os_linux.cpp revision 579:0fbdb4381b99
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 1521void os::dll_build_name( 1522 char* buffer, size_t buflen, const char* pname, const char* fname) { 1523 // copied from libhpi 1524 const size_t pnamelen = pname ? strlen(pname) : 0; 1525 1526 /* Quietly truncate on buffer overflow. Should be an error. */ 1527 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1528 *buffer = '\0'; 1529 return; 1530 } 1531 1532 if (pnamelen == 0) { 1533 sprintf(buffer, "lib%s.so", fname); 1534 } else { 1535 sprintf(buffer, "%s/lib%s.so", pname, fname); 1536 } 1537} 1538 1539const char* os::get_current_directory(char *buf, int buflen) { 1540 return getcwd(buf, buflen); 1541} 1542 1543// check if addr is inside libjvm[_g].so 1544bool os::address_is_in_vm(address addr) { 1545 static address libjvm_base_addr; 1546 Dl_info dlinfo; 1547 1548 if (libjvm_base_addr == NULL) { 1549 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); 1550 libjvm_base_addr = (address)dlinfo.dli_fbase; 1551 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1552 } 1553 1554 if (dladdr((void *)addr, &dlinfo)) { 1555 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1556 } 1557 1558 return false; 1559} 1560 1561bool os::dll_address_to_function_name(address addr, char *buf, 1562 int buflen, int *offset) { 1563 Dl_info dlinfo; 1564 1565 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { 1566 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1567 if (offset) *offset = addr - (address)dlinfo.dli_saddr; 1568 return true; 1569 } else { 1570 if (buf) buf[0] = '\0'; 1571 if (offset) *offset = -1; 1572 return false; 1573 } 1574} 1575 1576struct _address_to_library_name { 1577 address addr; // input : memory address 1578 size_t buflen; // size of fname 1579 char* fname; // output: library name 1580 address base; // library base addr 1581}; 1582 1583static int address_to_library_name_callback(struct dl_phdr_info *info, 1584 size_t size, void *data) { 1585 int i; 1586 bool found = false; 1587 address libbase = NULL; 1588 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1589 1590 // iterate through all loadable segments 1591 for (i = 0; i < info->dlpi_phnum; i++) { 1592 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1593 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1594 // base address of a library is the lowest address of its loaded 1595 // segments. 1596 if (libbase == NULL || libbase > segbase) { 1597 libbase = segbase; 1598 } 1599 // see if 'addr' is within current segment 1600 if (segbase <= d->addr && 1601 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1602 found = true; 1603 } 1604 } 1605 } 1606 1607 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1608 // so dll_address_to_library_name() can fall through to use dladdr() which 1609 // can figure out executable name from argv[0]. 1610 if (found && info->dlpi_name && info->dlpi_name[0]) { 1611 d->base = libbase; 1612 if (d->fname) { 1613 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1614 } 1615 return 1; 1616 } 1617 return 0; 1618} 1619 1620bool os::dll_address_to_library_name(address addr, char* buf, 1621 int buflen, int* offset) { 1622 Dl_info dlinfo; 1623 struct _address_to_library_name data; 1624 1625 // There is a bug in old glibc dladdr() implementation that it could resolve 1626 // to wrong library name if the .so file has a base address != NULL. Here 1627 // we iterate through the program headers of all loaded libraries to find 1628 // out which library 'addr' really belongs to. This workaround can be 1629 // removed once the minimum requirement for glibc is moved to 2.3.x. 1630 data.addr = addr; 1631 data.fname = buf; 1632 data.buflen = buflen; 1633 data.base = NULL; 1634 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1635 1636 if (rslt) { 1637 // buf already contains library name 1638 if (offset) *offset = addr - data.base; 1639 return true; 1640 } else if (dladdr((void*)addr, &dlinfo)){ 1641 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1642 if (offset) *offset = addr - (address)dlinfo.dli_fbase; 1643 return true; 1644 } else { 1645 if (buf) buf[0] = '\0'; 1646 if (offset) *offset = -1; 1647 return false; 1648 } 1649} 1650 1651 // Loads .dll/.so and 1652 // in case of error it checks if .dll/.so was built for the 1653 // same architecture as Hotspot is running on 1654 1655void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1656{ 1657 void * result= ::dlopen(filename, RTLD_LAZY); 1658 if (result != NULL) { 1659 // Successful loading 1660 return result; 1661 } 1662 1663 Elf32_Ehdr elf_head; 1664 1665 // Read system error message into ebuf 1666 // It may or may not be overwritten below 1667 ::strncpy(ebuf, ::dlerror(), ebuflen-1); 1668 ebuf[ebuflen-1]='\0'; 1669 int diag_msg_max_length=ebuflen-strlen(ebuf); 1670 char* diag_msg_buf=ebuf+strlen(ebuf); 1671 1672 if (diag_msg_max_length==0) { 1673 // No more space in ebuf for additional diagnostics message 1674 return NULL; 1675 } 1676 1677 1678 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1679 1680 if (file_descriptor < 0) { 1681 // Can't open library, report dlerror() message 1682 return NULL; 1683 } 1684 1685 bool failed_to_read_elf_head= 1686 (sizeof(elf_head)!= 1687 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1688 1689 ::close(file_descriptor); 1690 if (failed_to_read_elf_head) { 1691 // file i/o error - report dlerror() msg 1692 return NULL; 1693 } 1694 1695 typedef struct { 1696 Elf32_Half code; // Actual value as defined in elf.h 1697 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1698 char elf_class; // 32 or 64 bit 1699 char endianess; // MSB or LSB 1700 char* name; // String representation 1701 } arch_t; 1702 1703 #ifndef EM_486 1704 #define EM_486 6 /* Intel 80486 */ 1705 #endif 1706 1707 static const arch_t arch_array[]={ 1708 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1709 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1710 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1711 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1712 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1713 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1714 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1715 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1716 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"} 1717 }; 1718 1719 #if (defined IA32) 1720 static Elf32_Half running_arch_code=EM_386; 1721 #elif (defined AMD64) 1722 static Elf32_Half running_arch_code=EM_X86_64; 1723 #elif (defined IA64) 1724 static Elf32_Half running_arch_code=EM_IA_64; 1725 #elif (defined __sparc) && (defined _LP64) 1726 static Elf32_Half running_arch_code=EM_SPARCV9; 1727 #elif (defined __sparc) && (!defined _LP64) 1728 static Elf32_Half running_arch_code=EM_SPARC; 1729 #elif (defined __powerpc64__) 1730 static Elf32_Half running_arch_code=EM_PPC64; 1731 #elif (defined __powerpc__) 1732 static Elf32_Half running_arch_code=EM_PPC; 1733 #else 1734 #error Method os::dll_load requires that one of following is defined:\ 1735 IA32, AMD64, IA64, __sparc, __powerpc__ 1736 #endif 1737 1738 // Identify compatability class for VM's architecture and library's architecture 1739 // Obtain string descriptions for architectures 1740 1741 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1742 int running_arch_index=-1; 1743 1744 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1745 if (running_arch_code == arch_array[i].code) { 1746 running_arch_index = i; 1747 } 1748 if (lib_arch.code == arch_array[i].code) { 1749 lib_arch.compat_class = arch_array[i].compat_class; 1750 lib_arch.name = arch_array[i].name; 1751 } 1752 } 1753 1754 assert(running_arch_index != -1, 1755 "Didn't find running architecture code (running_arch_code) in arch_array"); 1756 if (running_arch_index == -1) { 1757 // Even though running architecture detection failed 1758 // we may still continue with reporting dlerror() message 1759 return NULL; 1760 } 1761 1762 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1763 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1764 return NULL; 1765 } 1766 1767 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1768 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1769 return NULL; 1770 } 1771 1772 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1773 if ( lib_arch.name!=NULL ) { 1774 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1775 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1776 lib_arch.name, arch_array[running_arch_index].name); 1777 } else { 1778 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1779 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1780 lib_arch.code, 1781 arch_array[running_arch_index].name); 1782 } 1783 } 1784 1785 return NULL; 1786} 1787 1788/* 1789 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 1790 * chances are you might want to run the generated bits against glibc-2.0 1791 * libdl.so, so always use locking for any version of glibc. 1792 */ 1793void* os::dll_lookup(void* handle, const char* name) { 1794 pthread_mutex_lock(&dl_mutex); 1795 void* res = dlsym(handle, name); 1796 pthread_mutex_unlock(&dl_mutex); 1797 return res; 1798} 1799 1800 1801bool _print_ascii_file(const char* filename, outputStream* st) { 1802 int fd = open(filename, O_RDONLY); 1803 if (fd == -1) { 1804 return false; 1805 } 1806 1807 char buf[32]; 1808 int bytes; 1809 while ((bytes = read(fd, buf, sizeof(buf))) > 0) { 1810 st->print_raw(buf, bytes); 1811 } 1812 1813 close(fd); 1814 1815 return true; 1816} 1817 1818void os::print_dll_info(outputStream *st) { 1819 st->print_cr("Dynamic libraries:"); 1820 1821 char fname[32]; 1822 pid_t pid = os::Linux::gettid(); 1823 1824 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1825 1826 if (!_print_ascii_file(fname, st)) { 1827 st->print("Can not get library information for pid = %d\n", pid); 1828 } 1829} 1830 1831 1832void os::print_os_info(outputStream* st) { 1833 st->print("OS:"); 1834 1835 // Try to identify popular distros. 1836 // Most Linux distributions have /etc/XXX-release file, which contains 1837 // the OS version string. Some have more than one /etc/XXX-release file 1838 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 1839 // so the order is important. 1840 if (!_print_ascii_file("/etc/mandrake-release", st) && 1841 !_print_ascii_file("/etc/sun-release", st) && 1842 !_print_ascii_file("/etc/redhat-release", st) && 1843 !_print_ascii_file("/etc/SuSE-release", st) && 1844 !_print_ascii_file("/etc/turbolinux-release", st) && 1845 !_print_ascii_file("/etc/gentoo-release", st) && 1846 !_print_ascii_file("/etc/debian_version", st)) { 1847 st->print("Linux"); 1848 } 1849 st->cr(); 1850 1851 // kernel 1852 st->print("uname:"); 1853 struct utsname name; 1854 uname(&name); 1855 st->print(name.sysname); st->print(" "); 1856 st->print(name.release); st->print(" "); 1857 st->print(name.version); st->print(" "); 1858 st->print(name.machine); 1859 st->cr(); 1860 1861 // Print warning if unsafe chroot environment detected 1862 if (unsafe_chroot_detected) { 1863 st->print("WARNING!! "); 1864 st->print_cr(unstable_chroot_error); 1865 } 1866 1867 // libc, pthread 1868 st->print("libc:"); 1869 st->print(os::Linux::glibc_version()); st->print(" "); 1870 st->print(os::Linux::libpthread_version()); st->print(" "); 1871 if (os::Linux::is_LinuxThreads()) { 1872 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 1873 } 1874 st->cr(); 1875 1876 // rlimit 1877 st->print("rlimit:"); 1878 struct rlimit rlim; 1879 1880 st->print(" STACK "); 1881 getrlimit(RLIMIT_STACK, &rlim); 1882 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1883 else st->print("%uk", rlim.rlim_cur >> 10); 1884 1885 st->print(", CORE "); 1886 getrlimit(RLIMIT_CORE, &rlim); 1887 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1888 else st->print("%uk", rlim.rlim_cur >> 10); 1889 1890 st->print(", NPROC "); 1891 getrlimit(RLIMIT_NPROC, &rlim); 1892 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1893 else st->print("%d", rlim.rlim_cur); 1894 1895 st->print(", NOFILE "); 1896 getrlimit(RLIMIT_NOFILE, &rlim); 1897 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1898 else st->print("%d", rlim.rlim_cur); 1899 1900 st->print(", AS "); 1901 getrlimit(RLIMIT_AS, &rlim); 1902 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1903 else st->print("%uk", rlim.rlim_cur >> 10); 1904 st->cr(); 1905 1906 // load average 1907 st->print("load average:"); 1908 double loadavg[3]; 1909 os::loadavg(loadavg, 3); 1910 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]); 1911 st->cr(); 1912} 1913 1914void os::print_memory_info(outputStream* st) { 1915 1916 st->print("Memory:"); 1917 st->print(" %dk page", os::vm_page_size()>>10); 1918 1919 // values in struct sysinfo are "unsigned long" 1920 struct sysinfo si; 1921 sysinfo(&si); 1922 1923 st->print(", physical " UINT64_FORMAT "k", 1924 os::physical_memory() >> 10); 1925 st->print("(" UINT64_FORMAT "k free)", 1926 os::available_memory() >> 10); 1927 st->print(", swap " UINT64_FORMAT "k", 1928 ((jlong)si.totalswap * si.mem_unit) >> 10); 1929 st->print("(" UINT64_FORMAT "k free)", 1930 ((jlong)si.freeswap * si.mem_unit) >> 10); 1931 st->cr(); 1932} 1933 1934// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 1935// but they're the same for all the linux arch that we support 1936// and they're the same for solaris but there's no common place to put this. 1937const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 1938 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 1939 "ILL_COPROC", "ILL_BADSTK" }; 1940 1941const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 1942 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 1943 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 1944 1945const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 1946 