os_linux.cpp revision 270:b7f01ad69d30
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/* Used to protect dlsym() calls */ 98static pthread_mutex_t dl_mutex; 99 100//////////////////////////////////////////////////////////////////////////////// 101// utility functions 102 103static int SR_initialize(); 104static int SR_finalize(); 105 106julong os::available_memory() { 107 return Linux::available_memory(); 108} 109 110julong os::Linux::available_memory() { 111 // values in struct sysinfo are "unsigned long" 112 struct sysinfo si; 113 sysinfo(&si); 114 115 return (julong)si.freeram * si.mem_unit; 116} 117 118julong os::physical_memory() { 119 return Linux::physical_memory(); 120} 121 122julong os::allocatable_physical_memory(julong size) { 123#ifdef _LP64 124 return size; 125#else 126 julong result = MIN2(size, (julong)3800*M); 127 if (!is_allocatable(result)) { 128 // See comments under solaris for alignment considerations 129 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); 130 result = MIN2(size, reasonable_size); 131 } 132 return result; 133#endif // _LP64 134} 135 136//////////////////////////////////////////////////////////////////////////////// 137// environment support 138 139bool os::getenv(const char* name, char* buf, int len) { 140 const char* val = ::getenv(name); 141 if (val != NULL && strlen(val) < (size_t)len) { 142 strcpy(buf, val); 143 return true; 144 } 145 if (len > 0) buf[0] = 0; // return a null string 146 return false; 147} 148 149 150// Return true if user is running as root. 151 152bool os::have_special_privileges() { 153 static bool init = false; 154 static bool privileges = false; 155 if (!init) { 156 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 157 init = true; 158 } 159 return privileges; 160} 161 162 163#ifndef SYS_gettid 164// i386: 224, ia64: 1105, amd64: 186, sparc 143 165#ifdef __ia64__ 166#define SYS_gettid 1105 167#elif __i386__ 168#define SYS_gettid 224 169#elif __amd64__ 170#define SYS_gettid 186 171#elif __sparc__ 172#define SYS_gettid 143 173#else 174#error define gettid for the arch 175#endif 176#endif 177 178// Cpu architecture string 179#if defined(IA64) 180static char cpu_arch[] = "ia64"; 181#elif defined(IA32) 182static char cpu_arch[] = "i386"; 183#elif defined(AMD64) 184static char cpu_arch[] = "amd64"; 185#elif defined(SPARC) 186# ifdef _LP64 187static char cpu_arch[] = "sparcv9"; 188# else 189static char cpu_arch[] = "sparc"; 190# endif 191#else 192#error Add appropriate cpu_arch setting 193#endif 194 195 196// pid_t gettid() 197// 198// Returns the kernel thread id of the currently running thread. Kernel 199// thread id is used to access /proc. 200// 201// (Note that getpid() on LinuxThreads returns kernel thread id too; but 202// on NPTL, it returns the same pid for all threads, as required by POSIX.) 203// 204pid_t os::Linux::gettid() { 205 int rslt = syscall(SYS_gettid); 206 if (rslt == -1) { 207 // old kernel, no NPTL support 208 return getpid(); 209 } else { 210 return (pid_t)rslt; 211 } 212} 213 214// Most versions of linux have a bug where the number of processors are 215// determined by looking at the /proc file system. In a chroot environment, 216// the system call returns 1. This causes the VM to act as if it is 217// a single processor and elide locking (see is_MP() call). 218static bool unsafe_chroot_detected = false; 219static const char *unstable_chroot_error = "/proc file system not found.\n" 220 "Java may be unstable running multithreaded in a chroot " 221 "environment on Linux when /proc filesystem is not mounted."; 222 223void os::Linux::initialize_system_info() { 224 _processor_count = sysconf(_SC_NPROCESSORS_CONF); 225 if (_processor_count == 1) { 226 pid_t pid = os::Linux::gettid(); 227 char fname[32]; 228 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 229 FILE *fp = fopen(fname, "r"); 230 if (fp == NULL) { 231 unsafe_chroot_detected = true; 232 } else { 233 fclose(fp); 234 } 235 } 236 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 237 assert(_processor_count > 0, "linux error"); 238} 239 240void os::init_system_properties_values() { 241// char arch[12]; 242// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 243 244 // The next steps are taken in the product version: 245 // 246 // Obtain the JAVA_HOME value from the location of libjvm[_g].so. 247 // This library should be located at: 248 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so. 249 // 250 // If "/jre/lib/" appears at the right place in the path, then we 251 // assume libjvm[_g].so is installed in a JDK and we use this path. 252 // 253 // Otherwise exit with message: "Could not create the Java virtual machine." 254 // 255 // The following extra steps are taken in the debugging version: 256 // 257 // If "/jre/lib/" does NOT appear at the right place in the path 258 // instead of exit check for $JAVA_HOME environment variable. 259 // 260 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 261 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so 262 // it looks like libjvm[_g].so is installed there 263 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so. 264 // 265 // Otherwise exit. 266 // 267 // Important note: if the location of libjvm.so changes this 268 // code needs to be changed accordingly. 269 270 // The next few definitions allow the code to be verbatim: 271#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n)) 272#define getenv(n) ::getenv(n) 273 274/* 275 * See ld(1): 276 * The linker uses the following search paths to locate required 277 * shared libraries: 278 * 1: ... 279 * ... 280 * 7: The default directories, normally /lib and /usr/lib. 281 */ 282#define DEFAULT_LIBPATH "/lib:/usr/lib" 283 284#define EXTENSIONS_DIR "/lib/ext" 285#define ENDORSED_DIR "/lib/endorsed" 286#define REG_DIR "/usr/java/packages" 287 288 { 289 /* sysclasspath, java_home, dll_dir */ 290 { 291 char *home_path; 292 char *dll_path; 293 char *pslash; 294 char buf[MAXPATHLEN]; 295 os::jvm_path(buf, sizeof(buf)); 296 297 // Found the full path to libjvm.so. 298 // Now cut the path to <java_home>/jre if we can. 299 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 300 pslash = strrchr(buf, '/'); 301 if (pslash != NULL) 302 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 303 dll_path = malloc(strlen(buf) + 1); 304 if (dll_path == NULL) 305 return; 306 strcpy(dll_path, buf); 307 Arguments::set_dll_dir(dll_path); 308 309 if (pslash != NULL) { 310 pslash = strrchr(buf, '/'); 311 if (pslash != NULL) { 312 *pslash = '\0'; /* get rid of /<arch> */ 313 pslash = strrchr(buf, '/'); 314 if (pslash != NULL) 315 *pslash = '\0'; /* get rid of /lib */ 316 } 317 } 318 319 home_path = malloc(strlen(buf) + 1); 320 if (home_path == NULL) 321 return; 322 strcpy(home_path, buf); 323 Arguments::set_java_home(home_path); 324 325 if (!set_boot_path('/', ':')) 326 return; 327 } 328 329 /* 330 * Where to look for native libraries 331 * 332 * Note: Due to a legacy implementation, most of the library path 333 * is set in the launcher. This was to accomodate linking restrictions 334 * on legacy Linux implementations (which are no longer supported). 335 * Eventually, all the library path setting will be done here. 336 * 337 * However, to prevent the proliferation of improperly built native 338 * libraries, the new path component /usr/java/packages is added here. 339 * Eventually, all the library path setting will be done here. 340 */ 341 { 342 char *ld_library_path; 343 344 /* 345 * Construct the invariant part of ld_library_path. Note that the 346 * space for the colon and the trailing null are provided by the 347 * nulls included by the sizeof operator (so actually we allocate 348 * a byte more than necessary). 349 */ 350 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 351 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 352 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 353 354 /* 355 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 356 * should always exist (until the legacy problem cited above is 357 * addressed). 358 */ 359 char *v = getenv("LD_LIBRARY_PATH"); 360 if (v != NULL) { 361 char *t = ld_library_path; 362 /* That's +1 for the colon and +1 for the trailing '\0' */ 363 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 364 sprintf(ld_library_path, "%s:%s", v, t); 365 } 366 Arguments::set_library_path(ld_library_path); 367 } 368 369 /* 370 * Extensions directories. 371 * 372 * Note that the space for the colon and the trailing null are provided 373 * by the nulls included by the sizeof operator (so actually one byte more 374 * than necessary is allocated). 375 */ 376 { 377 char *buf = malloc(strlen(Arguments::get_java_home()) + 378 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 379 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 380 Arguments::get_java_home()); 381 Arguments::set_ext_dirs(buf); 382 } 383 384 /* Endorsed standards default directory. */ 385 { 386 char * buf; 387 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 388 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 389 Arguments::set_endorsed_dirs(buf); 390 } 391 } 392 393#undef malloc 394#undef getenv 395#undef EXTENSIONS_DIR 396#undef ENDORSED_DIR 397 398 // Done 399 return; 400} 401 402//////////////////////////////////////////////////////////////////////////////// 403// breakpoint support 404 405void os::breakpoint() { 406 BREAKPOINT; 407} 408 409extern "C" void breakpoint() { 410 // use debugger to set breakpoint here 411} 412 413//////////////////////////////////////////////////////////////////////////////// 414// signal support 415 416debug_only(static bool signal_sets_initialized = false); 417static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 418 419bool os::Linux::is_sig_ignored(int sig) { 420 struct sigaction oact; 421 sigaction(sig, (struct sigaction*)NULL, &oact); 422 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 423 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 424 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 425 return true; 426 else 427 return false; 428} 429 430void os::Linux::signal_sets_init() { 431 // Should also have an assertion stating we are still single-threaded. 