os_linux.cpp revision 235:9c2ecc2ffb12
1/* 2 * Copyright 1999-2008 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//////////////////////////////////////////////////////////////////////////////// 98// utility functions 99 100static int SR_initialize(); 101static int SR_finalize(); 102 103julong os::available_memory() { 104 return Linux::available_memory(); 105} 106 107julong os::Linux::available_memory() { 108 // values in struct sysinfo are "unsigned long" 109 struct sysinfo si; 110 sysinfo(&si); 111 112 return (julong)si.freeram * si.mem_unit; 113} 114 115julong os::physical_memory() { 116 return Linux::physical_memory(); 117} 118 119julong os::allocatable_physical_memory(julong size) { 120#ifdef _LP64 121 return size; 122#else 123 julong result = MIN2(size, (julong)3800*M); 124 if (!is_allocatable(result)) { 125 // See comments under solaris for alignment considerations 126 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); 127 result = MIN2(size, reasonable_size); 128 } 129 return result; 130#endif // _LP64 131} 132 133//////////////////////////////////////////////////////////////////////////////// 134// environment support 135 136bool os::getenv(const char* name, char* buf, int len) { 137 const char* val = ::getenv(name); 138 if (val != NULL && strlen(val) < (size_t)len) { 139 strcpy(buf, val); 140 return true; 141 } 142 if (len > 0) buf[0] = 0; // return a null string 143 return false; 144} 145 146 147// Return true if user is running as root. 148 149bool os::have_special_privileges() { 150 static bool init = false; 151 static bool privileges = false; 152 if (!init) { 153 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 154 init = true; 155 } 156 return privileges; 157} 158 159 160#ifndef SYS_gettid 161// i386: 224, ia64: 1105, amd64: 186, sparc 143 162#ifdef __ia64__ 163#define SYS_gettid 1105 164#elif __i386__ 165#define SYS_gettid 224 166#elif __amd64__ 167#define SYS_gettid 186 168#elif __sparc__ 169#define SYS_gettid 143 170#else 171#error define gettid for the arch 172#endif 173#endif 174 175// Cpu architecture string 176#if defined(IA64) 177static char cpu_arch[] = "ia64"; 178#elif defined(IA32) 179static char cpu_arch[] = "i386"; 180#elif defined(AMD64) 181static char cpu_arch[] = "amd64"; 182#elif defined(SPARC) 183# ifdef _LP64 184static char cpu_arch[] = "sparcv9"; 185# else 186static char cpu_arch[] = "sparc"; 187# endif 188#else 189#error Add appropriate cpu_arch setting 190#endif 191 192 193// pid_t gettid() 194// 195// Returns the kernel thread id of the currently running thread. Kernel 196// thread id is used to access /proc. 197// 198// (Note that getpid() on LinuxThreads returns kernel thread id too; but 199// on NPTL, it returns the same pid for all threads, as required by POSIX.) 200// 201pid_t os::Linux::gettid() { 202 int rslt = syscall(SYS_gettid); 203 if (rslt == -1) { 204 // old kernel, no NPTL support 205 return getpid(); 206 } else { 207 return (pid_t)rslt; 208 } 209} 210 211// Most versions of linux have a bug where the number of processors are 212// determined by looking at the /proc file system. In a chroot environment, 213// the system call returns 1. This causes the VM to act as if it is 214// a single processor and elide locking (see is_MP() call). 215static bool unsafe_chroot_detected = false; 216static const char *unstable_chroot_error = "/proc file system not found.\n" 217 "Java may be unstable running multithreaded in a chroot " 218 "environment on Linux when /proc filesystem is not mounted."; 219 220void os::Linux::initialize_system_info() { 221 _processor_count = sysconf(_SC_NPROCESSORS_CONF); 222 if (_processor_count == 1) { 223 pid_t pid = os::Linux::gettid(); 224 char fname[32]; 225 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 226 FILE *fp = fopen(fname, "r"); 227 if (fp == NULL) { 228 unsafe_chroot_detected = true; 229 } else { 230 fclose(fp); 231 } 232 } 233 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 234 assert(_processor_count > 0, "linux error"); 235} 236 237void os::init_system_properties_values() { 238// char arch[12]; 239// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 240 241 // The next steps are taken in the product version: 242 // 243 // Obtain the JAVA_HOME value from the location of libjvm[_g].so. 244 // This library should be located at: 245 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so. 246 // 247 // If "/jre/lib/" appears at the right place in the path, then we 248 // assume libjvm[_g].so is installed in a JDK and we use this path. 249 // 250 // Otherwise exit with message: "Could not create the Java virtual machine." 251 // 252 // The following extra steps are taken in the debugging version: 253 // 254 // If "/jre/lib/" does NOT appear at the right place in the path 255 // instead of exit check for $JAVA_HOME environment variable. 256 // 257 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 258 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so 259 // it looks like libjvm[_g].so is installed there 260 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so. 261 // 262 // Otherwise exit. 263 // 264 // Important note: if the location of libjvm.so changes this 265 // code needs to be changed accordingly. 266 267 // The next few definitions allow the code to be verbatim: 268#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n)) 269#define getenv(n) ::getenv(n) 270 271/* 272 * See ld(1): 273 * The linker uses the following search paths to locate required 274 * shared libraries: 275 * 1: ... 276 * ... 277 * 7: The default directories, normally /lib and /usr/lib. 278 */ 279#define DEFAULT_LIBPATH "/lib:/usr/lib" 280 281#define EXTENSIONS_DIR "/lib/ext" 282#define ENDORSED_DIR "/lib/endorsed" 283#define REG_DIR "/usr/java/packages" 284 285 { 286 /* sysclasspath, java_home, dll_dir */ 287 { 288 char *home_path; 289 char *dll_path; 290 char *pslash; 291 char buf[MAXPATHLEN]; 292 os::jvm_path(buf, sizeof(buf)); 293 294 // Found the full path to libjvm.so. 295 // Now cut the path to <java_home>/jre if we can. 296 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 297 pslash = strrchr(buf, '/'); 298 if (pslash != NULL) 299 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 300 dll_path = malloc(strlen(buf) + 1); 301 if (dll_path == NULL) 302 return; 303 strcpy(dll_path, buf); 304 Arguments::set_dll_dir(dll_path); 305 306 if (pslash != NULL) { 307 pslash = strrchr(buf, '/'); 308 if (pslash != NULL) { 309 *pslash = '\0'; /* get rid of /<arch> */ 310 pslash = strrchr(buf, '/'); 311 if (pslash != NULL) 312 *pslash = '\0'; /* get rid of /lib */ 313 } 314 } 315 316 home_path = malloc(strlen(buf) + 1); 317 if (home_path == NULL) 318 return; 319 strcpy(home_path, buf); 320 Arguments::set_java_home(home_path); 321 322 if (!set_boot_path('/', ':')) 323 return; 324 } 325 326 /* 327 * Where to look for native libraries 328 * 329 * Note: Due to a legacy implementation, most of the library path 330 * is set in the launcher. This was to accomodate linking restrictions 331 * on legacy Linux implementations (which are no longer supported). 332 * Eventually, all the library path setting will be done here. 333 * 334 * However, to prevent the proliferation of improperly built native 335 * libraries, the new path component /usr/java/packages is added here. 336 * Eventually, all the library path setting will be done here. 337 */ 338 { 339 char *ld_library_path; 340 341 /* 342 * Construct the invariant part of ld_library_path. Note that the 343 * space for the colon and the trailing null are provided by the 344 * nulls included by the sizeof operator (so actually we allocate 345 * a byte more than necessary). 346 */ 347 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 348 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 349 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 350 351 /* 352 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 353 * should always exist (until the legacy problem cited above is 354 * addressed). 355 */ 356 char *v = getenv("LD_LIBRARY_PATH"); 357 if (v != NULL) { 358 char *t = ld_library_path; 359 /* That's +1 for the colon and +1 for the trailing '\0' */ 360 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 361 sprintf(ld_library_path, "%s:%s", v, t); 362 } 363 Arguments::set_library_path(ld_library_path); 364 } 365 366 /* 367 * Extensions directories. 368 * 369 * Note that the space for the colon and the trailing null are provided 370 * by the nulls included by the sizeof operator (so actually one byte more 371 * than necessary is allocated). 372 */ 373 { 374 char *buf = malloc(strlen(Arguments::get_java_home()) + 375 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 376 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 377 Arguments::get_java_home()); 378 Arguments::set_ext_dirs(buf); 379 } 380 381 /* Endorsed standards default directory. */ 382 { 383 char * buf; 384 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 385 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 386 Arguments::set_endorsed_dirs(buf); 387 } 388 } 389 390#undef malloc 391#undef getenv 392#undef EXTENSIONS_DIR 393#undef ENDORSED_DIR 394 395 // Done 396 return; 397} 398 399//////////////////////////////////////////////////////////////////////////////// 400// breakpoint support 401 402void os::breakpoint() { 403 BREAKPOINT; 404} 405 406extern "C" void breakpoint() { 407 // use debugger to set breakpoint here 408} 409 410//////////////////////////////////////////////////////////////////////////////// 411// signal support 412 413debug_only(static bool signal_sets_initialized = false); 414static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 415 416bool os::Linux::is_sig_ignored(int sig) { 417 struct sigaction oact; 418 sigaction(sig, (struct sigaction*)NULL, &oact); 419 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 420 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 421 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 422 return true; 423 else 424 return false; 425} 426 427void os::Linux::signal_sets_init() { 428 // Should also have an assertion stating we are still single-threaded. 429 assert(!signal_sets_initialized, "Already initialized"); 430 // Fill in signals that are necessarily unblocked for all threads in 431 // the VM. Currently, we unblock the following signals: 432 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 433 // by -Xrs (=ReduceSignalUsage)); 434 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 435 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 436 // the dispositions or masks wrt these signals. 437 // Programs embedding the VM that want to use the above signals for their 438 // own purposes must, at this time, use the "-Xrs" option to prevent 439 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 440 // (See bug 4345157, and other related bugs). 441 // In reality, though, unblocking these signals is really a nop, since 442 // these signals are not blocked by default. 443 sigemptyset(&unblocked_sigs); 444 sigemptyset(&allowdebug_blocked_sigs); 445 sigaddset(&unblocked_sigs, SIGILL); 446 sigaddset(&unblocked_sigs, SIGSEGV); 447 sigaddset(&unblocked_sigs, SIGBUS); 448 sigaddset(&unblocked_sigs, SIGFPE); 449 sigaddset(&unblocked_sigs, SR_signum); 450 451 if (!ReduceSignalUsage) { 452 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 453 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 454 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 455 } 456 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 457 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 458 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 459 } 460 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 461 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 462 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 463 } 464 } 465 // Fill in signals that are blocked by all but the VM thread. 466 sigemptyset(&vm_sigs); 467 if (!ReduceSignalUsage) 468 sigaddset(&vm_sigs, BREAK_SIGNAL); 469 debug_only(signal_sets_initialized = true); 470 471} 472 473// These are signals that are unblocked while a thread is running Java. 474// (For some reason, they get blocked by default.) 475sigset_t* os::Linux::unblocked_signals() { 476 assert(signal_sets_initialized, "Not initialized"); 477 return &unblocked_sigs; 478} 479 480// These are the signals that are blocked while a (non-VM) thread is 481// running Java. Only the VM thread handles these signals. 482sigset_t* os::Linux::vm_signals() { 483 assert(signal_sets_initialized, "Not initialized"); 484 return &vm_sigs; 485} 486 487// These are signals that are blocked during cond_wait to allow debugger in 488sigset_t* os::Linux::allowdebug_blocked_signals() { 489 assert(signal_sets_initialized, "Not initialized"); 490 return &allowdebug_blocked_sigs; 491} 492 493void os::Linux::hotspot_sigmask(Thread* thread) { 494 495 //Save caller's signal mask before setting VM signal mask 496 sigset_t caller_sigmask; 497 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 498 499 OSThread* osthread = thread->osthread(); 500 osthread->set_caller_sigmask(caller_sigmask); 501 502 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 503 504 if (!ReduceSignalUsage) { 505 if (thread->is_VM_thread()) { 506 // Only the VM thread handles BREAK_SIGNAL ... 507 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 508 } else { 509 // ... all other threads block BREAK_SIGNAL 510 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 511 } 512 } 513} 514 515////////////////////////////////////////////////////////////////////////////// 516// detecting pthread library 517 518void os::Linux::libpthread_init() { 519 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 520 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 521 // generic name for earlier versions. 522 // Define macros here so we can build HotSpot on old systems. 523# ifndef _CS_GNU_LIBC_VERSION 524# define _CS_GNU_LIBC_VERSION 2 525# endif 526# ifndef _CS_GNU_LIBPTHREAD_VERSION 527# define _CS_GNU_LIBPTHREAD_VERSION 3 528# endif 529 530 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 531 if (n > 0) { 532 char *str = (char *)malloc(n); 533 confstr(_CS_GNU_LIBC_VERSION, str, n); 534 os::Linux::set_glibc_version(str); 535 } else { 536 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 537 static char _gnu_libc_version[32]; 538 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 539 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 540 os::Linux::set_glibc_version(_gnu_libc_version); 541 } 542 543 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 544 if (n > 0) { 545 char *str = (char *)malloc(n); 546 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 547 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 548 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 549 // is the case. LinuxThreads has a hard limit on max number of threads. 550 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 551 // On the other hand, NPTL does not have such a limit, sysconf() 552 // will return -1 and errno is not changed. Check if it is really NPTL. 553 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 554 strstr(str, "NPTL") && 555 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 556 free(str); 557 os::Linux::set_libpthread_version("linuxthreads"); 558 } else { 559 os::Linux::set_libpthread_version(str); 560 } 561 } else { 562 // glibc before 2.3.2 only has LinuxThreads. 563 os::Linux::set_libpthread_version("linuxthreads"); 564 } 565 566 if (strstr(libpthread_version(), "NPTL")) { 567 os::Linux::set_is_NPTL(); 568 } else { 569 os::Linux::set_is_LinuxThreads(); 570 } 571 572 // LinuxThreads have two flavors: floating-stack mode, which allows variable 573 // stack size; and fixed-stack mode. NPTL is always floating-stack. 574 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 575 os::Linux::set_is_floating_stack(); 576 } 577} 578 579///////////////////////////////////////////////////////////////////////////// 580// thread stack 581 582// Force Linux kernel to expand current thread stack. If "bottom" is close 583// to the stack guard, caller should block all signals. 584// 585// MAP_GROWSDOWN: 586// A special mmap() flag that is used to implement thread stacks. It tells 587// kernel that the memory region should extend downwards when needed. This 588// allows early versions of LinuxThreads to only mmap the first few pages 589// when creating a new thread. Linux kernel will automatically expand thread 590// stack as needed (on page faults). 591// 592// However, because the memory region of a MAP_GROWSDOWN stack can grow on 593// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 594// region, it's hard to tell if the fault is due to a legitimate stack 595// access or because of reading/writing non-exist memory (e.g. buffer 596// overrun). As a rule, if the fault happens below current stack pointer, 597// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 598// application (see Linux kernel fault.c). 599// 600// This Linux feature can cause SIGSEGV when VM bangs thread stack for 601// stack overflow detection. 602// 603// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 604// not use this flag. However, the stack of initial thread is not created 605// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 606// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 607// and then attach the thread to JVM. 608// 609// To get around the problem and allow stack banging on Linux, we need to 610// manually expand thread stack after receiving the SIGSEGV. 611// 612// There are two ways to expand thread stack to address "bottom", we used 613// both of them in JVM before 1.5: 614// 1. adjust stack pointer first so that it is below "bottom", and then 615// touch "bottom" 616// 2. mmap() the page in question 617// 618// Now alternate signal stack is gone, it's harder to use 2. For instance, 619// if current sp is already near the lower end of page 101, and we need to 620// call mmap() to map page 100, it is possible that part of the mmap() frame 621// will be placed in page 100. When page 100 is mapped, it is zero-filled. 622// That will destroy the mmap() frame and cause VM to crash. 623// 624// The following code works by adjusting sp first, then accessing the "bottom" 625// page to force a page fault. Linux kernel will then automatically expand the 626// stack mapping. 627// 628// _expand_stack_to() assumes its frame size is less than page size, which 629// should always be true if the function is not inlined. 630 631#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 632#define NOINLINE 633#else 634#define NOINLINE __attribute__ ((noinline)) 635#endif 636 637static void _expand_stack_to(address bottom) NOINLINE; 638 639static void _expand_stack_to(address bottom) { 640 address sp; 641 size_t size; 642 volatile char *p; 643 644 // Adjust bottom to point to the largest address within the same page, it 645 // gives us a one-page buffer if alloca() allocates slightly more memory. 646 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 647 bottom += os::Linux::page_size() - 1; 648 649 // sp might be slightly above current stack pointer; if that's the case, we 650 // will alloca() a little more space than necessary, which is OK. Don't use 651 // os::current_stack_pointer(), as its result can be slightly below current 652 // stack pointer, causing us to not alloca enough to reach "bottom". 