1947const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 1948 1949void os::print_siginfo(outputStream* st, void* siginfo) { 1950 st->print("siginfo:"); 1951 1952 const int buflen = 100; 1953 char buf[buflen]; 1954 siginfo_t *si = (siginfo_t*)siginfo; 1955 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 1956 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 1957 st->print("si_errno=%s", buf); 1958 } else { 1959 st->print("si_errno=%d", si->si_errno); 1960 } 1961 const int c = si->si_code; 1962 assert(c > 0, "unexpected si_code"); 1963 switch (si->si_signo) { 1964 case SIGILL: 1965 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 1966 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1967 break; 1968 case SIGFPE: 1969 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 1970 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1971 break; 1972 case SIGSEGV: 1973 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 1974 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1975 break; 1976 case SIGBUS: 1977 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 1978 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1979 break; 1980 default: 1981 st->print(", si_code=%d", si->si_code); 1982 // no si_addr 1983 } 1984 1985 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 1986 UseSharedSpaces) { 1987 FileMapInfo* mapinfo = FileMapInfo::current_info(); 1988 if (mapinfo->is_in_shared_space(si->si_addr)) { 1989 st->print("\n\nError accessing class data sharing archive." \ 1990 " Mapped file inaccessible during execution, " \ 1991 " possible disk/network problem."); 1992 } 1993 } 1994 st->cr(); 1995} 1996 1997 1998static void print_signal_handler(outputStream* st, int sig, 1999 char* buf, size_t buflen); 2000 2001void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2002 st->print_cr("Signal Handlers:"); 2003 print_signal_handler(st, SIGSEGV, buf, buflen); 2004 print_signal_handler(st, SIGBUS , buf, buflen); 2005 print_signal_handler(st, SIGFPE , buf, buflen); 2006 print_signal_handler(st, SIGPIPE, buf, buflen); 2007 print_signal_handler(st, SIGXFSZ, buf, buflen); 2008 print_signal_handler(st, SIGILL , buf, buflen); 2009 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2010 print_signal_handler(st, SR_signum, buf, buflen); 2011 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2012 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2013 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2014 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2015} 2016 2017static char saved_jvm_path[MAXPATHLEN] = {0}; 2018 2019// Find the full path to the current module, libjvm.so or libjvm_g.so 2020void os::jvm_path(char *buf, jint len) { 2021 // Error checking. 2022 if (len < MAXPATHLEN) { 2023 assert(false, "must use a large-enough buffer"); 2024 buf[0] = '\0'; 2025 return; 2026 } 2027 // Lazy resolve the path to current module. 2028 if (saved_jvm_path[0] != 0) { 2029 strcpy(buf, saved_jvm_path); 2030 return; 2031 } 2032 2033 char dli_fname[MAXPATHLEN]; 2034 bool ret = dll_address_to_library_name( 2035 CAST_FROM_FN_PTR(address, os::jvm_path), 2036 dli_fname, sizeof(dli_fname), NULL); 2037 assert(ret != 0, "cannot locate libjvm"); 2038 if (realpath(dli_fname, buf) == NULL) 2039 return; 2040 2041 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) { 2042 // Support for the gamma launcher. Typical value for buf is 2043 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 2044 // the right place in the string, then assume we are installed in a JDK and 2045 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2046 // up the path so it looks like libjvm.so is installed there (append a 2047 // fake suffix hotspot/libjvm.so). 2048 const char *p = buf + strlen(buf) - 1; 2049 for (int count = 0; p > buf && count < 5; ++count) { 2050 for (--p; p > buf && *p != '/'; --p) 2051 /* empty */ ; 2052 } 2053 2054 if (strncmp(p, "/jre/lib/", 9) != 0) { 2055 // Look for JAVA_HOME in the environment. 2056 char* java_home_var = ::getenv("JAVA_HOME"); 2057 if (java_home_var != NULL && java_home_var[0] != 0) { 2058 // Check the current module name "libjvm.so" or "libjvm_g.so". 2059 p = strrchr(buf, '/'); 2060 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2061 p = strstr(p, "_g") ? "_g" : ""; 2062 2063 if (realpath(java_home_var, buf) == NULL) 2064 return; 2065 sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch); 2066 if (0 == access(buf, F_OK)) { 2067 // Use current module name "libjvm[_g].so" instead of 2068 // "libjvm"debug_only("_g")".so" since for fastdebug version 2069 // we should have "libjvm.so" but debug_only("_g") adds "_g"! 2070 // It is used when we are choosing the HPI library's name 2071 // "libhpi[_g].so" in hpi::initialize_get_interface(). 2072 sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p); 2073 } else { 2074 // Go back to path of .so 2075 if (realpath(dli_fname, buf) == NULL) 2076 return; 2077 } 2078 } 2079 } 2080 } 2081 2082 strcpy(saved_jvm_path, buf); 2083} 2084 2085void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2086 // no prefix required, not even "_" 2087} 2088 2089void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2090 // no suffix required 2091} 2092 2093//////////////////////////////////////////////////////////////////////////////// 2094// sun.misc.Signal support 2095 2096static volatile jint sigint_count = 0; 2097 2098static void 2099UserHandler(int sig, void *siginfo, void *context) { 2100 // 4511530 - sem_post is serialized and handled by the manager thread. When 2101 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2102 // don't want to flood the manager thread with sem_post requests. 2103 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2104 return; 2105 2106 // Ctrl-C is pressed during error reporting, likely because the error 2107 // handler fails to abort. Let VM die immediately. 2108 if (sig == SIGINT && is_error_reported()) { 2109 os::die(); 2110 } 2111 2112 os::signal_notify(sig); 2113} 2114 2115void* os::user_handler() { 2116 return CAST_FROM_FN_PTR(void*, UserHandler); 2117} 2118 2119extern "C" { 2120 typedef void (*sa_handler_t)(int); 2121 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2122} 2123 2124void* os::signal(int signal_number, void* handler) { 2125 struct sigaction sigAct, oldSigAct; 2126 2127 sigfillset(&(sigAct.sa_mask)); 2128 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2129 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2130 2131 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2132 // -1 means registration failed 2133 return (void *)-1; 2134 } 2135 2136 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2137} 2138 2139void os::signal_raise(int signal_number) { 2140 ::raise(signal_number); 2141} 2142 2143/* 2144 * The following code is moved from os.cpp for making this 2145 * code platform specific, which it is by its very nature. 2146 */ 2147 2148// Will be modified when max signal is changed to be dynamic 2149int os::sigexitnum_pd() { 2150 return NSIG; 2151} 2152 2153// a counter for each possible signal value 2154static volatile jint pending_signals[NSIG+1] = { 0 }; 2155 2156// Linux(POSIX) specific hand shaking semaphore. 2157static sem_t sig_sem; 2158 2159void os::signal_init_pd() { 2160 // Initialize signal structures 2161 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2162 2163 // Initialize signal semaphore 2164 ::sem_init(&sig_sem, 0, 0); 2165} 2166 2167void os::signal_notify(int sig) { 2168 Atomic::inc(&pending_signals[sig]); 2169 ::sem_post(&sig_sem); 2170} 2171 2172static int check_pending_signals(bool wait) { 2173 Atomic::store(0, &sigint_count); 2174 for (;;) { 2175 for (int i = 0; i < NSIG + 1; i++) { 2176 jint n = pending_signals[i]; 2177 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2178 return i; 2179 } 2180 } 2181 if (!wait) { 2182 return -1; 2183 } 2184 JavaThread *thread = JavaThread::current(); 2185 ThreadBlockInVM tbivm(thread); 2186 2187 bool threadIsSuspended; 2188 do { 2189 thread->set_suspend_equivalent(); 2190 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2191 ::sem_wait(&sig_sem); 2192 2193 // were we externally suspended while we were waiting? 2194 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2195 if (threadIsSuspended) { 2196 // 2197 // The semaphore has been incremented, but while we were waiting 2198 // another thread suspended us. We don't want to continue running 2199 // while suspended because that would surprise the thread that 2200 // suspended us. 2201 // 2202 ::sem_post(&sig_sem); 2203 2204 thread->java_suspend_self(); 2205 } 2206 } while (threadIsSuspended); 2207 } 2208} 2209 2210int os::signal_lookup() { 2211 return check_pending_signals(false); 2212} 2213 2214int os::signal_wait() { 2215 return check_pending_signals(true); 2216} 2217 2218//////////////////////////////////////////////////////////////////////////////// 2219// Virtual Memory 2220 2221int os::vm_page_size() { 2222 // Seems redundant as all get out 2223 assert(os::Linux::page_size() != -1, "must call os::init"); 2224 return os::Linux::page_size(); 2225} 2226 2227// Solaris allocates memory by pages. 2228int os::vm_allocation_granularity() { 2229 assert(os::Linux::page_size() != -1, "must call os::init"); 2230 return os::Linux::page_size(); 2231} 2232 2233// Rationale behind this function: 2234// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2235// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2236// samples for JITted code. Here we create private executable mapping over the code cache 2237// and then we can use standard (well, almost, as mapping can change) way to provide 2238// info for the reporting script by storing timestamp and location of symbol 2239void linux_wrap_code(char* base, size_t size) { 2240 static volatile jint cnt = 0; 2241 2242 if (!UseOprofile) { 2243 return; 2244 } 2245 2246 char buf[40]; 2247 int num = Atomic::add(1, &cnt); 2248 2249 sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num); 2250 unlink(buf); 2251 2252 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU); 2253 2254 if (fd != -1) { 2255 off_t rv = lseek(fd, size-2, SEEK_SET); 2256 if (rv != (off_t)-1) { 2257 if (write(fd, "", 1) == 1) { 2258 mmap(base, size, 2259 PROT_READ|PROT_WRITE|PROT_EXEC, 2260 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2261 } 2262 } 2263 close(fd); 2264 unlink(buf); 2265 } 2266} 2267 2268// NOTE: Linux kernel does not really reserve the pages for us. 2269// All it does is to check if there are enough free pages 2270// left at the time of mmap(). This could be a potential 2271// problem. 2272bool os::commit_memory(char* addr, size_t size) { 2273 uintptr_t res = (uintptr_t) ::mmap(addr, size, 2274 PROT_READ|PROT_WRITE|PROT_EXEC, 2275 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2276 return res != (uintptr_t) MAP_FAILED; 2277} 2278 2279bool os::commit_memory(char* addr, size_t size, size_t alignment_hint) { 2280 return commit_memory(addr, size); 2281} 2282 2283void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { } 2284 2285void os::free_memory(char *addr, size_t bytes) { 2286 uncommit_memory(addr, bytes); 2287} 2288 2289void os::numa_make_global(char *addr, size_t bytes) { 2290 Linux::numa_interleave_memory(addr, bytes); 2291} 2292 2293void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2294 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2295} 2296 2297bool os::numa_topology_changed() { return false; } 2298 2299size_t os::numa_get_groups_num() { 2300 int max_node = Linux::numa_max_node(); 2301 return max_node > 0 ? max_node + 1 : 1; 2302} 2303 2304int os::numa_get_group_id() { 2305 int cpu_id = Linux::sched_getcpu(); 2306 if (cpu_id != -1) { 2307 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2308 if (lgrp_id != -1) { 2309 return lgrp_id; 2310 } 2311 } 2312 return 0; 2313} 2314 2315size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2316 for (size_t i = 0; i < size; i++) { 2317 ids[i] = i; 2318 } 2319 return size; 2320} 2321 2322bool os::get_page_info(char *start, page_info* info) { 2323 return false; 2324} 2325 2326char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2327 return end; 2328} 2329 2330extern "C" void numa_warn(int number, char *where, ...) { } 2331extern "C" void numa_error(char *where) { } 2332 2333bool os::Linux::libnuma_init() { 2334 // sched_getcpu() should be in libc. 2335 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2336 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2337 2338 if (sched_getcpu() != -1) { // Does it work? 2339 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2340 if (handle != NULL) { 2341 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2342 dlsym(handle, "numa_node_to_cpus"))); 2343 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2344 dlsym(handle, "numa_max_node"))); 2345 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2346 dlsym(handle, "numa_available"))); 2347 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2348 dlsym(handle, "numa_tonode_memory"))); 2349 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2350 dlsym(handle, "numa_interleave_memory"))); 2351 2352 2353 if (numa_available() != -1) { 2354 set_numa_all_nodes((unsigned long*)dlsym(handle, "numa_all_nodes")); 2355 // Create a cpu -> node mapping 2356 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true); 2357 rebuild_cpu_to_node_map(); 2358 return true; 2359 } 2360 } 2361 } 2362 return false; 2363} 2364 2365// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2366// The table is later used in get_node_by_cpu(). 2367void os::Linux::rebuild_cpu_to_node_map() { 2368 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2369 // in libnuma (possible values are starting from 16, 2370 // and continuing up with every other power of 2, but less 2371 // than the maximum number of CPUs supported by kernel), and 2372 // is a subject to change (in libnuma version 2 the requirements 2373 // are more reasonable) we'll just hardcode the number they use 2374 // in the library. 