432 assert(!signal_sets_initialized, "Already initialized"); 433 // Fill in signals that are necessarily unblocked for all threads in 434 // the VM. Currently, we unblock the following signals: 435 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 436 // by -Xrs (=ReduceSignalUsage)); 437 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 438 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 439 // the dispositions or masks wrt these signals. 440 // Programs embedding the VM that want to use the above signals for their 441 // own purposes must, at this time, use the "-Xrs" option to prevent 442 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 443 // (See bug 4345157, and other related bugs). 444 // In reality, though, unblocking these signals is really a nop, since 445 // these signals are not blocked by default. 446 sigemptyset(&unblocked_sigs); 447 sigemptyset(&allowdebug_blocked_sigs); 448 sigaddset(&unblocked_sigs, SIGILL); 449 sigaddset(&unblocked_sigs, SIGSEGV); 450 sigaddset(&unblocked_sigs, SIGBUS); 451 sigaddset(&unblocked_sigs, SIGFPE); 452 sigaddset(&unblocked_sigs, SR_signum); 453 454 if (!ReduceSignalUsage) { 455 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 456 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 457 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 458 } 459 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 460 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 461 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 462 } 463 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 464 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 465 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 466 } 467 } 468 // Fill in signals that are blocked by all but the VM thread. 469 sigemptyset(&vm_sigs); 470 if (!ReduceSignalUsage) 471 sigaddset(&vm_sigs, BREAK_SIGNAL); 472 debug_only(signal_sets_initialized = true); 473 474} 475 476// These are signals that are unblocked while a thread is running Java. 477// (For some reason, they get blocked by default.) 478sigset_t* os::Linux::unblocked_signals() { 479 assert(signal_sets_initialized, "Not initialized"); 480 return &unblocked_sigs; 481} 482 483// These are the signals that are blocked while a (non-VM) thread is 484// running Java. Only the VM thread handles these signals. 485sigset_t* os::Linux::vm_signals() { 486 assert(signal_sets_initialized, "Not initialized"); 487 return &vm_sigs; 488} 489 490// These are signals that are blocked during cond_wait to allow debugger in 491sigset_t* os::Linux::allowdebug_blocked_signals() { 492 assert(signal_sets_initialized, "Not initialized"); 493 return &allowdebug_blocked_sigs; 494} 495 496void os::Linux::hotspot_sigmask(Thread* thread) { 497 498 //Save caller's signal mask before setting VM signal mask 499 sigset_t caller_sigmask; 500 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 501 502 OSThread* osthread = thread->osthread(); 503 osthread->set_caller_sigmask(caller_sigmask); 504 505 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 506 507 if (!ReduceSignalUsage) { 508 if (thread->is_VM_thread()) { 509 // Only the VM thread handles BREAK_SIGNAL ... 510 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 511 } else { 512 // ... all other threads block BREAK_SIGNAL 513 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 514 } 515 } 516} 517 518////////////////////////////////////////////////////////////////////////////// 519// detecting pthread library 520 521void os::Linux::libpthread_init() { 522 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 523 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 524 // generic name for earlier versions. 525 // Define macros here so we can build HotSpot on old systems. 526# ifndef _CS_GNU_LIBC_VERSION 527# define _CS_GNU_LIBC_VERSION 2 528# endif 529# ifndef _CS_GNU_LIBPTHREAD_VERSION 530# define _CS_GNU_LIBPTHREAD_VERSION 3 531# endif 532 533 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 534 if (n > 0) { 535 char *str = (char *)malloc(n); 536 confstr(_CS_GNU_LIBC_VERSION, str, n); 537 os::Linux::set_glibc_version(str); 538 } else { 539 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 540 static char _gnu_libc_version[32]; 541 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 542 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 543 os::Linux::set_glibc_version(_gnu_libc_version); 544 } 545 546 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 547 if (n > 0) { 548 char *str = (char *)malloc(n); 549 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 550 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 551 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 552 // is the case. LinuxThreads has a hard limit on max number of threads. 553 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 554 // On the other hand, NPTL does not have such a limit, sysconf() 555 // will return -1 and errno is not changed. Check if it is really NPTL. 556 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 557 strstr(str, "NPTL") && 558 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 559 free(str); 560 os::Linux::set_libpthread_version("linuxthreads"); 561 } else { 562 os::Linux::set_libpthread_version(str); 563 } 564 } else { 565 // glibc before 2.3.2 only has LinuxThreads. 566 os::Linux::set_libpthread_version("linuxthreads"); 567 } 568 569 if (strstr(libpthread_version(), "NPTL")) { 570 os::Linux::set_is_NPTL(); 571 } else { 572 os::Linux::set_is_LinuxThreads(); 573 } 574 575 // LinuxThreads have two flavors: floating-stack mode, which allows variable 576 // stack size; and fixed-stack mode. NPTL is always floating-stack. 577 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 578 os::Linux::set_is_floating_stack(); 579 } 580} 581 582///////////////////////////////////////////////////////////////////////////// 583// thread stack 584 585// Force Linux kernel to expand current thread stack. If "bottom" is close 586// to the stack guard, caller should block all signals. 587// 588// MAP_GROWSDOWN: 589// A special mmap() flag that is used to implement thread stacks. It tells 590// kernel that the memory region should extend downwards when needed. This 591// allows early versions of LinuxThreads to only mmap the first few pages 592// when creating a new thread. Linux kernel will automatically expand thread 593// stack as needed (on page faults). 594// 595// However, because the memory region of a MAP_GROWSDOWN stack can grow on 596// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 597// region, it's hard to tell if the fault is due to a legitimate stack 598// access or because of reading/writing non-exist memory (e.g. buffer 599// overrun). As a rule, if the fault happens below current stack pointer, 600// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 601// application (see Linux kernel fault.c). 602// 603// This Linux feature can cause SIGSEGV when VM bangs thread stack for 604// stack overflow detection. 605// 606// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 607// not use this flag. However, the stack of initial thread is not created 608// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 609// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 610// and then attach the thread to JVM. 611// 612// To get around the problem and allow stack banging on Linux, we need to 613// manually expand thread stack after receiving the SIGSEGV. 614// 615// There are two ways to expand thread stack to address "bottom", we used 616// both of them in JVM before 1.5: 617// 1. adjust stack pointer first so that it is below "bottom", and then 618// touch "bottom" 619// 2. mmap() the page in question 620// 621// Now alternate signal stack is gone, it's harder to use 2. For instance, 622// if current sp is already near the lower end of page 101, and we need to 623// call mmap() to map page 100, it is possible that part of the mmap() frame 624// will be placed in page 100. When page 100 is mapped, it is zero-filled. 625// That will destroy the mmap() frame and cause VM to crash. 626// 627// The following code works by adjusting sp first, then accessing the "bottom" 628// page to force a page fault. Linux kernel will then automatically expand the 629// stack mapping. 630// 631// _expand_stack_to() assumes its frame size is less than page size, which 632// should always be true if the function is not inlined. 633 634#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 635#define NOINLINE 636#else 637#define NOINLINE __attribute__ ((noinline)) 638#endif 639 640static void _expand_stack_to(address bottom) NOINLINE; 641 642static void _expand_stack_to(address bottom) { 643 address sp; 644 size_t size; 645 volatile char *p; 646 647 // Adjust bottom to point to the largest address within the same page, it 648 // gives us a one-page buffer if alloca() allocates slightly more memory. 649 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 650 bottom += os::Linux::page_size() - 1; 651 652 // sp might be slightly above current stack pointer; if that's the case, we 653 // will alloca() a little more space than necessary, which is OK. Don't use 654 // os::current_stack_pointer(), as its result can be slightly below current 655 // stack pointer, causing us to not alloca enough to reach "bottom". 656 sp = (address)&sp; 657 658 if (sp > bottom) { 659 size = sp - bottom; 660 p = (volatile char *)alloca(size); 661 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 662 p[0] = '\0'; 663 } 664} 665 666bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 667 assert(t!