (int)'\n') break; 2592 } 2593 } 2594 } 2595 fclose(fp); 2596 } 2597 } 2598 2599 const size_t default_page_size = (size_t)Linux::page_size(); 2600 if (_large_page_size > default_page_size) { 2601 _page_sizes[0] = _large_page_size; 2602 _page_sizes[1] = default_page_size; 2603 _page_sizes[2] = 0; 2604 } 2605 2606 // Large page support is available on 2.6 or newer kernel, some vendors 2607 // (e.g. Redhat) have backported it to their 2.4 based distributions. 2608 // We optimistically assume the support is available. If later it turns out 2609 // not true, VM will automatically switch to use regular page size. 2610 return true; 2611} 2612 2613#ifndef SHM_HUGETLB 2614#define SHM_HUGETLB 04000 2615#endif 2616 2617char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) { 2618 // "exec" is passed in but not used. Creating the shared image for 2619 // the code cache doesn't have an SHM_X executable permission to check. 2620 assert(UseLargePages, "only for large pages"); 2621 2622 key_t key = IPC_PRIVATE; 2623 char *addr; 2624 2625 bool warn_on_failure = UseLargePages && 2626 (!FLAG_IS_DEFAULT(UseLargePages) || 2627 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 2628 ); 2629 char msg[128]; 2630 2631 // Create a large shared memory region to attach to based on size. 2632 // Currently, size is the total size of the heap 2633 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 2634 if (shmid == -1) { 2635 // Possible reasons for shmget failure: 2636 // 1. shmmax is too small for Java heap. 2637 // > check shmmax value: cat /proc/sys/kernel/shmmax 2638 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 2639 // 2. not enough large page memory. 2640 // > check available large pages: cat /proc/meminfo 2641 // > increase amount of large pages: 2642 // echo new_value > /proc/sys/vm/nr_hugepages 2643 // Note 1: different Linux may use different name for this property, 2644 // e.g. on Redhat AS-3 it is "hugetlb_pool". 2645 // Note 2: it's possible there's enough physical memory available but 2646 // they are so fragmented after a long run that they can't 2647 // coalesce into large pages. Try to reserve large pages when 2648 // the system is still "fresh". 2649 if (warn_on_failure) { 2650 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 2651 warning(msg); 2652 } 2653 return NULL; 2654 } 2655 2656 // attach to the region 2657 addr = (char*)shmat(shmid, NULL, 0); 2658 int err = errno; 2659 2660 // Remove shmid. If shmat() is successful, the actual shared memory segment 2661 // will be deleted when it's detached by shmdt() or when the process 2662 // terminates. If shmat() is not successful this will remove the shared 2663 // segment immediately. 2664 shmctl(shmid, IPC_RMID, NULL); 2665 2666 if ((intptr_t)addr == -1) { 2667 if (warn_on_failure) { 2668 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 2669 warning(msg); 2670 } 2671 return NULL; 2672 } 2673 2674 return addr; 2675} 2676 2677bool os::release_memory_special(char* base, size_t bytes) { 2678 // detaching the SHM segment will also delete it, see reserve_memory_special() 2679 int rslt = shmdt(base); 2680 return rslt == 0; 2681} 2682 2683size_t os::large_page_size() { 2684 return _large_page_size; 2685} 2686 2687// Linux does not support anonymous mmap with large page memory. The only way 2688// to reserve large page memory without file backing is through SysV shared 2689// memory API. The entire memory region is committed and pinned upfront. 2690// Hopefully this will change in the future... 2691bool os::can_commit_large_page_memory() { 2692 return false; 2693} 2694 2695bool os::can_execute_large_page_memory() { 2696 return false; 2697} 2698 2699// Reserve memory at an arbitrary address, only if that area is 2700// available (and not reserved for something else). 2701 2702char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 2703 const int max_tries = 10; 2704 char* base[max_tries]; 2705 size_t size[max_tries]; 2706 const size_t gap = 0x000000; 2707 2708 // Assert only that the size is a multiple of the page size, since 2709 // that's all that mmap requires, and since that's all we really know 2710 // about at this low abstraction level. If we need higher alignment, 2711 // we can either pass an alignment to this method or verify alignment 2712 // in one of the methods further up the call chain. See bug 5044738. 2713 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 2714 2715 // Repeatedly allocate blocks until the block is allocated at the 2716 // right spot. Give up after max_tries. Note that reserve_memory() will 2717 // automatically update _highest_vm_reserved_address if the call is 2718 // successful. The variable tracks the highest memory address every reserved 2719 // by JVM. It is used to detect heap-stack collision if running with 2720 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 2721 // space than needed, it could confuse the collision detecting code. To 2722 // solve the problem, save current _highest_vm_reserved_address and 2723 // calculate the correct value before return. 2724 address old_highest = _highest_vm_reserved_address; 2725 2726 // Linux mmap allows caller to pass an address as hint; give it a try first, 2727 // if kernel honors the hint then we can return immediately. 2728 char * addr = anon_mmap(requested_addr, bytes, false); 2729 if (addr == requested_addr) { 2730 return requested_addr; 2731 } 2732 2733 if (addr != NULL) { 2734 // mmap() is successful but it fails to reserve at the requested address 2735 anon_munmap(addr, bytes); 2736 } 2737 2738 int i; 2739 for (i = 0; i < max_tries; ++i) { 2740 base[i] = reserve_memory(bytes); 2741 2742 if (base[i] != NULL) { 2743 // Is this the block we wanted? 2744 if (base[i] == requested_addr) { 2745 size[i] = bytes; 2746 break; 2747 } 2748 2749 // Does this overlap the block we wanted? Give back the overlapped 2750 // parts and try again. 2751 2752 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 2753 if (top_overlap >= 0 && top_overlap < bytes) { 2754 unmap_memory(base[i], top_overlap); 2755 base[i] += top_overlap; 2756 size[i] = bytes - top_overlap; 2757 } else { 2758 size_t bottom_overlap = base[i] + bytes - requested_addr; 2759 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 2760 unmap_memory(requested_addr, bottom_overlap); 2761 size[i] = bytes - bottom_overlap; 2762 } else { 2763 size[i] = bytes; 2764 } 2765 } 2766 } 2767 } 2768 2769 // Give back the unused reserved pieces. 2770 2771 for (int j = 0; j < i; ++j) { 2772 if (base[j] != NULL) { 2773 unmap_memory(base[j], size[j]); 2774 } 2775 } 2776 2777 if (i < max_tries) { 2778 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 2779 return requested_addr; 2780 } else { 2781 _highest_vm_reserved_address = old_highest; 2782 return NULL; 2783 } 2784} 2785 2786size_t os::read(int fd, void *buf, unsigned int nBytes) { 2787 return ::read(fd, buf, nBytes); 2788} 2789 2790// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 2791// Solaris uses poll(), linux uses park(). 2792// Poll() is likely a better choice, assuming that Thread.interrupt() 2793// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 2794// SIGSEGV, see 4355769. 2795 2796const int NANOSECS_PER_MILLISECS = 1000000; 2797 2798int os::sleep(Thread* thread, jlong millis, bool interruptible) { 2799 assert(thread == Thread::current(), "thread consistency check"); 2800 2801 ParkEvent * const slp = thread->_SleepEvent ; 2802 slp->reset() ; 2803 OrderAccess::fence() ; 2804 2805 if (interruptible) { 2806 jlong prevtime = javaTimeNanos(); 2807 2808 for (;;) { 2809 if (os::is_interrupted(thread, true)) { 2810 return OS_INTRPT; 2811 } 2812 2813 jlong newtime = javaTimeNanos(); 2814 2815 if (newtime - prevtime < 0) { 2816 // time moving backwards, should only happen if no monotonic clock 2817 // not a guarantee() because JVM should not abort on kernel/glibc bugs 2818 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 2819 } else { 2820 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 2821 } 2822 2823 if(millis <= 0) { 2824 return OS_OK; 2825 } 2826 2827 prevtime = newtime; 2828 2829 { 2830 assert(thread->is_Java_thread(), "sanity check"); 2831 JavaThread *jt = (JavaThread *) thread; 2832 ThreadBlockInVM tbivm(jt); 2833 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 2834 2835 jt->set_suspend_equivalent(); 2836 // cleared by handle_special_suspend_equivalent_condition() or 2837 // java_suspend_self() via check_and_wait_while_suspended() 2838 2839 slp->park(millis); 2840 2841 // were we externally suspended while we were waiting? 2842 jt->check_and_wait_while_suspended(); 2843 } 2844 } 2845 } else { 2846 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 2847 jlong prevtime = javaTimeNanos(); 2848 2849 for (;;) { 2850 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 2851 // the 1st iteration ... 2852 jlong newtime = javaTimeNanos(); 2853 2854 if (newtime - prevtime < 0) { 2855 // time moving backwards, should only happen if no monotonic clock 2856 // not a guarantee() because JVM should not abort on kernel/glibc bugs 2857 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 2858 } else { 2859 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 2860 } 2861 2862 if(millis <= 0) break ; 2863 2864 prevtime = newtime; 2865 slp->park(millis); 2866 } 2867 return OS_OK ; 2868 } 2869} 2870 2871int os::naked_sleep() { 2872 // %% make the sleep time an integer flag. for now use 1 millisec. 2873 return os::sleep(Thread::current(), 1, false); 2874} 2875 2876// Sleep forever; naked call to OS-specific sleep; use with CAUTION 2877void os::infinite_sleep() { 2878 while (true) { // sleep forever ... 