2375 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2376 2377 size_t cpu_num = os::active_processor_count(); 2378 size_t cpu_map_size = NCPUS / BitsPerCLong; 2379 size_t cpu_map_valid_size = 2380 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2381 2382 cpu_to_node()->clear(); 2383 cpu_to_node()->at_grow(cpu_num - 1); 2384 size_t node_num = numa_get_groups_num(); 2385 2386 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size); 2387 for (size_t i = 0; i < node_num; i++) { 2388 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2389 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2390 if (cpu_map[j] != 0) { 2391 for (size_t k = 0; k < BitsPerCLong; k++) { 2392 if (cpu_map[j] & (1UL << k)) { 2393 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2394 } 2395 } 2396 } 2397 } 2398 } 2399 } 2400 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2401} 2402 2403int os::Linux::get_node_by_cpu(int cpu_id) { 2404 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2405 return cpu_to_node()->at(cpu_id); 2406 } 2407 return -1; 2408} 2409 2410GrowableArray<int>* os::Linux::_cpu_to_node; 2411os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2412os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2413os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2414os::Linux::numa_available_func_t os::Linux::_numa_available; 2415os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2416os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2417unsigned long* os::Linux::_numa_all_nodes; 2418 2419bool os::uncommit_memory(char* addr, size_t size) { 2420 return ::mmap(addr, size, 2421 PROT_READ|PROT_WRITE|PROT_EXEC, 2422 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0) 2423 != MAP_FAILED; 2424} 2425 2426static address _highest_vm_reserved_address = NULL; 2427 2428// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2429// at 'requested_addr'. If there are existing memory mappings at the same 2430// location, however, they will be overwritten. If 'fixed' is false, 2431// 'requested_addr' is only treated as a hint, the return value may or 2432// may not start from the requested address. Unlike Linux mmap(), this 2433// function returns NULL to indicate failure. 2434static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2435 char * addr; 2436 int flags; 2437 2438 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2439 if (fixed) { 2440 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2441 flags |= MAP_FIXED; 2442 } 2443 2444 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE|PROT_EXEC, 2445 flags, -1, 0); 2446 2447 if (addr != MAP_FAILED) { 2448 // anon_mmap() should only get called during VM initialization, 2449 // don't need lock (actually we can skip locking even it can be called 2450 // from multiple threads, because _highest_vm_reserved_address is just a 2451 // hint about the upper limit of non-stack memory regions.) 2452 if ((address)addr + bytes > _highest_vm_reserved_address) { 2453 _highest_vm_reserved_address = (address)addr + bytes; 2454 } 2455 } 2456 2457 return addr == MAP_FAILED ? NULL : addr; 2458} 2459 2460// Don't update _highest_vm_reserved_address, because there might be memory 2461// regions above addr + size. If so, releasing a memory region only creates 2462// a hole in the address space, it doesn't help prevent heap-stack collision. 2463// 2464static int anon_munmap(char * addr, size_t size) { 2465 return ::munmap(addr, size) == 0; 2466} 2467 2468char* os::reserve_memory(size_t bytes, char* requested_addr, 2469 size_t alignment_hint) { 2470 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 2471} 2472 2473bool os::release_memory(char* addr, size_t size) { 2474 return anon_munmap(addr, size); 2475} 2476 2477static address highest_vm_reserved_address() { 2478 return _highest_vm_reserved_address; 2479} 2480 2481static bool linux_mprotect(char* addr, size_t size, int prot) { 2482 // Linux wants the mprotect address argument to be page aligned. 2483 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 2484 2485 // According to SUSv3, mprotect() should only be used with mappings 2486 // established by mmap(), and mmap() always maps whole pages. Unaligned 2487 // 'addr' likely indicates problem in the VM (e.g. trying to change 2488 // protection of malloc'ed or statically allocated memory). Check the 2489 // caller if you hit this assert. 2490 assert(addr == bottom, "sanity check"); 2491 2492 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 2493 return ::mprotect(bottom, size, prot) == 0; 2494} 2495 2496// Set protections specified 2497bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 2498 bool is_committed) { 2499 unsigned int p = 0; 2500 switch (prot) { 2501 case MEM_PROT_NONE: p = PROT_NONE; break; 2502 case MEM_PROT_READ: p = PROT_READ; break; 2503 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 2504 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 2505 default: 2506 ShouldNotReachHere(); 2507 } 2508 // is_committed is unused. 2509 return linux_mprotect(addr, bytes, p); 2510} 2511 2512bool os::guard_memory(char* addr, size_t size) { 2513 return linux_mprotect(addr, size, PROT_NONE); 2514} 2515 2516bool os::unguard_memory(char* addr, size_t size) { 2517 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 2518} 2519 2520// Large page support 2521 2522static size_t _large_page_size = 0; 2523 2524bool os::large_page_init() { 2525 if (!UseLargePages) return false; 2526 2527 if (LargePageSizeInBytes) { 2528 _large_page_size = LargePageSizeInBytes; 2529 } else { 2530 // large_page_size on Linux is used to round up heap size. x86 uses either 2531 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 2532 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 2533 // page as large as 256M. 2534 // 2535 // Here we try to figure out page size by parsing /proc/meminfo and looking 2536 // for a line with the following format: 2537 // Hugepagesize: 2048 kB 2538 // 2539 // If we can't determine the value (e.g. /proc is not mounted, or the text 2540 // format has been changed), we'll use the largest page size supported by 2541 // the processor. 2542 2543 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M); 2544 2545 FILE *fp = fopen("/proc/meminfo", "r"); 2546 if (fp) { 2547 while (!feof(fp)) { 2548 int x = 0; 2549 char buf[16]; 2550 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 2551 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 2552 _large_page_size = x * K; 2553 break; 2554 } 2555 } else { 2556 // skip to next line 2557 for (;;) { 2558 int ch = fgetc(fp); 2559 if (ch == EOF || ch == (int)'\n') break; 2560 } 2561 } 2562 } 2563 fclose(fp); 2564 } 2565 } 2566 2567 const size_t default_page_size = (size_t)Linux::page_size(); 2568 if (_large_page_size > default_page_size) { 2569 _page_sizes[0] = _large_page_size; 2570 _page_sizes[1] = default_page_size; 2571 _page_sizes[2] = 0; 2572 } 2573 2574 // Large page support is available on 2.6 or newer kernel, some vendors 2575 // (e.g. Redhat) have backported it to their 2.4 based distributions. 2576 // We optimistically assume the support is available. If later it turns out 2577 // not true, VM will automatically switch to use regular page size. 2578 return true; 2579} 2580 2581#ifndef SHM_HUGETLB 2582#define SHM_HUGETLB 04000 2583#endif 2584 2585char* os::reserve_memory_special(size_t bytes) { 2586 assert(UseLargePages, "only for large pages"); 2587 2588 key_t key = IPC_PRIVATE; 2589 char *addr; 2590 2591 bool warn_on_failure = UseLargePages && 2592 (!FLAG_IS_DEFAULT(UseLargePages) || 2593 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 2594 ); 2595 char msg[128]; 2596 2597 // Create a large shared memory region to attach to based on size. 2598 // Currently, size is the total size of the heap 2599 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 2600 if (shmid == -1) { 2601 // Possible reasons for shmget failure: 2602 // 1. shmmax is too small for Java heap. 2603 // > check shmmax value: cat /proc/sys/kernel/shmmax 2604 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 2605 // 2. not enough large page memory. 2606 // > check available large pages: cat /proc/meminfo 2607 // > increase amount of large pages: 2608 // echo new_value > /proc/sys/vm/nr_hugepages 2609 // Note 1: different Linux may use different name for this property, 2610 // e.g. on Redhat AS-3 it is "hugetlb_pool". 2611 // Note 2: it's possible there's enough physical memory available but 2612 // they are so fragmented after a long run that they can't 2613 // coalesce into large pages. Try to reserve large pages when 2614 // the system is still "fresh". 2615 if (warn_on_failure) { 2616 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 2617 warning(msg); 2618 } 2619 return NULL; 2620 } 2621 2622 // attach to the region 2623 addr = (char*)shmat(shmid, NULL, 0); 2624 int err = errno; 2625 2626 // Remove shmid. If shmat() is successful, the actual shared memory segment 2627 // will be deleted when it's detached by shmdt() or when the process 2628 // terminates. If shmat() is not successful this will remove the shared 2629 // segment immediately. 2630 shmctl(shmid, IPC_RMID, NULL); 2631 2632 if ((intptr_t)addr == -1) { 2633 if (warn_on_failure) { 2634 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 2635 warning(msg); 2636 } 2637 return NULL; 2638 } 2639 2640 return addr; 2641} 2642 2643bool os::release_memory_special(char* base, size_t bytes) { 2644 // detaching the SHM segment will also delete it, see reserve_memory_special() 2645 int rslt = shmdt(base); 2646 return rslt == 0; 2647} 2648 2649size_t os::large_page_size() { 2650 return _large_page_size; 2651} 2652 2653// Linux does not support anonymous mmap with large page memory. The only way 2654// to reserve large page memory without file backing is through SysV shared 2655// memory API. The entire memory region is committed and pinned upfront. 2656// Hopefully this will change in the future... 2657bool os::can_commit_large_page_memory() { 2658 return false; 2659} 2660 2661bool os::can_execute_large_page_memory() { 2662 return false; 2663} 2664 2665// Reserve memory at an arbitrary address, only if that area is 2666// available (and not reserved for something else). 2667 2668char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 2669 const int max_tries = 10; 2670 char* base[max_tries]; 2671 size_t size[max_tries]; 2672 const size_t gap = 0x000000; 2673 2674 // Assert only that the size is a multiple of the page size, since 2675 // that's all that mmap requires, and since that's all we really know 2676 // about at this low abstraction level. If we need higher alignment, 2677 // we can either pass an alignment to this method or verify alignment 2678 // in one of the methods further up the call chain. See bug 5044738. 2679 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 2680 2681 // Repeatedly allocate blocks until the block is allocated at the 2682 // right spot. Give up after max_tries. Note that reserve_memory() will 2683 // automatically update _highest_vm_reserved_address if the call is 2684 // successful. The variable tracks the highest memory address every reserved 2685 // by JVM. It is used to detect heap-stack collision if running with 2686 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 2687 // space than needed, it could confuse the collision detecting code. To 2688 // solve the problem, save current _highest_vm_reserved_address and 2689 // calculate the correct value before return. 2690 address old_highest = _highest_vm_reserved_address; 2691 2692 // Linux mmap allows caller to pass an address as hint; give it a try first, 2693 // if kernel honors the hint then we can return immediately. 2694 char * addr = anon_mmap(requested_addr, bytes, false); 2695 if (addr == requested_addr) { 2696 return requested_addr; 2697 } 2698 2699 if (addr != NULL) { 2700 // mmap() is successful but it fails to reserve at the requested address 2701 anon_munmap(addr, bytes); 2702 } 2703 2704 int i; 2705 for (i = 0; i < max_tries; ++i) { 2706 base[i] = reserve_memory(bytes); 2707 2708 if (base[i] != NULL) { 2709 // Is this the block we wanted? 2710 if (base[i] == requested_addr) { 2711 size[i] = bytes; 2712 break; 2713 } 2714 2715 // Does this overlap the block we wanted? Give back the overlapped 2716 // parts and try again. 2717 2718 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 2719 if (top_overlap >= 0 && top_overlap < bytes) { 2720 unmap_memory(base[i], top_overlap); 2721 base[i] += top_overlap; 2722 size[i] = bytes - top_overlap; 2723 } else { 2724 size_t bottom_overlap = base[i] + bytes - requested_addr; 2725 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 2726 unmap_memory(requested_addr, bottom_overlap); 2727 size[i] = bytes - bottom_overlap; 2728 } else { 2729 size[i] = bytes; 2730 } 2731 } 2732 } 2733 } 2734 2735 // Give back the unused reserved pieces. 2736 2737 for (int j = 0; j < i; ++j) { 2738 if (base[j] != NULL) { 2739 unmap_memory(base[j], size[j]); 2740 } 2741 } 2742 2743 if (i < max_tries) { 2744 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 2745 return requested_addr; 2746 } else { 2747 _highest_vm_reserved_address = old_highest; 2748 return NULL; 2749 } 2750} 2751 2752size_t os::read(int fd, void *buf, unsigned int nBytes) { 2753 return ::read(fd, buf, nBytes); 2754} 2755 2756// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 2757// Solaris uses poll(), linux uses park(). 2758// Poll() is likely a better choice, assuming that Thread.interrupt() 2759// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 2760// SIGSEGV, see 4355769. 