=NULL, "just checking"); 668 assert(t->osthread()->expanding_stack(), "expand should be set"); 669 assert(t->stack_base() != NULL, "stack_base was not initialized"); 670 671 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 672 sigset_t mask_all, old_sigset; 673 sigfillset(&mask_all); 674 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 675 _expand_stack_to(addr); 676 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 677 return true; 678 } 679 return false; 680} 681 682////////////////////////////////////////////////////////////////////////////// 683// create new thread 684 685static address highest_vm_reserved_address(); 686 687// check if it's safe to start a new thread 688static bool _thread_safety_check(Thread* thread) { 689 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 690 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 691 // Heap is mmap'ed at lower end of memory space. Thread stacks are 692 // allocated (MAP_FIXED) from high address space. Every thread stack 693 // occupies a fixed size slot (usually 2Mbytes, but user can change 694 // it to other values if they rebuild LinuxThreads). 695 // 696 // Problem with MAP_FIXED is that mmap() can still succeed even part of 697 // the memory region has already been mmap'ed. That means if we have too 698 // many threads and/or very large heap, eventually thread stack will 699 // collide with heap. 700 // 701 // Here we try to prevent heap/stack collision by comparing current 702 // stack bottom with the highest address that has been mmap'ed by JVM 703 // plus a safety margin for memory maps created by native code. 704 // 705 // This feature can be disabled by setting ThreadSafetyMargin to 0 706 // 707 if (ThreadSafetyMargin > 0) { 708 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 709 710 // not safe if our stack extends below the safety margin 711 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 712 } else { 713 return true; 714 } 715 } else { 716 // Floating stack LinuxThreads or NPTL: 717 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 718 // there's not enough space left, pthread_create() will fail. If we come 719 // here, that means enough space has been reserved for stack. 720 return true; 721 } 722} 723 724// Thread start routine for all newly created threads 725static void *java_start(Thread *thread) { 726 // Try to randomize the cache line index of hot stack frames. 727 // This helps when threads of the same stack traces evict each other's 728 // cache lines. The threads can be either from the same JVM instance, or 729 // from different JVM instances. The benefit is especially true for 730 // processors with hyperthreading technology. 731 static int counter = 0; 732 int pid = os::current_process_id(); 733 alloca(((pid ^ counter++) & 7) * 128); 734 735 ThreadLocalStorage::set_thread(thread); 736 737 OSThread* osthread = thread->osthread(); 738 Monitor* sync = osthread->startThread_lock(); 739 740 // non floating stack LinuxThreads needs extra check, see above 741 if (!_thread_safety_check(thread)) { 742 // notify parent thread 743 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 744 osthread->set_state(ZOMBIE); 745 sync->notify_all(); 746 return NULL; 747 } 748 749 // thread_id is kernel thread id (similar to Solaris LWP id) 750 osthread->set_thread_id(os::Linux::gettid()); 751 752 if (UseNUMA) { 753 int lgrp_id = os::numa_get_group_id(); 754 if (lgrp_id != -1) { 755 thread->set_lgrp_id(lgrp_id); 756 } 757 } 758 // initialize signal mask for this thread 759 os::Linux::hotspot_sigmask(thread); 760 761 // initialize floating point control register 762 os::Linux::init_thread_fpu_state(); 763 764 // handshaking with parent thread 765 { 766 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 767 768 // notify parent thread 769 osthread->set_state(INITIALIZED); 770 sync->notify_all(); 771 772 // wait until os::start_thread() 773 while (osthread->get_state() == INITIALIZED) { 774 sync->wait(Mutex::_no_safepoint_check_flag); 775 } 776 } 777 778 // call one more level start routine 779 thread->run(); 780 781 return 0; 782} 783 784bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 785 assert(thread->osthread() == NULL, "caller responsible"); 786 787 // Allocate the OSThread object 788 OSThread* osthread = new OSThread(NULL, NULL); 789 if (osthread == NULL) { 790 return false; 791 } 792 793 // set the correct thread state 794 osthread->set_thread_type(thr_type); 795 796 // Initial state is ALLOCATED but not INITIALIZED 797 osthread->set_state(ALLOCATED); 798 799 thread->set_osthread(osthread); 800 801 // init thread attributes 802 pthread_attr_t attr; 803 pthread_attr_init(&attr); 804 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 805 806 // stack size 807 if (os::Linux::supports_variable_stack_size()) { 808 // calculate stack size if it's not specified by caller 809 if (stack_size == 0) { 810 stack_size = os::Linux::default_stack_size(thr_type); 811 812 switch (thr_type) { 813 case os::java_thread: 814 // Java threads use ThreadStackSize which default value can be changed with the flag -Xss 815 if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create(); 816 break; 817 case os::compiler_thread: 818 if (CompilerThreadStackSize > 0) { 819 stack_size = (size_t)(CompilerThreadStackSize * K); 820 break; 821 } // else fall through: 822 // use VMThreadStackSize if CompilerThreadStackSize is not defined 823 case os::vm_thread: 824 case os::pgc_thread: 825 case os::cgc_thread: 826 case os::watcher_thread: 827 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 828 break; 829 } 830 } 831 832 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 833 pthread_attr_setstacksize(&attr, stack_size); 834 } else { 835 // let pthread_create() pick the default value. 836 } 837 838 // glibc guard page 839 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 840 841 ThreadState state; 842 843 { 844 // Serialize thread creation if we are running with fixed stack LinuxThreads 845 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 846 if (lock) { 847 os::Linux::createThread_lock()->lock_without_safepoint_check(); 848 } 849 850 pthread_t tid; 851 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 852 853 pthread_attr_destroy(&attr); 854 855 if (ret != 0) { 856 if (PrintMiscellaneous && (Verbose || WizardMode)) { 857 perror("pthread_create()"); 858 } 859 // Need to clean up stuff we've allocated so far 860 thread->set_osthread(NULL); 861 delete osthread; 862 if (lock) os::Linux::createThread_lock()->unlock(); 863 return false; 864 } 865 866 // Store pthread info into the OSThread 867 osthread->set_pthread_id(tid); 868 869 // Wait until child thread is either initialized or aborted 870 { 871 Monitor* sync_with_child = osthread->startThread_lock(); 872 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 873 while ((state = osthread->get_state()) == ALLOCATED) { 874 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 875 } 876 } 877 878 if (lock) { 879 os::Linux::createThread_lock()->unlock(); 880 } 881 } 882 883 // Aborted due to thread limit being reached 884 if (state == ZOMBIE) { 885 thread->set_osthread(NULL); 886 delete osthread; 887 return false; 888 } 889 890 // The thread is returned suspended (in state INITIALIZED), 891 // and is started higher up in the call chain 892 assert(state == INITIALIZED, "race condition"); 893 return true; 894} 895 896///////////////////////////////////////////////////////////////////////////// 897// attach existing thread 898 899// bootstrap the main thread 900bool os::create_main_thread(JavaThread* thread) { 901 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 902 return create_attached_thread(thread); 903} 904 905bool os::create_attached_thread(JavaThread* thread) { 906#ifdef ASSERT 907 thread->verify_not_published(); 908#endif 909 910 // Allocate the OSThread object 911 OSThread* osthread = new OSThread(NULL, NULL); 912 913 if (osthread == NULL) { 914 return false; 915 } 916 917 // Store pthread info into the OSThread 918 osthread->set_thread_id(os::Linux::gettid()); 919 osthread->set_pthread_id(::pthread_self()); 920 921 // initialize floating point control register 922 os::Linux::init_thread_fpu_state(); 923 924 // Initial thread state is RUNNABLE 925 osthread->set_state(RUNNABLE); 926 927 thread->set_osthread(osthread); 928 929 if (UseNUMA) { 930 int lgrp_id = os::numa_get_group_id(); 931 if (lgrp_id != -1) { 932 thread->set_lgrp_id(lgrp_id); 933 } 934 } 935 936 if (os::Linux::is_initial_thread()) { 937 // If current thread is initial thread, its stack is mapped on demand, 938 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 939 // the entire stack region to avoid SEGV in stack banging. 940 // It is also useful to get around the heap-stack-gap problem on SuSE 941 // kernel (see 4821821 for details). We first expand stack to the top 942 // of yellow zone, then enable stack yellow zone (order is significant, 943 // enabling yellow zone first will crash JVM on SuSE Linux), so there 944 // is no gap between the last two virtual memory regions. 