653 sp = (address)&sp; 654 655 if (sp > bottom) { 656 size = sp - bottom; 657 p = (volatile char *)alloca(size); 658 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 659 p[0] = '\0'; 660 } 661} 662 663bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 664 assert(t!=NULL, "just checking"); 665 assert(t->osthread()->expanding_stack(), "expand should be set"); 666 assert(t->stack_base() != NULL, "stack_base was not initialized"); 667 668 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 669 sigset_t mask_all, old_sigset; 670 sigfillset(&mask_all); 671 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 672 _expand_stack_to(addr); 673 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 674 return true; 675 } 676 return false; 677} 678 679////////////////////////////////////////////////////////////////////////////// 680// create new thread 681 682static address highest_vm_reserved_address(); 683 684// check if it's safe to start a new thread 685static bool _thread_safety_check(Thread* thread) { 686 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 687 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 688 // Heap is mmap'ed at lower end of memory space. Thread stacks are 689 // allocated (MAP_FIXED) from high address space. Every thread stack 690 // occupies a fixed size slot (usually 2Mbytes, but user can change 691 // it to other values if they rebuild LinuxThreads). 692 // 693 // Problem with MAP_FIXED is that mmap() can still succeed even part of 694 // the memory region has already been mmap'ed. That means if we have too 695 // many threads and/or very large heap, eventually thread stack will 696 // collide with heap. 697 // 698 // Here we try to prevent heap/stack collision by comparing current 699 // stack bottom with the highest address that has been mmap'ed by JVM 700 // plus a safety margin for memory maps created by native code. 701 // 702 // This feature can be disabled by setting ThreadSafetyMargin to 0 703 // 704 if (ThreadSafetyMargin > 0) { 705 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 706 707 // not safe if our stack extends below the safety margin 708 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 709 } else { 710 return true; 711 } 712 } else { 713 // Floating stack LinuxThreads or NPTL: 714 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 715 // there's not enough space left, pthread_create() will fail. If we come 716 // here, that means enough space has been reserved for stack. 717 return true; 718 } 719} 720 721// Thread start routine for all newly created threads 722static void *java_start(Thread *thread) { 723 // Try to randomize the cache line index of hot stack frames. 724 // This helps when threads of the same stack traces evict each other's 725 // cache lines. The threads can be either from the same JVM instance, or 726 // from different JVM instances. The benefit is especially true for 727 // processors with hyperthreading technology. 728 static int counter = 0; 729 int pid = os::current_process_id(); 730 alloca(((pid ^ counter++) & 7) * 128); 731 732 ThreadLocalStorage::set_thread(thread); 733 734 OSThread* osthread = thread->osthread(); 735 Monitor* sync = osthread->startThread_lock(); 736 737 // non floating stack LinuxThreads needs extra check, see above 738 if (!_thread_safety_check(thread)) { 739 // notify parent thread 740 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 741 osthread->set_state(ZOMBIE); 742 sync->notify_all(); 743 return NULL; 744 } 745 746 // thread_id is kernel thread id (similar to Solaris LWP id) 747 osthread->set_thread_id(os::Linux::gettid()); 748 749 if (UseNUMA) { 750 int lgrp_id = os::numa_get_group_id(); 751 if (lgrp_id != -1) { 752 thread->set_lgrp_id(lgrp_id); 753 } 754 } 755 // initialize signal mask for this thread 756 os::Linux::hotspot_sigmask(thread); 757 758 // initialize floating point control register 759 os::Linux::init_thread_fpu_state(); 760 761 // handshaking with parent thread 762 { 763 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 764 765 // notify parent thread 766 osthread->set_state(INITIALIZED); 767 sync->notify_all(); 768 769 // wait until os::start_thread() 770 while (osthread->get_state() == INITIALIZED) { 771 sync->wait(Mutex::_no_safepoint_check_flag); 772 } 773 } 774 775 // call one more level start routine 776 thread->run(); 777 778 return 0; 779} 780 781bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 782 assert(thread->osthread() == NULL, "caller responsible"); 783 784 // Allocate the OSThread object 785 OSThread* osthread = new OSThread(NULL, NULL); 786 if (osthread == NULL) { 787 return false; 788 } 789 790 // set the correct thread state 791 osthread->set_thread_type(thr_type); 792 793 // Initial state is ALLOCATED but not INITIALIZED 794 osthread->set_state(ALLOCATED); 795 796 thread->set_osthread(osthread); 797 798 // init thread attributes 799 pthread_attr_t attr; 800 pthread_attr_init(&attr); 801 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 802 803 // stack size 804 if (os::Linux::supports_variable_stack_size()) { 805 // calculate stack size if it's not specified by caller 806 if (stack_size == 0) { 807 stack_size = os::Linux::default_stack_size(thr_type); 808 809 switch (thr_type) { 810 case os::java_thread: 811 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss 812 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create(); 813 break; 814 case os::compiler_thread: 815 if (CompilerThreadStackSize > 0) { 816 stack_size = (size_t)(CompilerThreadStackSize * K); 817 break; 818 } // else fall through: 819 // use VMThreadStackSize if CompilerThreadStackSize is not defined 820 case os::vm_thread: 821 case os::pgc_thread: 822 case os::cgc_thread: 823 case os::watcher_thread: 824 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 825 break; 826 } 827 } 828 829 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 830 pthread_attr_setstacksize(&attr, stack_size); 831 } else { 832 // let pthread_create() pick the default value. 833 } 834 835 // glibc guard page 836 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 837 838 ThreadState state; 839 840 { 841 // Serialize thread creation if we are running with fixed stack LinuxThreads 842 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 843 if (lock) { 844 os::Linux::createThread_lock()->lock_without_safepoint_check(); 845 } 846 847 pthread_t tid; 848 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 849 850 pthread_attr_destroy(&attr); 851 852 if (ret != 0) { 853 if (PrintMiscellaneous && (Verbose || WizardMode)) { 854 perror("pthread_create()"); 855 } 856 // Need to clean up stuff we've allocated so far 857 thread->set_osthread(NULL); 858 delete osthread; 859 if (lock) os::Linux::createThread_lock()->unlock(); 860 return false; 861 } 862 863 // Store pthread info into the OSThread 864 osthread->set_pthread_id(tid); 865 866 // Wait until child thread is either initialized or aborted 867 { 868 Monitor* sync_with_child = osthread->startThread_lock(); 869 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 870 while ((state = osthread->get_state()) == ALLOCATED) { 871 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 872 } 873 } 874 875 if (lock) { 876 os::Linux::createThread_lock()->unlock(); 877 } 878 } 879 880 // Aborted due to thread limit being reached 881 if (state == ZOMBIE) { 882 thread->set_osthread(NULL); 883 delete osthread; 884 return false; 885 } 886 887 // The thread is returned suspended (in state INITIALIZED), 888 // and is started higher up in the call chain 889 assert(state == INITIALIZED, "race condition"); 890 return true; 891} 892 893///////////////////////////////////////////////////////////////////////////// 894// attach existing thread 895 896// bootstrap the main thread 897bool os::create_main_thread(JavaThread* thread) { 898 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 899 return create_attached_thread(thread); 900} 901 902bool os::create_attached_thread(JavaThread* thread) { 903#ifdef ASSERT 904 thread->verify_not_published(); 905#endif 906 907 // Allocate the OSThread object 908 OSThread* osthread = new OSThread(NULL, NULL); 909 910 if (osthread == NULL) { 911 return false; 912 } 913 914 // Store pthread info into the OSThread 915 osthread->set_thread_id(os::Linux::gettid()); 916 osthread->set_pthread_id(::pthread_self()); 917 918 // initialize floating point control register 919 os::Linux::init_thread_fpu_state(); 920 921 // Initial thread state is RUNNABLE 922 osthread->set_state(RUNNABLE); 923 924 thread->set_osthread(osthread); 925 926 if (UseNUMA) { 927 int lgrp_id = os::numa_get_group_id(); 928 if (lgrp_id != -1) { 929 thread->set_lgrp_id(lgrp_id); 930 } 931 } 932 933 if (os::Linux::is_initial_thread()) { 934 // If current thread is initial thread, its stack is mapped on demand, 935 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 936 // the entire stack region to avoid SEGV in stack banging. 937 // It is also useful to get around the heap-stack-gap problem on SuSE 938 // kernel (see 4821821 for details). We first expand stack to the top 939 // of yellow zone, then enable stack yellow zone (order is significant, 940 // enabling yellow zone first will crash JVM on SuSE Linux), so there 941 // is no gap between the last two virtual memory regions. 942 943 JavaThread *jt = (JavaThread *)thread; 944 address addr = jt->stack_yellow_zone_base(); 945 assert(addr != NULL, "initialization problem?"); 946 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 947 948 osthread->set_expanding_stack(); 949 os::Linux::manually_expand_stack(jt, addr); 950 osthread->clear_expanding_stack(); 951 } 952 953 // initialize signal mask for this thread 954 // and save the caller's signal mask 955 os::Linux::hotspot_sigmask(thread); 956 957 return true; 958} 959 960void os::pd_start_thread(Thread* thread) { 961 OSThread * osthread = thread->osthread(); 962 assert(osthread->get_state() != INITIALIZED, "just checking"); 963 Monitor* sync_with_child = osthread->startThread_lock(); 964 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 965 sync_with_child->notify(); 966} 967 968// Free Linux resources related to the OSThread 969void os::free_thread(OSThread* osthread) { 970 assert(osthread != NULL, "osthread not set"); 971 972 if (Thread::current()->osthread() == osthread) { 973 // Restore caller's signal mask 974 sigset_t sigmask = osthread->caller_sigmask(); 975 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 976 } 977 978 delete osthread; 979} 980 981////////////////////////////////////////////////////////////////////////////// 982// thread local storage 983 984int os::allocate_thread_local_storage() { 985 pthread_key_t key; 986 int rslt = pthread_key_create(&key, NULL); 987 assert(rslt == 0, "cannot allocate thread local storage"); 988 return (int)key; 989} 990 991// Note: This is currently not used by VM, as we don't destroy TLS key 992// on VM exit. 993void os::free_thread_local_storage(int index) { 994 int rslt = pthread_key_delete((pthread_key_t)index); 995 assert(rslt == 0, "invalid index"); 996} 997 998void os::thread_local_storage_at_put(int index, void* value) { 999 int rslt = pthread_setspecific((pthread_key_t)index, value); 1000 assert(rslt == 0, "pthread_setspecific failed"); 1001} 1002 1003extern "C" Thread* get_thread() { 1004 return ThreadLocalStorage::thread(); 1005} 1006 1007////////////////////////////////////////////////////////////////////////////// 1008// initial thread 1009 1010// Check if current thread is the initial thread, similar to Solaris thr_main. 1011bool os::Linux::is_initial_thread(void) { 1012 char dummy; 1013 // If called before init complete, thread stack bottom will be null. 1014 // Can be called if fatal error occurs before initialization. 1015 if (initial_thread_stack_bottom() == NULL) return false; 1016 assert(initial_thread_stack_bottom() != NULL && 1017 initial_thread_stack_size() != 0, 1018 "os::init did not locate initial thread's stack region"); 1019 if ((address)&dummy >= initial_thread_stack_bottom() && 1020 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1021 return true; 1022 else return false; 1023} 1024 1025// Find the virtual memory area that contains addr 1026static bool find_vma(address addr, address* vma_low, address* vma_high) { 1027 FILE *fp = fopen("/proc/self/maps", "r"); 1028 if (fp) { 1029 address low, high; 1030 while (!feof(fp)) { 1031 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1032 if (low <= addr && addr < high) { 1033 if (vma_low) *vma_low = low; 1034 if (vma_high) *vma_high = high; 1035 fclose (fp); 1036 return true; 1037 } 1038 } 1039 for (;;) { 1040 int ch = fgetc(fp); 1041 if (ch == EOF || ch == (int)'\n') break; 1042 } 1043 } 1044 fclose(fp); 1045 } 1046 return false; 1047} 1048 1049// Locate initial thread stack. This special handling of initial thread stack 1050// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1051// bogus value for initial thread. 1052void os::Linux::capture_initial_stack(size_t max_size) { 1053 // stack size is the easy part, get it from RLIMIT_STACK 1054 size_t stack_size; 1055 struct rlimit rlim; 1056 getrlimit(RLIMIT_STACK, &rlim); 1057 stack_size = rlim.rlim_cur; 1058 1059 // 6308388: a bug in ld.so will relocate its own .data section to the 1060 // lower end of primordial stack; reduce ulimit -s value a little bit 1061 // so we won't install guard page on ld.so's data section. 1062 stack_size -= 2 * page_size(); 1063 1064 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1065 // 7.1, in both cases we will get 2G in return value. 1066 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1067 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1068 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1069 // in case other parts in glibc still assumes 2M max stack size. 1070 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1071#ifndef IA64 1072 if (stack_size > 2 * K * K) stack_size = 2 * K * K; 1073#else 1074 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1075 if (stack_size > 4 * K * K) stack_size = 4 * K * K; 1076#endif 1077 1078 // Try to figure out where the stack base (top) is. This is harder. 1079 // 1080 // When an application is started, glibc saves the initial stack pointer in 1081 // a global variable "__libc_stack_end", which is then used by system 1082 // libraries. __libc_stack_end should be pretty close to stack top. The 1083 // variable is available since the very early days. However, because it is 1084 // a private interface, it could disappear in the future. 1085 // 1086 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1087 // to __libc_stack_end, it is very close to stack top, but isn't the real 1088 // stack top. Note that /proc may not exist if VM is running as a chroot 1089 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1090 // /proc/<pid>/stat could change in the future (though unlikely). 1091 // 1092 // We try __libc_stack_end first. If that doesn't work, look for 1093 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1094 // as a hint, which should work well in most cases. 1095 1096 uintptr_t stack_start; 1097 1098 // try __libc_stack_end first 1099 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1100 if (p && *p) { 1101 stack_start = *p; 1102 } else { 1103 // see if we can get the start_stack field from /proc/self/stat 1104 FILE *fp; 1105 int pid; 1106 char state; 1107 int ppid; 1108 int pgrp; 1109 int session; 1110 int nr; 1111 int tpgrp; 1112 unsigned long flags; 1113 unsigned long minflt; 1114 unsigned long cminflt; 1115 unsigned long majflt; 1116 unsigned long cmajflt; 1117 unsigned long utime; 1118 unsigned long stime; 1119 long cutime; 1120 long cstime; 1121 long prio; 1122 long nice; 1123 long junk; 1124 long it_real; 1125 uintptr_t start; 1126 uintptr_t vsize; 1127 uintptr_t rss; 1128 unsigned long rsslim; 1129 uintptr_t scodes; 1130 uintptr_t ecode; 1131 int i; 1132 1133 // Figure what the primordial thread stack base is. Code is inspired 1134 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1135 // followed by command name surrounded by parentheses, state, etc. 1136 char stat[2048]; 1137 int statlen; 1138 1139 fp = fopen("/proc/self/stat", "r"); 1140 if (fp) { 1141 statlen = fread(stat, 1, 2047, fp); 1142 stat[statlen] = '\0'; 1143 fclose(fp); 1144 1145 // Skip pid and the command string. Note that we could be dealing with 1146 // weird command names, e.g. user could decide to rename java launcher 1147 // to "java 1.4.2 :)", then the stat file would look like 1148 // 1234 (java 1.4.2 :)) R ... ... 1149 // We don't really need to know the command string, just find the last 1150 // occurrence of ")" and then start parsing from there. See bug 4726580. 1151 char * s = strrchr(stat, ')'); 1152 1153 i = 0; 1154 if (s) { 1155 // Skip blank chars 1156 do s++; while (isspace(*s)); 1157 1158 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1159 /* 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 */ 1160 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld %lu %lu %ld %lu %lu %lu %lu", 1161 &state, /* 3 %c */ 1162 &ppid, /* 4 %d */ 1163 &pgrp, /* 5 %d */ 1164 &session, /* 6 %d */ 1165 &nr, /* 7 %d */ 1166 &tpgrp, /* 8 %d */ 1167 &flags, /* 9 %lu */ 1168 &minflt, /* 10 %lu */ 1169 &cminflt, /* 11 %lu */ 1170 &majflt, /* 12 %lu */ 1171 &cmajflt, /* 13 %lu */ 1172 &utime, /* 14 %lu */ 1173 &stime, /* 15 %lu */ 1174 &cutime, /* 16 %ld */ 1175 &cstime, /* 17 %ld */ 1176 &prio, /* 18 %ld */ 1177 &nice, /* 19 %ld */ 1178 &junk, /* 20 %ld */ 1179 &it_real, /* 21 %ld */ 1180 &start, /* 22 %lu */ 1181 &vsize, /* 23 %lu */ 1182 &rss, /* 24 %ld */ 1183 &rsslim, /* 25 %lu */ 1184 &scodes, /* 26 %lu */ 1185 &ecode, /* 27 %lu */ 1186 &stack_start); /* 28 %lu */ 1187 } 1188 1189 if (i != 28 - 2) { 1190 assert(false, "Bad conversion from /proc/self/stat"); 1191 // product mode - assume we are the initial thread, good luck in the 1192 // embedded case. 1193 warning("Can't detect initial thread stack location - bad conversion"); 1194 stack_start = (uintptr_t) &rlim; 1195 } 1196 } else { 1197 // For some reason we can't open /proc/self/stat (for example, running on 1198 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1199 // most cases, so don't abort: 1200 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1201 stack_start = (uintptr_t) &rlim; 1202 } 1203 } 1204 1205 // Now we have a pointer (stack_start) very close to the stack top, the 1206 // next thing to do is to figure out the exact location of stack top. We 1207 // can find out the virtual memory area that contains stack_start by 1208 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1209 // and its upper limit is the real stack top. (again, this would fail if 1210 // running inside chroot, because /proc may not exist.) 1211 1212 uintptr_t stack_top; 1213 address low, high; 1214 if (find_vma((address)stack_start, &low, &high)) { 1215 // success, "high" is the true stack top. (ignore "low", because initial 1216 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1217 stack_top = (uintptr_t)high; 1218 } else { 1219 // failed, likely because /proc/self/maps does not exist 1220 warning("Can't detect initial thread stack location - find_vma failed"); 1221 // best effort: stack_start is normally within a few pages below the real 1222 // stack top, use it as stack top, and reduce stack size so we won't put 1223 // guard page outside stack. 