2761 2762const int NANOSECS_PER_MILLISECS = 1000000; 2763 2764int os::sleep(Thread* thread, jlong millis, bool interruptible) { 2765 assert(thread == Thread::current(), "thread consistency check"); 2766 2767 ParkEvent * const slp = thread->_SleepEvent ; 2768 slp->reset() ; 2769 OrderAccess::fence() ; 2770 2771 if (interruptible) { 2772 jlong prevtime = javaTimeNanos(); 2773 2774 for (;;) { 2775 if (os::is_interrupted(thread, true)) { 2776 return OS_INTRPT; 2777 } 2778 2779 jlong newtime = javaTimeNanos(); 2780 2781 if (newtime - prevtime < 0) { 2782 // time moving backwards, should only happen if no monotonic clock 2783 // not a guarantee() because JVM should not abort on kernel/glibc bugs 2784 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 2785 } else { 2786 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 2787 } 2788 2789 if(millis <= 0) { 2790 return OS_OK; 2791 } 2792 2793 prevtime = newtime; 2794 2795 { 2796 assert(thread->is_Java_thread(), "sanity check"); 2797 JavaThread *jt = (JavaThread *) thread; 2798 ThreadBlockInVM tbivm(jt); 2799 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 2800 2801 jt->set_suspend_equivalent(); 2802 // cleared by handle_special_suspend_equivalent_condition() or 2803 // java_suspend_self() via check_and_wait_while_suspended() 2804 2805 slp->park(millis); 2806 2807 // were we externally suspended while we were waiting? 2808 jt->check_and_wait_while_suspended(); 2809 } 2810 } 2811 } else { 2812 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 2813 jlong prevtime = javaTimeNanos(); 2814 2815 for (;;) { 2816 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 2817 // the 1st iteration ... 2818 jlong newtime = javaTimeNanos(); 2819 2820 if (newtime - prevtime < 0) { 2821 // time moving backwards, should only happen if no monotonic clock 2822 // not a guarantee() because JVM should not abort on kernel/glibc bugs 2823 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 2824 } else { 2825 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 2826 } 2827 2828 if(millis <= 0) break ; 2829 2830 prevtime = newtime; 2831 slp->park(millis); 2832 } 2833 return OS_OK ; 2834 } 2835} 2836 2837int os::naked_sleep() { 2838 // %% make the sleep time an integer flag. for now use 1 millisec. 2839 return os::sleep(Thread::current(), 1, false); 2840} 2841 2842// Sleep forever; naked call to OS-specific sleep; use with CAUTION 2843void os::infinite_sleep() { 2844 while (true) { // sleep forever ... 2845 ::sleep(100); // ... 100 seconds at a time 2846 } 2847} 2848 2849// Used to convert frequent JVM_Yield() to nops 2850bool os::dont_yield() { 2851 return DontYieldALot; 2852} 2853 2854void os::yield() { 2855 sched_yield(); 2856} 2857 2858os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 2859 2860void os::yield_all(int attempts) { 2861 // Yields to all threads, including threads with lower priorities 2862 // Threads on Linux are all with same priority. The Solaris style 2863 // os::yield_all() with nanosleep(1ms) is not necessary. 2864 sched_yield(); 2865} 2866 2867// Called from the tight loops to possibly influence time-sharing heuristics 2868void os::loop_breaker(int attempts) { 2869 os::yield_all(attempts); 2870} 2871 2872//////////////////////////////////////////////////////////////////////////////// 2873// thread priority support 2874 2875// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 2876// only supports dynamic priority, static priority must be zero. For real-time 2877// applications, Linux supports SCHED_RR which allows static priority (1-99). 2878// However, for large multi-threaded applications, SCHED_RR is not only slower 2879// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 2880// of 5 runs - Sep 2005). 2881// 2882// The following code actually changes the niceness of kernel-thread/LWP. It 2883// has an assumption that setpriority() only modifies one kernel-thread/LWP, 2884// not the entire user process, and user level threads are 1:1 mapped to kernel 2885// threads. It has always been the case, but could change in the future. For 2886// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 2887// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 2888 2889int os::java_to_os_priority[MaxPriority + 1] = { 2890 19, // 0 Entry should never be used 2891 2892 4, // 1 MinPriority 2893 3, // 2 2894 2, // 3 2895 2896 1, // 4 2897 0, // 5 NormPriority 2898 -1, // 6 2899 2900 -2, // 7 2901 -3, // 8 2902 -4, // 9 NearMaxPriority 2903 2904 -5 // 10 MaxPriority 2905}; 2906 2907static int prio_init() { 2908 if (ThreadPriorityPolicy == 1) { 2909 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 2910 // if effective uid is not root. Perhaps, a more elegant way of doing 2911 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 2912 if (geteuid() != 0) { 2913 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 2914 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 2915 } 2916 ThreadPriorityPolicy = 0; 2917 } 2918 } 2919 return 0; 2920} 2921 2922OSReturn os::set_native_priority(Thread* thread, int newpri) { 2923 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 2924 2925 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 2926 return (ret == 0) ? OS_OK : OS_ERR; 2927} 2928 2929OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 2930 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 2931 *priority_ptr = java_to_os_priority[NormPriority]; 2932 return OS_OK; 2933 } 2934 2935 errno = 0; 2936 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 2937 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 2938} 2939 2940// Hint to the underlying OS that a task switch would not be good. 2941// Void return because it's a hint and can fail. 2942void os::hint_no_preempt() {} 2943 2944//////////////////////////////////////////////////////////////////////////////// 2945// suspend/resume support 2946 2947// the low-level signal-based suspend/resume support is a remnant from the 2948// old VM-suspension that used to be for java-suspension, safepoints etc, 2949// within hotspot. Now there is a single use-case for this: 2950// - calling get_thread_pc() on the VMThread by the flat-profiler task 2951// that runs in the watcher thread. 2952// The remaining code is greatly simplified from the more general suspension 2953// code that used to be used. 2954// 2955// The protocol is quite simple: 2956// - suspend: 2957// - sends a signal to the target thread 2958// - polls the suspend state of the osthread using a yield loop 2959// - target thread signal handler (SR_handler) sets suspend state 2960// and blocks in sigsuspend until continued 2961// - resume: 2962// - sets target osthread state to continue 2963// - sends signal to end the sigsuspend loop in the SR_handler 2964// 2965// Note that the SR_lock plays no role in this suspend/resume protocol. 2966// 2967 2968static void resume_clear_context(OSThread *osthread) { 2969 osthread->set_ucontext(NULL); 2970 osthread->set_siginfo(NULL); 2971 2972 // notify the suspend action is completed, we have now resumed 2973 osthread->sr.clear_suspended(); 2974} 2975 2976static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 2977 osthread->set_ucontext(context); 2978 osthread->set_siginfo(siginfo); 2979} 2980 2981// 2982// Handler function invoked when a thread's execution is suspended or 2983// resumed. We have to be careful that only async-safe functions are 2984// called here (Note: most pthread functions are not async safe and 2985// should be avoided.) 2986// 2987// Note: sigwait() is a more natural fit than sigsuspend() from an 2988// interface point of view, but sigwait() prevents the signal hander 2989// from being run. libpthread would get very confused by not having 2990// its signal handlers run and prevents sigwait()'s use with the 2991// mutex granting granting signal. 2992// 2993// Currently only ever called on the VMThread 2994// 2995static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 2996 // Save and restore errno to avoid confusing native code with EINTR 2997 // after sigsuspend. 2998 int old_errno = errno; 2999 3000 Thread* thread = Thread::current(); 3001 OSThread* osthread = thread->osthread(); 3002 assert(thread->is_VM_thread(), "Must be VMThread"); 3003 // read current suspend action 3004 int action = osthread->sr.suspend_action(); 3005 if (action == SR_SUSPEND) { 3006 suspend_save_context(osthread, siginfo, context); 3007 3008 // Notify the suspend action is about to be completed. do_suspend() 3009 // waits until SR_SUSPENDED is set and then returns. We will wait 3010 // here for a resume signal and that completes the suspend-other 3011 // action. do_suspend/do_resume is always called as a pair from 3012 // the same thread - so there are no races 3013 3014 // notify the caller 3015 osthread->sr.set_suspended(); 3016 3017 sigset_t suspend_set; // signals for sigsuspend() 3018 3019 // get current set of blocked signals and unblock resume signal 3020 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3021 sigdelset(&suspend_set, SR_signum); 3022 3023 // wait here until we are resumed 3024 do { 3025 sigsuspend(&suspend_set); 3026 // ignore all returns until we get a resume signal 3027 } while (osthread->sr.suspend_action() != SR_CONTINUE); 3028 3029 resume_clear_context(osthread); 3030 3031 } else { 3032 assert(action == SR_CONTINUE, "unexpected sr action"); 3033 // nothing special to do - just leave the handler 3034 } 3035 3036 errno = old_errno; 3037} 3038 3039 3040static int SR_initialize() { 3041 struct sigaction act; 3042 char *s; 3043 /* Get signal number to use for suspend/resume */ 3044 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3045 int sig = ::strtol(s, 0, 10); 3046 if (sig > 0 || sig < _NSIG) { 3047 SR_signum = sig; 3048 } 3049 } 3050 3051 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3052 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3053 3054 sigemptyset(&SR_sigset); 3055 sigaddset(&SR_sigset, SR_signum); 3056 3057 /* Set up signal handler for suspend/resume */ 3058 act.sa_flags = SA_RESTART|SA_SIGINFO; 3059 act.sa_handler = (void (*)(int)) SR_handler; 3060 3061 // SR_signum is blocked by default. 3062 // 4528190 - We also need to block pthread restart signal (32 on all 3063 // supported Linux platforms). Note that LinuxThreads need to block 3064 // this signal for all threads to work properly. So we don't have 3065 // to use hard-coded signal number when setting up the mask. 3066 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 3067 3068 if (sigaction(SR_signum, &act, 0) == -1) { 3069 return -1; 3070 } 3071 3072 // Save signal flag 3073 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 3074 return 0; 3075} 3076 3077static int SR_finalize() { 3078 return 0; 3079} 3080 3081 3082// returns true on success and false on error - really an error is fatal 3083// but this seems the normal response to library errors 3084static bool do_suspend(OSThread* osthread) { 3085 // mark as suspended and send signal 3086 osthread->sr.set_suspend_action(SR_SUSPEND); 3087 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3088 assert_status(status == 0, status, "pthread_kill"); 3089 3090 // check status and wait until notified of suspension 3091 if (status == 0) { 3092 for (int i = 0; !osthread->sr.is_suspended(); i++) { 3093 os::yield_all(i); 3094 } 3095 osthread->sr.set_suspend_action(SR_NONE); 3096 return true; 3097 } 3098 else { 3099 osthread->sr.set_suspend_action(SR_NONE); 3100 return false; 3101 } 3102} 3103 3104static void do_resume(OSThread* osthread) { 3105 assert(osthread->sr.is_suspended(), "thread should be suspended"); 3106 osthread->sr.set_suspend_action(SR_CONTINUE); 3107 3108 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3109 assert_status(status == 0, status, "pthread_kill"); 3110 // check status and wait unit notified of resumption 3111 if (status == 0) { 3112 for (int i = 0; osthread->sr.is_suspended(); i++) { 3113 os::yield_all(i); 3114 } 3115 } 3116 osthread->sr.set_suspend_action(SR_NONE); 3117} 3118 3119//////////////////////////////////////////////////////////////////////////////// 3120// interrupt support 3121 3122void os::interrupt(Thread* thread) { 3123 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3124 "possibility of dangling Thread pointer"); 3125 3126 OSThread* osthread = thread->osthread(); 3127 3128 if (!osthread->interrupted()) { 3129 osthread->set_interrupted(true); 3130 // More than one thread can get here with the same value of osthread, 3131 // resulting in multiple notifications. We do, however, want the store 3132 // to interrupted() to be visible to other threads before we execute unpark(). 3133 OrderAccess::fence(); 3134 ParkEvent * const slp = thread->_SleepEvent ; 3135 if (slp != NULL) slp->unpark() ; 3136 } 3137 3138 // For JSR166. Unpark even if interrupt status already was set 3139 if (thread->is_Java_thread()) 3140 ((JavaThread*)thread)->parker()->unpark(); 3141 3142 ParkEvent * ev = thread->_ParkEvent ; 3143 if (ev != NULL) ev->unpark() ; 3144 3145} 3146 3147bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 3148 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3149 "possibility of dangling Thread pointer"); 3150 3151 OSThread* osthread = thread->osthread(); 3152 3153 bool interrupted = osthread->interrupted(); 3154 3155 if (interrupted && clear_interrupted) { 3156 osthread->set_interrupted(false); 3157 // consider thread->_SleepEvent->reset() ... optional optimization 3158 } 3159 3160 return interrupted; 3161} 3162 3163/////////////////////////////////////////////////////////////////////////////////// 3164// signal handling (except suspend/resume) 3165 3166// This routine may be used by user applications as a "hook" to catch signals. 3167// The user-defined signal handler must pass unrecognized signals to this 3168// routine, and if it returns true (non-zero), then the signal handler must 3169// return immediately. If the flag "abort_if_unrecognized" is true, then this 3170// routine will never retun false (zero), but instead will execute a VM panic 3171// routine kill the process. 