2879 ::sleep(100); // ... 100 seconds at a time 2880 } 2881} 2882 2883// Used to convert frequent JVM_Yield() to nops 2884bool os::dont_yield() { 2885 return DontYieldALot; 2886} 2887 2888void os::yield() { 2889 sched_yield(); 2890} 2891 2892os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 2893 2894void os::yield_all(int attempts) { 2895 // Yields to all threads, including threads with lower priorities 2896 // Threads on Linux are all with same priority. The Solaris style 2897 // os::yield_all() with nanosleep(1ms) is not necessary. 2898 sched_yield(); 2899} 2900 2901// Called from the tight loops to possibly influence time-sharing heuristics 2902void os::loop_breaker(int attempts) { 2903 os::yield_all(attempts); 2904} 2905 2906//////////////////////////////////////////////////////////////////////////////// 2907// thread priority support 2908 2909// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 2910// only supports dynamic priority, static priority must be zero. For real-time 2911// applications, Linux supports SCHED_RR which allows static priority (1-99). 2912// However, for large multi-threaded applications, SCHED_RR is not only slower 2913// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 2914// of 5 runs - Sep 2005). 2915// 2916// The following code actually changes the niceness of kernel-thread/LWP. It 2917// has an assumption that setpriority() only modifies one kernel-thread/LWP, 2918// not the entire user process, and user level threads are 1:1 mapped to kernel 2919// threads. It has always been the case, but could change in the future. For 2920// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 2921// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 2922 2923int os::java_to_os_priority[MaxPriority + 1] = { 2924 19, // 0 Entry should never be used 2925 2926 4, // 1 MinPriority 2927 3, // 2 2928 2, // 3 2929 2930 1, // 4 2931 0, // 5 NormPriority 2932 -1, // 6 2933 2934 -2, // 7 2935 -3, // 8 2936 -4, // 9 NearMaxPriority 2937 2938 -5 // 10 MaxPriority 2939}; 2940 2941static int prio_init() { 2942 if (ThreadPriorityPolicy == 1) { 2943 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 2944 // if effective uid is not root. Perhaps, a more elegant way of doing 2945 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 2946 if (geteuid() != 0) { 2947 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 2948 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 2949 } 2950 ThreadPriorityPolicy = 0; 2951 } 2952 } 2953 return 0; 2954} 2955 2956OSReturn os::set_native_priority(Thread* thread, int newpri) { 2957 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 2958 2959 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 2960 return (ret == 0) ? OS_OK : OS_ERR; 2961} 2962 2963OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 2964 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 2965 *priority_ptr = java_to_os_priority[NormPriority]; 2966 return OS_OK; 2967 } 2968 2969 errno = 0; 2970 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 2971 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 2972} 2973 2974// Hint to the underlying OS that a task switch would not be good. 2975// Void return because it's a hint and can fail. 2976void os::hint_no_preempt() {} 2977 2978//////////////////////////////////////////////////////////////////////////////// 2979// suspend/resume support 2980 2981// the low-level signal-based suspend/resume support is a remnant from the 2982// old VM-suspension that used to be for java-suspension, safepoints etc, 2983// within hotspot. Now there is a single use-case for this: 2984// - calling get_thread_pc() on the VMThread by the flat-profiler task 2985// that runs in the watcher thread. 2986// The remaining code is greatly simplified from the more general suspension 2987// code that used to be used. 2988// 2989// The protocol is quite simple: 2990// - suspend: 2991// - sends a signal to the target thread 2992// - polls the suspend state of the osthread using a yield loop 2993// - target thread signal handler (SR_handler) sets suspend state 2994// and blocks in sigsuspend until continued 2995// - resume: 2996// - sets target osthread state to continue 2997// - sends signal to end the sigsuspend loop in the SR_handler 2998// 2999// Note that the SR_lock plays no role in this suspend/resume protocol. 3000// 3001 3002static void resume_clear_context(OSThread *osthread) { 3003 osthread->set_ucontext(NULL); 3004 osthread->set_siginfo(NULL); 3005 3006 // notify the suspend action is completed, we have now resumed 3007 osthread->sr.clear_suspended(); 3008} 3009 3010static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3011 osthread->set_ucontext(context); 3012 osthread->set_siginfo(siginfo); 3013} 3014 3015// 3016// Handler function invoked when a thread's execution is suspended or 3017// resumed. We have to be careful that only async-safe functions are 3018// called here (Note: most pthread functions are not async safe and 3019// should be avoided.) 3020// 3021// Note: sigwait() is a more natural fit than sigsuspend() from an 3022// interface point of view, but sigwait() prevents the signal hander 3023// from being run. libpthread would get very confused by not having 3024// its signal handlers run and prevents sigwait()'s use with the 3025// mutex granting granting signal. 3026// 3027// Currently only ever called on the VMThread 3028// 3029static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3030 // Save and restore errno to avoid confusing native code with EINTR 3031 // after sigsuspend. 3032 int old_errno = errno; 3033 3034 Thread* thread = Thread::current(); 3035 OSThread* osthread = thread->osthread(); 3036 assert(thread->is_VM_thread(), "Must be VMThread"); 3037 // read current suspend action 3038 int action = osthread->sr.suspend_action(); 3039 if (action == SR_SUSPEND) { 3040 suspend_save_context(osthread, siginfo, context); 3041 3042 // Notify the suspend action is about to be completed. do_suspend() 3043 // waits until SR_SUSPENDED is set and then returns. We will wait 3044 // here for a resume signal and that completes the suspend-other 3045 // action. do_suspend/do_resume is always called as a pair from 3046 // the same thread - so there are no races 3047 3048 // notify the caller 3049 osthread->sr.set_suspended(); 3050 3051 sigset_t suspend_set; // signals for sigsuspend() 3052 3053 // get current set of blocked signals and unblock resume signal 3054 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3055 sigdelset(&suspend_set, SR_signum); 3056 3057 // wait here until we are resumed 3058 do { 3059 sigsuspend(&suspend_set); 3060 // ignore all returns until we get a resume signal 3061 } while (osthread->sr.suspend_action() != SR_CONTINUE); 3062 3063 resume_clear_context(osthread); 3064 3065 } else { 3066 assert(action == SR_CONTINUE, "unexpected sr action"); 3067 // nothing special to do - just leave the handler 3068 } 3069 3070 errno = old_errno; 3071} 3072 3073 3074static int SR_initialize() { 3075 struct sigaction act; 3076 char *s; 3077 /* Get signal number to use for suspend/resume */ 3078 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3079 int sig = ::strtol(s, 0, 10); 3080 if (sig > 0 || sig < _NSIG) { 3081 SR_signum = sig; 3082 } 3083 } 3084 3085 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3086 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3087 3088 sigemptyset(&SR_sigset); 3089 sigaddset(&SR_sigset, SR_signum); 3090 3091 /* Set up signal handler for suspend/resume */ 3092 act.sa_flags = SA_RESTART|SA_SIGINFO; 3093 act.sa_handler = (void (*)(int)) SR_handler; 3094 3095 // SR_signum is blocked by default. 3096 // 4528190 - We also need to block pthread restart signal (32 on all 3097 // supported Linux platforms). Note that LinuxThreads need to block 3098 // this signal for all threads to work properly. So we don't have 3099 // to use hard-coded signal number when setting up the mask. 3100 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 3101 3102 if (sigaction(SR_signum, &act, 0) == -1) { 3103 return -1; 3104 } 3105 3106 // Save signal flag 3107 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 3108 return 0; 3109} 3110 3111static int SR_finalize() { 3112 return 0; 3113} 3114 3115 3116// returns true on success and false on error - really an error is fatal 3117// but this seems the normal response to library errors 3118static bool do_suspend(OSThread* osthread) { 3119 // mark as suspended and send signal 3120 osthread->sr.set_suspend_action(SR_SUSPEND); 3121 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3122 assert_status(status == 0, status, "pthread_kill"); 3123 3124 // check status and wait until notified of suspension 3125 if (status == 0) { 3126 for (int i = 0; !osthread->sr.is_suspended(); i++) { 3127 os::yield_all(i); 3128 } 3129 osthread->sr.set_suspend_action(SR_NONE); 3130 return true; 3131 } 3132 else { 3133 osthread->sr.set_suspend_action(SR_NONE); 3134 return false; 3135 } 3136} 3137 3138static void do_resume(OSThread* osthread) { 3139 assert(osthread->sr.is_suspended(), "thread should be suspended"); 3140 osthread->sr.set_suspend_action(SR_CONTINUE); 3141 3142 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3143 assert_status(status == 0, status, "pthread_kill"); 3144 // check status and wait unit notified of resumption 3145 if (status == 0) { 3146 for (int i = 0; osthread->sr.is_suspended(); i++) { 3147 os::yield_all(i); 3148 } 3149 } 3150 osthread->sr.set_suspend_action(SR_NONE); 3151} 3152 3153//////////////////////////////////////////////////////////////////////////////// 3154// interrupt support 3155 3156void os::interrupt(Thread* thread) { 3157 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3158 "possibility of dangling Thread pointer"); 3159 3160 OSThread* osthread = thread->osthread(); 3161 3162 if (!osthread->interrupted()) { 3163 osthread->set_interrupted(true); 3164 // More than one thread can get here with the same value of osthread, 3165 // resulting in multiple notifications. We do, however, want the store 3166 // to interrupted() to be visible to other threads before we execute unpark(). 