945 946 JavaThread *jt = (JavaThread *)thread; 947 address addr = jt->stack_yellow_zone_base(); 948 assert(addr != NULL, "initialization problem?"); 949 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 950 951 osthread->set_expanding_stack(); 952 os::Linux::manually_expand_stack(jt, addr); 953 osthread->clear_expanding_stack(); 954 } 955 956 // initialize signal mask for this thread 957 // and save the caller's signal mask 958 os::Linux::hotspot_sigmask(thread); 959 960 return true; 961} 962 963void os::pd_start_thread(Thread* thread) { 964 OSThread * osthread = thread->osthread(); 965 assert(osthread->get_state() != INITIALIZED, "just checking"); 966 Monitor* sync_with_child = osthread->startThread_lock(); 967 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 968 sync_with_child->notify(); 969} 970 971// Free Linux resources related to the OSThread 972void os::free_thread(OSThread* osthread) { 973 assert(osthread != NULL, "osthread not set"); 974 975 if (Thread::current()->osthread() == osthread) { 976 // Restore caller's signal mask 977 sigset_t sigmask = osthread->caller_sigmask(); 978 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 979 } 980 981 delete osthread; 982} 983 984////////////////////////////////////////////////////////////////////////////// 985// thread local storage 986 987int os::allocate_thread_local_storage() { 988 pthread_key_t key; 989 int rslt = pthread_key_create(&key, NULL); 990 assert(rslt == 0, "cannot allocate thread local storage"); 991 return (int)key; 992} 993 994// Note: This is currently not used by VM, as we don't destroy TLS key 995// on VM exit. 996void os::free_thread_local_storage(int index) { 997 int rslt = pthread_key_delete((pthread_key_t)index); 998 assert(rslt == 0, "invalid index"); 999} 1000 1001void os::thread_local_storage_at_put(int index, void* value) { 1002 int rslt = pthread_setspecific((pthread_key_t)index, value); 1003 assert(rslt == 0, "pthread_setspecific failed"); 1004} 1005 1006extern "C" Thread* get_thread() { 1007 return ThreadLocalStorage::thread(); 1008} 1009 1010////////////////////////////////////////////////////////////////////////////// 1011// initial thread 1012 1013// Check if current thread is the initial thread, similar to Solaris thr_main. 1014bool os::Linux::is_initial_thread(void) { 1015 char dummy; 1016 // If called before init complete, thread stack bottom will be null. 1017 // Can be called if fatal error occurs before initialization. 1018 if (initial_thread_stack_bottom() == NULL) return false; 1019 assert(initial_thread_stack_bottom() != NULL && 1020 initial_thread_stack_size() != 0, 1021 "os::init did not locate initial thread's stack region"); 1022 if ((address)&dummy >= initial_thread_stack_bottom() && 1023 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1024 return true; 1025 else return false; 1026} 1027 1028// Find the virtual memory area that contains addr 1029static bool find_vma(address addr, address* vma_low, address* vma_high) { 1030 FILE *fp = fopen("/proc/self/maps", "r"); 1031 if (fp) { 1032 address low, high; 1033 while (!feof(fp)) { 1034 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1035 if (low <= addr && addr < high) { 1036 if (vma_low) *vma_low = low; 1037 if (vma_high) *vma_high = high; 1038 fclose (fp); 1039 return true; 1040 } 1041 } 1042 for (;;) { 1043 int ch = fgetc(fp); 1044 if (ch == EOF || ch == (int)'\n') break; 1045 } 1046 } 1047 fclose(fp); 1048 } 1049 return false; 1050} 1051 1052// Locate initial thread stack. This special handling of initial thread stack 1053// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1054// bogus value for initial thread. 1055void os::Linux::capture_initial_stack(size_t max_size) { 1056 // stack size is the easy part, get it from RLIMIT_STACK 1057 size_t stack_size; 1058 struct rlimit rlim; 1059 getrlimit(RLIMIT_STACK, &rlim); 1060 stack_size = rlim.rlim_cur; 1061 1062 // 6308388: a bug in ld.so will relocate its own .data section to the 1063 // lower end of primordial stack; reduce ulimit -s value a little bit 1064 // so we won't install guard page on ld.so's data section. 1065 stack_size -= 2 * page_size(); 1066 1067 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1068 // 7.1, in both cases we will get 2G in return value. 1069 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1070 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1071 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1072 // in case other parts in glibc still assumes 2M max stack size. 1073 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1074#ifndef IA64 1075 if (stack_size > 2 * K * K) stack_size = 2 * K * K; 1076#else 1077 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1078 if (stack_size > 4 * K * K) stack_size = 4 * K * K; 1079#endif 1080 1081 // Try to figure out where the stack base (top) is. This is harder. 1082 // 1083 // When an application is started, glibc saves the initial stack pointer in 1084 // a global variable "__libc_stack_end", which is then used by system 1085 // libraries. __libc_stack_end should be pretty close to stack top. The 1086 // variable is available since the very early days. However, because it is 1087 // a private interface, it could disappear in the future. 1088 // 1089 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1090 // to __libc_stack_end, it is very close to stack top, but isn't the real 1091 // stack top. Note that /proc may not exist if VM is running as a chroot 1092 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1093 // /proc/<pid>/stat could change in the future (though unlikely). 1094 // 1095 // We try __libc_stack_end first. If that doesn't work, look for 1096 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1097 // as a hint, which should work well in most cases. 1098 1099 uintptr_t stack_start; 1100 1101 // try __libc_stack_end first 1102 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1103 if (p && *p) { 1104 stack_start = *p; 1105 } else { 1106 // see if we can get the start_stack field from /proc/self/stat 1107 FILE *fp; 1108 int pid; 1109 char state; 1110 int ppid; 1111 int pgrp; 1112 int session; 1113 int nr; 1114 int tpgrp; 1115 unsigned long flags; 1116 unsigned long minflt; 1117 unsigned long cminflt; 1118 unsigned long majflt; 1119 unsigned long cmajflt; 1120 unsigned long utime; 1121 unsigned long stime; 1122 long cutime; 1123 long cstime; 1124 long prio; 1125 long nice; 1126 long junk; 1127 long it_real; 1128 uintptr_t start; 1129 uintptr_t vsize; 1130 uintptr_t rss; 1131 unsigned long rsslim; 1132 uintptr_t scodes; 1133 uintptr_t ecode; 1134 int i; 1135 1136 // Figure what the primordial thread stack base is. Code is inspired 1137 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1138 // followed by command name surrounded by parentheses, state, etc. 1139 char stat[2048]; 1140 int statlen; 1141 1142 fp = fopen("/proc/self/stat", "r"); 1143 if (fp) { 1144 statlen = fread(stat, 1, 2047, fp); 1145 stat[statlen] = '\0'; 1146 fclose(fp); 1147 1148 // Skip pid and the command string. Note that we could be dealing with 1149 // weird command names, e.g. user could decide to rename java launcher 1150 // to "java 1.4.2 :)", then the stat file would look like 1151 // 1234 (java 1.4.2 :)) R ... ... 1152 // We don't really need to know the command string, just find the last 1153 // occurrence of ")" and then start parsing from there. See bug 4726580. 1154 char * s = strrchr(stat, ')'); 1155 1156 i = 0; 1157 if (s) { 1158 // Skip blank chars 1159 do s++; while (isspace(*s)); 1160 1161 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1162 /* 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 */ 1163 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld %lu %lu %ld %lu %lu %lu %lu", 1164 &state, /* 3 %c */ 1165 &ppid, /* 4 %d */ 1166 &pgrp, /* 5 %d */ 1167 &session, /* 6 %d */ 1168 &nr, /* 7 %d */ 1169 &tpgrp, /* 8 %d */ 1170 &flags, /* 9 %lu */ 1171 &minflt, /* 10 %lu */ 1172 &cminflt, /* 11 %lu */ 1173 &majflt, /* 12 %lu */ 1174 &cmajflt, /* 13 %lu */ 1175 &utime, /* 14 %lu */ 1176 &stime, /* 15 %lu */ 1177 &cutime, /* 16 %ld */ 1178 &cstime, /* 17 %ld */ 1179 &prio, /* 18 %ld */ 1180 &nice, /* 19 %ld */ 1181 &junk, /* 20 %ld */ 1182 &it_real, /* 21 %ld */ 1183 &start, /* 22 %lu */ 1184 &vsize, /* 23 %lu */ 1185 &rss, /* 24 %ld */ 1186 &rsslim, /* 25 %lu */ 1187 &scodes, /* 26 %lu */ 1188 &ecode, /* 27 %lu */ 1189 &stack_start); /* 28 %lu */ 1190 } 1191 1192 if (i != 28 - 2) { 1193 assert(false, "Bad conversion from /proc/self/stat"); 1194 // product mode - assume we are the initial thread, good luck in the 1195 // embedded case. 1196 warning("Can't detect initial thread stack location - bad conversion"); 1197 stack_start = (uintptr_t) &rlim; 1198 } 1199 } else { 1200 // For some reason we can't open /proc/self/stat (for example, running on 1201 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1202 // most cases, so don't abort: 1203 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1204 stack_start = (uintptr_t) &rlim; 1205 } 1206 } 1207 1208 // Now we have a pointer (stack_start) very close to the stack top, the 1209 // next thing to do is to figure out the exact location of stack top. We 1210 // can find out the virtual memory area that contains stack_start by 1211 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1212 // and its upper limit is the real stack top. (again, this would fail if 1213 // running inside chroot, because /proc may not exist.) 1214 1215 uintptr_t stack_top; 1216 address low, high; 1217 if (find_vma((address)stack_start, &low, &high)) { 1218 // success, "high" is the true stack top. (ignore "low", because initial 1219 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1220 stack_top = (uintptr_t)high; 1221 } else { 1222 // failed, likely because /proc/self/maps does not exist 1223 warning("Can't detect initial thread stack location - find_vma failed"); 1224 // best effort: stack_start is normally within a few pages below the real 1225 // stack top, use it as stack top, and reduce stack size so we won't put 1226 // guard page outside stack. 1227 stack_top = stack_start; 1228 stack_size -= 16 * page_size(); 1229 } 1230 1231 // stack_top could be partially down the page so align it 1232 stack_top = align_size_up(stack_top, page_size()); 1233 1234 if (max_size && stack_size > max_size) { 1235 _initial_thread_stack_size = max_size; 1236 } else { 1237 _initial_thread_stack_size = stack_size; 1238 } 1239 1240 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1241 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1242} 1243 1244//////////////////////////////////////////////////////////////////////////////// 1245// time support 1246 1247// Time since start-up in seconds to a fine granularity. 1248// Used by VMSelfDestructTimer and the MemProfiler. 