3172// 3173// If this routine returns false, it is OK to call it again. This allows 3174// the user-defined signal handler to perform checks either before or after 3175// the VM performs its own checks. Naturally, the user code would be making 3176// a serious error if it tried to handle an exception (such as a null check 3177// or breakpoint) that the VM was generating for its own correct operation. 3178// 3179// This routine may recognize any of the following kinds of signals: 3180// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 3181// It should be consulted by handlers for any of those signals. 3182// 3183// The caller of this routine must pass in the three arguments supplied 3184// to the function referred to in the "sa_sigaction" (not the "sa_handler") 3185// field of the structure passed to sigaction(). This routine assumes that 3186// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 3187// 3188// Note that the VM will print warnings if it detects conflicting signal 3189// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 3190// 3191extern "C" int 3192JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 3193 void* ucontext, int abort_if_unrecognized); 3194 3195void signalHandler(int sig, siginfo_t* info, void* uc) { 3196 assert(info != NULL && uc != NULL, "it must be old kernel"); 3197 JVM_handle_linux_signal(sig, info, uc, true); 3198} 3199 3200 3201// This boolean allows users to forward their own non-matching signals 3202// to JVM_handle_linux_signal, harmlessly. 3203bool os::Linux::signal_handlers_are_installed = false; 3204 3205// For signal-chaining 3206struct sigaction os::Linux::sigact[MAXSIGNUM]; 3207unsigned int os::Linux::sigs = 0; 3208bool os::Linux::libjsig_is_loaded = false; 3209typedef struct sigaction *(*get_signal_t)(int); 3210get_signal_t os::Linux::get_signal_action = NULL; 3211 3212struct sigaction* os::Linux::get_chained_signal_action(int sig) { 3213 struct sigaction *actp = NULL; 3214 3215 if (libjsig_is_loaded) { 3216 // Retrieve the old signal handler from libjsig 3217 actp = (*get_signal_action)(sig); 3218 } 3219 if (actp == NULL) { 3220 // Retrieve the preinstalled signal handler from jvm 3221 actp = get_preinstalled_handler(sig); 3222 } 3223 3224 return actp; 3225} 3226 3227static bool call_chained_handler(struct sigaction *actp, int sig, 3228 siginfo_t *siginfo, void *context) { 3229 // Call the old signal handler 3230 if (actp->sa_handler == SIG_DFL) { 3231 // It's more reasonable to let jvm treat it as an unexpected exception 3232 // instead of taking the default action. 3233 return false; 3234 } else if (actp->sa_handler != SIG_IGN) { 3235 if ((actp->sa_flags & SA_NODEFER) == 0) { 3236 // automaticlly block the signal 3237 sigaddset(&(actp->sa_mask), sig); 3238 } 3239 3240 sa_handler_t hand; 3241 sa_sigaction_t sa; 3242 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 3243 // retrieve the chained handler 3244 if (siginfo_flag_set) { 3245 sa = actp->sa_sigaction; 3246 } else { 3247 hand = actp->sa_handler; 3248 } 3249 3250 if ((actp->sa_flags & SA_RESETHAND) != 0) { 3251 actp->sa_handler = SIG_DFL; 3252 } 3253 3254 // try to honor the signal mask 3255 sigset_t oset; 3256 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 3257 3258 // call into the chained handler 3259 if (siginfo_flag_set) { 3260 (*sa)(sig, siginfo, context); 3261 } else { 3262 (*hand)(sig); 3263 } 3264 3265 // restore the signal mask 3266 pthread_sigmask(SIG_SETMASK, &oset, 0); 3267 } 3268 // Tell jvm's signal handler the signal is taken care of. 3269 return true; 3270} 3271 3272bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 3273 bool chained = false; 3274 // signal-chaining 3275 if (UseSignalChaining) { 3276 struct sigaction *actp = get_chained_signal_action(sig); 3277 if (actp != NULL) { 3278 chained = call_chained_handler(actp, sig, siginfo, context); 3279 } 3280 } 3281 return chained; 3282} 3283 3284struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 3285 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 3286 return &sigact[sig]; 3287 } 3288 return NULL; 3289} 3290 3291void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 3292 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3293 sigact[sig] = oldAct; 3294 sigs |= (unsigned int)1 << sig; 3295} 3296 3297// for diagnostic 3298int os::Linux::sigflags[MAXSIGNUM]; 3299 3300int os::Linux::get_our_sigflags(int sig) { 3301 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3302 return sigflags[sig]; 3303} 3304 3305void os::Linux::set_our_sigflags(int sig, int flags) { 3306 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3307 sigflags[sig] = flags; 3308} 3309 3310void os::Linux::set_signal_handler(int sig, bool set_installed) { 3311 // Check for overwrite. 3312 struct sigaction oldAct; 3313 sigaction(sig, (struct sigaction*)NULL, &oldAct); 3314 3315 void* oldhand = oldAct.sa_sigaction 3316 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3317 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3318 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 3319 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 3320 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 3321 if (AllowUserSignalHandlers || !set_installed) { 3322 // Do not overwrite; user takes responsibility to forward to us. 3323 return; 3324 } else if (UseSignalChaining) { 3325 // save the old handler in jvm 3326 save_preinstalled_handler(sig, oldAct); 3327 // libjsig also interposes the sigaction() call below and saves the 3328 // old sigaction on it own. 3329 } else { 3330 fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig); 3331 } 3332 } 3333 3334 struct sigaction sigAct; 3335 sigfillset(&(sigAct.sa_mask)); 3336 sigAct.sa_handler = SIG_DFL; 3337 if (!set_installed) { 3338 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3339 } else { 3340 sigAct.sa_sigaction = signalHandler; 3341 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3342 } 3343 // Save flags, which are set by ours 3344 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3345 sigflags[sig] = sigAct.sa_flags; 3346 3347 int ret = sigaction(sig, &sigAct, &oldAct); 3348 assert(ret == 0, "check"); 3349 3350 void* oldhand2 = oldAct.sa_sigaction 3351 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3352 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3353 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 3354} 3355 3356// install signal handlers for signals that HotSpot needs to 3357// handle in order to support Java-level exception handling. 3358 3359void os::Linux::install_signal_handlers() { 3360 if (!signal_handlers_are_installed) { 3361 signal_handlers_are_installed = true; 3362 3363 // signal-chaining 3364 typedef void (*signal_setting_t)(); 3365 signal_setting_t begin_signal_setting = NULL; 3366 signal_setting_t end_signal_setting = NULL; 3367 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3368 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 3369 if (begin_signal_setting != NULL) { 3370 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3371 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 3372 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 3373 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 3374 libjsig_is_loaded = true; 3375 assert(UseSignalChaining, "should enable signal-chaining"); 3376 } 3377 if (libjsig_is_loaded) { 3378 // Tell libjsig jvm is setting signal handlers 3379 (*begin_signal_setting)(); 3380 } 3381 3382 set_signal_handler(SIGSEGV, true); 3383 set_signal_handler(SIGPIPE, true); 3384 set_signal_handler(SIGBUS, true); 3385 set_signal_handler(SIGILL, true); 3386 set_signal_handler(SIGFPE, true); 3387 set_signal_handler(SIGXFSZ, true); 3388 3389 if (libjsig_is_loaded) { 3390 // Tell libjsig jvm finishes setting signal handlers 3391 (*end_signal_setting)(); 3392 } 3393 3394 // We don't activate signal checker if libjsig is in place, we trust ourselves 3395 // and if UserSignalHandler is installed all bets are off 3396 if (CheckJNICalls) { 3397 if (libjsig_is_loaded) { 3398 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 3399 check_signals = false; 3400 } 3401 if (AllowUserSignalHandlers) { 3402 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 3403 check_signals = false; 3404 } 3405 } 3406 } 3407} 3408 3409// This is the fastest way to get thread cpu time on Linux. 3410// Returns cpu time (user+sys) for any thread, not only for current. 3411// POSIX compliant clocks are implemented in the kernels 2.6.16+. 3412// It might work on 2.6.10+ with a special kernel/glibc patch. 3413// For reference, please, see IEEE Std 1003.1-2004: 3414// http://www.unix.org/single_unix_specification 3415 3416jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 3417 struct timespec tp; 3418 int rc = os::Linux::clock_gettime(clockid, &tp); 3419 assert(rc == 0, "clock_gettime is expected to return 0 code"); 3420 3421 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec; 3422} 3423 3424///// 3425// glibc on Linux platform uses non-documented flag 3426// to indicate, that some special sort of signal 3427// trampoline is used. 3428// We will never set this flag, and we should 3429// ignore this flag in our diagnostic 3430#ifdef SIGNIFICANT_SIGNAL_MASK 3431#undef SIGNIFICANT_SIGNAL_MASK 3432#endif 3433#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 3434 3435static const char* get_signal_handler_name(address handler, 3436 char* buf, int buflen) { 3437 int offset; 3438 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 3439 if (found) { 3440 // skip directory names 3441 const char *p1, *p2; 3442 p1 = buf; 3443 size_t len = strlen(os::file_separator()); 3444 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 3445 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 3446 } else { 3447 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 3448 } 3449 return buf; 3450} 3451 3452static void print_signal_handler(outputStream* st, int sig, 3453 char* buf, size_t buflen) { 3454 struct sigaction sa; 3455 3456 sigaction(sig, NULL, &sa); 3457 3458 // See comment for SIGNIFICANT_SIGNAL_MASK define 3459 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3460 3461 st->print("%s: ", os::exception_name(sig, buf, buflen)); 3462 3463 address handler = (sa.sa_flags & SA_SIGINFO) 3464 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 3465 : CAST_FROM_FN_PTR(address, sa.sa_handler); 3466 3467 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 3468 st->print("SIG_DFL"); 3469 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 3470 st->print("SIG_IGN"); 3471 } else { 3472 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 3473 } 3474 3475 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 3476 3477 address rh = VMError::get_resetted_sighandler(sig); 3478 // May be, handler was resetted by VMError? 3479 if(rh != NULL) { 3480 handler = rh; 3481 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 3482 } 3483 3484 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 3485 3486 // Check: is it our handler? 3487 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 3488 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 3489 // It is our signal handler 3490 // check for flags, reset system-used one! 3491 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 3492 st->print( 3493 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 3494 os::Linux::get_our_sigflags(sig)); 3495 } 3496 } 3497 st->cr(); 3498} 3499 3500 3501#define DO_SIGNAL_CHECK(sig) \ 3502 if (!sigismember(&check_signal_done, sig)) \ 3503 os::Linux::check_signal_handler(sig) 3504 3505// This method is a periodic task to check for misbehaving JNI applications 3506// under CheckJNI, we can add any periodic checks here 3507 3508void os::run_periodic_checks() { 3509 3510 if (check_signals == false) return; 3511 3512 // SEGV and BUS if overridden could potentially prevent 3513 // generation of hs*.log in the event of a crash, debugging 3514 // such a case can be very challenging, so we absolutely 3515 // check the following for a good measure: 3516 DO_SIGNAL_CHECK(SIGSEGV); 3517 DO_SIGNAL_CHECK(SIGILL); 3518 DO_SIGNAL_CHECK(SIGFPE); 3519 DO_SIGNAL_CHECK(SIGBUS); 3520 DO_SIGNAL_CHECK(SIGPIPE); 3521 DO_SIGNAL_CHECK(SIGXFSZ); 3522 3523 3524 // ReduceSignalUsage allows the user to override these handlers 3525 // see comments at the very top and jvm_solaris.h 3526 if (!ReduceSignalUsage) { 3527 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 3528 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 3529 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 3530 DO_SIGNAL_CHECK(BREAK_SIGNAL); 3531 } 3532 3533 DO_SIGNAL_CHECK(SR_signum); 3534 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 3535} 3536 3537typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 3538 3539static os_sigaction_t os_sigaction = NULL; 3540 3541void os::Linux::check_signal_handler(int sig) { 3542 char buf[O_BUFLEN]; 3543 address jvmHandler = NULL; 3544 3545 3546 struct sigaction act; 3547 if (os_sigaction == NULL) { 3548 // only trust the default sigaction, in case it has been interposed 3549 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 3550 if (os_sigaction == NULL) return; 3551 } 3552 3553 os_sigaction(sig, (struct sigaction*)NULL, &act); 3554 3555 3556 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3557 3558 address thisHandler = (act.sa_flags & SA_SIGINFO) 3559 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 3560 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 3561 3562 3563 switch(sig) { 3564 case SIGSEGV: 3565 case SIGBUS: 3566 case SIGFPE: 3567 case SIGPIPE: 3568 case SIGILL: 3569 case SIGXFSZ: 3570 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 3571 break; 3572 3573 case SHUTDOWN1_SIGNAL: 3574 case SHUTDOWN2_SIGNAL: 3575 case SHUTDOWN3_SIGNAL: 3576 case BREAK_SIGNAL: 3577 jvmHandler = (address)user_handler(); 3578 break; 3579 3580 case INTERRUPT_SIGNAL: 3581 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 3582 break; 3583 3584 default: 3585 if (sig == SR_signum) { 3586 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 3587 } else { 3588 return; 3589 } 3590 break; 3591 } 3592 3593 if (thisHandler != jvmHandler) { 3594 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 3595 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 3596 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 3597 // No need to check this sig any longer 3598 sigaddset(&check_signal_done, sig); 3599 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 3600 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 3601 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 3602 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 3603 // No need to check this sig any longer 3604 sigaddset(&check_signal_done, sig); 3605 } 3606 3607 // Dump all the signal 3608 if (sigismember(&check_signal_done, sig)) { 3609 print_signal_handlers(tty, buf, O_BUFLEN); 3610 } 3611} 3612 3613extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 3614 3615extern bool signal_name(int signo, char* buf, size_t len); 3616 3617const char* os::exception_name(int exception_code, char* buf, size_t size) { 3618 if (0 < exception_code && exception_code <= SIGRTMAX) { 3619 // signal 3620 if (!signal_name(exception_code, buf, size)) { 3621 jio_snprintf(buf, size, "SIG%d", exception_code); 3622 } 3623 return buf; 3624 } else { 3625 return NULL; 3626 } 3627} 3628 3629// this is called _before_ the most of global arguments have been parsed 3630void os::init(void) { 3631 char dummy; /* used to get a guess on initial stack address */ 3632// first_hrtime = gethrtime(); 3633 3634 // With LinuxThreads the JavaMain thread pid (primordial thread) 3635 // is different than the pid of the java launcher thread. 