3167 OrderAccess::fence(); 3168 ParkEvent * const slp = thread->_SleepEvent ; 3169 if (slp != NULL) slp->unpark() ; 3170 } 3171 3172 // For JSR166. Unpark even if interrupt status already was set 3173 if (thread->is_Java_thread()) 3174 ((JavaThread*)thread)->parker()->unpark(); 3175 3176 ParkEvent * ev = thread->_ParkEvent ; 3177 if (ev != NULL) ev->unpark() ; 3178 3179} 3180 3181bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 3182 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3183 "possibility of dangling Thread pointer"); 3184 3185 OSThread* osthread = thread->osthread(); 3186 3187 bool interrupted = osthread->interrupted(); 3188 3189 if (interrupted && clear_interrupted) { 3190 osthread->set_interrupted(false); 3191 // consider thread->_SleepEvent->reset() ... optional optimization 3192 } 3193 3194 return interrupted; 3195} 3196 3197/////////////////////////////////////////////////////////////////////////////////// 3198// signal handling (except suspend/resume) 3199 3200// This routine may be used by user applications as a "hook" to catch signals. 3201// The user-defined signal handler must pass unrecognized signals to this 3202// routine, and if it returns true (non-zero), then the signal handler must 3203// return immediately. If the flag "abort_if_unrecognized" is true, then this 3204// routine will never retun false (zero), but instead will execute a VM panic 3205// routine kill the process. 3206// 3207// If this routine returns false, it is OK to call it again. This allows 3208// the user-defined signal handler to perform checks either before or after 3209// the VM performs its own checks. Naturally, the user code would be making 3210// a serious error if it tried to handle an exception (such as a null check 3211// or breakpoint) that the VM was generating for its own correct operation. 3212// 3213// This routine may recognize any of the following kinds of signals: 3214// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 3215// It should be consulted by handlers for any of those signals. 3216// 3217// The caller of this routine must pass in the three arguments supplied 3218// to the function referred to in the "sa_sigaction" (not the "sa_handler") 3219// field of the structure passed to sigaction(). This routine assumes that 3220// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 3221// 3222// Note that the VM will print warnings if it detects conflicting signal 3223// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 3224// 3225extern "C" int 3226JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 3227 void* ucontext, int abort_if_unrecognized); 3228 3229void signalHandler(int sig, siginfo_t* info, void* uc) { 3230 assert(info != NULL && uc != NULL, "it must be old kernel"); 3231 JVM_handle_linux_signal(sig, info, uc, true); 3232} 3233 3234 3235// This boolean allows users to forward their own non-matching signals 3236// to JVM_handle_linux_signal, harmlessly. 3237bool os::Linux::signal_handlers_are_installed = false; 3238 3239// For signal-chaining 3240struct sigaction os::Linux::sigact[MAXSIGNUM]; 3241unsigned int os::Linux::sigs = 0; 3242bool os::Linux::libjsig_is_loaded = false; 3243typedef struct sigaction *(*get_signal_t)(int); 3244get_signal_t os::Linux::get_signal_action = NULL; 3245 3246struct sigaction* os::Linux::get_chained_signal_action(int sig) { 3247 struct sigaction *actp = NULL; 3248 3249 if (libjsig_is_loaded) { 3250 // Retrieve the old signal handler from libjsig 3251 actp = (*get_signal_action)(sig); 3252 } 3253 if (actp == NULL) { 3254 // Retrieve the preinstalled signal handler from jvm 3255 actp = get_preinstalled_handler(sig); 3256 } 3257 3258 return actp; 3259} 3260 3261static bool call_chained_handler(struct sigaction *actp, int sig, 3262 siginfo_t *siginfo, void *context) { 3263 // Call the old signal handler 3264 if (actp->sa_handler == SIG_DFL) { 3265 // It's more reasonable to let jvm treat it as an unexpected exception 3266 // instead of taking the default action. 3267 return false; 3268 } else if (actp->sa_handler != SIG_IGN) { 3269 if ((actp->sa_flags & SA_NODEFER) == 0) { 3270 // automaticlly block the signal 3271 sigaddset(&(actp->sa_mask), sig); 3272 } 3273 3274 sa_handler_t hand; 3275 sa_sigaction_t sa; 3276 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 3277 // retrieve the chained handler 3278 if (siginfo_flag_set) { 3279 sa = actp->sa_sigaction; 3280 } else { 3281 hand = actp->sa_handler; 3282 } 3283 3284 if ((actp->sa_flags & SA_RESETHAND) != 0) { 3285 actp->sa_handler = SIG_DFL; 3286 } 3287 3288 // try to honor the signal mask 3289 sigset_t oset; 3290 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 3291 3292 // call into the chained handler 3293 if (siginfo_flag_set) { 3294 (*sa)(sig, siginfo, context); 3295 } else { 3296 (*hand)(sig); 3297 } 3298 3299 // restore the signal mask 3300 pthread_sigmask(SIG_SETMASK, &oset, 0); 3301 } 3302 // Tell jvm's signal handler the signal is taken care of. 3303 return true; 3304} 3305 3306bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 3307 bool chained = false; 3308 // signal-chaining 3309 if (UseSignalChaining) { 3310 struct sigaction *actp = get_chained_signal_action(sig); 3311 if (actp != NULL) { 3312 chained = call_chained_handler(actp, sig, siginfo, context); 3313 } 3314 } 3315 return chained; 3316} 3317 3318struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 3319 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 3320 return &sigact[sig]; 3321 } 3322 return NULL; 3323} 3324 3325void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 3326 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3327 sigact[sig] = oldAct; 3328 sigs |= (unsigned int)1 << sig; 3329} 3330 3331// for diagnostic 3332int os::Linux::sigflags[MAXSIGNUM]; 3333 3334int os::Linux::get_our_sigflags(int sig) { 3335 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3336 return sigflags[sig]; 3337} 3338 3339void os::Linux::set_our_sigflags(int sig, int flags) { 3340 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3341 sigflags[sig] = flags; 3342} 3343 3344void os::Linux::set_signal_handler(int sig, bool set_installed) { 3345 // Check for overwrite. 3346 struct sigaction oldAct; 3347 sigaction(sig, (struct sigaction*)NULL, &oldAct); 3348 3349 void* oldhand = oldAct.sa_sigaction 3350 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3351 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3352 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 3353 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 3354 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 3355 if (AllowUserSignalHandlers || !set_installed) { 3356 // Do not overwrite; user takes responsibility to forward to us. 3357 return; 3358 } else if (UseSignalChaining) { 3359 // save the old handler in jvm 3360 save_preinstalled_handler(sig, oldAct); 3361 // libjsig also interposes the sigaction() call below and saves the 3362 // old sigaction on it own. 3363 } else { 3364 fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig); 3365 } 3366 } 3367 3368 struct sigaction sigAct; 3369 sigfillset(&(sigAct.sa_mask)); 3370 sigAct.sa_handler = SIG_DFL; 3371 if (!set_installed) { 3372 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3373 } else { 3374 sigAct.sa_sigaction = signalHandler; 3375 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3376 } 3377 // Save flags, which are set by ours 3378 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3379 sigflags[sig] = sigAct.sa_flags; 3380 3381 int ret = sigaction(sig, &sigAct, &oldAct); 3382 assert(ret == 0, "check"); 3383 3384 void* oldhand2 = oldAct.sa_sigaction 3385 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3386 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3387 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 3388} 3389 3390// install signal handlers for signals that HotSpot needs to 3391// handle in order to support Java-level exception handling. 