1224 stack_top = stack_start; 1225 stack_size -= 16 * page_size(); 1226 } 1227 1228 // stack_top could be partially down the page so align it 1229 stack_top = align_size_up(stack_top, page_size()); 1230 1231 if (max_size && stack_size > max_size) { 1232 _initial_thread_stack_size = max_size; 1233 } else { 1234 _initial_thread_stack_size = stack_size; 1235 } 1236 1237 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1238 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1239} 1240 1241//////////////////////////////////////////////////////////////////////////////// 1242// time support 1243 1244// Time since start-up in seconds to a fine granularity. 1245// Used by VMSelfDestructTimer and the MemProfiler. 1246double os::elapsedTime() { 1247 1248 return (double)(os::elapsed_counter()) * 0.000001; 1249} 1250 1251jlong os::elapsed_counter() { 1252 timeval time; 1253 int status = gettimeofday(&time, NULL); 1254 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1255} 1256 1257jlong os::elapsed_frequency() { 1258 return (1000 * 1000); 1259} 1260 1261jlong os::javaTimeMillis() { 1262 timeval time; 1263 int status = gettimeofday(&time, NULL); 1264 assert(status != -1, "linux error"); 1265 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1266} 1267 1268#ifndef CLOCK_MONOTONIC 1269#define CLOCK_MONOTONIC (1) 1270#endif 1271 1272void os::Linux::clock_init() { 1273 // we do dlopen's in this particular order due to bug in linux 1274 // dynamical loader (see 6348968) leading to crash on exit 1275 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1276 if (handle == NULL) { 1277 handle = dlopen("librt.so", RTLD_LAZY); 1278 } 1279 1280 if (handle) { 1281 int (*clock_getres_func)(clockid_t, struct timespec*) = 1282 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1283 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1284 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1285 if (clock_getres_func && clock_gettime_func) { 1286 // See if monotonic clock is supported by the kernel. Note that some 1287 // early implementations simply return kernel jiffies (updated every 1288 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1289 // for nano time (though the monotonic property is still nice to have). 1290 // It's fixed in newer kernels, however clock_getres() still returns 1291 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1292 // resolution for now. Hopefully as people move to new kernels, this 1293 // won't be a problem. 1294 struct timespec res; 1295 struct timespec tp; 1296 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1297 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1298 // yes, monotonic clock is supported 1299 _clock_gettime = clock_gettime_func; 1300 } else { 1301 // close librt if there is no monotonic clock 1302 dlclose(handle); 1303 } 1304 } 1305 } 1306} 1307 1308#ifndef SYS_clock_getres 1309 1310#if defined(IA32) || defined(AMD64) 1311#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1312#else 1313#error Value of SYS_clock_getres not known on this platform 1314#endif 1315 1316#endif 1317 1318#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1319 1320void os::Linux::fast_thread_clock_init() { 1321 if (!UseLinuxPosixThreadCPUClocks) { 1322 return; 1323 } 1324 clockid_t clockid; 1325 struct timespec tp; 1326 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1327 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1328 1329 // Switch to using fast clocks for thread cpu time if 1330 // the sys_clock_getres() returns 0 error code. 1331 // Note, that some kernels may support the current thread 1332 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1333 // returned by the pthread_getcpuclockid(). 1334 // If the fast Posix clocks are supported then the sys_clock_getres() 1335 // must return at least tp.tv_sec == 0 which means a resolution 1336 // better than 1 sec. This is extra check for reliability. 1337 1338 if(pthread_getcpuclockid_func && 1339 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1340 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1341 1342 _supports_fast_thread_cpu_time = true; 1343 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1344 } 1345} 1346 1347jlong os::javaTimeNanos() { 1348 if (Linux::supports_monotonic_clock()) { 1349 struct timespec tp; 1350 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1351 assert(status == 0, "gettime error"); 1352 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1353 return result; 1354 } else { 1355 timeval time; 1356 int status = gettimeofday(&time, NULL); 1357 assert(status != -1, "linux error"); 1358 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1359 return 1000 * usecs; 1360 } 1361} 1362 1363void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1364 if (Linux::supports_monotonic_clock()) { 1365 info_ptr->max_value = ALL_64_BITS; 1366 1367 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1368 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1369 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1370 } else { 1371 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1372 info_ptr->max_value = ALL_64_BITS; 1373 1374 // gettimeofday is a real time clock so it skips 1375 info_ptr->may_skip_backward = true; 1376 info_ptr->may_skip_forward = true; 1377 } 1378 1379 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1380} 1381 1382// Return the real, user, and system times in seconds from an 1383// arbitrary fixed point in the past. 1384bool os::getTimesSecs(double* process_real_time, 1385 double* process_user_time, 1386 double* process_system_time) { 1387 struct tms ticks; 1388 clock_t real_ticks = times(&ticks); 1389 1390 if (real_ticks == (clock_t) (-1)) { 1391 return false; 1392 } else { 1393 double ticks_per_second = (double) clock_tics_per_sec; 1394 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1395 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1396 *process_real_time = ((double) real_ticks) / ticks_per_second; 1397 1398 return true; 1399 } 1400} 1401 1402 1403char * os::local_time_string(char *buf, size_t buflen) { 1404 struct tm t; 1405 time_t long_time; 1406 time(&long_time); 1407 localtime_r(&long_time, &t); 1408 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1409 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1410 t.tm_hour, t.tm_min, t.tm_sec); 1411 return buf; 1412} 1413 1414//////////////////////////////////////////////////////////////////////////////// 1415// runtime exit support 1416 1417// Note: os::shutdown() might be called very early during initialization, or 1418// called from signal handler. Before adding something to os::shutdown(), make 1419// sure it is async-safe and can handle partially initialized VM. 1420void os::shutdown() { 1421 1422 // allow PerfMemory to attempt cleanup of any persistent resources 1423 perfMemory_exit(); 1424 1425 // needs to remove object in file system 1426 AttachListener::abort(); 1427 1428 // flush buffered output, finish log files 1429 ostream_abort(); 1430 1431 // Check for abort hook 1432 abort_hook_t abort_hook = Arguments::abort_hook(); 1433 if (abort_hook != NULL) { 1434 abort_hook(); 1435 } 1436 1437} 1438 1439// Note: os::abort() might be called very early during initialization, or 1440// called from signal handler. Before adding something to os::abort(), make 1441// sure it is async-safe and can handle partially initialized VM. 1442void os::abort(bool dump_core) { 1443 os::shutdown(); 1444 if (dump_core) { 1445#ifndef PRODUCT 1446 fdStream out(defaultStream::output_fd()); 1447 out.print_raw("Current thread is "); 1448 char buf[16]; 1449 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1450 out.print_raw_cr(buf); 1451 out.print_raw_cr("Dumping core ..."); 1452#endif 1453 ::abort(); // dump core 1454 } 1455 1456 ::exit(1); 1457} 1458 1459// Die immediately, no exit hook, no abort hook, no cleanup. 1460void os::die() { 1461 // _exit() on LinuxThreads only kills current thread 1462 ::abort(); 1463} 1464 1465// unused on linux for now. 1466void os::set_error_file(const char *logfile) {} 1467 1468intx os::current_thread_id() { return (intx)pthread_self(); } 1469int os::current_process_id() { 1470 1471 // Under the old linux thread library, linux gives each thread 1472 // its own process id. Because of this each thread will return 1473 // a different pid if this method were to return the result 1474 // of getpid(2). Linux provides no api that returns the pid 1475 // of the launcher thread for the vm. This implementation 1476 // returns a unique pid, the pid of the launcher thread 1477 // that starts the vm 'process'. 1478 1479 // Under the NPTL, getpid() returns the same pid as the 1480 // launcher thread rather than a unique pid per thread. 1481 // Use gettid() if you want the old pre NPTL behaviour. 1482 1483 // if you are looking for the result of a call to getpid() that 1484 // returns a unique pid for the calling thread, then look at the 1485 // OSThread::thread_id() method in osThread_linux.hpp file 1486 1487 return (int)(_initial_pid ? _initial_pid : getpid()); 1488} 1489 1490// DLL functions 1491 1492const char* os::dll_file_extension() { return ".so"; } 1493 1494const char* os::get_temp_directory() { return "/tmp/"; } 1495 1496const char* os::get_current_directory(char *buf, int buflen) { 1497 return getcwd(buf, buflen); 1498} 1499 1500// check if addr is inside libjvm[_g].so 1501bool os::address_is_in_vm(address addr) { 1502 static address libjvm_base_addr; 1503 Dl_info dlinfo; 1504 1505 if (libjvm_base_addr == NULL) { 1506 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); 1507 libjvm_base_addr = (address)dlinfo.dli_fbase; 1508 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1509 } 1510 1511 if (dladdr((void *)addr, &dlinfo)) { 1512 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1513 } 1514 1515 return false; 1516} 1517 1518bool os::dll_address_to_function_name(address addr, char *buf, 1519 int buflen, int *offset) { 1520 Dl_info dlinfo; 1521 1522 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { 1523 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1524 if (offset) *offset = addr - (address)dlinfo.dli_saddr; 1525 return true; 1526 } else { 1527 if (buf) buf[0] = '\0'; 1528 if (offset) *offset = -1; 1529 return false; 1530 } 1531} 1532 1533struct _address_to_library_name { 1534 address addr; // input : memory address 1535 size_t buflen; // size of fname 1536 char* fname; // output: library name 1537 address base; // library base addr 1538}; 1539 1540static int address_to_library_name_callback(struct dl_phdr_info *info, 1541 size_t size, void *data) { 1542 int i; 1543 bool found = false; 1544 address libbase = NULL; 1545 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1546 1547 // iterate through all loadable segments 1548 for (i = 0; i < info->dlpi_phnum; i++) { 1549 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1550 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1551 // base address of a library is the lowest address of its loaded 1552 // segments. 1553 if (libbase == NULL || libbase > segbase) { 1554 libbase = segbase; 1555 } 1556 // see if 'addr' is within current segment 1557 if (segbase <= d->addr && 1558 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1559 found = true; 1560 } 1561 } 1562 } 1563 1564 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1565 // so dll_address_to_library_name() can fall through to use dladdr() which 1566 // can figure out executable name from argv[0]. 1567 if (found && info->dlpi_name && info->dlpi_name[0]) { 1568 d->base = libbase; 1569 if (d->fname) { 1570 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1571 } 1572 return 1; 1573 } 1574 return 0; 1575} 1576 1577bool os::dll_address_to_library_name(address addr, char* buf, 1578 int buflen, int* offset) { 1579 Dl_info dlinfo; 1580 struct _address_to_library_name data; 1581 1582 // There is a bug in old glibc dladdr() implementation that it could resolve 1583 // to wrong library name if the .so file has a base address != NULL. Here 1584 // we iterate through the program headers of all loaded libraries to find 1585 // out which library 'addr' really belongs to. This workaround can be 1586 // removed once the minimum requirement for glibc is moved to 2.3.x. 1587 data.addr = addr; 1588 data.fname = buf; 1589 data.buflen = buflen; 1590 data.base = NULL; 1591 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1592 1593 if (rslt) { 1594 // buf already contains library name 1595 if (offset) *offset = addr - data.base; 1596 return true; 1597 } else if (dladdr((void*)addr, &dlinfo)){ 1598 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1599 if (offset) *offset = addr - (address)dlinfo.dli_fbase; 1600 return true; 1601 } else { 1602 if (buf) buf[0] = '\0'; 1603 if (offset) *offset = -1; 1604 return false; 1605 } 1606} 1607 1608 // Loads .dll/.so and 1609 // in case of error it checks if .dll/.so was built for the 1610 // same architecture as Hotspot is running on 1611 1612void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1613{ 1614 void * result= ::dlopen(filename, RTLD_LAZY); 1615 if (result != NULL) { 1616 // Successful loading 1617 return result; 1618 } 1619 1620 Elf32_Ehdr elf_head; 1621 1622 // Read system error message into ebuf 1623 // It may or may not be overwritten below 1624 ::strncpy(ebuf, ::dlerror(), ebuflen-1); 1625 ebuf[ebuflen-1]='\0'; 1626 int diag_msg_max_length=ebuflen-strlen(ebuf); 1627 char* diag_msg_buf=ebuf+strlen(ebuf); 1628 1629 if (diag_msg_max_length==0) { 1630 // No more space in ebuf for additional diagnostics message 1631 return NULL; 1632 } 1633 1634 1635 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1636 1637 if (file_descriptor < 0) { 1638 // Can't open library, report dlerror() message 1639 return NULL; 1640 } 1641 1642 bool failed_to_read_elf_head= 1643 (sizeof(elf_head)!= 1644 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1645 1646 ::close(file_descriptor); 1647 if (failed_to_read_elf_head) { 1648 // file i/o error - report dlerror() msg 1649 return NULL; 1650 } 1651 1652 typedef struct { 1653 Elf32_Half code; // Actual value as defined in elf.h 1654 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1655 char elf_class; // 32 or 64 bit 1656 char endianess; // MSB or LSB 1657 char* name; // String representation 1658 } arch_t; 1659 1660 #ifndef EM_486 1661 #define EM_486 6 /* Intel 80486 */ 1662 #endif 1663 1664 static const arch_t arch_array[]={ 1665 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1666 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1667 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1668 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1669 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1670 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1671 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1672 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1673 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"} 1674 }; 1675 1676 #if (defined IA32) 1677 static Elf32_Half running_arch_code=EM_386; 1678 #elif (defined AMD64) 1679 static Elf32_Half running_arch_code=EM_X86_64; 1680 #elif (defined IA64) 1681 static Elf32_Half running_arch_code=EM_IA_64; 1682 #elif (defined __sparc) && (defined _LP64) 1683 static Elf32_Half running_arch_code=EM_SPARCV9; 1684 #elif (defined __sparc) && (!defined _LP64) 1685 static Elf32_Half running_arch_code=EM_SPARC; 1686 #elif (defined __powerpc64__) 1687 static Elf32_Half running_arch_code=EM_PPC64; 1688 #elif (defined __powerpc__) 1689 static Elf32_Half running_arch_code=EM_PPC; 1690 #else 1691 #error Method os::dll_load requires that one of following is defined:\ 1692 IA32, AMD64, IA64, __sparc, __powerpc__ 1693 #endif 1694 1695 // Identify compatability class for VM's architecture and library's architecture 1696 // Obtain string descriptions for architectures 1697 1698 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1699 int running_arch_index=-1; 1700 1701 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1702 if (running_arch_code == arch_array[i].code) { 1703 running_arch_index = i; 1704 } 1705 if (lib_arch.code == arch_array[i].code) { 1706 lib_arch.compat_class = arch_array[i].compat_class; 1707 lib_arch.name = arch_array[i].name; 1708 } 1709 } 1710 1711 assert(running_arch_index != -1, 1712 "Didn't find running architecture code (running_arch_code) in arch_array"); 1713 if (running_arch_index == -1) { 1714 // Even though running architecture detection failed 1715 // we may still continue with reporting dlerror() message 1716 return NULL; 1717 } 1718 1719 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1720 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1721 return NULL; 1722 } 1723 1724 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1725 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1726 return NULL; 1727 } 1728 1729 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1730 if ( lib_arch.name!=NULL ) { 1731 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1732 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1733 lib_arch.name, arch_array[running_arch_index].name); 1734 } else { 1735 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1736 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1737 lib_arch.code, 1738 arch_array[running_arch_index].name); 1739 } 1740 } 1741 1742 return NULL; 1743} 1744 1745 1746 1747 1748bool _print_ascii_file(const char* filename, outputStream* st) { 1749 int fd = open(filename, O_RDONLY); 1750 if (fd == -1) { 1751 return false; 1752 } 1753 1754 char buf[32]; 1755 int bytes; 1756 while ((bytes = read(fd, buf, sizeof(buf))) > 0) { 1757 st->print_raw(buf, bytes); 1758 } 1759 1760 close(fd); 1761 1762 return true; 1763} 1764 1765void os::print_dll_info(outputStream *st) { 1766 st->print_cr("Dynamic libraries:"); 1767 1768 char fname[32]; 1769 pid_t pid = os::Linux::gettid(); 1770 1771 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1772 1773 if (!_print_ascii_file(fname, st)) { 1774 st->print("Can not get library information for pid = %d\n", pid); 1775 } 1776} 1777 1778 1779void os::print_os_info(outputStream* st) { 1780 st->print("OS:"); 1781 1782 // Try to identify popular distros. 1783 // Most Linux distributions have /etc/XXX-release file, which contains 1784 // the OS version string. Some have more than one /etc/XXX-release file 1785 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 1786 // so the order is important. 1787 if (!_print_ascii_file("/etc/mandrake-release", st) && 1788 !_print_ascii_file("/etc/sun-release", st) && 1789 !_print_ascii_file("/etc/redhat-release", st) && 1790 !_print_ascii_file("/etc/SuSE-release", st) && 1791 !_print_ascii_file("/etc/turbolinux-release", st) && 1792 !_print_ascii_file("/etc/gentoo-release", st) && 1793 !