1249double os::elapsedTime() { 1250 1251 return (double)(os::elapsed_counter()) * 0.000001; 1252} 1253 1254jlong os::elapsed_counter() { 1255 timeval time; 1256 int status = gettimeofday(&time, NULL); 1257 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1258} 1259 1260jlong os::elapsed_frequency() { 1261 return (1000 * 1000); 1262} 1263 1264jlong os::javaTimeMillis() { 1265 timeval time; 1266 int status = gettimeofday(&time, NULL); 1267 assert(status != -1, "linux error"); 1268 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1269} 1270 1271#ifndef CLOCK_MONOTONIC 1272#define CLOCK_MONOTONIC (1) 1273#endif 1274 1275void os::Linux::clock_init() { 1276 // we do dlopen's in this particular order due to bug in linux 1277 // dynamical loader (see 6348968) leading to crash on exit 1278 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1279 if (handle == NULL) { 1280 handle = dlopen("librt.so", RTLD_LAZY); 1281 } 1282 1283 if (handle) { 1284 int (*clock_getres_func)(clockid_t, struct timespec*) = 1285 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1286 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1287 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1288 if (clock_getres_func && clock_gettime_func) { 1289 // See if monotonic clock is supported by the kernel. Note that some 1290 // early implementations simply return kernel jiffies (updated every 1291 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1292 // for nano time (though the monotonic property is still nice to have). 1293 // It's fixed in newer kernels, however clock_getres() still returns 1294 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1295 // resolution for now. Hopefully as people move to new kernels, this 1296 // won't be a problem. 1297 struct timespec res; 1298 struct timespec tp; 1299 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1300 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1301 // yes, monotonic clock is supported 1302 _clock_gettime = clock_gettime_func; 1303 } else { 1304 // close librt if there is no monotonic clock 1305 dlclose(handle); 1306 } 1307 } 1308 } 1309} 1310 1311#ifndef SYS_clock_getres 1312 1313#if defined(IA32) || defined(AMD64) 1314#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1315#else 1316#error Value of SYS_clock_getres not known on this platform 1317#endif 1318 1319#endif 1320 1321#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1322 1323void os::Linux::fast_thread_clock_init() { 1324 if (!UseLinuxPosixThreadCPUClocks) { 1325 return; 1326 } 1327 clockid_t clockid; 1328 struct timespec tp; 1329 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1330 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1331 1332 // Switch to using fast clocks for thread cpu time if 1333 // the sys_clock_getres() returns 0 error code. 1334 // Note, that some kernels may support the current thread 1335 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1336 // returned by the pthread_getcpuclockid(). 1337 // If the fast Posix clocks are supported then the sys_clock_getres() 1338 // must return at least tp.tv_sec == 0 which means a resolution 1339 // better than 1 sec. This is extra check for reliability. 1340 1341 if(pthread_getcpuclockid_func && 1342 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1343 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1344 1345 _supports_fast_thread_cpu_time = true; 1346 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1347 } 1348} 1349 1350jlong os::javaTimeNanos() { 1351 if (Linux::supports_monotonic_clock()) { 1352 struct timespec tp; 1353 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1354 assert(status == 0, "gettime error"); 1355 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1356 return result; 1357 } else { 1358 timeval time; 1359 int status = gettimeofday(&time, NULL); 1360 assert(status != -1, "linux error"); 1361 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1362 return 1000 * usecs; 1363 } 1364} 1365 1366void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1367 if (Linux::supports_monotonic_clock()) { 1368 info_ptr->max_value = ALL_64_BITS; 1369 1370 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1371 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1372 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1373 } else { 1374 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1375 info_ptr->max_value = ALL_64_BITS; 1376 1377 // gettimeofday is a real time clock so it skips 1378 info_ptr->may_skip_backward = true; 1379 info_ptr->may_skip_forward = true; 1380 } 1381 1382 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1383} 1384 1385// Return the real, user, and system times in seconds from an 1386// arbitrary fixed point in the past. 1387bool os::getTimesSecs(double* process_real_time, 1388 double* process_user_time, 1389 double* process_system_time) { 1390 struct tms ticks; 1391 clock_t real_ticks = times(&ticks); 1392 1393 if (real_ticks == (clock_t) (-1)) { 1394 return false; 1395 } else { 1396 double ticks_per_second = (double) clock_tics_per_sec; 1397 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1398 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1399 *process_real_time = ((double) real_ticks) / ticks_per_second; 1400 1401 return true; 1402 } 1403} 1404 1405 1406char * os::local_time_string(char *buf, size_t buflen) { 1407 struct tm t; 1408 time_t long_time; 1409 time(&long_time); 1410 localtime_r(&long_time, &t); 1411 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1412 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1413 t.tm_hour, t.tm_min, t.tm_sec); 1414 return buf; 1415} 1416 1417//////////////////////////////////////////////////////////////////////////////// 1418// runtime exit support 1419 1420// Note: os::shutdown() might be called very early during initialization, or 1421// called from signal handler. Before adding something to os::shutdown(), make 1422// sure it is async-safe and can handle partially initialized VM. 1423void os::shutdown() { 1424 1425 // allow PerfMemory to attempt cleanup of any persistent resources 1426 perfMemory_exit(); 1427 1428 // needs to remove object in file system 1429 AttachListener::abort(); 1430 1431 // flush buffered output, finish log files 1432 ostream_abort(); 1433 1434 // Check for abort hook 1435 abort_hook_t abort_hook = Arguments::abort_hook(); 1436 if (abort_hook != NULL) { 1437 abort_hook(); 1438 } 1439 1440} 1441 1442// Note: os::abort() might be called very early during initialization, or 1443// called from signal handler. Before adding something to os::abort(), make 1444// sure it is async-safe and can handle partially initialized VM. 1445void os::abort(bool dump_core) { 1446 os::shutdown(); 1447 if (dump_core) { 1448#ifndef PRODUCT 1449 fdStream out(defaultStream::output_fd()); 1450 out.print_raw("Current thread is "); 1451 char buf[16]; 1452 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1453 out.print_raw_cr(buf); 1454 out.print_raw_cr("Dumping core ..."); 1455#endif 1456 ::abort(); // dump core 1457 } 1458 1459 ::exit(1); 1460} 1461 1462// Die immediately, no exit hook, no abort hook, no cleanup. 1463void os::die() { 1464 // _exit() on LinuxThreads only kills current thread 1465 ::abort(); 1466} 1467 1468// unused on linux for now. 1469void os::set_error_file(const char *logfile) {} 1470 1471intx os::current_thread_id() { return (intx)pthread_self(); } 1472int os::current_process_id() { 1473 1474 // Under the old linux thread library, linux gives each thread 1475 // its own process id. Because of this each thread will return 1476 // a different pid if this method were to return the result 1477 // of getpid(2). Linux provides no api that returns the pid 1478 // of the launcher thread for the vm. This implementation 1479 // returns a unique pid, the pid of the launcher thread 1480 // that starts the vm 'process'. 1481 1482 // Under the NPTL, getpid() returns the same pid as the 1483 // launcher thread rather than a unique pid per thread. 1484 // Use gettid() if you want the old pre NPTL behaviour. 1485 1486 // if you are looking for the result of a call to getpid() that 1487 // returns a unique pid for the calling thread, then look at the 1488 // OSThread::thread_id() method in osThread_linux.hpp file 1489 1490 return (int)(_initial_pid ? _initial_pid : getpid()); 1491} 1492 1493// DLL functions 1494 1495const char* os::dll_file_extension() { return ".so"; } 1496 1497const char* os::get_temp_directory() { return "/tmp/"; } 1498 1499void os::dll_build_name( 1500 char* buffer, size_t buflen, const char* pname, const char* fname) { 1501 // copied from libhpi 1502 const size_t pnamelen = pname ? strlen(pname) : 0; 1503 1504 /* Quietly truncate on buffer overflow. Should be an error. */ 1505 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1506 *buffer = '\0'; 1507 return; 1508 } 1509 1510 if (pnamelen == 0) { 1511 sprintf(buffer, "lib%s.so", fname); 1512 } else { 1513 sprintf(buffer, "%s/lib%s.so", pname, fname); 1514 } 1515} 1516 1517const char* os::get_current_directory(char *buf, int buflen) { 1518 return getcwd(buf, buflen); 1519} 1520 1521// check if addr is inside libjvm[_g].so 1522bool os::address_is_in_vm(address addr) { 1523 static address libjvm_base_addr; 1524 Dl_info dlinfo; 1525 1526 if (libjvm_base_addr == NULL) { 1527 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); 1528 libjvm_base_addr = (address)dlinfo.dli_fbase; 1529 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1530 } 1531 1532 if (dladdr((void *)addr, &dlinfo)) { 1533 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1534 } 1535 1536 return false; 1537} 1538 1539bool os::dll_address_to_function_name(address addr, char *buf, 1540 int buflen, int *offset) { 1541 Dl_info dlinfo; 1542 1543 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { 1544 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1545 if (offset) *offset = addr - (address)dlinfo.dli_saddr; 1546 return true; 1547 } else { 1548 if (buf) buf[0] = '\0'; 1549 if (offset) *offset = -1; 1550 return false; 1551 } 1552} 1553 1554struct _address_to_library_name { 1555 address addr; // input : memory address 1556 size_t buflen; // size of fname 1557 char* fname; // output: library name 1558 address base; // library base addr 1559}; 1560 1561static int address_to_library_name_callback(struct dl_phdr_info *info, 1562 size_t size, void *data) { 1563 int i; 1564 bool found = false; 1565 address libbase = NULL; 1566 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1567 1568 // iterate through all loadable segments 1569 for (i = 0; i < info->dlpi_phnum; i++) { 1570 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1571 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1572 // base address of a library is the lowest address of its loaded 1573 // segments. 