3636 // So, on Linux, the launcher thread pid is passed to the VM 3637 // via the sun.java.launcher.pid property. 3638 // Use this property instead of getpid() if it was correctly passed. 3639 // See bug 6351349. 3640 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 3641 3642 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 3643 3644 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 3645 3646 init_random(1234567); 3647 3648 ThreadCritical::initialize(); 3649 3650 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 3651 if (Linux::page_size() == -1) { 3652 fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno)); 3653 } 3654 init_page_sizes((size_t) Linux::page_size()); 3655 3656 Linux::initialize_system_info(); 3657 3658 // main_thread points to the aboriginal thread 3659 Linux::_main_thread = pthread_self(); 3660 3661 Linux::clock_init(); 3662 initial_time_count = os::elapsed_counter(); 3663 pthread_mutex_init(&dl_mutex, NULL); 3664} 3665 3666// To install functions for atexit system call 3667extern "C" { 3668 static void perfMemory_exit_helper() { 3669 perfMemory_exit(); 3670 } 3671} 3672 3673// this is called _after_ the global arguments have been parsed 3674jint os::init_2(void) 3675{ 3676 Linux::fast_thread_clock_init(); 3677 3678 // Allocate a single page and mark it as readable for safepoint polling 3679 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3680 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 3681 3682 os::set_polling_page( polling_page ); 3683 3684#ifndef PRODUCT 3685 if(Verbose && PrintMiscellaneous) 3686 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 3687#endif 3688 3689 if (!UseMembar) { 3690 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3691 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); 3692 os::set_memory_serialize_page( mem_serialize_page ); 3693 3694#ifndef PRODUCT 3695 if(Verbose && PrintMiscellaneous) 3696 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 3697#endif 3698 } 3699 3700 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init()); 3701 3702 // initialize suspend/resume support - must do this before signal_sets_init() 3703 if (SR_initialize() != 0) { 3704 perror("SR_initialize failed"); 3705 return JNI_ERR; 3706 } 3707 3708 Linux::signal_sets_init(); 3709 Linux::install_signal_handlers(); 3710 3711 size_t threadStackSizeInBytes = ThreadStackSize * K; 3712 if (threadStackSizeInBytes != 0 && 3713 threadStackSizeInBytes < Linux::min_stack_allowed) { 3714 tty->print_cr("\nThe stack size specified is too small, " 3715 "Specify at least %dk", 3716 Linux::min_stack_allowed / K); 3717 return JNI_ERR; 3718 } 3719 3720 // Make the stack size a multiple of the page size so that 3721 // the yellow/red zones can be guarded. 3722 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 3723 vm_page_size())); 3724 3725 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 3726 3727 Linux::libpthread_init(); 3728 if (PrintMiscellaneous && (Verbose || WizardMode)) { 3729 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 3730 Linux::glibc_version(), Linux::libpthread_version(), 3731 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 3732 } 3733 3734 if (UseNUMA) { 3735 if (!Linux::libnuma_init()) { 3736 UseNUMA = false; 3737 } else { 3738 if ((Linux::numa_max_node() < 1)) { 3739 // There's only one node(they start from 0), disable NUMA. 3740 UseNUMA = false; 3741 } 3742 } 3743 if (!UseNUMA && ForceNUMA) { 3744 UseNUMA = true; 3745 } 3746 } 3747 3748 if (MaxFDLimit) { 3749 // set the number of file descriptors to max. print out error 3750 // if getrlimit/setrlimit fails but continue regardless. 3751 struct rlimit nbr_files; 3752 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 3753 if (status != 0) { 3754 if (PrintMiscellaneous && (Verbose || WizardMode)) 3755 perror("os::init_2 getrlimit failed"); 3756 } else { 3757 nbr_files.rlim_cur = nbr_files.rlim_max; 3758 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 3759 if (status != 0) { 3760 if (PrintMiscellaneous && (Verbose || WizardMode)) 3761 perror("os::init_2 setrlimit failed"); 3762 } 3763 } 3764 } 3765 3766 // Initialize lock used to serialize thread creation (see os::create_thread) 3767 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 3768 3769 // Initialize HPI. 3770 jint hpi_result = hpi::initialize(); 3771 if (hpi_result != JNI_OK) { 3772 tty->print_cr("There was an error trying to initialize the HPI library."); 3773 return hpi_result; 3774 } 3775 3776 // at-exit methods are called in the reverse order of their registration. 3777 // atexit functions are called on return from main or as a result of a 3778 // call to exit(3C). There can be only 32 of these functions registered 3779 // and atexit() does not set errno. 3780 3781 if (PerfAllowAtExitRegistration) { 3782 // only register atexit functions if PerfAllowAtExitRegistration is set. 3783 // atexit functions can be delayed until process exit time, which 3784 // can be problematic for embedded VM situations. Embedded VMs should 3785 // call DestroyJavaVM() to assure that VM resources are released. 3786 3787 // note: perfMemory_exit_helper atexit function may be removed in 3788 // the future if the appropriate cleanup code can be added to the 3789 // VM_Exit VMOperation's doit method. 3790 if (atexit(perfMemory_exit_helper) != 0) { 3791 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 3792 } 3793 } 3794 3795 // initialize thread priority policy 3796 prio_init(); 3797 3798 return JNI_OK; 3799} 3800 3801// Mark the polling page as unreadable 3802void os::make_polling_page_unreadable(void) { 3803 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 3804 fatal("Could not disable polling page"); 3805}; 3806 3807// Mark the polling page as readable 3808void os::make_polling_page_readable(void) { 3809 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 3810 fatal("Could not enable polling page"); 3811 } 3812}; 3813 3814int os::active_processor_count() { 3815 // Linux doesn't yet have a (official) notion of processor sets, 3816 // so just return the number of online processors. 3817 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 3818 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 3819 return online_cpus; 3820} 3821 3822bool os::distribute_processes(uint length, uint* distribution) { 3823 // Not yet implemented. 3824 return false; 3825} 3826 3827bool os::bind_to_processor(uint processor_id) { 3828 // Not yet implemented. 3829 return false; 3830} 3831 3832/// 3833 3834// Suspends the target using the signal mechanism and then grabs the PC before 3835// resuming the target. Used by the flat-profiler only 3836ExtendedPC os::get_thread_pc(Thread* thread) { 3837 // Make sure that it is called by the watcher for the VMThread 3838 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 3839 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 3840 3841 ExtendedPC epc; 3842 3843 OSThread* osthread = thread->osthread(); 3844 if (do_suspend(osthread)) { 3845 if (osthread->ucontext() != NULL) { 3846 epc = os::Linux::ucontext_get_pc(osthread->ucontext()); 3847 } else { 3848 // NULL context is unexpected, double-check this is the VMThread 3849 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 3850 } 3851 do_resume(osthread); 3852 } 3853 // failure means pthread_kill failed for some reason - arguably this is 3854 // a fatal problem, but such problems are ignored elsewhere 3855 3856 return epc; 3857} 3858 3859int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 3860{ 3861 if (is_NPTL()) { 3862 return pthread_cond_timedwait(_cond, _mutex, _abstime); 3863 } else { 3864#ifndef IA64 3865 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 3866 // word back to default 64bit precision if condvar is signaled. Java 3867 // wants 53bit precision. Save and restore current value. 3868 int fpu = get_fpu_control_word(); 3869#endif // IA64 3870 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 3871#ifndef IA64 3872 set_fpu_control_word(fpu); 3873#endif // IA64 3874 return status; 3875 } 3876} 3877 3878//////////////////////////////////////////////////////////////////////////////// 3879// debug support 3880 3881#ifndef PRODUCT 3882static address same_page(address x, address y) { 3883 int page_bits = -os::vm_page_size(); 3884 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) 3885 return x; 3886 else if (x > y) 3887 return (address)(intptr_t(y) | ~page_bits) + 1; 3888 else 3889 return (address)(intptr_t(y) & page_bits); 3890} 3891 3892bool os::find(address addr) { 3893 Dl_info dlinfo; 3894 memset(&dlinfo, 0, sizeof(dlinfo)); 3895 if (dladdr(addr, &dlinfo)) { 3896 tty->print(PTR_FORMAT ": ", addr); 3897 if (dlinfo.dli_sname != NULL) { 3898 tty->print("%s+%#x", dlinfo.dli_sname, 3899 addr - (intptr_t)dlinfo.dli_saddr); 3900 } else if (dlinfo.dli_fname) { 3901 tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 3902 } else { 3903 tty->print("<absolute address>"); 3904 } 3905 if (dlinfo.dli_fname) { 3906 tty->print(" in %s", dlinfo.dli_fname); 3907 } 3908 if (dlinfo.dli_fbase) { 3909 tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 3910 } 3911 tty->cr(); 3912 3913 if (Verbose) { 3914 // decode some bytes around the PC 3915 address begin = same_page(addr-40, addr); 3916 address end = same_page(addr+40, addr); 3917 address lowest = (address) dlinfo.dli_sname; 3918 if (!lowest) lowest = (address) dlinfo.dli_fbase; 3919 if (begin < lowest) begin = lowest; 3920 Dl_info dlinfo2; 3921 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr 3922 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 3923 end = (address) dlinfo2.dli_saddr; 3924 Disassembler::decode(begin, end); 3925 } 3926 return true; 3927 } 3928 return false; 3929} 3930 3931#endif 3932 3933//////////////////////////////////////////////////////////////////////////////// 3934// misc 3935 3936// This does not do anything on Linux. This is basically a hook for being 3937// able to use structured exception handling (thread-local exception filters) 3938// on, e.g., Win32. 3939void 3940os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 3941 JavaCallArguments* args, Thread* thread) { 3942 f(value, method, args, thread); 3943} 3944 3945void os::print_statistics() { 3946} 3947 3948int os::message_box(const char* title, const char* message) { 3949 int i; 3950 fdStream err(defaultStream::error_fd()); 3951 for (i = 0; i < 78; i++) err.print_raw("="); 3952 err.cr(); 3953 err.print_raw_cr(title); 3954 for (i = 0; i < 78; i++) err.print_raw("-"); 3955 err.cr(); 3956 err.print_raw_cr(message); 3957 for (i = 0; i < 78; i++) err.print_raw("="); 3958 err.cr(); 3959 3960 char buf[16]; 3961 // Prevent process from exiting upon "read error" without consuming all CPU 3962 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 3963 3964 return buf[0] == 'y' || buf[0] == 'Y'; 3965} 3966 3967int os::stat(const char *path, struct stat *sbuf) { 3968 char pathbuf[MAX_PATH]; 3969 if (strlen(path) > MAX_PATH - 1) { 3970 errno = ENAMETOOLONG; 3971 return -1; 3972 } 3973 hpi::native_path(strcpy(pathbuf, path)); 3974 return ::stat(pathbuf, sbuf); 3975} 3976 3977bool os::check_heap(bool force) { 3978 return true; 3979} 3980 3981int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 3982 return ::vsnprintf(buf, count, format, args); 3983} 3984 3985// Is a (classpath) directory empty? 