_print_ascii_file("/etc/debian_version", st)) { 1794 st->print("Linux"); 1795 } 1796 st->cr(); 1797 1798 // kernel 1799 st->print("uname:"); 1800 struct utsname name; 1801 uname(&name); 1802 st->print(name.sysname); st->print(" "); 1803 st->print(name.release); st->print(" "); 1804 st->print(name.version); st->print(" "); 1805 st->print(name.machine); 1806 st->cr(); 1807 1808 // Print warning if unsafe chroot environment detected 1809 if (unsafe_chroot_detected) { 1810 st->print("WARNING!! "); 1811 st->print_cr(unstable_chroot_error); 1812 } 1813 1814 // libc, pthread 1815 st->print("libc:"); 1816 st->print(os::Linux::glibc_version()); st->print(" "); 1817 st->print(os::Linux::libpthread_version()); st->print(" "); 1818 if (os::Linux::is_LinuxThreads()) { 1819 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 1820 } 1821 st->cr(); 1822 1823 // rlimit 1824 st->print("rlimit:"); 1825 struct rlimit rlim; 1826 1827 st->print(" STACK "); 1828 getrlimit(RLIMIT_STACK, &rlim); 1829 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1830 else st->print("%uk", rlim.rlim_cur >> 10); 1831 1832 st->print(", CORE "); 1833 getrlimit(RLIMIT_CORE, &rlim); 1834 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1835 else st->print("%uk", rlim.rlim_cur >> 10); 1836 1837 st->print(", NPROC "); 1838 getrlimit(RLIMIT_NPROC, &rlim); 1839 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1840 else st->print("%d", rlim.rlim_cur); 1841 1842 st->print(", NOFILE "); 1843 getrlimit(RLIMIT_NOFILE, &rlim); 1844 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1845 else st->print("%d", rlim.rlim_cur); 1846 1847 st->print(", AS "); 1848 getrlimit(RLIMIT_AS, &rlim); 1849 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1850 else st->print("%uk", rlim.rlim_cur >> 10); 1851 st->cr(); 1852 1853 // load average 1854 st->print("load average:"); 1855 double loadavg[3]; 1856 os::loadavg(loadavg, 3); 1857 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]); 1858 st->cr(); 1859} 1860 1861void os::print_memory_info(outputStream* st) { 1862 1863 st->print("Memory:"); 1864 st->print(" %dk page", os::vm_page_size()>>10); 1865 1866 // values in struct sysinfo are "unsigned long" 1867 struct sysinfo si; 1868 sysinfo(&si); 1869 1870 st->print(", physical " UINT64_FORMAT "k", 1871 os::physical_memory() >> 10); 1872 st->print("(" UINT64_FORMAT "k free)", 1873 os::available_memory() >> 10); 1874 st->print(", swap " UINT64_FORMAT "k", 1875 ((jlong)si.totalswap * si.mem_unit) >> 10); 1876 st->print("(" UINT64_FORMAT "k free)", 1877 ((jlong)si.freeswap * si.mem_unit) >> 10); 1878 st->cr(); 1879} 1880 1881// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 1882// but they're the same for all the linux arch that we support 1883// and they're the same for solaris but there's no common place to put this. 1884const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 1885 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 1886 "ILL_COPROC", "ILL_BADSTK" }; 1887 1888const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 1889 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 1890 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 1891 1892const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 1893 1894const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 1895 1896void os::print_siginfo(outputStream* st, void* siginfo) { 1897 st->print("siginfo:"); 1898 1899 const int buflen = 100; 1900 char buf[buflen]; 1901 siginfo_t *si = (siginfo_t*)siginfo; 1902 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 1903 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 1904 st->print("si_errno=%s", buf); 1905 } else { 1906 st->print("si_errno=%d", si->si_errno); 1907 } 1908 const int c = si->si_code; 1909 assert(c > 0, "unexpected si_code"); 1910 switch (si->si_signo) { 1911 case SIGILL: 1912 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 1913 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1914 break; 1915 case SIGFPE: 1916 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 1917 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1918 break; 1919 case SIGSEGV: 1920 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 1921 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1922 break; 1923 case SIGBUS: 1924 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 1925 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1926 break; 1927 default: 1928 st->print(", si_code=%d", si->si_code); 1929 // no si_addr 1930 } 1931 1932 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 1933 UseSharedSpaces) { 1934 FileMapInfo* mapinfo = FileMapInfo::current_info(); 1935 if (mapinfo->is_in_shared_space(si->si_addr)) { 1936 st->print("\n\nError accessing class data sharing archive." \ 1937 " Mapped file inaccessible during execution, " \ 1938 " possible disk/network problem."); 1939 } 1940 } 1941 st->cr(); 1942} 1943 1944 1945static void print_signal_handler(outputStream* st, int sig, 1946 char* buf, size_t buflen); 1947 1948void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 1949 st->print_cr("Signal Handlers:"); 1950 print_signal_handler(st, SIGSEGV, buf, buflen); 1951 print_signal_handler(st, SIGBUS , buf, buflen); 1952 print_signal_handler(st, SIGFPE , buf, buflen); 1953 print_signal_handler(st, SIGPIPE, buf, buflen); 1954 print_signal_handler(st, SIGXFSZ, buf, buflen); 1955 print_signal_handler(st, SIGILL , buf, buflen); 1956 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 1957 print_signal_handler(st, SR_signum, buf, buflen); 1958 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 1959 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 1960 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 1961 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 1962} 1963 1964static char saved_jvm_path[MAXPATHLEN] = {0}; 1965 1966// Find the full path to the current module, libjvm.so or libjvm_g.so 1967void os::jvm_path(char *buf, jint len) { 1968 // Error checking. 1969 if (len < MAXPATHLEN) { 1970 assert(false, "must use a large-enough buffer"); 1971 buf[0] = '\0'; 1972 return; 1973 } 1974 // Lazy resolve the path to current module. 1975 if (saved_jvm_path[0] != 0) { 1976 strcpy(buf, saved_jvm_path); 1977 return; 1978 } 1979 1980 char dli_fname[MAXPATHLEN]; 1981 bool ret = dll_address_to_library_name( 1982 CAST_FROM_FN_PTR(address, os::jvm_path), 1983 dli_fname, sizeof(dli_fname), NULL); 1984 assert(ret != 0, "cannot locate libjvm"); 1985 realpath(dli_fname, buf); 1986 1987 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) { 1988 // Support for the gamma launcher. Typical value for buf is 1989 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 1990 // the right place in the string, then assume we are installed in a JDK and 1991 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 1992 // up the path so it looks like libjvm.so is installed there (append a 1993 // fake suffix hotspot/libjvm.so). 1994 const char *p = buf + strlen(buf) - 1; 1995 for (int count = 0; p > buf && count < 5; ++count) { 1996 for (--p; p > buf && *p != '/'; --p) 1997 /* empty */ ; 1998 } 1999 2000 if (strncmp(p, "/jre/lib/", 9) != 0) { 2001 // Look for JAVA_HOME in the environment. 2002 char* java_home_var = ::getenv("JAVA_HOME"); 2003 if (java_home_var != NULL && java_home_var[0] != 0) { 2004 // Check the current module name "libjvm.so" or "libjvm_g.so". 2005 p = strrchr(buf, '/'); 2006 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2007 p = strstr(p, "_g") ? "_g" : ""; 2008 2009 realpath(java_home_var, buf); 2010 sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch); 2011 if (0 == access(buf, F_OK)) { 2012 // Use current module name "libjvm[_g].so" instead of 2013 // "libjvm"debug_only("_g")".so" since for fastdebug version 2014 // we should have "libjvm.so" but debug_only("_g") adds "_g"! 2015 // It is used when we are choosing the HPI library's name 2016 // "libhpi[_g].so" in hpi::initialize_get_interface(). 2017 sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p); 2018 } else { 2019 // Go back to path of .so 2020 realpath(dli_fname, buf); 2021 } 2022 } 2023 } 2024 } 2025 2026 strcpy(saved_jvm_path, buf); 2027} 2028 2029void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2030 // no prefix required, not even "_" 2031} 2032 2033void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2034 // no suffix required 2035} 2036 2037//////////////////////////////////////////////////////////////////////////////// 2038// sun.misc.Signal support 2039 2040static volatile jint sigint_count = 0; 2041 2042static void 2043UserHandler(int sig, void *siginfo, void *context) { 2044 // 4511530 - sem_post is serialized and handled by the manager thread. When 2045 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2046 // don't want to flood the manager thread with sem_post requests. 2047 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2048 return; 2049 2050 // Ctrl-C is pressed during error reporting, likely because the error 2051 // handler fails to abort. Let VM die immediately. 2052 if (sig == SIGINT && is_error_reported()) { 2053 os::die(); 2054 } 2055 2056 os::signal_notify(sig); 2057} 2058 2059void* os::user_handler() { 2060 return CAST_FROM_FN_PTR(void*, UserHandler); 2061} 2062 2063extern "C" { 2064 typedef void (*sa_handler_t)(int); 2065 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2066} 2067 2068void* os::signal(int signal_number, void* handler) { 2069 struct sigaction sigAct, oldSigAct; 2070 2071 sigfillset(&(sigAct.sa_mask)); 2072 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2073 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2074 2075 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2076 // -1 means registration failed 2077 return (void *)-1; 2078 } 2079 2080 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2081} 2082 2083void os::signal_raise(int signal_number) { 2084 ::raise(signal_number); 2085} 2086 2087/* 2088 * The following code is moved from os.cpp for making this 2089 * code platform specific, which it is by its very nature. 2090 */ 2091 2092// Will be modified when max signal is changed to be dynamic 2093int os::sigexitnum_pd() { 2094 return NSIG; 2095} 2096 2097// a counter for each possible signal value 2098static volatile jint pending_signals[NSIG+1] = { 0 }; 2099 2100// Linux(POSIX) specific hand shaking semaphore. 2101static sem_t sig_sem; 2102 2103void os::signal_init_pd() { 2104 // Initialize signal structures 2105 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2106 2107 // Initialize signal semaphore 2108 ::sem_init(&sig_sem, 0, 0); 2109} 2110 2111void os::signal_notify(int sig) { 2112 Atomic::inc(&pending_signals[sig]); 2113 ::sem_post(&sig_sem); 2114} 2115 2116static int check_pending_signals(bool wait) { 2117 Atomic::store(0, &sigint_count); 2118 for (;;) { 2119 for (int i = 0; i < NSIG + 1; i++) { 2120 jint n = pending_signals[i]; 2121 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2122 return i; 2123 } 2124 } 2125 if (!wait) { 2126 return -1; 2127 } 2128 JavaThread *thread = JavaThread::current(); 2129 ThreadBlockInVM tbivm(thread); 2130 2131 bool threadIsSuspended; 2132 do { 2133 thread->set_suspend_equivalent(); 2134 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2135 ::sem_wait(&sig_sem); 2136 2137 // were we externally suspended while we were waiting? 2138 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2139 if (threadIsSuspended) { 2140 // 2141 // The semaphore has been incremented, but while we were waiting 2142 // another thread suspended us. We don't want to continue running 2143 // while suspended because that would surprise the thread that 2144 // suspended us. 2145 // 2146 ::sem_post(&sig_sem); 2147 2148 thread->java_suspend_self(); 2149 } 2150 } while (threadIsSuspended); 2151 } 2152} 2153 2154int os::signal_lookup() { 2155 return check_pending_signals(false); 2156} 2157 2158int os::signal_wait() { 2159 return check_pending_signals(true); 2160} 2161 2162//////////////////////////////////////////////////////////////////////////////// 2163// Virtual Memory 2164 2165int os::vm_page_size() { 2166 // Seems redundant as all get out 2167 assert(os::Linux::page_size() != -1, "must call os::init"); 2168 return os::Linux::page_size(); 2169} 2170 2171// Solaris allocates memory by pages. 2172int os::vm_allocation_granularity() { 2173 assert(os::Linux::page_size() != -1, "must call os::init"); 2174 return os::Linux::page_size(); 2175} 2176 2177// Rationale behind this function: 2178// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2179// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2180// samples for JITted code. Here we create private executable mapping over the code cache 2181// and then we can use standard (well, almost, as mapping can change) way to provide 2182// info for the reporting script by storing timestamp and location of symbol 2183void linux_wrap_code(char* base, size_t size) { 2184 static volatile jint cnt = 0; 2185 2186 if (!UseOprofile) { 2187 return; 2188 } 2189 2190 char buf[40]; 2191 int num = Atomic::add(1, &cnt); 2192 2193 sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num); 2194 unlink(buf); 2195 2196 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU); 2197 2198 if (fd != -1) { 2199 off_t rv = lseek(fd, size-2, SEEK_SET); 2200 if (rv != (off_t)-1) { 2201 if (write(fd, "", 1) == 1) { 2202 mmap(base, size, 2203 PROT_READ|PROT_WRITE|PROT_EXEC, 2204 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2205 } 2206 } 2207 close(fd); 2208 unlink(buf); 2209 } 2210} 2211 2212// NOTE: Linux kernel does not really reserve the pages for us. 2213// All it does is to check if there are enough free pages 2214// left at the time of mmap(). This could be a potential 2215// problem. 2216bool os::commit_memory(char* addr, size_t size) { 2217 uintptr_t res = (uintptr_t) ::mmap(addr, size, 2218 PROT_READ|PROT_WRITE|PROT_EXEC, 2219 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2220 return res != (uintptr_t) MAP_FAILED; 2221} 2222 2223bool os::commit_memory(char* addr, size_t size, size_t alignment_hint) { 2224 return commit_memory(addr, size); 2225} 2226 2227void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { } 2228 2229void os::free_memory(char *addr, size_t bytes) { 2230 uncommit_memory(addr, bytes); 2231} 2232 2233void os::numa_make_global(char *addr, size_t bytes) { } 2234 2235void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2236 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2237} 2238 2239bool os::numa_topology_changed() { return false; } 2240 2241size_t os::numa_get_groups_num() { 2242 int max_node = Linux::numa_max_node(); 2243 return max_node > 0 ? max_node + 1 : 1; 2244} 2245 2246int os::numa_get_group_id() { 2247 int cpu_id = Linux::sched_getcpu(); 2248 if (cpu_id != -1) { 2249 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2250 if (lgrp_id != -1) { 2251 return lgrp_id; 2252 } 2253 } 2254 return 0; 2255} 2256 2257size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2258 for (size_t i = 0; i < size; i++) { 2259 ids[i] = i; 2260 } 2261 return size; 2262} 2263 2264bool os::get_page_info(char *start, page_info* info) { 2265 return false; 2266} 2267 2268char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2269 return end; 2270} 2271 2272extern "C" void numa_warn(int number, char *where, ...) { } 2273extern "C" void numa_error(char *where) { } 2274 2275void os::Linux::libnuma_init() { 2276 // sched_getcpu() should be in libc. 2277 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2278 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2279 2280 if (sched_getcpu() != -1) { // Does it work? 2281 void *handle = dlopen("libnuma.so", RTLD_LAZY); 2282 if (handle != NULL) { 2283 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2284 dlsym(handle, "numa_node_to_cpus"))); 2285 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2286 dlsym(handle, "numa_max_node"))); 2287 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2288 dlsym(handle, "numa_available"))); 2289 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2290 dlsym(handle, "numa_tonode_memory"))); 2291 if (numa_available() != -1) { 2292 // Create a cpu -> node mapping 2293 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true); 2294 rebuild_cpu_to_node_map(); 2295 } 2296 } 2297 } 2298} 2299 2300// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2301// The table is later used in get_node_by_cpu(). 2302void os::Linux::rebuild_cpu_to_node_map() { 2303 int cpu_num = os::active_processor_count(); 2304 cpu_to_node()->clear(); 2305 cpu_to_node()->at_grow(cpu_num - 1); 2306 int node_num = numa_get_groups_num(); 2307 int cpu_map_size = (cpu_num + BitsPerLong - 1) / BitsPerLong; 2308 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size); 2309 for (int i = 0; i < node_num; i++) { 2310 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2311 for (int j = 0; j < cpu_map_size; j++) { 2312 if (cpu_map[j] != 0) { 2313 for (int k = 0; k < BitsPerLong; k++) { 2314 if (cpu_map[j] & (1UL << k)) { 2315 cpu_to_node()->at_put(j * BitsPerLong + k, i); 2316 } 2317 } 2318 } 2319 } 2320 } 2321 } 2322 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2323} 2324 2325int os::Linux::get_node_by_cpu(int cpu_id) { 2326 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2327 return cpu_to_node()->at(cpu_id); 2328 } 2329 return -1; 2330} 2331 2332GrowableArray<int>* os::Linux::_cpu_to_node; 2333os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2334os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2335os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2336os::Linux::numa_available_func_t os::Linux::_numa_available; 2337os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2338 2339 2340bool os::uncommit_memory(char* addr, size_t size) { 2341 return ::mmap(addr, size, 2342 PROT_READ|PROT_WRITE|PROT_EXEC, 2343 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0) 2344 != MAP_FAILED; 2345} 2346 2347static address _highest_vm_reserved_address = NULL; 2348 2349// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2350// at 'requested_addr'. If there are existing memory mappings at the same 2351// location, however, they will be overwritten. If 'fixed' is false, 2352// 'requested_addr' is only treated as a hint, the return value may or 2353// may not start from the requested address. Unlike Linux mmap(), this 2354// function returns NULL to indicate failure. 2355static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2356 char * addr; 2357 int flags; 2358 2359 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2360 if (fixed) { 2361 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2362 flags |= MAP_FIXED; 2363 } 2364 2365 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE|PROT_EXEC, 2366 flags, -1, 0); 2367 2368 if (addr != MAP_FAILED) { 2369 // anon_mmap() should only get called during VM initialization, 2370 // don't need lock (actually we can skip locking even it can be called 2371 // from multiple threads, because _highest_vm_reserved_address is just a 2372 // hint about the upper limit of non-stack memory regions.) 2373 if ((address)addr + bytes > _highest_vm_reserved_address) { 2374 _highest_vm_reserved_address = (address)addr + bytes; 2375 } 2376 } 2377 2378 return addr == MAP_FAILED ? NULL : addr; 2379} 2380 2381// Don't update _highest_vm_reserved_address, because there might be memory 2382// regions above addr + size. If so, releasing a memory region only creates 2383// a hole in the address space, it doesn't help prevent heap-stack collision. 2384// 2385static int anon_munmap(char * addr, size_t size) { 2386 return ::munmap(addr, size) == 0; 2387} 2388 2389char* os::reserve_memory(size_t bytes, char* requested_addr, 2390 size_t alignment_hint) { 2391 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 2392} 2393 2394bool os::release_memory(char* addr, size_t size) { 2395 return anon_munmap(addr, size); 2396} 2397 2398static address highest_vm_reserved_address() { 2399 return _highest_vm_reserved_address; 2400} 2401 2402static bool linux_mprotect(char* addr, size_t size, int prot) { 2403 // Linux wants the mprotect address argument to be page aligned. 