1574 if (libbase == NULL || libbase > segbase) { 1575 libbase = segbase; 1576 } 1577 // see if 'addr' is within current segment 1578 if (segbase <= d->addr && 1579 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1580 found = true; 1581 } 1582 } 1583 } 1584 1585 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1586 // so dll_address_to_library_name() can fall through to use dladdr() which 1587 // can figure out executable name from argv[0]. 1588 if (found && info->dlpi_name && info->dlpi_name[0]) { 1589 d->base = libbase; 1590 if (d->fname) { 1591 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1592 } 1593 return 1; 1594 } 1595 return 0; 1596} 1597 1598bool os::dll_address_to_library_name(address addr, char* buf, 1599 int buflen, int* offset) { 1600 Dl_info dlinfo; 1601 struct _address_to_library_name data; 1602 1603 // There is a bug in old glibc dladdr() implementation that it could resolve 1604 // to wrong library name if the .so file has a base address != NULL. Here 1605 // we iterate through the program headers of all loaded libraries to find 1606 // out which library 'addr' really belongs to. This workaround can be 1607 // removed once the minimum requirement for glibc is moved to 2.3.x. 1608 data.addr = addr; 1609 data.fname = buf; 1610 data.buflen = buflen; 1611 data.base = NULL; 1612 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1613 1614 if (rslt) { 1615 // buf already contains library name 1616 if (offset) *offset = addr - data.base; 1617 return true; 1618 } else if (dladdr((void*)addr, &dlinfo)){ 1619 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1620 if (offset) *offset = addr - (address)dlinfo.dli_fbase; 1621 return true; 1622 } else { 1623 if (buf) buf[0] = '\0'; 1624 if (offset) *offset = -1; 1625 return false; 1626 } 1627} 1628 1629 // Loads .dll/.so and 1630 // in case of error it checks if .dll/.so was built for the 1631 // same architecture as Hotspot is running on 1632 1633void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1634{ 1635 void * result= ::dlopen(filename, RTLD_LAZY); 1636 if (result != NULL) { 1637 // Successful loading 1638 return result; 1639 } 1640 1641 Elf32_Ehdr elf_head; 1642 1643 // Read system error message into ebuf 1644 // It may or may not be overwritten below 1645 ::strncpy(ebuf, ::dlerror(), ebuflen-1); 1646 ebuf[ebuflen-1]='\0'; 1647 int diag_msg_max_length=ebuflen-strlen(ebuf); 1648 char* diag_msg_buf=ebuf+strlen(ebuf); 1649 1650 if (diag_msg_max_length==0) { 1651 // No more space in ebuf for additional diagnostics message 1652 return NULL; 1653 } 1654 1655 1656 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1657 1658 if (file_descriptor < 0) { 1659 // Can't open library, report dlerror() message 1660 return NULL; 1661 } 1662 1663 bool failed_to_read_elf_head= 1664 (sizeof(elf_head)!= 1665 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1666 1667 ::close(file_descriptor); 1668 if (failed_to_read_elf_head) { 1669 // file i/o error - report dlerror() msg 1670 return NULL; 1671 } 1672 1673 typedef struct { 1674 Elf32_Half code; // Actual value as defined in elf.h 1675 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1676 char elf_class; // 32 or 64 bit 1677 char endianess; // MSB or LSB 1678 char* name; // String representation 1679 } arch_t; 1680 1681 #ifndef EM_486 1682 #define EM_486 6 /* Intel 80486 */ 1683 #endif 1684 1685 static const arch_t arch_array[]={ 1686 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1687 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1688 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1689 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1690 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1691 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1692 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1693 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1694 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"} 1695 }; 1696 1697 #if (defined IA32) 1698 static Elf32_Half running_arch_code=EM_386; 1699 #elif (defined AMD64) 1700 static Elf32_Half running_arch_code=EM_X86_64; 1701 #elif (defined IA64) 1702 static Elf32_Half running_arch_code=EM_IA_64; 1703 #elif (defined __sparc) && (defined _LP64) 1704 static Elf32_Half running_arch_code=EM_SPARCV9; 1705 #elif (defined __sparc) && (!defined _LP64) 1706 static Elf32_Half running_arch_code=EM_SPARC; 1707 #elif (defined __powerpc64__) 1708 static Elf32_Half running_arch_code=EM_PPC64; 1709 #elif (defined __powerpc__) 1710 static Elf32_Half running_arch_code=EM_PPC; 1711 #else 1712 #error Method os::dll_load requires that one of following is defined:\ 1713 IA32, AMD64, IA64, __sparc, __powerpc__ 1714 #endif 1715 1716 // Identify compatability class for VM's architecture and library's architecture 1717 // Obtain string descriptions for architectures 1718 1719 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1720 int running_arch_index=-1; 1721 1722 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1723 if (running_arch_code == arch_array[i].code) { 1724 running_arch_index = i; 1725 } 1726 if (lib_arch.code == arch_array[i].code) { 1727 lib_arch.compat_class = arch_array[i].compat_class; 1728 lib_arch.name = arch_array[i].name; 1729 } 1730 } 1731 1732 assert(running_arch_index != -1, 1733 "Didn't find running architecture code (running_arch_code) in arch_array"); 1734 if (running_arch_index == -1) { 1735 // Even though running architecture detection failed 1736 // we may still continue with reporting dlerror() message 1737 return NULL; 1738 } 1739 1740 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1741 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1742 return NULL; 1743 } 1744 1745 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1746 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1747 return NULL; 1748 } 1749 1750 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1751 if ( lib_arch.name!=NULL ) { 1752 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1753 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1754 lib_arch.name, arch_array[running_arch_index].name); 1755 } else { 1756 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1757 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1758 lib_arch.code, 1759 arch_array[running_arch_index].name); 1760 } 1761 } 1762 1763 return NULL; 1764} 1765 1766/* 1767 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 1768 * chances are you might want to run the generated bits against glibc-2.0 1769 * libdl.so, so always use locking for any version of glibc. 1770 */ 1771void* os::dll_lookup(void* handle, const char* name) { 1772 pthread_mutex_lock(&dl_mutex); 1773 void* res = dlsym(handle, name); 1774 pthread_mutex_unlock(&dl_mutex); 1775 return res; 1776} 1777 1778 1779bool _print_ascii_file(const char* filename, outputStream* st) { 1780 int fd = open(filename, O_RDONLY); 1781 if (fd == -1) { 1782 return false; 1783 } 1784 1785 char buf[32]; 1786 int bytes; 1787 while ((bytes = read(fd, buf, sizeof(buf))) > 0) { 1788 st->print_raw(buf, bytes); 1789 } 1790 1791 close(fd); 1792 1793 return true; 1794} 1795 1796void os::print_dll_info(outputStream *st) { 1797 st->print_cr("Dynamic libraries:"); 1798 1799 char fname[32]; 1800 pid_t pid = os::Linux::gettid(); 1801 1802 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1803 1804 if (!_print_ascii_file(fname, st)) { 1805 st->print("Can not get library information for pid = %d\n", pid); 1806 } 1807} 1808 1809 1810void os::print_os_info(outputStream* st) { 1811 st->print("OS:"); 1812 1813 // Try to identify popular distros. 1814 // Most Linux distributions have /etc/XXX-release file, which contains 1815 // the OS version string. Some have more than one /etc/XXX-release file 1816 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 1817 // so the order is important. 1818 if (!_print_ascii_file("/etc/mandrake-release", st) && 1819 !_print_ascii_file("/etc/sun-release", st) && 1820 !_print_ascii_file("/etc/redhat-release", st) && 1821 !_print_ascii_file("/etc/SuSE-release", st) && 1822 !_print_ascii_file("/etc/turbolinux-release", st) && 1823 !_print_ascii_file("/etc/gentoo-release", st) && 1824 !