3392 3393void os::Linux::install_signal_handlers() { 3394 if (!signal_handlers_are_installed) { 3395 signal_handlers_are_installed = true; 3396 3397 // signal-chaining 3398 typedef void (*signal_setting_t)(); 3399 signal_setting_t begin_signal_setting = NULL; 3400 signal_setting_t end_signal_setting = NULL; 3401 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3402 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 3403 if (begin_signal_setting != NULL) { 3404 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3405 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 3406 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 3407 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 3408 libjsig_is_loaded = true; 3409 assert(UseSignalChaining, "should enable signal-chaining"); 3410 } 3411 if (libjsig_is_loaded) { 3412 // Tell libjsig jvm is setting signal handlers 3413 (*begin_signal_setting)(); 3414 } 3415 3416 set_signal_handler(SIGSEGV, true); 3417 set_signal_handler(SIGPIPE, true); 3418 set_signal_handler(SIGBUS, true); 3419 set_signal_handler(SIGILL, true); 3420 set_signal_handler(SIGFPE, true); 3421 set_signal_handler(SIGXFSZ, true); 3422 3423 if (libjsig_is_loaded) { 3424 // Tell libjsig jvm finishes setting signal handlers 3425 (*end_signal_setting)(); 3426 } 3427 3428 // We don't activate signal checker if libjsig is in place, we trust ourselves 3429 // and if UserSignalHandler is installed all bets are off 3430 if (CheckJNICalls) { 3431 if (libjsig_is_loaded) { 3432 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 3433 check_signals = false; 3434 } 3435 if (AllowUserSignalHandlers) { 3436 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 3437 check_signals = false; 3438 } 3439 } 3440 } 3441} 3442 3443// This is the fastest way to get thread cpu time on Linux. 3444// Returns cpu time (user+sys) for any thread, not only for current. 3445// POSIX compliant clocks are implemented in the kernels 2.6.16+. 3446// It might work on 2.6.10+ with a special kernel/glibc patch. 3447// For reference, please, see IEEE Std 1003.1-2004: 3448// http://www.unix.org/single_unix_specification 3449 3450jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 3451 struct timespec tp; 3452 int rc = os::Linux::clock_gettime(clockid, &tp); 3453 assert(rc == 0, "clock_gettime is expected to return 0 code"); 3454 3455 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec; 3456} 3457 3458///// 3459// glibc on Linux platform uses non-documented flag 3460// to indicate, that some special sort of signal 3461// trampoline is used. 3462// We will never set this flag, and we should 3463// ignore this flag in our diagnostic 3464#ifdef SIGNIFICANT_SIGNAL_MASK 3465#undef SIGNIFICANT_SIGNAL_MASK 3466#endif 3467#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 3468 3469static const char* get_signal_handler_name(address handler, 3470 char* buf, int buflen) { 3471 int offset; 3472 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 3473 if (found) { 3474 // skip directory names 3475 const char *p1, *p2; 3476 p1 = buf; 3477 size_t len = strlen(os::file_separator()); 3478 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 3479 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 3480 } else { 3481 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 3482 } 3483 return buf; 3484} 3485 3486static void print_signal_handler(outputStream* st, int sig, 3487 char* buf, size_t buflen) { 3488 struct sigaction sa; 3489 3490 sigaction(sig, NULL, &sa); 3491 3492 // See comment for SIGNIFICANT_SIGNAL_MASK define 3493 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3494 3495 st->print("%s: ", os::exception_name(sig, buf, buflen)); 3496 3497 address handler = (sa.sa_flags & SA_SIGINFO) 3498 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 3499 : CAST_FROM_FN_PTR(address, sa.sa_handler); 3500 3501 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 3502 st->print("SIG_DFL"); 3503 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 3504 st->print("SIG_IGN"); 3505 } else { 3506 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 3507 } 3508 3509 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 3510 3511 address rh = VMError::get_resetted_sighandler(sig); 3512 // May be, handler was resetted by VMError? 3513 if(rh != NULL) { 3514 handler = rh; 3515 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 3516 } 3517 3518 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 3519 3520 // Check: is it our handler? 3521 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 3522 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 3523 // It is our signal handler 3524 // check for flags, reset system-used one! 3525 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 3526 st->print( 3527 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 3528 os::Linux::get_our_sigflags(sig)); 3529 } 3530 } 3531 st->cr(); 3532} 3533 3534 3535#define DO_SIGNAL_CHECK(sig) \ 3536 if (!sigismember(&check_signal_done, sig)) \ 3537 os::Linux::check_signal_handler(sig) 3538 3539// This method is a periodic task to check for misbehaving JNI applications 3540// under CheckJNI, we can add any periodic checks here 3541 3542void os::run_periodic_checks() { 3543 3544 if (check_signals == false) return; 3545 3546 // SEGV and BUS if overridden could potentially prevent 3547 // generation of hs*.log in the event of a crash, debugging 3548 // such a case can be very challenging, so we absolutely 3549 // check the following for a good measure: 3550 DO_SIGNAL_CHECK(SIGSEGV); 3551 DO_SIGNAL_CHECK(SIGILL); 3552 DO_SIGNAL_CHECK(SIGFPE); 3553 DO_SIGNAL_CHECK(SIGBUS); 3554 DO_SIGNAL_CHECK(SIGPIPE); 3555 DO_SIGNAL_CHECK(SIGXFSZ); 3556 3557 3558 // ReduceSignalUsage allows the user to override these handlers 3559 // see comments at the very top and jvm_solaris.h 3560 if (!ReduceSignalUsage) { 3561 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 3562 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 3563 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 3564 DO_SIGNAL_CHECK(BREAK_SIGNAL); 3565 } 3566 3567 DO_SIGNAL_CHECK(SR_signum); 3568 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 3569} 3570 3571typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 3572 3573static os_sigaction_t os_sigaction = NULL; 3574 3575void os::Linux::check_signal_handler(int sig) { 3576 char buf[O_BUFLEN]; 3577 address jvmHandler = NULL; 3578 3579 3580 struct sigaction act; 3581 if (os_sigaction == NULL) { 3582 // only trust the default sigaction, in case it has been interposed 3583 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 3584 if (os_sigaction == NULL) return; 3585 } 3586 3587 os_sigaction(sig, (struct sigaction*)NULL, &act); 3588 3589 3590 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3591 3592 address thisHandler = (act.sa_flags & SA_SIGINFO) 3593 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 3594 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 3595 3596 3597 switch(sig) { 3598 case SIGSEGV: 3599 case SIGBUS: 3600 case SIGFPE: 3601 case SIGPIPE: 3602 case SIGILL: 3603 case SIGXFSZ: 3604 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 3605 break; 3606 3607 case SHUTDOWN1_SIGNAL: 3608 case SHUTDOWN2_SIGNAL: 3609 case SHUTDOWN3_SIGNAL: 3610 case BREAK_SIGNAL: 3611 jvmHandler = (address)user_handler(); 3612 break; 3613 3614 case INTERRUPT_SIGNAL: 3615 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 3616 break; 3617 3618 default: 3619 if (sig == SR_signum) { 3620 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 3621 } else { 3622 return; 3623 } 3624 break; 3625 } 3626 3627 if (thisHandler != jvmHandler) { 3628 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 3629 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 3630 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 3631 // No need to check this sig any longer 3632 sigaddset(&check_signal_done, sig); 3633 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 3634 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 3635 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 3636 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 3637 // No need to check this sig any longer 3638 sigaddset(&check_signal_done, sig); 3639 } 3640 3641 // Dump all the signal 3642 if (sigismember(&check_signal_done, sig)) { 3643 print_signal_handlers(tty, buf, O_BUFLEN); 3644 } 3645} 3646 3647extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 3648 3649extern bool signal_name(int signo, char* buf, size_t len); 3650 3651const char* os::exception_name(int exception_code, char* buf, size_t size) { 3652 if (0 < exception_code && exception_code <= SIGRTMAX) { 3653 // signal 3654 if (!signal_name(exception_code, buf, size)) { 3655 jio_snprintf(buf, size, "SIG%d", exception_code); 3656 } 3657 return buf; 3658 } else { 3659 return NULL; 3660 } 3661} 3662 3663// this is called _before_ the most of global arguments have been parsed 3664void os::init(void) { 3665 char dummy; /* used to get a guess on initial stack address */ 3666// first_hrtime = gethrtime(); 3667 3668 // With LinuxThreads the JavaMain thread pid (primordial thread) 3669 // is different than the pid of the java launcher thread. 3670 // So, on Linux, the launcher thread pid is passed to the VM 3671 // via the sun.java.launcher.pid property. 3672 // Use this property instead of getpid() if it was correctly passed. 3673 // See bug 6351349. 