3986bool os::dir_is_empty(const char* path) { 3987 DIR *dir = NULL; 3988 struct dirent *ptr; 3989 3990 dir = opendir(path); 3991 if (dir == NULL) return true; 3992 3993 /* Scan the directory */ 3994 bool result = true; 3995 char buf[sizeof(struct dirent) + MAX_PATH]; 3996 while (result && (ptr = ::readdir(dir)) != NULL) { 3997 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 3998 result = false; 3999 } 4000 } 4001 closedir(dir); 4002 return result; 4003} 4004 4005// create binary file, rewriting existing file if required 4006int os::create_binary_file(const char* path, bool rewrite_existing) { 4007 int oflags = O_WRONLY | O_CREAT; 4008 if (!rewrite_existing) { 4009 oflags |= O_EXCL; 4010 } 4011 return ::open64(path, oflags, S_IREAD | S_IWRITE); 4012} 4013 4014// return current position of file pointer 4015jlong os::current_file_offset(int fd) { 4016 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 4017} 4018 4019// move file pointer to the specified offset 4020jlong os::seek_to_file_offset(int fd, jlong offset) { 4021 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 4022} 4023 4024// Map a block of memory. 4025char* os::map_memory(int fd, const char* file_name, size_t file_offset, 4026 char *addr, size_t bytes, bool read_only, 4027 bool allow_exec) { 4028 int prot; 4029 int flags; 4030 4031 if (read_only) { 4032 prot = PROT_READ; 4033 flags = MAP_SHARED; 4034 } else { 4035 prot = PROT_READ | PROT_WRITE; 4036 flags = MAP_PRIVATE; 4037 } 4038 4039 if (allow_exec) { 4040 prot |= PROT_EXEC; 4041 } 4042 4043 if (addr != NULL) { 4044 flags |= MAP_FIXED; 4045 } 4046 4047 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 4048 fd, file_offset); 4049 if (mapped_address == MAP_FAILED) { 4050 return NULL; 4051 } 4052 return mapped_address; 4053} 4054 4055 4056// Remap a block of memory. 4057char* os::remap_memory(int fd, const char* file_name, size_t file_offset, 4058 char *addr, size_t bytes, bool read_only, 4059 bool allow_exec) { 4060 // same as map_memory() on this OS 4061 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 4062 allow_exec); 4063} 4064 4065 4066// Unmap a block of memory. 4067bool os::unmap_memory(char* addr, size_t bytes) { 4068 return munmap(addr, bytes) == 0; 4069} 4070 4071static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 4072 4073static clockid_t thread_cpu_clockid(Thread* thread) { 4074 pthread_t tid = thread->osthread()->pthread_id(); 4075 clockid_t clockid; 4076 4077 // Get thread clockid 4078 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 4079 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 4080 return clockid; 4081} 4082 4083// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 4084// are used by JVM M&M and JVMTI to get user+sys or user CPU time 4085// of a thread. 4086// 4087// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 4088// the fast estimate available on the platform. 4089 4090jlong os::current_thread_cpu_time() { 4091 if (os::Linux::supports_fast_thread_cpu_time()) { 4092 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4093 } else { 4094 // return user + sys since the cost is the same 4095 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 4096 } 4097} 4098 4099jlong os::thread_cpu_time(Thread* thread) { 4100 // consistent with what current_thread_cpu_time() returns 4101 if (os::Linux::supports_fast_thread_cpu_time()) { 4102 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4103 } else { 4104 return slow_thread_cpu_time(thread, true /* user + sys */); 4105 } 4106} 4107 4108jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 4109 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4110 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4111 } else { 4112 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 4113 } 4114} 4115 4116jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4117 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4118 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4119 } else { 4120 return slow_thread_cpu_time(thread, user_sys_cpu_time); 4121 } 4122} 4123 4124// 4125// -1 on error. 4126// 4127 4128static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4129 static bool proc_pid_cpu_avail = true; 4130 static bool proc_task_unchecked = true; 4131 static const char *proc_stat_path = "/proc/%d/stat"; 4132 pid_t tid = thread->osthread()->thread_id(); 4133 int i; 4134 char *s; 4135 char stat[2048]; 4136 int statlen; 4137 char proc_name[64]; 4138 int count; 4139 long sys_time, user_time; 4140 char string[64]; 4141 int idummy; 4142 long ldummy; 4143 FILE *fp; 4144 4145 // We first try accessing /proc/<pid>/cpu since this is faster to 4146 // process. If this file is not present (linux kernels 2.5 and above) 4147 // then we open /proc/<pid>/stat. 4148 if ( proc_pid_cpu_avail ) { 4149 sprintf(proc_name, "/proc/%d/cpu", tid); 4150 fp = fopen(proc_name, "r"); 4151 if ( fp != NULL ) { 4152 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time); 4153 fclose(fp); 4154 if ( count != 3 ) return -1; 4155 4156 if (user_sys_cpu_time) { 4157 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4158 } else { 4159 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4160 } 4161 } 4162 else proc_pid_cpu_avail = false; 4163 } 4164 4165 // The /proc/<tid>/stat aggregates per-process usage on 4166 // new Linux kernels 2.6+ where NPTL is supported. 4167 // The /proc/self/task/<tid>/stat still has the per-thread usage. 4168 // See bug 6328462. 4169 // There can be no directory /proc/self/task on kernels 2.4 with NPTL 4170 // and possibly in some other cases, so we check its availability. 4171 if (proc_task_unchecked && os::Linux::is_NPTL()) { 4172 // This is executed only once 4173 proc_task_unchecked = false; 4174 fp = fopen("/proc/self/task", "r"); 4175 if (fp != NULL) { 4176 proc_stat_path = "/proc/self/task/%d/stat"; 4177 fclose(fp); 4178 } 4179 } 4180 4181 sprintf(proc_name, proc_stat_path, tid); 4182 fp = fopen(proc_name, "r"); 4183 if ( fp == NULL ) return -1; 4184 statlen = fread(stat, 1, 2047, fp); 4185 stat[statlen] = '\0'; 4186 fclose(fp); 4187 4188 // Skip pid and the command string. Note that we could be dealing with 4189 // weird command names, e.g. user could decide to rename java launcher 4190 // to "java 1.4.2 :)", then the stat file would look like 4191 // 1234 (java 1.4.2 :)) R ... ... 4192 // We don't really need to know the command string, just find the last 4193 // occurrence of ")" and then start parsing from there. See bug 4726580. 4194 s = strrchr(stat, ')'); 4195 i = 0; 4196 if (s == NULL ) return -1; 4197 4198 // Skip blank chars 4199 do s++; while (isspace(*s)); 4200 4201 count = sscanf(s,"%*c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 4202 &idummy, &idummy, &idummy, &idummy, &idummy, 4203 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 4204 &user_time, &sys_time); 4205 if ( count != 12 ) return -1; 4206 if (user_sys_cpu_time) { 4207 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4208 } else { 4209 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4210 } 4211} 4212 4213void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4214 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4215 info_ptr->may_skip_backward = false; // elapsed time not wall time 4216 info_ptr->may_skip_forward = false; // elapsed time not wall time 4217 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4218} 4219 4220void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4221 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4222 info_ptr->may_skip_backward = false; // elapsed time not wall time 4223 info_ptr->may_skip_forward = false; // elapsed time not wall time 4224 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4225} 4226 4227bool os::is_thread_cpu_time_supported() { 4228 return true; 4229} 4230 4231// System loadavg support. Returns -1 if load average cannot be obtained. 4232// Linux doesn't yet have a (official) notion of processor sets, 4233// so just return the system wide load average. 4234int os::loadavg(double loadavg[], int nelem) { 4235 return ::getloadavg(loadavg, nelem); 4236} 4237 4238void os::pause() { 4239 char filename[MAX_PATH]; 4240 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 4241 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 4242 } else { 4243 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 4244 } 4245 4246 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 4247 if (fd != -1) { 4248 struct stat buf; 4249 close(fd); 4250 while (::stat(filename, &buf) == 0) { 4251 (void)::poll(NULL, 0, 100); 4252 } 4253 } else { 4254 jio_fprintf(stderr, 4255 "Could not open pause file '%s', continuing immediately.\n", filename); 4256 } 4257} 4258 4259extern "C" { 4260 4261/** 4262 * NOTE: the following code is to keep the green threads code 4263 * in the libjava.so happy. Once the green threads is removed, 4264 * these code will no longer be needed. 4265 */ 4266int 4267jdk_waitpid(pid_t pid, int* status, int options) { 4268 return waitpid(pid, status, options); 4269} 4270 4271int 4272fork1() { 4273 return fork(); 4274} 4275 4276int 4277jdk_sem_init(sem_t *sem, int pshared, unsigned int value) { 4278 return sem_init(sem, pshared, value); 4279} 4280 4281int 4282jdk_sem_post(sem_t *sem) { 4283 return sem_post(sem); 4284} 4285 4286int 4287jdk_sem_wait(sem_t *sem) { 4288 return sem_wait(sem); 4289} 4290 4291int 4292jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) { 4293 return pthread_sigmask(how , newmask, oldmask); 4294} 4295 4296} 4297 4298// Refer to the comments in os_solaris.cpp park-unpark. 4299// 4300// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 4301// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 4302// For specifics regarding the bug see GLIBC BUGID 261237 : 4303// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 4304// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 4305// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 4306// is used. (The simple C test-case provided in the GLIBC bug report manifests the 4307// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 4308// and monitorenter when we're using 1-0 locking. All those operations may result in 4309// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 4310// of libpthread avoids the problem, but isn't practical. 4311// 4312// Possible remedies: 4313// 4314// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 4315// This is palliative and probabilistic, however. If the thread is preempted 4316// between the call to compute_abstime() and pthread_cond_timedwait(), more 4317// than the minimum period may have passed, and the abstime may be stale (in the 4318// past) resultin in a hang. Using this technique reduces the odds of a hang 4319// but the JVM is still vulnerable, particularly on heavily loaded systems. 4320// 4321// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 4322// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 4323// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 4324// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 4325// thread. 4326// 4327// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 4328// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 4329// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 4330// This also works well. In fact it avoids kernel-level scalability impediments 4331// on certain platforms that don't handle lots of active pthread_cond_timedwait() 4332// timers in a graceful fashion. 4333// 4334// 4. When the abstime value is in the past it appears that control returns 4335// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 4336// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 4337// can avoid the problem by reinitializing the condvar -- by cond_destroy() 4338// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 4339// It may be possible to avoid reinitialization by checking the return 4340// value from pthread_cond_timedwait(). In addition to reinitializing the 4341// condvar we must establish the invariant that cond_signal() is only called 4342// within critical sections protected by the adjunct mutex. This prevents 4343// cond_signal() from "seeing" a condvar that's in the midst of being 4344// reinitialized or that is corrupt. Sadly, this invariant obviates the 4345// desirable signal-after-unlock optimization that avoids futile context switching. 4346// 4347// I'm also concerned that some versions of NTPL might allocate an auxilliary 4348// structure when a condvar is used or initialized. cond_destroy() would 4349// release the helper structure. Our reinitialize-after-timedwait fix 4350// put excessive stress on malloc/free and locks protecting the c-heap. 4351// 4352// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 4353// It may be possible to refine (4) by checking the kernel and NTPL verisons 4354// and only enabling the work-around for vulnerable environments. 4355 4356// utility to compute the abstime argument to timedwait: 4357// millis is the relative timeout time 4358// abstime will be the absolute timeout time 4359// TODO: replace compute_abstime() with unpackTime() 4360 4361static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 4362 if (millis < 0) millis = 0; 4363 struct timeval now; 4364 int status = gettimeofday(&now, NULL); 4365 assert(status == 0, "gettimeofday"); 4366 jlong seconds = millis / 1000; 4367 millis %= 1000; 4368 if (seconds > 50000000) { // see man cond_timedwait(3T) 4369 seconds = 50000000; 4370 } 4371 abstime->tv_sec = now.tv_sec + seconds; 4372 long usec = now.tv_usec + millis * 1000; 4373 if (usec >= 1000000) { 4374 abstime->tv_sec += 1; 4375 usec -= 1000000; 4376 } 4377 abstime->tv_nsec = usec * 1000; 4378 return abstime; 4379} 4380 4381 4382// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 4383// Conceptually TryPark() should be equivalent to park(0). 4384 4385int os::PlatformEvent::TryPark() { 4386 for (;;) { 4387 const int v = _Event ; 4388 guarantee ((v == 0) || (v == 1), "invariant") ; 4389 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 4390 } 4391} 4392 4393void os::PlatformEvent::park() { // AKA "down()" 4394 // Invariant: Only the thread associated with the Event/PlatformEvent 4395 // may call park(). 