2404 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 2405 2406 // According to SUSv3, mprotect() should only be used with mappings 2407 // established by mmap(), and mmap() always maps whole pages. Unaligned 2408 // 'addr' likely indicates problem in the VM (e.g. trying to change 2409 // protection of malloc'ed or statically allocated memory). Check the 2410 // caller if you hit this assert. 2411 assert(addr == bottom, "sanity check"); 2412 2413 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 2414 return ::mprotect(bottom, size, prot) == 0; 2415} 2416 2417bool os::protect_memory(char* addr, size_t size) { 2418 return linux_mprotect(addr, size, PROT_READ); 2419} 2420 2421bool os::guard_memory(char* addr, size_t size) { 2422 return linux_mprotect(addr, size, PROT_NONE); 2423} 2424 2425bool os::unguard_memory(char* addr, size_t size) { 2426 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE|PROT_EXEC); 2427} 2428 2429// Large page support 2430 2431static size_t _large_page_size = 0; 2432 2433bool os::large_page_init() { 2434 if (!UseLargePages) return false; 2435 2436 if (LargePageSizeInBytes) { 2437 _large_page_size = LargePageSizeInBytes; 2438 } else { 2439 // large_page_size on Linux is used to round up heap size. x86 uses either 2440 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 2441 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 2442 // page as large as 256M. 2443 // 2444 // Here we try to figure out page size by parsing /proc/meminfo and looking 2445 // for a line with the following format: 2446 // Hugepagesize: 2048 kB 2447 // 2448 // If we can't determine the value (e.g. /proc is not mounted, or the text 2449 // format has been changed), we'll use the largest page size supported by 2450 // the processor. 2451 2452 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M); 2453 2454 FILE *fp = fopen("/proc/meminfo", "r"); 2455 if (fp) { 2456 while (!feof(fp)) { 2457 int x = 0; 2458 char buf[16]; 2459 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 2460 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 2461 _large_page_size = x * K; 2462 break; 2463 } 2464 } else { 2465 // skip to next line 2466 for (;;) { 2467 int ch = fgetc(fp); 2468 if (ch == EOF || ch == (int)'\n') break; 2469 } 2470 } 2471 } 2472 fclose(fp); 2473 } 2474 } 2475 2476 const size_t default_page_size = (size_t)Linux::page_size(); 2477 if (_large_page_size > default_page_size) { 2478 _page_sizes[0] = _large_page_size; 2479 _page_sizes[1] = default_page_size; 2480 _page_sizes[2] = 0; 2481 } 2482 2483 // Large page support is available on 2.6 or newer kernel, some vendors 2484 // (e.g. Redhat) have backported it to their 2.4 based distributions. 2485 // We optimistically assume the support is available. If later it turns out 2486 // not true, VM will automatically switch to use regular page size. 2487 return true; 2488} 2489 2490#ifndef SHM_HUGETLB 2491#define SHM_HUGETLB 04000 2492#endif 2493 2494char* os::reserve_memory_special(size_t bytes) { 2495 assert(UseLargePages, "only for large pages"); 2496 2497 key_t key = IPC_PRIVATE; 2498 char *addr; 2499 2500 bool warn_on_failure = UseLargePages && 2501 (!FLAG_IS_DEFAULT(UseLargePages) || 2502 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 2503 ); 2504 char msg[128]; 2505 2506 // Create a large shared memory region to attach to based on size. 2507 // Currently, size is the total size of the heap 2508 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 2509 if (shmid == -1) { 2510 // Possible reasons for shmget failure: 2511 // 1. shmmax is too small for Java heap. 2512 // > check shmmax value: cat /proc/sys/kernel/shmmax 2513 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 2514 // 2. not enough large page memory. 2515 // > check available large pages: cat /proc/meminfo 2516 // > increase amount of large pages: 2517 // echo new_value > /proc/sys/vm/nr_hugepages 2518 // Note 1: different Linux may use different name for this property, 2519 // e.g. on Redhat AS-3 it is "hugetlb_pool". 2520 // Note 2: it's possible there's enough physical memory available but 2521 // they are so fragmented after a long run that they can't 2522 // coalesce into large pages. Try to reserve large pages when 2523 // the system is still "fresh". 2524 if (warn_on_failure) { 2525 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 2526 warning(msg); 2527 } 2528 return NULL; 2529 } 2530 2531 // attach to the region 2532 addr = (char*)shmat(shmid, NULL, 0); 2533 int err = errno; 2534 2535 // Remove shmid. If shmat() is successful, the actual shared memory segment 2536 // will be deleted when it's detached by shmdt() or when the process 2537 // terminates. If shmat() is not successful this will remove the shared 2538 // segment immediately. 2539 shmctl(shmid, IPC_RMID, NULL); 2540 2541 if ((intptr_t)addr == -1) { 2542 if (warn_on_failure) { 2543 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 2544 warning(msg); 2545 } 2546 return NULL; 2547 } 2548 2549 return addr; 2550} 2551 2552bool os::release_memory_special(char* base, size_t bytes) { 2553 // detaching the SHM segment will also delete it, see reserve_memory_special() 2554 int rslt = shmdt(base); 2555 return rslt == 0; 2556} 2557 2558size_t os::large_page_size() { 2559 return _large_page_size; 2560} 2561 2562// Linux does not support anonymous mmap with large page memory. The only way 2563// to reserve large page memory without file backing is through SysV shared 2564// memory API. The entire memory region is committed and pinned upfront. 2565// Hopefully this will change in the future... 2566bool os::can_commit_large_page_memory() { 2567 return false; 2568} 2569 2570bool os::can_execute_large_page_memory() { 2571 return false; 2572} 2573 2574// Reserve memory at an arbitrary address, only if that area is 2575// available (and not reserved for something else). 2576 2577char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 2578 const int max_tries = 10; 2579 char* base[max_tries]; 2580 size_t size[max_tries]; 2581 const size_t gap = 0x000000; 2582 2583 // Assert only that the size is a multiple of the page size, since 2584 // that's all that mmap requires, and since that's all we really know 2585 // about at this low abstraction level. If we need higher alignment, 2586 // we can either pass an alignment to this method or verify alignment 2587 // in one of the methods further up the call chain. See bug 5044738. 2588 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 2589 2590 // Repeatedly allocate blocks until the block is allocated at the 2591 // right spot. Give up after max_tries. Note that reserve_memory() will 2592 // automatically update _highest_vm_reserved_address if the call is 2593 // successful. The variable tracks the highest memory address every reserved 2594 // by JVM. It is used to detect heap-stack collision if running with 2595 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 2596 // space than needed, it could confuse the collision detecting code. To 2597 // solve the problem, save current _highest_vm_reserved_address and 2598 // calculate the correct value before return. 2599 address old_highest = _highest_vm_reserved_address; 2600 2601 // Linux mmap allows caller to pass an address as hint; give it a try first, 2602 // if kernel honors the hint then we can return immediately. 2603 char * addr = anon_mmap(requested_addr, bytes, false); 2604 if (addr == requested_addr) { 2605 return requested_addr; 2606 } 2607 2608 if (addr != NULL) { 2609 // mmap() is successful but it fails to reserve at the requested address 2610 anon_munmap(addr, bytes); 2611 } 2612 2613 int i; 2614 for (i = 0; i < max_tries; ++i) { 2615 base[i] = reserve_memory(bytes); 2616 2617 if (base[i] != NULL) { 2618 // Is this the block we wanted? 2619 if (base[i] == requested_addr) { 2620 size[i] = bytes; 2621 break; 2622 } 2623 2624 // Does this overlap the block we wanted? Give back the overlapped 2625 // parts and try again. 2626 2627 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 2628 if (top_overlap >= 0 && top_overlap < bytes) { 2629 unmap_memory(base[i], top_overlap); 2630 base[i] += top_overlap; 2631 size[i] = bytes - top_overlap; 2632 } else { 2633 size_t bottom_overlap = base[i] + bytes - requested_addr; 2634 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 2635 unmap_memory(requested_addr, bottom_overlap); 2636 size[i] = bytes - bottom_overlap; 2637 } else { 2638 size[i] = bytes; 2639 } 2640 } 2641 } 2642 } 2643 2644 // Give back the unused reserved pieces. 2645 2646 for (int j = 0; j < i; ++j) { 2647 if (base[j] != NULL) { 2648 unmap_memory(base[j], size[j]); 2649 } 2650 } 2651 2652 if (i < max_tries) { 2653 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 2654 return requested_addr; 2655 } else { 2656 _highest_vm_reserved_address = old_highest; 2657 return NULL; 2658 } 2659} 2660 2661size_t os::read(int fd, void *buf, unsigned int nBytes) { 2662 return ::read(fd, buf, nBytes); 2663} 2664 2665// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 2666// Solaris uses poll(), linux uses park(). 2667// Poll() is likely a better choice, assuming that Thread.interrupt() 2668// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 2669// SIGSEGV, see 4355769. 2670 2671const int NANOSECS_PER_MILLISECS = 1000000; 2672 2673int os::sleep(Thread* thread, jlong millis, bool interruptible) { 2674 assert(thread == Thread::current(), "thread consistency check"); 2675 2676 ParkEvent * const slp = thread->_SleepEvent ; 2677 slp->reset() ; 2678 OrderAccess::fence() ; 2679 2680 if (interruptible) { 2681 jlong prevtime = javaTimeNanos(); 2682 2683 for (;;) { 2684 if (os::is_interrupted(thread, true)) { 2685 return OS_INTRPT; 2686 } 2687 2688 jlong newtime = javaTimeNanos(); 2689 2690 if (newtime - prevtime < 0) { 2691 // time moving backwards, should only happen if no monotonic clock 2692 // not a guarantee() because JVM should not abort on kernel/glibc bugs 2693 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 2694 } else { 2695 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 2696 } 2697 2698 if(millis <= 0) { 2699 return OS_OK; 2700 } 2701 2702 prevtime = newtime; 2703 2704 { 2705 assert(thread->is_Java_thread(), "sanity check"); 2706 JavaThread *jt = (JavaThread *) thread; 2707 ThreadBlockInVM tbivm(jt); 2708 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 2709 2710 jt->set_suspend_equivalent(); 2711 // cleared by handle_special_suspend_equivalent_condition() or 2712 // java_suspend_self() via check_and_wait_while_suspended() 2713 2714 slp->park(millis); 2715 2716 // were we externally suspended while we were waiting? 2717 jt->check_and_wait_while_suspended(); 2718 } 2719 } 2720 } else { 2721 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 2722 jlong prevtime = javaTimeNanos(); 2723 2724 for (;;) { 2725 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 2726 // the 1st iteration ... 2727 jlong newtime = javaTimeNanos(); 2728 2729 if (newtime - prevtime < 0) { 2730 // time moving backwards, should only happen if no monotonic clock 2731 // not a guarantee() because JVM should not abort on kernel/glibc bugs 2732 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 2733 } else { 2734 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 2735 } 2736 2737 if(millis <= 0) break ; 2738 2739 prevtime = newtime; 2740 slp->park(millis); 2741 } 2742 return OS_OK ; 2743 } 2744} 2745 2746int os::naked_sleep() { 2747 // %% make the sleep time an integer flag. for now use 1 millisec. 2748 return os::sleep(Thread::current(), 1, false); 2749} 2750 2751// Sleep forever; naked call to OS-specific sleep; use with CAUTION 2752void os::infinite_sleep() { 2753 while (true) { // sleep forever ... 2754 ::sleep(100); // ... 100 seconds at a time 2755 } 2756} 2757 2758// Used to convert frequent JVM_Yield() to nops 2759bool os::dont_yield() { 2760 return DontYieldALot; 2761} 2762 2763void os::yield() { 2764 sched_yield(); 2765} 2766 2767os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 2768 2769void os::yield_all(int attempts) { 2770 // Yields to all threads, including threads with lower priorities 2771 // Threads on Linux are all with same priority. The Solaris style 2772 // os::yield_all() with nanosleep(1ms) is not necessary. 2773 sched_yield(); 2774} 2775 2776// Called from the tight loops to possibly influence time-sharing heuristics 2777void os::loop_breaker(int attempts) { 2778 os::yield_all(attempts); 2779} 2780 2781//////////////////////////////////////////////////////////////////////////////// 2782// thread priority support 2783 2784// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 2785// only supports dynamic priority, static priority must be zero. For real-time 2786// applications, Linux supports SCHED_RR which allows static priority (1-99). 2787// However, for large multi-threaded applications, SCHED_RR is not only slower 2788// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 2789// of 5 runs - Sep 2005). 2790// 2791// The following code actually changes the niceness of kernel-thread/LWP. It 2792// has an assumption that setpriority() only modifies one kernel-thread/LWP, 2793// not the entire user process, and user level threads are 1:1 mapped to kernel 2794// threads. It has always been the case, but could change in the future. For 2795// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 2796// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 2797 2798int os::java_to_os_priority[MaxPriority + 1] = { 2799 19, // 0 Entry should never be used 2800 2801 4, // 1 MinPriority 2802 3, // 2 2803 2, // 3 2804 2805 1, // 4 2806 0, // 5 NormPriority 2807 -1, // 6 2808 2809 -2, // 7 2810 -3, // 8 2811 -4, // 9 NearMaxPriority 2812 2813 -5 // 10 MaxPriority 2814}; 2815 2816static int prio_init() { 2817 if (ThreadPriorityPolicy == 1) { 2818 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 2819 // if effective uid is not root. Perhaps, a more elegant way of doing 2820 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 2821 if (geteuid() != 0) { 2822 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 2823 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 2824 } 2825 ThreadPriorityPolicy = 0; 2826 } 2827 } 2828 return 0; 2829} 2830 2831OSReturn os::set_native_priority(Thread* thread, int newpri) { 2832 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 2833 2834 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 2835 return (ret == 0) ? OS_OK : OS_ERR; 2836} 2837 2838OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 2839 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 2840 *priority_ptr = java_to_os_priority[NormPriority]; 2841 return OS_OK; 2842 } 2843 2844 errno = 0; 2845 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 2846 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 2847} 2848 2849// Hint to the underlying OS that a task switch would not be good. 2850// Void return because it's a hint and can fail. 2851void os::hint_no_preempt() {} 2852 2853//////////////////////////////////////////////////////////////////////////////// 2854// suspend/resume support 2855 2856// the low-level signal-based suspend/resume support is a remnant from the 2857// old VM-suspension that used to be for java-suspension, safepoints etc, 2858// within hotspot. Now there is a single use-case for this: 2859// - calling get_thread_pc() on the VMThread by the flat-profiler task 2860// that runs in the watcher thread. 2861// The remaining code is greatly simplified from the more general suspension 2862// code that used to be used. 2863// 2864// The protocol is quite simple: 2865// - suspend: 2866// - sends a signal to the target thread 2867// - polls the suspend state of the osthread using a yield loop 2868// - target thread signal handler (SR_handler) sets suspend state 2869// and blocks in sigsuspend until continued 2870// - resume: 2871// - sets target osthread state to continue 2872// - sends signal to end the sigsuspend loop in the SR_handler 2873// 2874// Note that the SR_lock plays no role in this suspend/resume protocol. 2875// 2876 2877static void resume_clear_context(OSThread *osthread) { 2878 osthread->set_ucontext(NULL); 2879 osthread->set_siginfo(NULL); 2880 2881 // notify the suspend action is completed, we have now resumed 2882 osthread->sr.clear_suspended(); 2883} 2884 2885static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 2886 osthread->set_ucontext(context); 2887 osthread->set_siginfo(siginfo); 2888} 2889 2890// 2891// Handler function invoked when a thread's execution is suspended or 2892// resumed. We have to be careful that only async-safe functions are 2893// called here (Note: most pthread functions are not async safe and 2894// should be avoided.) 2895// 2896// Note: sigwait() is a more natural fit than sigsuspend() from an 2897// interface point of view, but sigwait() prevents the signal hander 2898// from being run. libpthread would get very confused by not having 2899// its signal handlers run and prevents sigwait()'s use with the 2900// mutex granting granting signal. 2901// 2902// Currently only ever called on the VMThread 2903// 2904static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 2905 // Save and restore errno to avoid confusing native code with EINTR 2906 // after sigsuspend. 2907 int old_errno = errno; 2908 2909 Thread* thread = Thread::current(); 2910 OSThread* osthread = thread->osthread(); 2911 assert(thread->is_VM_thread(), "Must be VMThread"); 2912 // read current suspend action 2913 int action = osthread->sr.suspend_action(); 2914 if (action == SR_SUSPEND) { 2915 suspend_save_context(osthread, siginfo, context); 2916 2917 // Notify the suspend action is about to be completed. do_suspend() 2918 // waits until SR_SUSPENDED is set and then returns. We will wait 2919 // here for a resume signal and that completes the suspend-other 2920 // action. do_suspend/do_resume is always called as a pair from 2921 // the same thread - so there are no races 2922 2923 // notify the caller 2924 osthread->sr.set_suspended(); 2925 2926 sigset_t suspend_set; // signals for sigsuspend() 2927 2928 // get current set of blocked signals and unblock resume signal 2929 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 2930 sigdelset(&suspend_set, SR_signum); 2931 2932 // wait here until we are resumed 2933 do { 2934 sigsuspend(&suspend_set); 2935 // ignore all returns until we get a resume signal 2936 } while (osthread->sr.suspend_action() != SR_CONTINUE); 2937 2938 resume_clear_context(osthread); 2939 2940 } else { 2941 assert(action == SR_CONTINUE, "unexpected sr action"); 2942 // nothing special to do - just leave the handler 2943 } 2944 2945 errno = old_errno; 2946} 2947 2948 2949static int SR_initialize() { 2950 struct sigaction act; 2951 char *s; 2952 /* Get signal number to use for suspend/resume */ 2953 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 2954 int sig = ::strtol(s, 0, 10); 2955 if (sig > 0 || sig < _NSIG) { 2956 SR_signum = sig; 2957 } 2958 } 2959 2960 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 2961 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 2962 2963 sigemptyset(&SR_sigset); 2964 sigaddset(&SR_sigset, SR_signum); 2965 2966 /* Set up signal handler for suspend/resume */ 2967 act.sa_flags = SA_RESTART|SA_SIGINFO; 2968 act.sa_handler = (void (*)(int)) SR_handler; 2969 2970 // SR_signum is blocked by default. 