3674 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 3675 3676 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 3677 3678 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 3679 3680 init_random(1234567); 3681 3682 ThreadCritical::initialize(); 3683 3684 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 3685 if (Linux::page_size() == -1) { 3686 fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno)); 3687 } 3688 init_page_sizes((size_t) Linux::page_size()); 3689 3690 Linux::initialize_system_info(); 3691 3692 // main_thread points to the aboriginal thread 3693 Linux::_main_thread = pthread_self(); 3694 3695 Linux::clock_init(); 3696 initial_time_count = os::elapsed_counter(); 3697 pthread_mutex_init(&dl_mutex, NULL); 3698} 3699 3700// To install functions for atexit system call 3701extern "C" { 3702 static void perfMemory_exit_helper() { 3703 perfMemory_exit(); 3704 } 3705} 3706 3707// this is called _after_ the global arguments have been parsed 3708jint os::init_2(void) 3709{ 3710 Linux::fast_thread_clock_init(); 3711 3712 // Allocate a single page and mark it as readable for safepoint polling 3713 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3714 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 3715 3716 os::set_polling_page( polling_page ); 3717 3718#ifndef PRODUCT 3719 if(Verbose && PrintMiscellaneous) 3720 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 3721#endif 3722 3723 if (!UseMembar) { 3724 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3725 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); 3726 os::set_memory_serialize_page( mem_serialize_page ); 3727 3728#ifndef PRODUCT 3729 if(Verbose && PrintMiscellaneous) 3730 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 3731#endif 3732 } 3733 3734 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init()); 3735 3736 // initialize suspend/resume support - must do this before signal_sets_init() 3737 if (SR_initialize() != 0) { 3738 perror("SR_initialize failed"); 3739 return JNI_ERR; 3740 } 3741 3742 Linux::signal_sets_init(); 3743 Linux::install_signal_handlers(); 3744 3745 size_t threadStackSizeInBytes = ThreadStackSize * K; 3746 if (threadStackSizeInBytes != 0 && 3747 threadStackSizeInBytes < Linux::min_stack_allowed) { 3748 tty->print_cr("\nThe stack size specified is too small, " 3749 "Specify at least %dk", 3750 Linux::min_stack_allowed / K); 3751 return JNI_ERR; 3752 } 3753 3754 // Make the stack size a multiple of the page size so that 3755 // the yellow/red zones can be guarded. 3756 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 3757 vm_page_size())); 3758 3759 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 3760 3761 Linux::libpthread_init(); 3762 if (PrintMiscellaneous && (Verbose || WizardMode)) { 3763 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 3764 Linux::glibc_version(), Linux::libpthread_version(), 3765 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 3766 } 3767 3768 if (UseNUMA) { 3769 if (!Linux::libnuma_init()) { 3770 UseNUMA = false; 3771 } else { 3772 if ((Linux::numa_max_node() < 1)) { 3773 // There's only one node(they start from 0), disable NUMA. 3774 UseNUMA = false; 3775 } 3776 } 3777 if (!UseNUMA && ForceNUMA) { 3778 UseNUMA = true; 3779 } 3780 } 3781 3782 if (MaxFDLimit) { 3783 // set the number of file descriptors to max. print out error 3784 // if getrlimit/setrlimit fails but continue regardless. 3785 struct rlimit nbr_files; 3786 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 3787 if (status != 0) { 3788 if (PrintMiscellaneous && (Verbose || WizardMode)) 3789 perror("os::init_2 getrlimit failed"); 3790 } else { 3791 nbr_files.rlim_cur = nbr_files.rlim_max; 3792 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 3793 if (status != 0) { 3794 if (PrintMiscellaneous && (Verbose || WizardMode)) 3795 perror("os::init_2 setrlimit failed"); 3796 } 3797 } 3798 } 3799 3800 // Initialize lock used to serialize thread creation (see os::create_thread) 3801 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 3802 3803 // Initialize HPI. 3804 jint hpi_result = hpi::initialize(); 3805 if (hpi_result != JNI_OK) { 3806 tty->print_cr("There was an error trying to initialize the HPI library."); 3807 return hpi_result; 3808 } 3809 3810 // at-exit methods are called in the reverse order of their registration. 3811 // atexit functions are called on return from main or as a result of a 3812 // call to exit(3C). There can be only 32 of these functions registered 3813 // and atexit() does not set errno. 3814 3815 if (PerfAllowAtExitRegistration) { 3816 // only register atexit functions if PerfAllowAtExitRegistration is set. 3817 // atexit functions can be delayed until process exit time, which 3818 // can be problematic for embedded VM situations. Embedded VMs should 3819 // call DestroyJavaVM() to assure that VM resources are released. 3820 3821 // note: perfMemory_exit_helper atexit function may be removed in 3822 // the future if the appropriate cleanup code can be added to the 3823 // VM_Exit VMOperation's doit method. 3824 if (atexit(perfMemory_exit_helper) != 0) { 3825 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 3826 } 3827 } 3828 3829 // initialize thread priority policy 3830 prio_init(); 3831 3832 return JNI_OK; 3833} 3834 3835// Mark the polling page as unreadable 3836void os::make_polling_page_unreadable(void) { 3837 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 3838 fatal("Could not disable polling page"); 3839}; 3840 3841// Mark the polling page as readable 3842void os::make_polling_page_readable(void) { 3843 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 3844 fatal("Could not enable polling page"); 3845 } 3846}; 3847 3848int os::active_processor_count() { 3849 // Linux doesn't yet have a (official) notion of processor sets, 3850 // so just return the number of online processors. 3851 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 3852 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 3853 return online_cpus; 3854} 3855 3856bool os::distribute_processes(uint length, uint* distribution) { 3857 // Not yet implemented. 3858 return false; 3859} 3860 3861bool os::bind_to_processor(uint processor_id) { 3862 // Not yet implemented. 3863 return false; 3864} 3865 3866/// 3867 3868// Suspends the target using the signal mechanism and then grabs the PC before 3869// resuming the target. Used by the flat-profiler only 3870ExtendedPC os::get_thread_pc(Thread* thread) { 3871 // Make sure that it is called by the watcher for the VMThread 3872 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 3873 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 3874 3875 ExtendedPC epc; 3876 3877 OSThread* osthread = thread->osthread(); 3878 if (do_suspend(osthread)) { 3879 if (osthread->ucontext() != NULL) { 3880 epc = os::Linux::ucontext_get_pc(osthread->ucontext()); 3881 } else { 3882 // NULL context is unexpected, double-check this is the VMThread 3883 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 3884 } 3885 do_resume(osthread); 3886 } 3887 // failure means pthread_kill failed for some reason - arguably this is 3888 // a fatal problem, but such problems are ignored elsewhere 3889 3890 return epc; 3891} 3892 3893int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 3894{ 3895 if (is_NPTL()) { 3896 return pthread_cond_timedwait(_cond, _mutex, _abstime); 3897 } else { 3898#ifndef IA64 3899 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 3900 // word back to default 64bit precision if condvar is signaled. Java 3901 // wants 53bit precision. Save and restore current value. 3902 int fpu = get_fpu_control_word(); 3903#endif // IA64 3904 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 3905#ifndef IA64 3906 set_fpu_control_word(fpu); 3907#endif // IA64 3908 return status; 3909 } 3910} 3911 3912//////////////////////////////////////////////////////////////////////////////// 3913// debug support 3914 3915#ifndef PRODUCT 3916static address same_page(address x, address y) { 3917 int page_bits = -os::vm_page_size(); 3918 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) 3919 return x; 3920 else if (x > y) 3921 return (address)(intptr_t(y) | ~page_bits) + 1; 3922 else 3923 return (address)(intptr_t(y) & page_bits); 3924} 3925 3926bool os::find(address addr) { 3927 Dl_info dlinfo; 3928 memset(&dlinfo, 0, sizeof(dlinfo)); 3929 if (dladdr(addr, &dlinfo)) { 3930 tty->print(PTR_FORMAT ": ", addr); 3931 if (dlinfo.dli_sname != NULL) { 3932 tty->print("%s+%#x", dlinfo.dli_sname, 3933 addr - (intptr_t)dlinfo.dli_saddr); 3934 } else if (dlinfo.dli_fname) { 3935 tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 3936 } else { 3937 tty->print("<absolute address>"); 3938 } 3939 if (dlinfo.dli_fname) { 3940 tty->print(" in %s", dlinfo.dli_fname); 3941 } 3942 if (dlinfo.dli_fbase) { 3943 tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 3944 } 3945 tty->cr(); 3946 3947 if (Verbose) { 3948 // decode some bytes around the PC 3949 address begin = same_page(addr-40, addr); 3950 address end = same_page(addr+40, addr); 3951 address lowest = (address) dlinfo.dli_sname; 3952 if (!lowest) lowest = (address) dlinfo.dli_fbase; 3953 if (begin < lowest) begin = lowest; 3954 Dl_info dlinfo2; 3955 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr 3956 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 3957 end = (address) dlinfo2.dli_saddr; 3958 Disassembler::decode(begin, end); 3959 } 3960 return true; 3961 } 3962 return false; 3963} 3964 3965#endif 3966 3967//////////////////////////////////////////////////////////////////////////////// 3968// misc 3969 3970// This does not do anything on Linux. This is basically a hook for being 3971// able to use structured exception handling (thread-local exception filters) 3972// on, e.g., Win32. 3973void 3974os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 3975 JavaCallArguments* args, Thread* thread) { 3976 f(value, method, args, thread); 3977} 3978 3979void os::print_statistics() { 3980} 3981 3982int os::message_box(const char* title, const char* message) { 3983 int i; 3984 fdStream err(defaultStream::error_fd()); 3985 for (i = 0; i < 78; i++) err.print_raw("="); 3986 err.cr(); 3987 err.print_raw_cr(title); 3988 for (i = 0; i < 78; i++) err.print_raw("-"); 3989 err.cr(); 3990 err.print_raw_cr(message); 3991 for (i = 0; i < 78; i++) err.print_raw("="); 3992 err.cr(); 3993 3994 char buf[16]; 3995 // Prevent process from exiting upon "read error" without consuming all CPU 3996 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 3997 3998 return buf[0] == 'y' || buf[0] == 'Y'; 3999} 4000 4001int os::stat(const char *path, struct stat *sbuf) { 4002 char pathbuf[MAX_PATH]; 4003 if (strlen(path) > MAX_PATH - 1) { 4004 errno = ENAMETOOLONG; 4005 return -1; 4006 } 4007 hpi::native_path(strcpy(pathbuf, path)); 4008 return ::stat(pathbuf, sbuf); 4009} 4010 4011bool os::check_heap(bool force) { 4012 return true; 4013} 4014 4015int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 4016 return ::vsnprintf(buf, count, format, args); 4017} 4018 4019// Is a (classpath) directory empty? 