_print_ascii_file("/etc/debian_version", st)) { 1825 st->print("Linux"); 1826 } 1827 st->cr(); 1828 1829 // kernel 1830 st->print("uname:"); 1831 struct utsname name; 1832 uname(&name); 1833 st->print(name.sysname); st->print(" "); 1834 st->print(name.release); st->print(" "); 1835 st->print(name.version); st->print(" "); 1836 st->print(name.machine); 1837 st->cr(); 1838 1839 // Print warning if unsafe chroot environment detected 1840 if (unsafe_chroot_detected) { 1841 st->print("WARNING!! "); 1842 st->print_cr(unstable_chroot_error); 1843 } 1844 1845 // libc, pthread 1846 st->print("libc:"); 1847 st->print(os::Linux::glibc_version()); st->print(" "); 1848 st->print(os::Linux::libpthread_version()); st->print(" "); 1849 if (os::Linux::is_LinuxThreads()) { 1850 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 1851 } 1852 st->cr(); 1853 1854 // rlimit 1855 st->print("rlimit:"); 1856 struct rlimit rlim; 1857 1858 st->print(" STACK "); 1859 getrlimit(RLIMIT_STACK, &rlim); 1860 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1861 else st->print("%uk", rlim.rlim_cur >> 10); 1862 1863 st->print(", CORE "); 1864 getrlimit(RLIMIT_CORE, &rlim); 1865 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1866 else st->print("%uk", rlim.rlim_cur >> 10); 1867 1868 st->print(", NPROC "); 1869 getrlimit(RLIMIT_NPROC, &rlim); 1870 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1871 else st->print("%d", rlim.rlim_cur); 1872 1873 st->print(", NOFILE "); 1874 getrlimit(RLIMIT_NOFILE, &rlim); 1875 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1876 else st->print("%d", rlim.rlim_cur); 1877 1878 st->print(", AS "); 1879 getrlimit(RLIMIT_AS, &rlim); 1880 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 1881 else st->print("%uk", rlim.rlim_cur >> 10); 1882 st->cr(); 1883 1884 // load average 1885 st->print("load average:"); 1886 double loadavg[3]; 1887 os::loadavg(loadavg, 3); 1888 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]); 1889 st->cr(); 1890} 1891 1892void os::print_memory_info(outputStream* st) { 1893 1894 st->print("Memory:"); 1895 st->print(" %dk page", os::vm_page_size()>>10); 1896 1897 // values in struct sysinfo are "unsigned long" 1898 struct sysinfo si; 1899 sysinfo(&si); 1900 1901 st->print(", physical " UINT64_FORMAT "k", 1902 os::physical_memory() >> 10); 1903 st->print("(" UINT64_FORMAT "k free)", 1904 os::available_memory() >> 10); 1905 st->print(", swap " UINT64_FORMAT "k", 1906 ((jlong)si.totalswap * si.mem_unit) >> 10); 1907 st->print("(" UINT64_FORMAT "k free)", 1908 ((jlong)si.freeswap * si.mem_unit) >> 10); 1909 st->cr(); 1910} 1911 1912// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 1913// but they're the same for all the linux arch that we support 1914// and they're the same for solaris but there's no common place to put this. 1915const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 1916 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 1917 "ILL_COPROC", "ILL_BADSTK" }; 1918 1919const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 1920 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 1921 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 1922 1923const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 1924 1925const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 1926 1927void os::print_siginfo(outputStream* st, void* siginfo) { 1928 st->print("siginfo:"); 1929 1930 const int buflen = 100; 1931 char buf[buflen]; 1932 siginfo_t *si = (siginfo_t*)siginfo; 1933 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 1934 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 1935 st->print("si_errno=%s", buf); 1936 } else { 1937 st->print("si_errno=%d", si->si_errno); 1938 } 1939 const int c = si->si_code; 1940 assert(c > 0, "unexpected si_code"); 1941 switch (si->si_signo) { 1942 case SIGILL: 1943 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 1944 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1945 break; 1946 case SIGFPE: 1947 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 1948 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1949 break; 1950 case SIGSEGV: 1951 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 1952 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1953 break; 1954 case SIGBUS: 1955 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 1956 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 1957 break; 1958 default: 1959 st->print(", si_code=%d", si->si_code); 1960 // no si_addr 1961 } 1962 1963 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 1964 UseSharedSpaces) { 1965 FileMapInfo* mapinfo = FileMapInfo::current_info(); 1966 if (mapinfo->is_in_shared_space(si->si_addr)) { 1967 st->print("\n\nError accessing class data sharing archive." \ 1968 " Mapped file inaccessible during execution, " \ 1969 " possible disk/network problem."); 1970 } 1971 } 1972 st->cr(); 1973} 1974 1975 1976static void print_signal_handler(outputStream* st, int sig, 1977 char* buf, size_t buflen); 1978 1979void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 1980 st->print_cr("Signal Handlers:"); 1981 print_signal_handler(st, SIGSEGV, buf, buflen); 1982 print_signal_handler(st, SIGBUS , buf, buflen); 1983 print_signal_handler(st, SIGFPE , buf, buflen); 1984 print_signal_handler(st, SIGPIPE, buf, buflen); 1985 print_signal_handler(st, SIGXFSZ, buf, buflen); 1986 print_signal_handler(st, SIGILL , buf, buflen); 1987 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 1988 print_signal_handler(st, SR_signum, buf, buflen); 1989 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 1990 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 1991 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 1992 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 1993} 1994 1995static char saved_jvm_path[MAXPATHLEN] = {0}; 1996 1997// Find the full path to the current module, libjvm.so or libjvm_g.so 1998void os::jvm_path(char *buf, jint len) { 1999 // Error checking. 2000 if (len < MAXPATHLEN) { 2001 assert(false, "must use a large-enough buffer"); 2002 buf[0] = '\0'; 2003 return; 2004 } 2005 // Lazy resolve the path to current module. 2006 if (saved_jvm_path[0] != 0) { 2007 strcpy(buf, saved_jvm_path); 2008 return; 2009 } 2010 2011 char dli_fname[MAXPATHLEN]; 2012 bool ret = dll_address_to_library_name( 2013 CAST_FROM_FN_PTR(address, os::jvm_path), 2014 dli_fname, sizeof(dli_fname), NULL); 2015 assert(ret != 0, "cannot locate libjvm"); 2016 realpath(dli_fname, buf); 2017 2018 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) { 2019 // Support for the gamma launcher. Typical value for buf is 2020 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 2021 // the right place in the string, then assume we are installed in a JDK and 2022 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2023 // up the path so it looks like libjvm.so is installed there (append a 2024 // fake suffix hotspot/libjvm.so). 2025 const char *p = buf + strlen(buf) - 1; 2026 for (int count = 0; p > buf && count < 5; ++count) { 2027 for (--p; p > buf && *p != '/'; --p) 2028 /* empty */ ; 2029 } 2030 2031 if (strncmp(p, "/jre/lib/", 9) != 0) { 2032 // Look for JAVA_HOME in the environment. 2033 char* java_home_var = ::getenv("JAVA_HOME"); 2034 if (java_home_var != NULL && java_home_var[0] != 0) { 2035 // Check the current module name "libjvm.so" or "libjvm_g.so". 2036 p = strrchr(buf, '/'); 2037 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2038 p = strstr(p, "_g") ? "_g" : ""; 2039 2040 realpath(java_home_var, buf); 2041 sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch); 2042 if (0 == access(buf, F_OK)) { 2043 // Use current module name "libjvm[_g].so" instead of 2044 // "libjvm"debug_only("_g")".so" since for fastdebug version 2045 // we should have "libjvm.so" but debug_only("_g") adds "_g"! 2046 // It is used when we are choosing the HPI library's name 2047 // "libhpi[_g].so" in hpi::initialize_get_interface(). 2048 sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p); 2049 } else { 2050 // Go back to path of .so 2051 realpath(dli_fname, buf); 2052 } 2053 } 2054 } 2055 } 2056 2057 strcpy(saved_jvm_path, buf); 2058} 2059 2060void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2061 // no prefix required, not even "_" 2062} 2063 2064void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2065 // no suffix required 2066} 2067 2068//////////////////////////////////////////////////////////////////////////////// 2069// sun.misc.Signal support 2070 2071static volatile jint sigint_count = 0; 2072 2073static void 2074UserHandler(int sig, void *siginfo, void *context) { 2075 // 4511530 - sem_post is serialized and handled by the manager thread. When 2076 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2077 // don't want to flood the manager thread with sem_post requests. 2078 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2079 return; 2080 2081 // Ctrl-C is pressed during error reporting, likely because the error 2082 // handler fails to abort. Let VM die immediately. 2083 if (sig == SIGINT && is_error_reported()) { 2084 os::die(); 2085 } 2086 2087 os::signal_notify(sig); 2088} 2089 2090void* os::user_handler() { 2091 return CAST_FROM_FN_PTR(void*, UserHandler); 2092} 2093 2094extern "C" { 2095 typedef void (*sa_handler_t)(int); 2096 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2097} 2098 2099void* os::signal(int signal_number, void* handler) { 2100 struct sigaction sigAct, oldSigAct; 2101 2102 sigfillset(&(sigAct.sa_mask)); 2103 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2104 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2105 2106 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2107 // -1 means registration failed 2108 return (void *)-1; 2109 } 2110 2111 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2112} 2113 2114void os::signal_raise(int signal_number) { 2115 ::raise(signal_number); 2116} 2117 2118/* 2119 * The following code is moved from os.cpp for making this 2120 * code platform specific, which it is by its very nature. 2121 */ 2122 2123// Will be modified when max signal is changed to be dynamic 2124int os::sigexitnum_pd() { 2125 return NSIG; 2126} 2127 2128// a counter for each possible signal value 2129static volatile jint pending_signals[NSIG+1] = { 0 }; 2130 2131// Linux(POSIX) specific hand shaking semaphore. 2132static sem_t sig_sem; 2133 2134void os::signal_init_pd() { 2135 // Initialize signal structures 2136 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2137 2138 // Initialize signal semaphore 2139 ::sem_init(&sig_sem, 0, 0); 2140} 2141 2142void os::signal_notify(int sig) { 2143 Atomic::inc(&pending_signals[sig]); 2144 ::sem_post(&sig_sem); 2145} 2146 2147static int check_pending_signals(bool wait) { 2148 Atomic::store(0, &sigint_count); 2149 for (;;) { 2150 for (int i = 0; i < NSIG + 1; i++) { 2151 jint n = pending_signals[i]; 2152 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2153 return i; 2154 } 2155 } 2156 if (!wait) { 2157 return -1; 2158 } 2159 JavaThread *thread = JavaThread::current(); 2160 ThreadBlockInVM tbivm(thread); 2161 2162 bool threadIsSuspended; 2163 do { 2164 thread->set_suspend_equivalent(); 2165 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2166 ::sem_wait(&sig_sem); 2167 2168 // were we externally suspended while we were waiting? 