2971 // 4528190 - We also need to block pthread restart signal (32 on all 2972 // supported Linux platforms). Note that LinuxThreads need to block 2973 // this signal for all threads to work properly. So we don't have 2974 // to use hard-coded signal number when setting up the mask. 2975 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 2976 2977 if (sigaction(SR_signum, &act, 0) == -1) { 2978 return -1; 2979 } 2980 2981 // Save signal flag 2982 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 2983 return 0; 2984} 2985 2986static int SR_finalize() { 2987 return 0; 2988} 2989 2990 2991// returns true on success and false on error - really an error is fatal 2992// but this seems the normal response to library errors 2993static bool do_suspend(OSThread* osthread) { 2994 // mark as suspended and send signal 2995 osthread->sr.set_suspend_action(SR_SUSPEND); 2996 int status = pthread_kill(osthread->pthread_id(), SR_signum); 2997 assert_status(status == 0, status, "pthread_kill"); 2998 2999 // check status and wait until notified of suspension 3000 if (status == 0) { 3001 for (int i = 0; !osthread->sr.is_suspended(); i++) { 3002 os::yield_all(i); 3003 } 3004 osthread->sr.set_suspend_action(SR_NONE); 3005 return true; 3006 } 3007 else { 3008 osthread->sr.set_suspend_action(SR_NONE); 3009 return false; 3010 } 3011} 3012 3013static void do_resume(OSThread* osthread) { 3014 assert(osthread->sr.is_suspended(), "thread should be suspended"); 3015 osthread->sr.set_suspend_action(SR_CONTINUE); 3016 3017 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3018 assert_status(status == 0, status, "pthread_kill"); 3019 // check status and wait unit notified of resumption 3020 if (status == 0) { 3021 for (int i = 0; osthread->sr.is_suspended(); i++) { 3022 os::yield_all(i); 3023 } 3024 } 3025 osthread->sr.set_suspend_action(SR_NONE); 3026} 3027 3028//////////////////////////////////////////////////////////////////////////////// 3029// interrupt support 3030 3031void os::interrupt(Thread* thread) { 3032 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3033 "possibility of dangling Thread pointer"); 3034 3035 OSThread* osthread = thread->osthread(); 3036 3037 if (!osthread->interrupted()) { 3038 osthread->set_interrupted(true); 3039 // More than one thread can get here with the same value of osthread, 3040 // resulting in multiple notifications. We do, however, want the store 3041 // to interrupted() to be visible to other threads before we execute unpark(). 3042 OrderAccess::fence(); 3043 ParkEvent * const slp = thread->_SleepEvent ; 3044 if (slp != NULL) slp->unpark() ; 3045 } 3046 3047 // For JSR166. Unpark even if interrupt status already was set 3048 if (thread->is_Java_thread()) 3049 ((JavaThread*)thread)->parker()->unpark(); 3050 3051 ParkEvent * ev = thread->_ParkEvent ; 3052 if (ev != NULL) ev->unpark() ; 3053 3054} 3055 3056bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 3057 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3058 "possibility of dangling Thread pointer"); 3059 3060 OSThread* osthread = thread->osthread(); 3061 3062 bool interrupted = osthread->interrupted(); 3063 3064 if (interrupted && clear_interrupted) { 3065 osthread->set_interrupted(false); 3066 // consider thread->_SleepEvent->reset() ... optional optimization 3067 } 3068 3069 return interrupted; 3070} 3071 3072/////////////////////////////////////////////////////////////////////////////////// 3073// signal handling (except suspend/resume) 3074 3075// This routine may be used by user applications as a "hook" to catch signals. 3076// The user-defined signal handler must pass unrecognized signals to this 3077// routine, and if it returns true (non-zero), then the signal handler must 3078// return immediately. If the flag "abort_if_unrecognized" is true, then this 3079// routine will never retun false (zero), but instead will execute a VM panic 3080// routine kill the process. 3081// 3082// If this routine returns false, it is OK to call it again. This allows 3083// the user-defined signal handler to perform checks either before or after 3084// the VM performs its own checks. Naturally, the user code would be making 3085// a serious error if it tried to handle an exception (such as a null check 3086// or breakpoint) that the VM was generating for its own correct operation. 3087// 3088// This routine may recognize any of the following kinds of signals: 3089// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 3090// It should be consulted by handlers for any of those signals. 3091// 3092// The caller of this routine must pass in the three arguments supplied 3093// to the function referred to in the "sa_sigaction" (not the "sa_handler") 3094// field of the structure passed to sigaction(). This routine assumes that 3095// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 3096// 3097// Note that the VM will print warnings if it detects conflicting signal 3098// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 3099// 3100extern "C" int 3101JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 3102 void* ucontext, int abort_if_unrecognized); 3103 3104void signalHandler(int sig, siginfo_t* info, void* uc) { 3105 assert(info != NULL && uc != NULL, "it must be old kernel"); 3106 JVM_handle_linux_signal(sig, info, uc, true); 3107} 3108 3109 3110// This boolean allows users to forward their own non-matching signals 3111// to JVM_handle_linux_signal, harmlessly. 3112bool os::Linux::signal_handlers_are_installed = false; 3113 3114// For signal-chaining 3115struct sigaction os::Linux::sigact[MAXSIGNUM]; 3116unsigned int os::Linux::sigs = 0; 3117bool os::Linux::libjsig_is_loaded = false; 3118typedef struct sigaction *(*get_signal_t)(int); 3119get_signal_t os::Linux::get_signal_action = NULL; 3120 3121struct sigaction* os::Linux::get_chained_signal_action(int sig) { 3122 struct sigaction *actp = NULL; 3123 3124 if (libjsig_is_loaded) { 3125 // Retrieve the old signal handler from libjsig 3126 actp = (*get_signal_action)(sig); 3127 } 3128 if (actp == NULL) { 3129 // Retrieve the preinstalled signal handler from jvm 3130 actp = get_preinstalled_handler(sig); 3131 } 3132 3133 return actp; 3134} 3135 3136static bool call_chained_handler(struct sigaction *actp, int sig, 3137 siginfo_t *siginfo, void *context) { 3138 // Call the old signal handler 3139 if (actp->sa_handler == SIG_DFL) { 3140 // It's more reasonable to let jvm treat it as an unexpected exception 3141 // instead of taking the default action. 3142 return false; 3143 } else if (actp->sa_handler != SIG_IGN) { 3144 if ((actp->sa_flags & SA_NODEFER) == 0) { 3145 // automaticlly block the signal 3146 sigaddset(&(actp->sa_mask), sig); 3147 } 3148 3149 sa_handler_t hand; 3150 sa_sigaction_t sa; 3151 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 3152 // retrieve the chained handler 3153 if (siginfo_flag_set) { 3154 sa = actp->sa_sigaction; 3155 } else { 3156 hand = actp->sa_handler; 3157 } 3158 3159 if ((actp->sa_flags & SA_RESETHAND) != 0) { 3160 actp->sa_handler = SIG_DFL; 3161 } 3162 3163 // try to honor the signal mask 3164 sigset_t oset; 3165 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 3166 3167 // call into the chained handler 3168 if (siginfo_flag_set) { 3169 (*sa)(sig, siginfo, context); 3170 } else { 3171 (*hand)(sig); 3172 } 3173 3174 // restore the signal mask 3175 pthread_sigmask(SIG_SETMASK, &oset, 0); 3176 } 3177 // Tell jvm's signal handler the signal is taken care of. 3178 return true; 3179} 3180 3181bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 3182 bool chained = false; 3183 // signal-chaining 3184 if (UseSignalChaining) { 3185 struct sigaction *actp = get_chained_signal_action(sig); 3186 if (actp != NULL) { 3187 chained = call_chained_handler(actp, sig, siginfo, context); 3188 } 3189 } 3190 return chained; 3191} 3192 3193struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 3194 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 3195 return &sigact[sig]; 3196 } 3197 return NULL; 3198} 3199 3200void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 3201 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3202 sigact[sig] = oldAct; 3203 sigs |= (unsigned int)1 << sig; 3204} 3205 3206// for diagnostic 3207int os::Linux::sigflags[MAXSIGNUM]; 3208 3209int os::Linux::get_our_sigflags(int sig) { 3210 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3211 return sigflags[sig]; 3212} 3213 3214void os::Linux::set_our_sigflags(int sig, int flags) { 3215 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3216 sigflags[sig] = flags; 3217} 3218 3219void os::Linux::set_signal_handler(int sig, bool set_installed) { 3220 // Check for overwrite. 3221 struct sigaction oldAct; 3222 sigaction(sig, (struct sigaction*)NULL, &oldAct); 3223 3224 void* oldhand = oldAct.sa_sigaction 3225 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3226 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3227 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 3228 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 3229 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 3230 if (AllowUserSignalHandlers || !set_installed) { 3231 // Do not overwrite; user takes responsibility to forward to us. 3232 return; 3233 } else if (UseSignalChaining) { 3234 // save the old handler in jvm 3235 save_preinstalled_handler(sig, oldAct); 3236 // libjsig also interposes the sigaction() call below and saves the 3237 // old sigaction on it own. 3238 } else { 3239 fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig); 3240 } 3241 } 3242 3243 struct sigaction sigAct; 3244 sigfillset(&(sigAct.sa_mask)); 3245 sigAct.sa_handler = SIG_DFL; 3246 if (!set_installed) { 3247 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3248 } else { 3249 sigAct.sa_sigaction = signalHandler; 3250 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3251 } 3252 // Save flags, which are set by ours 3253 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3254 sigflags[sig] = sigAct.sa_flags; 3255 3256 int ret = sigaction(sig, &sigAct, &oldAct); 3257 assert(ret == 0, "check"); 3258 3259 void* oldhand2 = oldAct.sa_sigaction 3260 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3261 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3262 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 3263} 3264 3265// install signal handlers for signals that HotSpot needs to 3266// handle in order to support Java-level exception handling. 3267 3268void os::Linux::install_signal_handlers() { 3269 if (!signal_handlers_are_installed) { 3270 signal_handlers_are_installed = true; 3271 3272 // signal-chaining 3273 typedef void (*signal_setting_t)(); 3274 signal_setting_t begin_signal_setting = NULL; 3275 signal_setting_t end_signal_setting = NULL; 3276 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3277 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 3278 if (begin_signal_setting != NULL) { 3279 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3280 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 3281 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 3282 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 3283 libjsig_is_loaded = true; 3284 assert(UseSignalChaining, "should enable signal-chaining"); 3285 } 3286 if (libjsig_is_loaded) { 3287 // Tell libjsig jvm is setting signal handlers 3288 (*begin_signal_setting)(); 3289 } 3290 3291 set_signal_handler(SIGSEGV, true); 3292 set_signal_handler(SIGPIPE, true); 3293 set_signal_handler(SIGBUS, true); 3294 set_signal_handler(SIGILL, true); 3295 set_signal_handler(SIGFPE, true); 3296 set_signal_handler(SIGXFSZ, true); 3297 3298 if (libjsig_is_loaded) { 3299 // Tell libjsig jvm finishes setting signal handlers 3300 (*end_signal_setting)(); 3301 } 3302 3303 // We don't activate signal checker if libjsig is in place, we trust ourselves 3304 // and if UserSignalHandler is installed all bets are off 3305 if (CheckJNICalls) { 3306 if (libjsig_is_loaded) { 3307 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 3308 check_signals = false; 3309 } 3310 if (AllowUserSignalHandlers) { 3311 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 3312 check_signals = false; 3313 } 3314 } 3315 } 3316} 3317 3318// This is the fastest way to get thread cpu time on Linux. 3319// Returns cpu time (user+sys) for any thread, not only for current. 3320// POSIX compliant clocks are implemented in the kernels 2.6.16+. 3321// It might work on 2.6.10+ with a special kernel/glibc patch. 3322// For reference, please, see IEEE Std 1003.1-2004: 3323// http://www.unix.org/single_unix_specification 3324 3325jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 3326 struct timespec tp; 3327 int rc = os::Linux::clock_gettime(clockid, &tp); 3328 assert(rc == 0, "clock_gettime is expected to return 0 code"); 3329 3330 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec; 3331} 3332 3333///// 3334// glibc on Linux platform uses non-documented flag 3335// to indicate, that some special sort of signal 3336// trampoline is used. 3337// We will never set this flag, and we should 3338// ignore this flag in our diagnostic 3339#ifdef SIGNIFICANT_SIGNAL_MASK 3340#undef SIGNIFICANT_SIGNAL_MASK 3341#endif 3342#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 3343 3344static const char* get_signal_handler_name(address handler, 3345 char* buf, int buflen) { 3346 int offset; 3347 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 3348 if (found) { 3349 // skip directory names 3350 const char *p1, *p2; 3351 p1 = buf; 3352 size_t len = strlen(os::file_separator()); 3353 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 3354 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 3355 } else { 3356 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 3357 } 3358 return buf; 3359} 3360 3361static void print_signal_handler(outputStream* st, int sig, 3362 char* buf, size_t buflen) { 3363 struct sigaction sa; 3364 3365 sigaction(sig, NULL, &sa); 3366 3367 // See comment for SIGNIFICANT_SIGNAL_MASK define 3368 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3369 3370 st->print("%s: ", os::exception_name(sig, buf, buflen)); 3371 3372 address handler = (sa.sa_flags & SA_SIGINFO) 3373 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 3374 : CAST_FROM_FN_PTR(address, sa.sa_handler); 3375 3376 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 3377 st->print("SIG_DFL"); 3378 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 3379 st->print("SIG_IGN"); 3380 } else { 3381 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 3382 } 3383 3384 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 3385 3386 address rh = VMError::get_resetted_sighandler(sig); 3387 // May be, handler was resetted by VMError? 3388 if(rh != NULL) { 3389 handler = rh; 3390 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 3391 } 3392 3393 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 3394 3395 // Check: is it our handler? 3396 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 3397 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 3398 // It is our signal handler 3399 // check for flags, reset system-used one! 3400 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 3401 st->print( 3402 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 3403 os::Linux::get_our_sigflags(sig)); 3404 } 3405 } 3406 st->cr(); 3407} 3408 3409 3410#define DO_SIGNAL_CHECK(sig) \ 3411 if (!sigismember(&check_signal_done, sig)) \ 3412 os::Linux::check_signal_handler(sig) 3413 3414// This method is a periodic task to check for misbehaving JNI applications 3415// under CheckJNI, we can add any periodic checks here 3416 3417void os::run_periodic_checks() { 3418 3419 if (check_signals == false) return; 3420 3421 // SEGV and BUS if overridden could potentially prevent 3422 // generation of hs*.log in the event of a crash, debugging 3423 // such a case can be very challenging, so we absolutely 3424 // check the following for a good measure: 3425 DO_SIGNAL_CHECK(SIGSEGV); 3426 DO_SIGNAL_CHECK(SIGILL); 3427 DO_SIGNAL_CHECK(SIGFPE); 3428 DO_SIGNAL_CHECK(SIGBUS); 3429 DO_SIGNAL_CHECK(SIGPIPE); 3430 DO_SIGNAL_CHECK(SIGXFSZ); 3431 3432 3433 // ReduceSignalUsage allows the user to override these handlers 3434 // see comments at the very top and jvm_solaris.h 3435 if (!ReduceSignalUsage) { 3436 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 3437 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 3438 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 3439 DO_SIGNAL_CHECK(BREAK_SIGNAL); 3440 } 3441 3442 DO_SIGNAL_CHECK(SR_signum); 3443 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 3444} 3445 3446typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 3447 3448static os_sigaction_t os_sigaction = NULL; 3449 3450void os::Linux::check_signal_handler(int sig) { 3451 char buf[O_BUFLEN]; 3452 address jvmHandler = NULL; 3453 3454 3455 struct sigaction act; 3456 if (os_sigaction == NULL) { 3457 // only trust the default sigaction, in case it has been interposed 3458 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 3459 if (os_sigaction == NULL) return; 3460 } 3461 3462 os_sigaction(sig, (struct sigaction*)NULL, &act); 3463 3464 3465 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3466 3467 address thisHandler = (act.sa_flags & SA_SIGINFO) 3468 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 3469 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 3470 3471 3472 switch(sig) { 3473 case SIGSEGV: 3474 case SIGBUS: 3475 case SIGFPE: 3476 case SIGPIPE: 3477 case SIGILL: 3478 case SIGXFSZ: 3479 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 3480 break; 3481 3482 case SHUTDOWN1_SIGNAL: 3483 case SHUTDOWN2_SIGNAL: 3484 case SHUTDOWN3_SIGNAL: 3485 case BREAK_SIGNAL: 3486 jvmHandler = (address)user_handler(); 3487 break; 3488 3489 case INTERRUPT_SIGNAL: 3490 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 3491 break; 3492 3493 default: 3494 if (sig == SR_signum) { 3495 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 3496 } else { 3497 return; 3498 } 3499 break; 3500 } 3501 3502 if (thisHandler != jvmHandler) { 3503 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 3504 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 3505 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 3506 // No need to check this sig any longer 3507 sigaddset(&check_signal_done, sig); 3508 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 3509 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 3510 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 3511 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 3512 // No need to check this sig any longer 3513 sigaddset(&check_signal_done, sig); 3514 } 3515 3516 // Dump all the signal 3517 if (sigismember(&check_signal_done, sig)) { 3518 print_signal_handlers(tty, buf, O_BUFLEN); 3519 } 3520} 3521 3522extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 3523 3524extern bool signal_name(int signo, char* buf, size_t len); 3525 3526const char* os::exception_name(int exception_code, char* buf, size_t size) { 3527 if (0 < exception_code && exception_code <= SIGRTMAX) { 3528 // signal 3529 if (!signal_name(exception_code, buf, size)) { 3530 jio_snprintf(buf, size, "SIG%d", exception_code); 3531 } 3532 return buf; 3533 } else { 3534 return NULL; 3535 } 3536} 3537 3538// this is called _before_ the most of global arguments have been parsed 3539void os::init(void) { 3540 char dummy; /* used to get a guess on initial stack address */ 3541// first_hrtime = gethrtime(); 3542 3543 // With LinuxThreads the JavaMain thread pid (primordial thread) 3544 // is different than the pid of the java launcher thread. 3545 // So, on Linux, the launcher thread pid is passed to the VM 3546 // via the sun.java.launcher.pid property. 