4020bool os::dir_is_empty(const char* path) { 4021 DIR *dir = NULL; 4022 struct dirent *ptr; 4023 4024 dir = opendir(path); 4025 if (dir == NULL) return true; 4026 4027 /* Scan the directory */ 4028 bool result = true; 4029 char buf[sizeof(struct dirent) + MAX_PATH]; 4030 while (result && (ptr = ::readdir(dir)) != NULL) { 4031 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 4032 result = false; 4033 } 4034 } 4035 closedir(dir); 4036 return result; 4037} 4038 4039// create binary file, rewriting existing file if required 4040int os::create_binary_file(const char* path, bool rewrite_existing) { 4041 int oflags = O_WRONLY | O_CREAT; 4042 if (!rewrite_existing) { 4043 oflags |= O_EXCL; 4044 } 4045 return ::open64(path, oflags, S_IREAD | S_IWRITE); 4046} 4047 4048// return current position of file pointer 4049jlong os::current_file_offset(int fd) { 4050 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 4051} 4052 4053// move file pointer to the specified offset 4054jlong os::seek_to_file_offset(int fd, jlong offset) { 4055 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 4056} 4057 4058// Map a block of memory. 4059char* os::map_memory(int fd, const char* file_name, size_t file_offset, 4060 char *addr, size_t bytes, bool read_only, 4061 bool allow_exec) { 4062 int prot; 4063 int flags; 4064 4065 if (read_only) { 4066 prot = PROT_READ; 4067 flags = MAP_SHARED; 4068 } else { 4069 prot = PROT_READ | PROT_WRITE; 4070 flags = MAP_PRIVATE; 4071 } 4072 4073 if (allow_exec) { 4074 prot |= PROT_EXEC; 4075 } 4076 4077 if (addr != NULL) { 4078 flags |= MAP_FIXED; 4079 } 4080 4081 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 4082 fd, file_offset); 4083 if (mapped_address == MAP_FAILED) { 4084 return NULL; 4085 } 4086 return mapped_address; 4087} 4088 4089 4090// Remap a block of memory. 4091char* os::remap_memory(int fd, const char* file_name, size_t file_offset, 4092 char *addr, size_t bytes, bool read_only, 4093 bool allow_exec) { 4094 // same as map_memory() on this OS 4095 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 4096 allow_exec); 4097} 4098 4099 4100// Unmap a block of memory. 4101bool os::unmap_memory(char* addr, size_t bytes) { 4102 return munmap(addr, bytes) == 0; 4103} 4104 4105static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 4106 4107static clockid_t thread_cpu_clockid(Thread* thread) { 4108 pthread_t tid = thread->osthread()->pthread_id(); 4109 clockid_t clockid; 4110 4111 // Get thread clockid 4112 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 4113 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 4114 return clockid; 4115} 4116 4117// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 4118// are used by JVM M&M and JVMTI to get user+sys or user CPU time 4119// of a thread. 4120// 4121// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 4122// the fast estimate available on the platform. 4123 4124jlong os::current_thread_cpu_time() { 4125 if (os::Linux::supports_fast_thread_cpu_time()) { 4126 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4127 } else { 4128 // return user + sys since the cost is the same 4129 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 4130 } 4131} 4132 4133jlong os::thread_cpu_time(Thread* thread) { 4134 // consistent with what current_thread_cpu_time() returns 4135 if (os::Linux::supports_fast_thread_cpu_time()) { 4136 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4137 } else { 4138 return slow_thread_cpu_time(thread, true /* user + sys */); 4139 } 4140} 4141 4142jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 4143 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4144 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4145 } else { 4146 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 4147 } 4148} 4149 4150jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4151 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4152 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4153 } else { 4154 return slow_thread_cpu_time(thread, user_sys_cpu_time); 4155 } 4156} 4157 4158// 4159// -1 on error. 4160// 4161 4162static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4163 static bool proc_pid_cpu_avail = true; 4164 static bool proc_task_unchecked = true; 4165 static const char *proc_stat_path = "/proc/%d/stat"; 4166 pid_t tid = thread->osthread()->thread_id(); 4167 int i; 4168 char *s; 4169 char stat[2048]; 4170 int statlen; 4171 char proc_name[64]; 4172 int count; 4173 long sys_time, user_time; 4174 char string[64]; 4175 int idummy; 4176 long ldummy; 4177 FILE *fp; 4178 4179 // We first try accessing /proc/<pid>/cpu since this is faster to 4180 // process. If this file is not present (linux kernels 2.5 and above) 4181 // then we open /proc/<pid>/stat. 4182 if ( proc_pid_cpu_avail ) { 4183 sprintf(proc_name, "/proc/%d/cpu", tid); 4184 fp = fopen(proc_name, "r"); 4185 if ( fp != NULL ) { 4186 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time); 4187 fclose(fp); 4188 if ( count != 3 ) return -1; 4189 4190 if (user_sys_cpu_time) { 4191 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4192 } else { 4193 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4194 } 4195 } 4196 else proc_pid_cpu_avail = false; 4197 } 4198 4199 // The /proc/<tid>/stat aggregates per-process usage on 4200 // new Linux kernels 2.6+ where NPTL is supported. 4201 // The /proc/self/task/<tid>/stat still has the per-thread usage. 4202 // See bug 6328462. 4203 // There can be no directory /proc/self/task on kernels 2.4 with NPTL 4204 // and possibly in some other cases, so we check its availability. 4205 if (proc_task_unchecked && os::Linux::is_NPTL()) { 4206 // This is executed only once 4207 proc_task_unchecked = false; 4208 fp = fopen("/proc/self/task", "r"); 4209 if (fp != NULL) { 4210 proc_stat_path = "/proc/self/task/%d/stat"; 4211 fclose(fp); 4212 } 4213 } 4214 4215 sprintf(proc_name, proc_stat_path, tid); 4216 fp = fopen(proc_name, "r"); 4217 if ( fp == NULL ) return -1; 4218 statlen = fread(stat, 1, 2047, fp); 4219 stat[statlen] = '\0'; 4220 fclose(fp); 4221 4222 // Skip pid and the command string. Note that we could be dealing with 4223 // weird command names, e.g. user could decide to rename java launcher 4224 // to "java 1.4.2 :)", then the stat file would look like 4225 // 1234 (java 1.4.2 :)) R ... ... 4226 // We don't really need to know the command string, just find the last 4227 // occurrence of ")" and then start parsing from there. See bug 4726580. 4228 s = strrchr(stat, ')'); 4229 i = 0; 4230 if (s == NULL ) return -1; 4231 4232 // Skip blank chars 4233 do s++; while (isspace(*s)); 4234 4235 count = sscanf(s,"%*c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 4236 &idummy, &idummy, &idummy, &idummy, &idummy, 4237 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 4238 &user_time, &sys_time); 4239 if ( count != 12 ) return -1; 4240 if (user_sys_cpu_time) { 4241 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4242 } else { 4243 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4244 } 4245} 4246 4247void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4248 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4249 info_ptr->may_skip_backward = false; // elapsed time not wall time 4250 info_ptr->may_skip_forward = false; // elapsed time not wall time 4251 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4252} 4253 4254void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4255 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4256 info_ptr->may_skip_backward = false; // elapsed time not wall time 4257 info_ptr->may_skip_forward = false; // elapsed time not wall time 4258 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4259} 4260 4261bool os::is_thread_cpu_time_supported() { 4262 return true; 4263} 4264 4265// System loadavg support. Returns -1 if load average cannot be obtained. 4266// Linux doesn't yet have a (official) notion of processor sets, 4267// so just return the system wide load average. 4268int os::loadavg(double loadavg[], int nelem) { 4269 return ::getloadavg(loadavg, nelem); 4270} 4271 4272void os::pause() { 4273 char filename[MAX_PATH]; 4274 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 4275 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 4276 } else { 4277 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 4278 } 4279 4280 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 4281 if (fd != -1) { 4282 struct stat buf; 4283 close(fd); 4284 while (::stat(filename, &buf) == 0) { 4285 (void)::poll(NULL, 0, 100); 4286 } 4287 } else { 4288 jio_fprintf(stderr, 4289 "Could not open pause file '%s', continuing immediately.\n", filename); 4290 } 4291} 4292 4293extern "C" { 4294 4295/** 4296 * NOTE: the following code is to keep the green threads code 4297 * in the libjava.so happy. Once the green threads is removed, 4298 * these code will no longer be needed. 