4396 // TODO: assert that _Assoc != NULL or _Assoc == Self 4397 int v ; 4398 for (;;) { 4399 v = _Event ; 4400 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4401 } 4402 guarantee (v >= 0, "invariant") ; 4403 if (v == 0) { 4404 // Do this the hard way by blocking ... 4405 int status = pthread_mutex_lock(_mutex); 4406 assert_status(status == 0, status, "mutex_lock"); 4407 guarantee (_nParked == 0, "invariant") ; 4408 ++ _nParked ; 4409 while (_Event < 0) { 4410 status = pthread_cond_wait(_cond, _mutex); 4411 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 4412 // Treat this the same as if the wait was interrupted 4413 if (status == ETIME) { status = EINTR; } 4414 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 4415 } 4416 -- _nParked ; 4417 4418 // In theory we could move the ST of 0 into _Event past the unlock(), 4419 // but then we'd need a MEMBAR after the ST. 4420 _Event = 0 ; 4421 status = pthread_mutex_unlock(_mutex); 4422 assert_status(status == 0, status, "mutex_unlock"); 4423 } 4424 guarantee (_Event >= 0, "invariant") ; 4425} 4426 4427int os::PlatformEvent::park(jlong millis) { 4428 guarantee (_nParked == 0, "invariant") ; 4429 4430 int v ; 4431 for (;;) { 4432 v = _Event ; 4433 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4434 } 4435 guarantee (v >= 0, "invariant") ; 4436 if (v != 0) return OS_OK ; 4437 4438 // We do this the hard way, by blocking the thread. 4439 // Consider enforcing a minimum timeout value. 4440 struct timespec abst; 4441 compute_abstime(&abst, millis); 4442 4443 int ret = OS_TIMEOUT; 4444 int status = pthread_mutex_lock(_mutex); 4445 assert_status(status == 0, status, "mutex_lock"); 4446 guarantee (_nParked == 0, "invariant") ; 4447 ++_nParked ; 4448 4449 // Object.wait(timo) will return because of 4450 // (a) notification 4451 // (b) timeout 4452 // (c) thread.interrupt 4453 // 4454 // Thread.interrupt and object.notify{All} both call Event::set. 4455 // That is, we treat thread.interrupt as a special case of notification. 4456 // The underlying Solaris implementation, cond_timedwait, admits 4457 // spurious/premature wakeups, but the JLS/JVM spec prevents the 4458 // JVM from making those visible to Java code. As such, we must 4459 // filter out spurious wakeups. We assume all ETIME returns are valid. 4460 // 4461 // TODO: properly differentiate simultaneous notify+interrupt. 4462 // In that case, we should propagate the notify to another waiter. 4463 4464 while (_Event < 0) { 4465 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 4466 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4467 pthread_cond_destroy (_cond); 4468 pthread_cond_init (_cond, NULL) ; 4469 } 4470 assert_status(status == 0 || status == EINTR || 4471 status == ETIME || status == ETIMEDOUT, 4472 status, "cond_timedwait"); 4473 if (!FilterSpuriousWakeups) break ; // previous semantics 4474 if (status == ETIME || status == ETIMEDOUT) break ; 4475 // We consume and ignore EINTR and spurious wakeups. 4476 } 4477 --_nParked ; 4478 if (_Event >= 0) { 4479 ret = OS_OK; 4480 } 4481 _Event = 0 ; 4482 status = pthread_mutex_unlock(_mutex); 4483 assert_status(status == 0, status, "mutex_unlock"); 4484 assert (_nParked == 0, "invariant") ; 4485 return ret; 4486} 4487 4488void os::PlatformEvent::unpark() { 4489 int v, AnyWaiters ; 4490 for (;;) { 4491 v = _Event ; 4492 if (v > 0) { 4493 // The LD of _Event could have reordered or be satisfied 4494 // by a read-aside from this processor's write buffer. 4495 // To avoid problems execute a barrier and then 4496 // ratify the value. 4497 OrderAccess::fence() ; 4498 if (_Event == v) return ; 4499 continue ; 4500 } 4501 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ; 4502 } 4503 if (v < 0) { 4504 // Wait for the thread associated with the event to vacate 4505 int status = pthread_mutex_lock(_mutex); 4506 assert_status(status == 0, status, "mutex_lock"); 4507 AnyWaiters = _nParked ; 4508 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ; 4509 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 4510 AnyWaiters = 0 ; 4511 pthread_cond_signal (_cond); 4512 } 4513 status = pthread_mutex_unlock(_mutex); 4514 assert_status(status == 0, status, "mutex_unlock"); 4515 if (AnyWaiters != 0) { 4516 status = pthread_cond_signal(_cond); 4517 assert_status(status == 0, status, "cond_signal"); 4518 } 4519 } 4520 4521 // Note that we signal() _after dropping the lock for "immortal" Events. 4522 // This is safe and avoids a common class of futile wakeups. In rare 4523 // circumstances this can cause a thread to return prematurely from 4524 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 4525 // simply re-test the condition and re-park itself. 4526} 4527 4528 4529// JSR166 4530// ------------------------------------------------------- 4531 4532/* 4533 * The solaris and linux implementations of park/unpark are fairly 4534 * conservative for now, but can be improved. They currently use a 4535 * mutex/condvar pair, plus a a count. 4536 * Park decrements count if > 0, else does a condvar wait. Unpark 4537 * sets count to 1 and signals condvar. Only one thread ever waits 4538 * on the condvar. Contention seen when trying to park implies that someone 4539 * is unparking you, so don't wait. And spurious returns are fine, so there 4540 * is no need to track notifications. 4541 */ 4542 4543 4544#define NANOSECS_PER_SEC 1000000000 4545#define NANOSECS_PER_MILLISEC 1000000 4546#define MAX_SECS 100000000 4547/* 4548 * This code is common to linux and solaris and will be moved to a 4549 * common place in dolphin. 4550 * 4551 * The passed in time value is either a relative time in nanoseconds 4552 * or an absolute time in milliseconds. Either way it has to be unpacked 4553 * into suitable seconds and nanoseconds components and stored in the 4554 * given timespec structure. 4555 * Given time is a 64-bit value and the time_t used in the timespec is only 4556 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 4557 * overflow if times way in the future are given. Further on Solaris versions 4558 * prior to 10 there is a restriction (see cond_timedwait) that the specified 4559 * number of seconds, in abstime, is less than current_time + 100,000,000. 4560 * As it will be 28 years before "now + 100000000" will overflow we can 4561 * ignore overflow and just impose a hard-limit on seconds using the value 4562 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 4563 * years from "now". 4564 */ 4565 4566static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 4567 assert (time > 0, "convertTime"); 4568 4569 struct timeval now; 4570 int status = gettimeofday(&now, NULL); 4571 assert(status == 0, "gettimeofday"); 4572 4573 time_t max_secs = now.tv_sec + MAX_SECS; 4574 4575 if (isAbsolute) { 4576 jlong secs = time / 1000; 4577 if (secs > max_secs) { 4578 absTime->tv_sec = max_secs; 4579 } 4580 else { 4581 absTime->tv_sec = secs; 4582 } 4583 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 4584 } 4585 else { 4586 jlong secs = time / NANOSECS_PER_SEC; 4587 if (secs >= MAX_SECS) { 4588 absTime->tv_sec = max_secs; 4589 absTime->tv_nsec = 0; 4590 } 4591 else { 4592 absTime->tv_sec = now.tv_sec + secs; 4593 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 4594 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 4595 absTime->tv_nsec -= NANOSECS_PER_SEC; 4596 ++absTime->tv_sec; // note: this must be <= max_secs 4597 } 4598 } 4599 } 4600 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 4601 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 4602 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 4603 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 4604} 4605 4606void Parker::park(bool isAbsolute, jlong time) { 4607 // Optional fast-path check: 4608 // Return immediately if a permit is available. 4609 if (_counter > 0) { 4610 _counter = 0 ; 4611 return ; 4612 } 4613 4614 Thread* thread = Thread::current(); 4615 assert(thread->is_Java_thread(), "Must be JavaThread"); 4616 JavaThread *jt = (JavaThread *)thread; 4617 4618 // Optional optimization -- avoid state transitions if there's an interrupt pending. 4619 // Check interrupt before trying to wait 4620 if (Thread::is_interrupted(thread, false)) { 4621 return; 4622 } 4623 4624 // Next, demultiplex/decode time arguments 4625 timespec absTime; 4626 if (time < 0) { // don't wait at all 4627 return; 4628 } 4629 if (time > 0) { 4630 unpackTime(&absTime, isAbsolute, time); 4631 } 4632 4633 4634 // Enter safepoint region 4635 // Beware of deadlocks such as 6317397. 4636 // The per-thread Parker:: mutex is a classic leaf-lock. 4637 // In particular a thread must never block on the Threads_lock while 4638 // holding the Parker:: mutex. If safepoints are pending both the 4639 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 4640 ThreadBlockInVM tbivm(jt); 4641 4642 // Don't wait if cannot get lock since interference arises from 4643 // unblocking. Also. check interrupt before trying wait 4644 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 4645 return; 4646 } 4647 4648 int status ; 4649 if (_counter > 0) { // no wait needed 4650 _counter = 0; 4651 status = pthread_mutex_unlock(_mutex); 4652 assert (status == 0, "invariant") ; 4653 return; 4654 } 4655 4656#ifdef ASSERT 4657 // Don't catch signals while blocked; let the running threads have the signals. 4658 // (This allows a debugger to break into the running thread.) 4659 sigset_t oldsigs; 4660 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 4661 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 4662#endif 4663 4664 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 4665 jt->set_suspend_equivalent(); 4666 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 4667 4668 if (time == 0) { 4669 status = pthread_cond_wait (_cond, _mutex) ; 4670 } else { 4671 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 4672 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4673 pthread_cond_destroy (_cond) ; 4674 pthread_cond_init (_cond, NULL); 4675 } 4676 } 4677 assert_status(status == 0 || status == EINTR || 4678 status == ETIME || status == ETIMEDOUT, 4679 status, "cond_timedwait"); 4680 4681#ifdef ASSERT 4682 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 4683#endif 4684 4685 _counter = 0 ; 4686 status = pthread_mutex_unlock(_mutex) ; 4687 assert_status(status == 0, status, "invariant") ; 4688 // If externally suspended while waiting, re-suspend 4689 if (jt->handle_special_suspend_equivalent_condition()) { 4690 jt->java_suspend_self(); 4691 } 4692 4693} 4694 4695void Parker::unpark() { 4696 int s, status ; 4697 status = pthread_mutex_lock(_mutex); 4698 assert (status == 0, "invariant") ; 4699 s = _counter; 4700 _counter = 1; 4701 if (s < 1) { 4702 if (WorkAroundNPTLTimedWaitHang) { 4703 status = pthread_cond_signal (_cond) ; 4704 assert (status == 0, "invariant") ; 4705 status = pthread_mutex_unlock(_mutex); 4706 assert (status == 0, "invariant") ; 4707 } else { 4708 status = pthread_mutex_unlock(_mutex); 4709 assert (status == 0, "invariant") ; 4710 status = pthread_cond_signal (_cond) ; 4711 assert (status == 0, "invariant") ; 4712 } 4713 } else { 4714 pthread_mutex_unlock(_mutex); 4715 assert (status == 0, "invariant") ; 4716 } 4717} 4718 4719 4720extern char** environ; 4721 4722#ifndef __NR_fork 4723#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 4724#endif 4725 4726#ifndef __NR_execve 4727#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 4728#endif 4729 4730// Run the specified command in a separate process. Return its exit value, 4731// or -1 on failure (e.g. can't fork a new process). 4732// Unlike system(), this function can be called from signal handler. It 4733// doesn't block SIGINT et al. 4734int os::fork_and_exec(char* cmd) { 4735 const char * argv[4] = {"sh", "-c", cmd, NULL}; 4736 4737 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 4738 // pthread_atfork handlers and reset pthread library. All we need is a 4739 // separate process to execve. Make a direct syscall to fork process. 4740 // On IA64 there's no fork syscall, we have to use fork() and hope for 4741 // the best... 4742 pid_t pid = NOT_IA64(syscall(__NR_fork);) 4743 IA64_ONLY(fork();) 4744 4745 if (pid < 0) { 4746 // fork failed 4747 return -1; 4748 4749 } else if (pid == 0) { 4750 // child process 4751 4752 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 4753 // first to kill every thread on the thread list. Because this list is 4754 // not reset by fork() (see notes above), execve() will instead kill 4755 // every thread in the parent process. We know this is the only thread 4756 // in the new process, so make a system call directly. 4757 // IA64 should use normal execve() from glibc to match the glibc fork() 4758 // above. 4759 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 4760 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 4761 4762 // execve failed 4763 _exit(-1); 4764 4765 } else { 4766 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 4767 // care about the actual exit code, for now. 4768 4769 int status; 4770 4771 // Wait for the child process to exit. This returns immediately if 4772 // the child has already exited. */ 4773 while (waitpid(pid, &status, 0) < 0) { 4774 switch (errno) { 4775 case ECHILD: return 0; 4776 case EINTR: break; 4777 default: return -1; 4778 } 4779 } 4780 4781 if (WIFEXITED(status)) { 4782 // The child exited normally; get its exit code. 4783 return WEXITSTATUS(status); 4784 } else if (WIFSIGNALED(status)) { 4785 // The child exited because of a signal 4786 // The best value to return is 0x80 + signal number, 4787 // because that is what all Unix shells do, and because 4788 // it allows callers to distinguish between process exit and 4789 // process death by signal. 4790 return 0x80 + WTERMSIG(status); 4791 } else { 4792 // Unknown exit code; pass it through 4793 return status; 4794 } 4795 } 4796} 4797