3547 // Use this property instead of getpid() if it was correctly passed. 3548 // See bug 6351349. 3549 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 3550 3551 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 3552 3553 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 3554 3555 init_random(1234567); 3556 3557 ThreadCritical::initialize(); 3558 3559 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 3560 if (Linux::page_size() == -1) { 3561 fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno)); 3562 } 3563 init_page_sizes((size_t) Linux::page_size()); 3564 3565 Linux::initialize_system_info(); 3566 3567 // main_thread points to the aboriginal thread 3568 Linux::_main_thread = pthread_self(); 3569 3570 Linux::clock_init(); 3571 initial_time_count = os::elapsed_counter(); 3572} 3573 3574// To install functions for atexit system call 3575extern "C" { 3576 static void perfMemory_exit_helper() { 3577 perfMemory_exit(); 3578 } 3579} 3580 3581// this is called _after_ the global arguments have been parsed 3582jint os::init_2(void) 3583{ 3584 Linux::fast_thread_clock_init(); 3585 3586 // Allocate a single page and mark it as readable for safepoint polling 3587 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3588 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 3589 3590 os::set_polling_page( polling_page ); 3591 3592#ifndef PRODUCT 3593 if(Verbose && PrintMiscellaneous) 3594 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 3595#endif 3596 3597 if (!UseMembar) { 3598 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3599 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); 3600 os::set_memory_serialize_page( mem_serialize_page ); 3601 3602#ifndef PRODUCT 3603 if(Verbose && PrintMiscellaneous) 3604 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 3605#endif 3606 } 3607 3608 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init()); 3609 3610 // initialize suspend/resume support - must do this before signal_sets_init() 3611 if (SR_initialize() != 0) { 3612 perror("SR_initialize failed"); 3613 return JNI_ERR; 3614 } 3615 3616 Linux::signal_sets_init(); 3617 Linux::install_signal_handlers(); 3618 3619 size_t threadStackSizeInBytes = ThreadStackSize * K; 3620 if (threadStackSizeInBytes != 0 && 3621 threadStackSizeInBytes < Linux::min_stack_allowed) { 3622 tty->print_cr("\nThe stack size specified is too small, " 3623 "Specify at least %dk", 3624 Linux::min_stack_allowed / K); 3625 return JNI_ERR; 3626 } 3627 3628 // Make the stack size a multiple of the page size so that 3629 // the yellow/red zones can be guarded. 3630 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 3631 vm_page_size())); 3632 3633 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 3634 3635 Linux::libpthread_init(); 3636 if (PrintMiscellaneous && (Verbose || WizardMode)) { 3637 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 3638 Linux::glibc_version(), Linux::libpthread_version(), 3639 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 3640 } 3641 3642 if (UseNUMA) { 3643 Linux::libnuma_init(); 3644 } 3645 3646 if (MaxFDLimit) { 3647 // set the number of file descriptors to max. print out error 3648 // if getrlimit/setrlimit fails but continue regardless. 3649 struct rlimit nbr_files; 3650 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 3651 if (status != 0) { 3652 if (PrintMiscellaneous && (Verbose || WizardMode)) 3653 perror("os::init_2 getrlimit failed"); 3654 } else { 3655 nbr_files.rlim_cur = nbr_files.rlim_max; 3656 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 3657 if (status != 0) { 3658 if (PrintMiscellaneous && (Verbose || WizardMode)) 3659 perror("os::init_2 setrlimit failed"); 3660 } 3661 } 3662 } 3663 3664 // Initialize lock used to serialize thread creation (see os::create_thread) 3665 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 3666 3667 // Initialize HPI. 3668 jint hpi_result = hpi::initialize(); 3669 if (hpi_result != JNI_OK) { 3670 tty->print_cr("There was an error trying to initialize the HPI library."); 3671 return hpi_result; 3672 } 3673 3674 // at-exit methods are called in the reverse order of their registration. 3675 // atexit functions are called on return from main or as a result of a 3676 // call to exit(3C). There can be only 32 of these functions registered 3677 // and atexit() does not set errno. 3678 3679 if (PerfAllowAtExitRegistration) { 3680 // only register atexit functions if PerfAllowAtExitRegistration is set. 3681 // atexit functions can be delayed until process exit time, which 3682 // can be problematic for embedded VM situations. Embedded VMs should 3683 // call DestroyJavaVM() to assure that VM resources are released. 3684 3685 // note: perfMemory_exit_helper atexit function may be removed in 3686 // the future if the appropriate cleanup code can be added to the 3687 // VM_Exit VMOperation's doit method. 3688 if (atexit(perfMemory_exit_helper) != 0) { 3689 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 3690 } 3691 } 3692 3693 // initialize thread priority policy 3694 prio_init(); 3695 3696 return JNI_OK; 3697} 3698 3699// Mark the polling page as unreadable 3700void os::make_polling_page_unreadable(void) { 3701 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 3702 fatal("Could not disable polling page"); 3703}; 3704 3705// Mark the polling page as readable 3706void os::make_polling_page_readable(void) { 3707 if( !protect_memory((char *)_polling_page, Linux::page_size()) ) 3708 fatal("Could not enable polling page"); 3709}; 3710 3711int os::active_processor_count() { 3712 // Linux doesn't yet have a (official) notion of processor sets, 3713 // so just return the number of online processors. 3714 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 3715 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 3716 return online_cpus; 3717} 3718 3719bool os::distribute_processes(uint length, uint* distribution) { 3720 // Not yet implemented. 3721 return false; 3722} 3723 3724bool os::bind_to_processor(uint processor_id) { 3725 // Not yet implemented. 3726 return false; 3727} 3728 3729/// 3730 3731// Suspends the target using the signal mechanism and then grabs the PC before 3732// resuming the target. Used by the flat-profiler only 3733ExtendedPC os::get_thread_pc(Thread* thread) { 3734 // Make sure that it is called by the watcher for the VMThread 3735 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 3736 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 3737 3738 ExtendedPC epc; 3739 3740 OSThread* osthread = thread->osthread(); 3741 if (do_suspend(osthread)) { 3742 if (osthread->ucontext() != NULL) { 3743 epc = os::Linux::ucontext_get_pc(osthread->ucontext()); 3744 } else { 3745 // NULL context is unexpected, double-check this is the VMThread 3746 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 3747 } 3748 do_resume(osthread); 3749 } 3750 // failure means pthread_kill failed for some reason - arguably this is 3751 // a fatal problem, but such problems are ignored elsewhere 3752 3753 return epc; 3754} 3755 3756int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 3757{ 3758 if (is_NPTL()) { 3759 return pthread_cond_timedwait(_cond, _mutex, _abstime); 3760 } else { 3761#ifndef IA64 3762 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 3763 // word back to default 64bit precision if condvar is signaled. Java 3764 // wants 53bit precision. Save and restore current value. 3765 int fpu = get_fpu_control_word(); 3766#endif // IA64 3767 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 3768#ifndef IA64 3769 set_fpu_control_word(fpu); 3770#endif // IA64 3771 return status; 3772 } 3773} 3774 3775//////////////////////////////////////////////////////////////////////////////// 3776// debug support 3777 3778#ifndef PRODUCT 3779static address same_page(address x, address y) { 3780 int page_bits = -os::vm_page_size(); 3781 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) 3782 return x; 3783 else if (x > y) 3784 return (address)(intptr_t(y) | ~page_bits) + 1; 3785 else 3786 return (address)(intptr_t(y) & page_bits); 3787} 3788 3789bool os::find(address addr) { 3790 Dl_info dlinfo; 3791 memset(&dlinfo, 0, sizeof(dlinfo)); 3792 if (dladdr(addr, &dlinfo)) { 3793 tty->print(PTR_FORMAT ": ", addr); 3794 if (dlinfo.dli_sname != NULL) { 3795 tty->print("%s+%#x", dlinfo.dli_sname, 3796 addr - (intptr_t)dlinfo.dli_saddr); 3797 } else if (dlinfo.dli_fname) { 3798 tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 3799 } else { 3800 tty->print("<absolute address>"); 3801 } 3802 if (dlinfo.dli_fname) { 3803 tty->print(" in %s", dlinfo.dli_fname); 3804 } 3805 if (dlinfo.dli_fbase) { 3806 tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 3807 } 3808 tty->cr(); 3809 3810 if (Verbose) { 3811 // decode some bytes around the PC 3812 address begin = same_page(addr-40, addr); 3813 address end = same_page(addr+40, addr); 3814 address lowest = (address) dlinfo.dli_sname; 3815 if (!lowest) lowest = (address) dlinfo.dli_fbase; 3816 if (begin < lowest) begin = lowest; 3817 Dl_info dlinfo2; 3818 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr 3819 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 3820 end = (address) dlinfo2.dli_saddr; 3821 Disassembler::decode(begin, end); 3822 } 3823 return true; 3824 } 3825 return false; 3826} 3827 3828#endif 3829 3830//////////////////////////////////////////////////////////////////////////////// 3831// misc 3832 3833// This does not do anything on Linux. This is basically a hook for being 3834// able to use structured exception handling (thread-local exception filters) 3835// on, e.g., Win32. 3836void 3837os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 3838 JavaCallArguments* args, Thread* thread) { 3839 f(value, method, args, thread); 3840} 3841 3842void os::print_statistics() { 3843} 3844 3845int os::message_box(const char* title, const char* message) { 3846 int i; 3847 fdStream err(defaultStream::error_fd()); 3848 for (i = 0; i < 78; i++) err.print_raw("="); 3849 err.cr(); 3850 err.print_raw_cr(title); 3851 for (i = 0; i < 78; i++) err.print_raw("-"); 3852 err.cr(); 3853 err.print_raw_cr(message); 3854 for (i = 0; i < 78; i++) err.print_raw("="); 3855 err.cr(); 3856 3857 char buf[16]; 3858 // Prevent process from exiting upon "read error" without consuming all CPU 3859 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 3860 3861 return buf[0] == 'y' || buf[0] == 'Y'; 3862} 3863 3864int os::stat(const char *path, struct stat *sbuf) { 3865 char pathbuf[MAX_PATH]; 3866 if (strlen(path) > MAX_PATH - 1) { 3867 errno = ENAMETOOLONG; 3868 return -1; 3869 } 3870 hpi::native_path(strcpy(pathbuf, path)); 3871 return ::stat(pathbuf, sbuf); 3872} 3873 3874bool os::check_heap(bool force) { 3875 return true; 3876} 3877 3878int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 3879 return ::vsnprintf(buf, count, format, args); 3880} 3881 3882// Is a (classpath) directory empty? 3883bool os::dir_is_empty(const char* path) { 3884 DIR *dir = NULL; 3885 struct dirent *ptr; 3886 3887 dir = opendir(path); 3888 if (dir == NULL) return true; 3889 3890 /* Scan the directory */ 3891 bool result = true; 3892 char buf[sizeof(struct dirent) + MAX_PATH]; 3893 while (result && (ptr = ::readdir(dir)) != NULL) { 3894 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 3895 result = false; 3896 } 3897 } 3898 closedir(dir); 3899 return result; 3900} 3901 3902// create binary file, rewriting existing file if required 3903int os::create_binary_file(const char* path, bool rewrite_existing) { 3904 int oflags = O_WRONLY | O_CREAT; 3905 if (!rewrite_existing) { 3906 oflags |= O_EXCL; 3907 } 3908 return ::open64(path, oflags, S_IREAD | S_IWRITE); 3909} 3910 3911// return current position of file pointer 3912jlong os::current_file_offset(int fd) { 3913 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 3914} 3915 3916// move file pointer to the specified offset 3917jlong os::seek_to_file_offset(int fd, jlong offset) { 3918 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 3919} 3920 3921// Map a block of memory. 3922char* os::map_memory(int fd, const char* file_name, size_t file_offset, 3923 char *addr, size_t bytes, bool read_only, 3924 bool allow_exec) { 3925 int prot; 3926 int flags; 3927 3928 if (read_only) { 3929 prot = PROT_READ; 3930 flags = MAP_SHARED; 3931 } else { 3932 prot = PROT_READ | PROT_WRITE; 3933 flags = MAP_PRIVATE; 3934 } 3935 3936 if (allow_exec) { 3937 prot |= PROT_EXEC; 3938 } 3939 3940 if (addr != NULL) { 3941 flags |= MAP_FIXED; 3942 } 3943 3944 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 3945 fd, file_offset); 3946 if (mapped_address == MAP_FAILED) { 3947 return NULL; 3948 } 3949 return mapped_address; 3950} 3951 3952 3953// Remap a block of memory. 3954char* os::remap_memory(int fd, const char* file_name, size_t file_offset, 3955 char *addr, size_t bytes, bool read_only, 3956 bool allow_exec) { 3957 // same as map_memory() on this OS 3958 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 3959 allow_exec); 3960} 3961 3962 3963// Unmap a block of memory. 3964bool os::unmap_memory(char* addr, size_t bytes) { 3965 return munmap(addr, bytes) == 0; 3966} 3967 3968static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 3969 3970static clockid_t thread_cpu_clockid(Thread* thread) { 3971 pthread_t tid = thread->osthread()->pthread_id(); 3972 clockid_t clockid; 3973 3974 // Get thread clockid 3975 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 3976 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 3977 return clockid; 3978} 3979 3980// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 3981// are used by JVM M&M and JVMTI to get user+sys or user CPU time 3982// of a thread. 3983// 3984// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 3985// the fast estimate available on the platform. 3986 3987jlong os::current_thread_cpu_time() { 3988 if (os::Linux::supports_fast_thread_cpu_time()) { 3989 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 3990 } else { 3991 // return user + sys since the cost is the same 3992 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 3993 } 3994} 3995 3996jlong os::thread_cpu_time(Thread* thread) { 3997 // consistent with what current_thread_cpu_time() returns 3998 if (os::Linux::supports_fast_thread_cpu_time()) { 3999 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4000 } else { 4001 return slow_thread_cpu_time(thread, true /* user + sys */); 4002 } 4003} 4004 4005jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 4006 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4007 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4008 } else { 4009 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 4010 } 4011} 4012 4013jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4014 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4015 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4016 } else { 4017 return slow_thread_cpu_time(thread, user_sys_cpu_time); 4018 } 4019} 4020 4021// 4022// -1 on error. 4023// 4024 4025static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4026 static bool proc_pid_cpu_avail = true; 4027 static bool proc_task_unchecked = true; 4028 static const char *proc_stat_path = "/proc/%d/stat"; 4029 pid_t tid = thread->osthread()->thread_id(); 4030 int i; 4031 char *s; 4032 char stat[2048]; 4033 int statlen; 4034 char proc_name[64]; 4035 int count; 4036 long sys_time, user_time; 4037 char string[64]; 4038 int idummy; 4039 long ldummy; 4040 FILE *fp; 4041 4042 // We first try accessing /proc/<pid>/cpu since this is faster to 4043 // process. If this file is not present (linux kernels 2.5 and above) 4044 // then we open /proc/<pid>/stat. 4045 if ( proc_pid_cpu_avail ) { 4046 sprintf(proc_name, "/proc/%d/cpu", tid); 4047 fp = fopen(proc_name, "r"); 4048 if ( fp != NULL ) { 4049 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time); 4050 fclose(fp); 4051 if ( count != 3 ) return -1; 4052 4053 if (user_sys_cpu_time) { 4054 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4055 } else { 4056 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4057 } 4058 } 4059 else proc_pid_cpu_avail = false; 4060 } 4061 4062 // The /proc/<tid>/stat aggregates per-process usage on 4063 // new Linux kernels 2.6+ where NPTL is supported. 4064 // The /proc/self/task/<tid>/stat still has the per-thread usage. 4065 // See bug 6328462. 4066 // There can be no directory /proc/self/task on kernels 2.4 with NPTL 4067 // and possibly in some other cases, so we check its availability. 4068 if (proc_task_unchecked && os::Linux::is_NPTL()) { 4069 // This is executed only once 4070 proc_task_unchecked = false; 4071 fp = fopen("/proc/self/task", "r"); 4072 if (fp != NULL) { 4073 proc_stat_path = "/proc/self/task/%d/stat"; 4074 fclose(fp); 4075 } 4076 } 4077 4078 sprintf(proc_name, proc_stat_path, tid); 4079 fp = fopen(proc_name, "r"); 4080 if ( fp == NULL ) return -1; 4081 statlen = fread(stat, 1, 2047, fp); 4082 stat[statlen] = '\0'; 4083 fclose(fp); 4084 4085 // Skip pid and the command string. Note that we could be dealing with 4086 // weird command names, e.g. user could decide to rename java launcher 4087 // to "java 1.4.2 :)", then the stat file would look like 4088 // 1234 (java 1.4.2 :)) R ... ... 4089 // We don't really need to know the command string, just find the last 4090 // occurrence of ")" and then start parsing from there. See bug 4726580. 4091 s = strrchr(stat, ')'); 4092 i = 0; 4093 if (s == NULL ) return -1; 4094 4095 // Skip blank chars 4096 do s++; while (isspace(*s)); 4097 4098 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 4099 &idummy, &idummy, &idummy, &idummy, &idummy, &idummy, 4100 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 4101 &user_time, &sys_time); 4102 if ( count != 13 ) return -1; 4103 if (user_sys_cpu_time) { 4104 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4105 } else { 4106 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4107 } 4108} 4109 4110void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4111 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4112 info_ptr->may_skip_backward = false; // elapsed time not wall time 4113 info_ptr->may_skip_forward = false; // elapsed time not wall time 4114 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4115} 4116 4117void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4118 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4119 info_ptr->may_skip_backward = false; // elapsed time not wall time 4120 info_ptr->may_skip_forward = false; // elapsed time not wall time 4121 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4122} 4123 4124bool os::is_thread_cpu_time_supported() { 4125 return true; 4126} 4127 4128// System loadavg support. Returns -1 if load average cannot be obtained. 4129// Linux doesn't yet have a (official) notion of processor sets, 4130// so just return the system wide load average. 4131int os::loadavg(double loadavg[], int nelem) { 4132 return ::getloadavg(loadavg, nelem); 4133} 4134 4135void os::pause() { 4136 char filename[MAX_PATH]; 4137 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 4138 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 4139 } else { 4140 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 4141 } 4142 4143 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 4144 if (fd != -1) { 4145 struct stat buf; 4146 close(fd); 4147 while (::stat(filename, &buf) == 0) { 4148 (void)::poll(NULL, 0, 100); 4149 } 4150 } else { 4151 jio_fprintf(stderr, 4152 "Could not open pause file '%s', continuing immediately.