4299 */ 4300int 4301jdk_waitpid(pid_t pid, int* status, int options) { 4302 return waitpid(pid, status, options); 4303} 4304 4305int 4306fork1() { 4307 return fork(); 4308} 4309 4310int 4311jdk_sem_init(sem_t *sem, int pshared, unsigned int value) { 4312 return sem_init(sem, pshared, value); 4313} 4314 4315int 4316jdk_sem_post(sem_t *sem) { 4317 return sem_post(sem); 4318} 4319 4320int 4321jdk_sem_wait(sem_t *sem) { 4322 return sem_wait(sem); 4323} 4324 4325int 4326jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) { 4327 return pthread_sigmask(how , newmask, oldmask); 4328} 4329 4330} 4331 4332// Refer to the comments in os_solaris.cpp park-unpark. 4333// 4334// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 4335// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 4336// For specifics regarding the bug see GLIBC BUGID 261237 : 4337// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 4338// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 4339// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 4340// is used. (The simple C test-case provided in the GLIBC bug report manifests the 4341// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 4342// and monitorenter when we're using 1-0 locking. All those operations may result in 4343// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 4344// of libpthread avoids the problem, but isn't practical. 4345// 4346// Possible remedies: 4347// 4348// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 4349// This is palliative and probabilistic, however. If the thread is preempted 4350// between the call to compute_abstime() and pthread_cond_timedwait(), more 4351// than the minimum period may have passed, and the abstime may be stale (in the 4352// past) resultin in a hang. Using this technique reduces the odds of a hang 4353// but the JVM is still vulnerable, particularly on heavily loaded systems. 4354// 4355// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 4356// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 4357// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 4358// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 4359// thread. 4360// 4361// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 4362// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 4363// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 4364// This also works well. In fact it avoids kernel-level scalability impediments 4365// on certain platforms that don't handle lots of active pthread_cond_timedwait() 4366// timers in a graceful fashion. 4367// 4368// 4. When the abstime value is in the past it appears that control returns 4369// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 4370// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 4371// can avoid the problem by reinitializing the condvar -- by cond_destroy() 4372// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 4373// It may be possible to avoid reinitialization by checking the return 4374// value from pthread_cond_timedwait(). In addition to reinitializing the 4375// condvar we must establish the invariant that cond_signal() is only called 4376// within critical sections protected by the adjunct mutex. This prevents 4377// cond_signal() from "seeing" a condvar that's in the midst of being 4378// reinitialized or that is corrupt. Sadly, this invariant obviates the 4379// desirable signal-after-unlock optimization that avoids futile context switching. 4380// 4381// I'm also concerned that some versions of NTPL might allocate an auxilliary 4382// structure when a condvar is used or initialized. cond_destroy() would 4383// release the helper structure. Our reinitialize-after-timedwait fix 4384// put excessive stress on malloc/free and locks protecting the c-heap. 4385// 4386// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 4387// It may be possible to refine (4) by checking the kernel and NTPL verisons 4388// and only enabling the work-around for vulnerable environments. 4389 4390// utility to compute the abstime argument to timedwait: 4391// millis is the relative timeout time 4392// abstime will be the absolute timeout time 4393// TODO: replace compute_abstime() with unpackTime() 4394 4395static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 4396 if (millis < 0) millis = 0; 4397 struct timeval now; 4398 int status = gettimeofday(&now, NULL); 4399 assert(status == 0, "gettimeofday"); 4400 jlong seconds = millis / 1000; 4401 millis %= 1000; 4402 if (seconds > 50000000) { // see man cond_timedwait(3T) 4403 seconds = 50000000; 4404 } 4405 abstime->tv_sec = now.tv_sec + seconds; 4406 long usec = now.tv_usec + millis * 1000; 4407 if (usec >= 1000000) { 4408 abstime->tv_sec += 1; 4409 usec -= 1000000; 4410 } 4411 abstime->tv_nsec = usec * 1000; 4412 return abstime; 4413} 4414 4415 4416// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 4417// Conceptually TryPark() should be equivalent to park(0). 4418 4419int os::PlatformEvent::TryPark() { 4420 for (;;) { 4421 const int v = _Event ; 4422 guarantee ((v == 0) || (v == 1), "invariant") ; 4423 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 4424 } 4425} 4426 4427void os::PlatformEvent::park() { // AKA "down()" 4428 // Invariant: Only the thread associated with the Event/PlatformEvent 4429 // may call park(). 4430 // TODO: assert that _Assoc != NULL or _Assoc == Self 4431 int v ; 4432 for (;;) { 4433 v = _Event ; 4434 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4435 } 4436 guarantee (v >= 0, "invariant") ; 4437 if (v == 0) { 4438 // Do this the hard way by blocking ... 4439 int status = pthread_mutex_lock(_mutex); 4440 assert_status(status == 0, status, "mutex_lock"); 4441 guarantee (_nParked == 0, "invariant") ; 4442 ++ _nParked ; 4443 while (_Event < 0) { 4444 status = pthread_cond_wait(_cond, _mutex); 4445 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 4446 // Treat this the same as if the wait was interrupted 4447 if (status == ETIME) { status = EINTR; } 4448 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 4449 } 4450 -- _nParked ; 4451 4452 // In theory we could move the ST of 0 into _Event past the unlock(), 4453 // but then we'd need a MEMBAR after the ST. 4454 _Event = 0 ; 4455 status = pthread_mutex_unlock(_mutex); 4456 assert_status(status == 0, status, "mutex_unlock"); 4457 } 4458 guarantee (_Event >= 0, "invariant") ; 4459} 4460 4461int os::PlatformEvent::park(jlong millis) { 4462 guarantee (_nParked == 0, "invariant") ; 4463 4464 int v ; 4465 for (;;) { 4466 v = _Event ; 4467 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4468 } 4469 guarantee (v >= 0, "invariant") ; 4470 if (v != 0) return OS_OK ; 4471 4472 // We do this the hard way, by blocking the thread. 4473 // Consider enforcing a minimum timeout value. 4474 struct timespec abst; 4475 compute_abstime(&abst, millis); 4476 4477 int ret = OS_TIMEOUT; 4478 int status = pthread_mutex_lock(_mutex); 4479 assert_status(status == 0, status, "mutex_lock"); 4480 guarantee (_nParked == 0, "invariant") ; 4481 ++_nParked ; 4482 4483 // Object.wait(timo) will return because of 4484 // (a) notification 4485 // (b) timeout 4486 // (c) thread.interrupt 4487 // 4488 // Thread.interrupt and object.notify{All} both call Event::set. 4489 // That is, we treat thread.interrupt as a special case of notification. 4490 // The underlying Solaris implementation, cond_timedwait, admits 4491 // spurious/premature wakeups, but the JLS/JVM spec prevents the 4492 // JVM from making those visible to Java code. As such, we must 4493 // filter out spurious wakeups. We assume all ETIME returns are valid. 4494 // 4495 // TODO: properly differentiate simultaneous notify+interrupt. 4496 // In that case, we should propagate the notify to another waiter. 4497 4498 while (_Event < 0) { 4499 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 4500 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4501 pthread_cond_destroy (_cond); 4502 pthread_cond_init (_cond, NULL) ; 4503 } 4504 assert_status(status == 0 || status == EINTR || 4505 status == ETIME || status == ETIMEDOUT, 4506 status, "cond_timedwait"); 4507 if (!FilterSpuriousWakeups) break ; // previous semantics 4508 if (status == ETIME || status == ETIMEDOUT) break ; 4509 // We consume and ignore EINTR and spurious wakeups. 4510 } 4511 --_nParked ; 4512 if (_Event >= 0) { 4513 ret = OS_OK; 4514 } 4515 _Event = 0 ; 4516 status = pthread_mutex_unlock(_mutex); 4517 assert_status(status == 0, status, "mutex_unlock"); 4518 assert (_nParked == 0, "invariant") ; 4519 return ret; 4520} 4521 4522void os::PlatformEvent::unpark() { 4523 int v, AnyWaiters ; 4524 for (;;) { 4525 v = _Event ; 4526 if (v > 0) { 4527 // The LD of _Event could have reordered or be satisfied 4528 // by a read-aside from this processor's write buffer. 4529 // To avoid problems execute a barrier and then 4530 // ratify the value. 4531 OrderAccess::fence() ; 4532 if (_Event == v) return ; 4533 continue ; 4534 } 4535 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ; 4536 } 4537 if (v < 0) { 4538 // Wait for the thread associated with the event to vacate 4539 int status = pthread_mutex_lock(_mutex); 4540 assert_status(status == 0, status, "mutex_lock"); 4541 AnyWaiters = _nParked ; 4542 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ; 4543 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 4544 AnyWaiters = 0 ; 4545 pthread_cond_signal (_cond); 4546 } 4547 status = pthread_mutex_unlock(_mutex); 4548 assert_status(status == 0, status, "mutex_unlock"); 4549 if (AnyWaiters != 0) { 4550 status = pthread_cond_signal(_cond); 4551 assert_status(status == 0, status, "cond_signal"); 4552 } 4553 } 4554 4555 // Note that we signal() _after dropping the lock for "immortal" Events. 4556 // This is safe and avoids a common class of futile wakeups. In rare 4557 // circumstances this can cause a thread to return prematurely from 4558 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 4559 // simply re-test the condition and re-park itself. 4560} 4561 4562 4563// JSR166 4564// ------------------------------------------------------- 4565 4566/* 4567 * The solaris and linux implementations of park/unpark are fairly 4568 * conservative for now, but can be improved. They currently use a 4569 * mutex/condvar pair, plus a a count. 4570 * Park decrements count if > 0, else does a condvar wait. Unpark 4571 * sets count to 1 and signals condvar. Only one thread ever waits 4572 * on the condvar. Contention seen when trying to park implies that someone 4573 * is unparking you, so don't wait. And spurious returns are fine, so there 4574 * is no need to track notifications. 4575 */ 4576 4577 4578#define NANOSECS_PER_SEC 1000000000 4579#define NANOSECS_PER_MILLISEC 1000000 4580#define MAX_SECS 100000000 4581/* 4582 * This code is common to linux and solaris and will be moved to a 4583 * common place in dolphin. 4584 * 4585 * The passed in time value is either a relative time in nanoseconds 4586 * or an absolute time in milliseconds. Either way it has to be unpacked 4587 * into suitable seconds and nanoseconds components and stored in the 4588 * given timespec structure. 4589 * Given time is a 64-bit value and the time_t used in the timespec is only 4590 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 4591 * overflow if times way in the future are given. Further on Solaris versions 4592 * prior to 10 there is a restriction (see cond_timedwait) that the specified 4593 * number of seconds, in abstime, is less than current_time + 100,000,000. 4594 * As it will be 28 years before "now + 100000000" will overflow we can 4595 * ignore overflow and just impose a hard-limit on seconds using the value 4596 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 4597 * years from "now". 4598 */ 4599 4600static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 4601 assert (time > 0, "convertTime"); 4602 4603 struct timeval now; 4604 int status = gettimeofday(&now, NULL); 4605 assert(status == 0, "gettimeofday"); 4606 4607 time_t max_secs = now.tv_sec + MAX_SECS; 4608 4609 if (isAbsolute) { 4610 jlong secs = time / 1000; 4611 if (secs > max_secs) { 4612 absTime->tv_sec = max_secs; 4613 } 4614 else { 4615 absTime->tv_sec = secs; 4616 } 4617 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 4618 } 4619 else { 4620 jlong secs = time / NANOSECS_PER_SEC; 4621 if (secs >= MAX_SECS) { 4622 absTime->tv_sec = max_secs; 4623 absTime->tv_nsec = 0; 4624 } 4625 else { 4626 absTime->tv_sec = now.tv_sec + secs; 4627 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 4628 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 4629 absTime->tv_nsec -= NANOSECS_PER_SEC; 4630 ++absTime->tv_sec; // note: this must be <= max_secs 4631 } 4632 } 4633 } 4634 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 4635 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 4636 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 4637 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 4638} 4639 4640void Parker::park(bool isAbsolute, jlong time) { 4641 // Optional fast-path check: 4642 // Return immediately if a permit is available. 4643 if (_counter > 0) { 4644 _counter = 0 ; 4645 return ; 4646 } 4647 4648 Thread* thread = Thread::current(); 4649 assert(thread->is_Java_thread(), "Must be JavaThread"); 4650 JavaThread *jt = (JavaThread *)thread; 4651 4652 // Optional optimization -- avoid state transitions if there's an interrupt pending. 4653 // Check interrupt before trying to wait 4654 if (Thread::is_interrupted(thread, false)) { 4655 return; 4656 } 4657 4658 // Next, demultiplex/decode time arguments 4659 timespec absTime; 4660 if (time < 0) { // don't wait at all 4661 return; 4662 } 4663 if (time > 0) { 4664 unpackTime(&absTime, isAbsolute, time); 4665 } 4666 4667 4668 // Enter safepoint region 4669 // Beware of deadlocks such as 6317397. 4670 // The per-thread Parker:: mutex is a classic leaf-lock. 4671 // In particular a thread must never block on the Threads_lock while 4672 // holding the Parker:: mutex. If safepoints are pending both the 4673 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 4674 ThreadBlockInVM tbivm(jt); 4675 4676 // Don't wait if cannot get lock since interference arises from 4677 // unblocking. Also. check interrupt before trying wait 4678 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 4679 return; 4680 } 4681 4682 int status ; 4683 if (_counter > 0) { // no wait needed 4684 _counter = 0; 4685 status = pthread_mutex_unlock(_mutex); 4686 assert (status == 0, "invariant") ; 4687 return; 4688 } 4689 4690#ifdef ASSERT 4691 // Don't catch signals while blocked; let the running threads have the signals. 4692 // (This allows a debugger to break into the running thread.) 4693 sigset_t oldsigs; 4694 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 4695 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 4696#endif 4697 4698 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 4699 jt->set_suspend_equivalent(); 4700 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 4701 4702 if (time == 0) { 4703 status = pthread_cond_wait (_cond, _mutex) ; 4704 } else { 4705 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 4706 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4707 pthread_cond_destroy (_cond) ; 4708 pthread_cond_init (_cond, NULL); 4709 } 4710 } 4711 assert_status(status == 0 || status == EINTR || 4712 status == ETIME || status == ETIMEDOUT, 4713 status, "cond_timedwait"); 4714 4715#ifdef ASSERT 4716 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 4717#endif 4718 4719 _counter = 0 ; 4720 status = pthread_mutex_unlock(_mutex) ; 4721 assert_status(status == 0, status, "invariant") ; 4722 // If externally suspended while waiting, re-suspend 4723 if (jt->handle_special_suspend_equivalent_condition()) { 4724 jt->java_suspend_self(); 4725 } 4726 4727} 4728 4729void Parker::unpark() { 4730 int s, status ; 4731 status = pthread_mutex_lock(_mutex); 4732 assert (status == 0, "invariant") ; 4733 s = _counter; 4734 _counter = 1; 4735 if (s < 1) { 4736 if (WorkAroundNPTLTimedWaitHang) { 4737 status = pthread_cond_signal (_cond) ; 4738 assert (status == 0, "invariant") ; 4739 status = pthread_mutex_unlock(_mutex); 4740 assert (status == 0, "invariant") ; 4741 } else { 4742 status = pthread_mutex_unlock(_mutex); 4743 assert (status == 0, "invariant") ; 4744 status = pthread_cond_signal (_cond) ; 4745 assert (status == 0, "invariant") ; 4746 } 4747 } else { 4748 pthread_mutex_unlock(_mutex); 4749 assert (status == 0, "invariant") ; 4750 } 4751} 4752 4753 4754extern char** environ; 4755 4756#ifndef __NR_fork 4757#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 4758#endif 4759 4760#ifndef __NR_execve 4761#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 4762#endif 4763 4764// Run the specified command in a separate process. Return its exit value, 4765// or -1 on failure (e.g. can't fork a new process). 4766// Unlike system(), this function can be called from signal handler. It 4767// doesn't block SIGINT et al. 4768int os::fork_and_exec(char* cmd) { 4769 const char * argv[4] = {"sh", "-c", cmd, NULL}; 4770 4771 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 4772 // pthread_atfork handlers and reset pthread library. All we need is a 4773 // separate process to execve. Make a direct syscall to fork process. 4774 // On IA64 there's no fork syscall, we have to use fork() and hope for 4775 // the best... 4776 pid_t pid = NOT_IA64(syscall(__NR_fork);) 4777 IA64_ONLY(fork();) 4778 4779 if (pid < 0) { 4780 // fork failed 4781 return -1; 4782 4783 } else if (pid == 0) { 4784 // child process 4785 4786 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 4787 // first to kill every thread on the thread list. Because this list is 4788 // not reset by fork() (see notes above), execve() will instead kill 4789 // every thread in the parent process. We know this is the only thread 4790 // in the new process, so make a system call directly. 4791 // IA64 should use normal execve() from glibc to match the glibc fork() 4792 // above. 4793 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 4794 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 4795 4796 // execve failed 4797 _exit(-1); 4798 4799 } else { 4800 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 4801 // care about the actual exit code, for now. 4802 4803 int status; 4804 4805 // Wait for the child process to exit. This returns immediately if 4806 // the child has already exited. */ 4807 while (waitpid(pid, &status, 0) < 0) { 4808 switch (errno) { 4809 case ECHILD: return 0; 4810 case EINTR: break; 4811 default: return -1; 4812 } 4813 } 4814 4815 if (WIFEXITED(status)) { 4816 // The child exited normally; get its exit code. 4817 return WEXITSTATUS(status); 4818 } else if (WIFSIGNALED(status)) { 4819 // The child exited because of a signal 4820 // The best value to return is 0x80 + signal number, 4821 // because that is what all Unix shells do, and because 4822 // it allows callers to distinguish between process exit and 4823 // process death by signal. 4824 return 0x80 + WTERMSIG(status); 4825 } else { 4826 // Unknown exit code; pass it through 4827 return status; 4828 } 4829 } 4830} 4831