\n", filename); 4153 } 4154} 4155 4156extern "C" { 4157 4158/** 4159 * NOTE: the following code is to keep the green threads code 4160 * in the libjava.so happy. Once the green threads is removed, 4161 * these code will no longer be needed. 4162 */ 4163int 4164jdk_waitpid(pid_t pid, int* status, int options) { 4165 return waitpid(pid, status, options); 4166} 4167 4168int 4169fork1() { 4170 return fork(); 4171} 4172 4173int 4174jdk_sem_init(sem_t *sem, int pshared, unsigned int value) { 4175 return sem_init(sem, pshared, value); 4176} 4177 4178int 4179jdk_sem_post(sem_t *sem) { 4180 return sem_post(sem); 4181} 4182 4183int 4184jdk_sem_wait(sem_t *sem) { 4185 return sem_wait(sem); 4186} 4187 4188int 4189jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) { 4190 return pthread_sigmask(how , newmask, oldmask); 4191} 4192 4193} 4194 4195// Refer to the comments in os_solaris.cpp park-unpark. 4196// 4197// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 4198// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 4199// For specifics regarding the bug see GLIBC BUGID 261237 : 4200// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 4201// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 4202// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 4203// is used. (The simple C test-case provided in the GLIBC bug report manifests the 4204// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 4205// and monitorenter when we're using 1-0 locking. All those operations may result in 4206// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 4207// of libpthread avoids the problem, but isn't practical. 4208// 4209// Possible remedies: 4210// 4211// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 4212// This is palliative and probabilistic, however. If the thread is preempted 4213// between the call to compute_abstime() and pthread_cond_timedwait(), more 4214// than the minimum period may have passed, and the abstime may be stale (in the 4215// past) resultin in a hang. Using this technique reduces the odds of a hang 4216// but the JVM is still vulnerable, particularly on heavily loaded systems. 4217// 4218// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 4219// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 4220// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 4221// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 4222// thread. 4223// 4224// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 4225// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 4226// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 4227// This also works well. In fact it avoids kernel-level scalability impediments 4228// on certain platforms that don't handle lots of active pthread_cond_timedwait() 4229// timers in a graceful fashion. 4230// 4231// 4. When the abstime value is in the past it appears that control returns 4232// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 4233// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 4234// can avoid the problem by reinitializing the condvar -- by cond_destroy() 4235// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 4236// It may be possible to avoid reinitialization by checking the return 4237// value from pthread_cond_timedwait(). In addition to reinitializing the 4238// condvar we must establish the invariant that cond_signal() is only called 4239// within critical sections protected by the adjunct mutex. This prevents 4240// cond_signal() from "seeing" a condvar that's in the midst of being 4241// reinitialized or that is corrupt. Sadly, this invariant obviates the 4242// desirable signal-after-unlock optimization that avoids futile context switching. 4243// 4244// I'm also concerned that some versions of NTPL might allocate an auxilliary 4245// structure when a condvar is used or initialized. cond_destroy() would 4246// release the helper structure. Our reinitialize-after-timedwait fix 4247// put excessive stress on malloc/free and locks protecting the c-heap. 4248// 4249// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 4250// It may be possible to refine (4) by checking the kernel and NTPL verisons 4251// and only enabling the work-around for vulnerable environments. 4252 4253// utility to compute the abstime argument to timedwait: 4254// millis is the relative timeout time 4255// abstime will be the absolute timeout time 4256// TODO: replace compute_abstime() with unpackTime() 4257 4258static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 4259 if (millis < 0) millis = 0; 4260 struct timeval now; 4261 int status = gettimeofday(&now, NULL); 4262 assert(status == 0, "gettimeofday"); 4263 jlong seconds = millis / 1000; 4264 millis %= 1000; 4265 if (seconds > 50000000) { // see man cond_timedwait(3T) 4266 seconds = 50000000; 4267 } 4268 abstime->tv_sec = now.tv_sec + seconds; 4269 long usec = now.tv_usec + millis * 1000; 4270 if (usec >= 1000000) { 4271 abstime->tv_sec += 1; 4272 usec -= 1000000; 4273 } 4274 abstime->tv_nsec = usec * 1000; 4275 return abstime; 4276} 4277 4278 4279// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 4280// Conceptually TryPark() should be equivalent to park(0). 4281 4282int os::PlatformEvent::TryPark() { 4283 for (;;) { 4284 const int v = _Event ; 4285 guarantee ((v == 0) || (v == 1), "invariant") ; 4286 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 4287 } 4288} 4289 4290void os::PlatformEvent::park() { // AKA "down()" 4291 // Invariant: Only the thread associated with the Event/PlatformEvent 4292 // may call park(). 4293 // TODO: assert that _Assoc != NULL or _Assoc == Self 4294 int v ; 4295 for (;;) { 4296 v = _Event ; 4297 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4298 } 4299 guarantee (v >= 0, "invariant") ; 4300 if (v == 0) { 4301 // Do this the hard way by blocking ... 4302 int status = pthread_mutex_lock(_mutex); 4303 assert_status(status == 0, status, "mutex_lock"); 4304 guarantee (_nParked == 0, "invariant") ; 4305 ++ _nParked ; 4306 while (_Event < 0) { 4307 status = pthread_cond_wait(_cond, _mutex); 4308 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 4309 // Treat this the same as if the wait was interrupted 4310 if (status == ETIME) { status = EINTR; } 4311 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 4312 } 4313 -- _nParked ; 4314 4315 // In theory we could move the ST of 0 into _Event past the unlock(), 4316 // but then we'd need a MEMBAR after the ST. 4317 _Event = 0 ; 4318 status = pthread_mutex_unlock(_mutex); 4319 assert_status(status == 0, status, "mutex_unlock"); 4320 } 4321 guarantee (_Event >= 0, "invariant") ; 4322} 4323 4324int os::PlatformEvent::park(jlong millis) { 4325 guarantee (_nParked == 0, "invariant") ; 4326 4327 int v ; 4328 for (;;) { 4329 v = _Event ; 4330 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4331 } 4332 guarantee (v >= 0, "invariant") ; 4333 if (v != 0) return OS_OK ; 4334 4335 // We do this the hard way, by blocking the thread. 4336 // Consider enforcing a minimum timeout value. 4337 struct timespec abst; 4338 compute_abstime(&abst, millis); 4339 4340 int ret = OS_TIMEOUT; 4341 int status = pthread_mutex_lock(_mutex); 4342 assert_status(status == 0, status, "mutex_lock"); 4343 guarantee (_nParked == 0, "invariant") ; 4344 ++_nParked ; 4345 4346 // Object.wait(timo) will return because of 4347 // (a) notification 4348 // (b) timeout 4349 // (c) thread.interrupt 4350 // 4351 // Thread.interrupt and object.notify{All} both call Event::set. 4352 // That is, we treat thread.interrupt as a special case of notification. 4353 // The underlying Solaris implementation, cond_timedwait, admits 4354 // spurious/premature wakeups, but the JLS/JVM spec prevents the 4355 // JVM from making those visible to Java code. As such, we must 4356 // filter out spurious wakeups. We assume all ETIME returns are valid. 4357 // 4358 // TODO: properly differentiate simultaneous notify+interrupt. 4359 // In that case, we should propagate the notify to another waiter. 4360 4361 while (_Event < 0) { 4362 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 4363 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4364 pthread_cond_destroy (_cond); 4365 pthread_cond_init (_cond, NULL) ; 4366 } 4367 assert_status(status == 0 || status == EINTR || 4368 status == ETIME || status == ETIMEDOUT, 4369 status, "cond_timedwait"); 4370 if (!FilterSpuriousWakeups) break ; // previous semantics 4371 if (status == ETIME || status == ETIMEDOUT) break ; 4372 // We consume and ignore EINTR and spurious wakeups. 4373 } 4374 --_nParked ; 4375 if (_Event >= 0) { 4376 ret = OS_OK; 4377 } 4378 _Event = 0 ; 4379 status = pthread_mutex_unlock(_mutex); 4380 assert_status(status == 0, status, "mutex_unlock"); 4381 assert (_nParked == 0, "invariant") ; 4382 return ret; 4383} 4384 4385void os::PlatformEvent::unpark() { 4386 int v, AnyWaiters ; 4387 for (;;) { 4388 v = _Event ; 4389 if (v > 0) { 4390 // The LD of _Event could have reordered or be satisfied 4391 // by a read-aside from this processor's write buffer. 4392 // To avoid problems execute a barrier and then 4393 // ratify the value. 4394 OrderAccess::fence() ; 4395 if (_Event == v) return ; 4396 continue ; 4397 } 4398 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ; 4399 } 4400 if (v < 0) { 4401 // Wait for the thread associated with the event to vacate 4402 int status = pthread_mutex_lock(_mutex); 4403 assert_status(status == 0, status, "mutex_lock"); 4404 AnyWaiters = _nParked ; 4405 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ; 4406 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 4407 AnyWaiters = 0 ; 4408 pthread_cond_signal (_cond); 4409 } 4410 status = pthread_mutex_unlock(_mutex); 4411 assert_status(status == 0, status, "mutex_unlock"); 4412 if (AnyWaiters != 0) { 4413 status = pthread_cond_signal(_cond); 4414 assert_status(status == 0, status, "cond_signal"); 4415 } 4416 } 4417 4418 // Note that we signal() _after dropping the lock for "immortal" Events. 4419 // This is safe and avoids a common class of futile wakeups. In rare 4420 // circumstances this can cause a thread to return prematurely from 4421 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 4422 // simply re-test the condition and re-park itself. 4423} 4424 4425 4426// JSR166 4427// ------------------------------------------------------- 4428 4429/* 4430 * The solaris and linux implementations of park/unpark are fairly 4431 * conservative for now, but can be improved. They currently use a 4432 * mutex/condvar pair, plus a a count. 4433 * Park decrements count if > 0, else does a condvar wait. Unpark 4434 * sets count to 1 and signals condvar. Only one thread ever waits 4435 * on the condvar. Contention seen when trying to park implies that someone 4436 * is unparking you, so don't wait. And spurious returns are fine, so there 4437 * is no need to track notifications. 4438 */ 4439 4440 4441#define NANOSECS_PER_SEC 1000000000 4442#define NANOSECS_PER_MILLISEC 1000000 4443#define MAX_SECS 100000000 4444/* 4445 * This code is common to linux and solaris and will be moved to a 4446 * common place in dolphin. 4447 * 4448 * The passed in time value is either a relative time in nanoseconds 4449 * or an absolute time in milliseconds. Either way it has to be unpacked 4450 * into suitable seconds and nanoseconds components and stored in the 4451 * given timespec structure. 4452 * Given time is a 64-bit value and the time_t used in the timespec is only 4453 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 4454 * overflow if times way in the future are given. Further on Solaris versions 4455 * prior to 10 there is a restriction (see cond_timedwait) that the specified 4456 * number of seconds, in abstime, is less than current_time + 100,000,000. 4457 * As it will be 28 years before "now + 100000000" will overflow we can 4458 * ignore overflow and just impose a hard-limit on seconds using the value 4459 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 4460 * years from "now". 4461 */ 4462 4463static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 4464 assert (time > 0, "convertTime"); 4465 4466 struct timeval now; 4467 int status = gettimeofday(&now, NULL); 4468 assert(status == 0, "gettimeofday"); 4469 4470 time_t max_secs = now.tv_sec + MAX_SECS; 4471 4472 if (isAbsolute) { 4473 jlong secs = time / 1000; 4474 if (secs > max_secs) { 4475 absTime->tv_sec = max_secs; 4476 } 4477 else { 4478 absTime->tv_sec = secs; 4479 } 4480 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 4481 } 4482 else { 4483 jlong secs = time / NANOSECS_PER_SEC; 4484 if (secs >= MAX_SECS) { 4485 absTime->tv_sec = max_secs; 4486 absTime->tv_nsec = 0; 4487 } 4488 else { 4489 absTime->tv_sec = now.tv_sec + secs; 4490 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 4491 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 4492 absTime->tv_nsec -= NANOSECS_PER_SEC; 4493 ++absTime->tv_sec; // note: this must be <= max_secs 4494 } 4495 } 4496 } 4497 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 4498 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 4499 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 4500 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 4501} 4502 4503void Parker::park(bool isAbsolute, jlong time) { 4504 // Optional fast-path check: 4505 // Return immediately if a permit is available. 4506 if (_counter > 0) { 4507 _counter = 0 ; 4508 return ; 4509 } 4510 4511 Thread* thread = Thread::current(); 4512 assert(thread->is_Java_thread(), "Must be JavaThread"); 4513 JavaThread *jt = (JavaThread *)thread; 4514 4515 // Optional optimization -- avoid state transitions if there's an interrupt pending. 4516 // Check interrupt before trying to wait 4517 if (Thread::is_interrupted(thread, false)) { 4518 return; 4519 } 4520 4521 // Next, demultiplex/decode time arguments 4522 timespec absTime; 4523 if (time < 0) { // don't wait at all 4524 return; 4525 } 4526 if (time > 0) { 4527 unpackTime(&absTime, isAbsolute, time); 4528 } 4529 4530 4531 // Enter safepoint region 4532 // Beware of deadlocks such as 6317397. 4533 // The per-thread Parker:: mutex is a classic leaf-lock. 4534 // In particular a thread must never block on the Threads_lock while 4535 // holding the Parker:: mutex. If safepoints are pending both the 4536 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 4537 ThreadBlockInVM tbivm(jt); 4538 4539 // Don't wait if cannot get lock since interference arises from 4540 // unblocking. Also. check interrupt before trying wait 4541 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 4542 return; 4543 } 4544 4545 int status ; 4546 if (_counter > 0) { // no wait needed 4547 _counter = 0; 4548 status = pthread_mutex_unlock(_mutex); 4549 assert (status == 0, "invariant") ; 4550 return; 4551 } 4552 4553#ifdef ASSERT 4554 // Don't catch signals while blocked; let the running threads have the signals. 4555 // (This allows a debugger to break into the running thread.) 4556 sigset_t oldsigs; 4557 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 4558 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 4559#endif 4560 4561 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 4562 jt->set_suspend_equivalent(); 4563 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 4564 4565 if (time == 0) { 4566 status = pthread_cond_wait (_cond, _mutex) ; 4567 } else { 4568 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 4569 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4570 pthread_cond_destroy (_cond) ; 4571 pthread_cond_init (_cond, NULL); 4572 } 4573 } 4574 assert_status(status == 0 || status == EINTR || 4575 status == ETIME || status == ETIMEDOUT, 4576 status, "cond_timedwait"); 4577 4578#ifdef ASSERT 4579 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 4580#endif 4581 4582 _counter = 0 ; 4583 status = pthread_mutex_unlock(_mutex) ; 4584 assert_status(status == 0, status, "invariant") ; 4585 // If externally suspended while waiting, re-suspend 4586 if (jt->handle_special_suspend_equivalent_condition()) { 4587 jt->java_suspend_self(); 4588 } 4589 4590} 4591 4592void Parker::unpark() { 4593 int s, status ; 4594 status = pthread_mutex_lock(_mutex); 4595 assert (status == 0, "invariant") ; 4596 s = _counter; 4597 _counter = 1; 4598 if (s < 1) { 4599 if (WorkAroundNPTLTimedWaitHang) { 4600 status = pthread_cond_signal (_cond) ; 4601 assert (status == 0, "invariant") ; 4602 status = pthread_mutex_unlock(_mutex); 4603 assert (status == 0, "invariant") ; 4604 } else { 4605 status = pthread_mutex_unlock(_mutex); 4606 assert (status == 0, "invariant") ; 4607 status = pthread_cond_signal (_cond) ; 4608 assert (status == 0, "invariant") ; 4609 } 4610 } else { 4611 pthread_mutex_unlock(_mutex); 4612 assert (status == 0, "invariant") ; 4613 } 4614} 4615 4616 4617extern char** environ; 4618 4619#ifndef __NR_fork 4620#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 4621#endif 4622 4623#ifndef __NR_execve 4624#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 4625#endif 4626 4627// Run the specified command in a separate process. Return its exit value, 4628// or -1 on failure (e.g. can't fork a new process). 4629// Unlike system(), this function can be called from signal handler. It 4630// doesn't block SIGINT et al. 4631int os::fork_and_exec(char* cmd) { 4632 const char * argv[4] = {"sh", "-c", cmd, NULL}; 4633 4634 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 4635 // pthread_atfork handlers and reset pthread library. All we need is a 4636 // separate process to execve. Make a direct syscall to fork process. 4637 // On IA64 there's no fork syscall, we have to use fork() and hope for 4638 // the best... 4639 pid_t pid = NOT_IA64(syscall(__NR_fork);) 4640 IA64_ONLY(fork();) 4641 4642 if (pid < 0) { 4643 // fork failed 4644 return -1; 4645 4646 } else if (pid == 0) { 4647 // child process 4648 4649 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 4650 // first to kill every thread on the thread list. Because this list is 4651 // not reset by fork() (see notes above), execve() will instead kill 4652 // every thread in the parent process. We know this is the only thread 4653 // in the new process, so make a system call directly. 4654 // IA64 should use normal execve() from glibc to match the glibc fork() 4655 // above. 4656 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 4657 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 4658 4659 // execve failed 4660 _exit(-1); 4661 4662 } else { 4663 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 4664 // care about the actual exit code, for now. 4665 4666 int status; 4667 4668 // Wait for the child process to exit. This returns immediately if 4669 // the child has already exited. */ 4670 while (waitpid(pid, &status, 0) < 0) { 4671 switch (errno) { 4672 case ECHILD: return 0; 4673 case EINTR: break; 4674 default: return -1; 4675 } 4676 } 4677 4678 if (WIFEXITED(status)) { 4679 // The child exited normally; get its exit code. 4680 return WEXITSTATUS(status); 4681 } else if (WIFSIGNALED(status)) { 4682 // The child exited because of a signal 4683 // The best value to return is 0x80 + signal number, 4684 // because that is what all Unix shells do, and because 4685 // it allows callers to distinguish between process exit and 4686 // process death by signal. 4687 return 0x80 + WTERMSIG(status); 4688 } else { 4689 // Unknown exit code; pass it through 4690 return status; 4691 } 4692 } 4693} 4694