2169 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2170 if (threadIsSuspended) { 2171 // 2172 // The semaphore has been incremented, but while we were waiting 2173 // another thread suspended us. We don't want to continue running 2174 // while suspended because that would surprise the thread that 2175 // suspended us. 2176 // 2177 ::sem_post(&sig_sem); 2178 2179 thread->java_suspend_self(); 2180 } 2181 } while (threadIsSuspended); 2182 } 2183} 2184 2185int os::signal_lookup() { 2186 return check_pending_signals(false); 2187} 2188 2189int os::signal_wait() { 2190 return check_pending_signals(true); 2191} 2192 2193//////////////////////////////////////////////////////////////////////////////// 2194// Virtual Memory 2195 2196int os::vm_page_size() { 2197 // Seems redundant as all get out 2198 assert(os::Linux::page_size() != -1, "must call os::init"); 2199 return os::Linux::page_size(); 2200} 2201 2202// Solaris allocates memory by pages. 2203int os::vm_allocation_granularity() { 2204 assert(os::Linux::page_size() != -1, "must call os::init"); 2205 return os::Linux::page_size(); 2206} 2207 2208// Rationale behind this function: 2209// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2210// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2211// samples for JITted code. Here we create private executable mapping over the code cache 2212// and then we can use standard (well, almost, as mapping can change) way to provide 2213// info for the reporting script by storing timestamp and location of symbol 2214void linux_wrap_code(char* base, size_t size) { 2215 static volatile jint cnt = 0; 2216 2217 if (!UseOprofile) { 2218 return; 2219 } 2220 2221 char buf[40]; 2222 int num = Atomic::add(1, &cnt); 2223 2224 sprintf(buf, "/tmp/hs-vm-%d-%d", os::current_process_id(), num); 2225 unlink(buf); 2226 2227 int fd = open(buf, O_CREAT | O_RDWR, S_IRWXU); 2228 2229 if (fd != -1) { 2230 off_t rv = lseek(fd, size-2, SEEK_SET); 2231 if (rv != (off_t)-1) { 2232 if (write(fd, "", 1) == 1) { 2233 mmap(base, size, 2234 PROT_READ|PROT_WRITE|PROT_EXEC, 2235 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2236 } 2237 } 2238 close(fd); 2239 unlink(buf); 2240 } 2241} 2242 2243// NOTE: Linux kernel does not really reserve the pages for us. 2244// All it does is to check if there are enough free pages 2245// left at the time of mmap(). This could be a potential 2246// problem. 2247bool os::commit_memory(char* addr, size_t size) { 2248 uintptr_t res = (uintptr_t) ::mmap(addr, size, 2249 PROT_READ|PROT_WRITE|PROT_EXEC, 2250 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2251 return res != (uintptr_t) MAP_FAILED; 2252} 2253 2254bool os::commit_memory(char* addr, size_t size, size_t alignment_hint) { 2255 return commit_memory(addr, size); 2256} 2257 2258void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { } 2259 2260void os::free_memory(char *addr, size_t bytes) { 2261 uncommit_memory(addr, bytes); 2262} 2263 2264void os::numa_make_global(char *addr, size_t bytes) { } 2265 2266void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2267 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2268} 2269 2270bool os::numa_topology_changed() { return false; } 2271 2272size_t os::numa_get_groups_num() { 2273 int max_node = Linux::numa_max_node(); 2274 return max_node > 0 ? max_node + 1 : 1; 2275} 2276 2277int os::numa_get_group_id() { 2278 int cpu_id = Linux::sched_getcpu(); 2279 if (cpu_id != -1) { 2280 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2281 if (lgrp_id != -1) { 2282 return lgrp_id; 2283 } 2284 } 2285 return 0; 2286} 2287 2288size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2289 for (size_t i = 0; i < size; i++) { 2290 ids[i] = i; 2291 } 2292 return size; 2293} 2294 2295bool os::get_page_info(char *start, page_info* info) { 2296 return false; 2297} 2298 2299char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2300 return end; 2301} 2302 2303extern "C" void numa_warn(int number, char *where, ...) { } 2304extern "C" void numa_error(char *where) { } 2305 2306void os::Linux::libnuma_init() { 2307 // sched_getcpu() should be in libc. 2308 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2309 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2310 2311 if (sched_getcpu() != -1) { // Does it work? 2312 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2313 if (handle != NULL) { 2314 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2315 dlsym(handle, "numa_node_to_cpus"))); 2316 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2317 dlsym(handle, "numa_max_node"))); 2318 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2319 dlsym(handle, "numa_available"))); 2320 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2321 dlsym(handle, "numa_tonode_memory"))); 2322 if (numa_available() != -1) { 2323 // Create a cpu -> node mapping 2324 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true); 2325 rebuild_cpu_to_node_map(); 2326 } 2327 } 2328 } 2329} 2330 2331// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2332// The table is later used in get_node_by_cpu(). 2333void os::Linux::rebuild_cpu_to_node_map() { 2334 int cpu_num = os::active_processor_count(); 2335 cpu_to_node()->clear(); 2336 cpu_to_node()->at_grow(cpu_num - 1); 2337 int node_num = numa_get_groups_num(); 2338 int cpu_map_size = (cpu_num + BitsPerLong - 1) / BitsPerLong; 2339 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size); 2340 for (int i = 0; i < node_num; i++) { 2341 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2342 for (int j = 0; j < cpu_map_size; j++) { 2343 if (cpu_map[j] != 0) { 2344 for (int k = 0; k < BitsPerLong; k++) { 2345 if (cpu_map[j] & (1UL << k)) { 2346 cpu_to_node()->at_put(j * BitsPerLong + k, i); 2347 } 2348 } 2349 } 2350 } 2351 } 2352 } 2353 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2354} 2355 2356int os::Linux::get_node_by_cpu(int cpu_id) { 2357 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2358 return cpu_to_node()->at(cpu_id); 2359 } 2360 return -1; 2361} 2362 2363GrowableArray<int>* os::Linux::_cpu_to_node; 2364os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2365os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2366os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2367os::Linux::numa_available_func_t os::Linux::_numa_available; 2368os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2369 2370 2371bool os::uncommit_memory(char* addr, size_t size) { 2372 return ::mmap(addr, size, 2373 PROT_READ|PROT_WRITE|PROT_EXEC, 2374 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0) 2375 != MAP_FAILED; 2376} 2377 2378static address _highest_vm_reserved_address = NULL; 2379 2380// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2381// at 'requested_addr'. If there are existing memory mappings at the same 2382// location, however, they will be overwritten. If 'fixed' is false, 2383// 'requested_addr' is only treated as a hint, the return value may or 2384// may not start from the requested address. Unlike Linux mmap(), this 2385// function returns NULL to indicate failure. 2386static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2387 char * addr; 2388 int flags; 2389 2390 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2391 if (fixed) { 2392 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2393 flags |= MAP_FIXED; 2394 } 2395 2396 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE|PROT_EXEC, 2397 flags, -1, 0); 2398 2399 if (addr != MAP_FAILED) { 2400 // anon_mmap() should only get called during VM initialization, 2401 // don't need lock (actually we can skip locking even it can be called 2402 // from multiple threads, because _highest_vm_reserved_address is just a 2403 // hint about the upper limit of non-stack memory regions.) 2404 if ((address)addr + bytes > _highest_vm_reserved_address) { 2405 _highest_vm_reserved_address = (address)addr + bytes; 2406 } 2407 } 2408 2409 return addr == MAP_FAILED ? NULL : addr; 2410} 2411 2412// Don't update _highest_vm_reserved_address, because there might be memory 2413// regions above addr + size. If so, releasing a memory region only creates 2414// a hole in the address space, it doesn't help prevent heap-stack collision. 2415// 2416static int anon_munmap(char * addr, size_t size) { 2417 return ::munmap(addr, size) == 0; 2418} 2419 2420char* os::reserve_memory(size_t bytes, char* requested_addr, 2421 size_t alignment_hint) { 2422 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 2423} 2424 2425bool os::release_memory(char* addr, size_t size) { 2426 return anon_munmap(addr, size); 2427} 2428 2429static address highest_vm_reserved_address() { 2430 return _highest_vm_reserved_address; 2431} 2432 2433static bool linux_mprotect(char* addr, size_t size, int prot) { 2434 // Linux wants the mprotect address argument to be page aligned. 2435 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 2436 2437 // According to SUSv3, mprotect() should only be used with mappings 2438 // established by mmap(), and mmap() always maps whole pages. Unaligned 2439 // 'addr' likely indicates problem in the VM (e.g. trying to change 2440 // protection of malloc'ed or statically allocated memory). Check the 2441 // caller if you hit this assert. 2442 assert(addr == bottom, "sanity check"); 2443 2444 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 2445 return ::mprotect(bottom, size, prot) == 0; 2446} 2447 2448// Set protections specified 2449bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 2450 bool is_committed) { 2451 unsigned int p = 0; 2452 switch (prot) { 2453 case MEM_PROT_NONE: p = PROT_NONE; break; 2454 case MEM_PROT_READ: p = PROT_READ; break; 2455 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 2456 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 2457 default: 2458 ShouldNotReachHere(); 2459 } 2460 // is_committed is unused. 2461 return linux_mprotect(addr, bytes, p); 2462} 2463 2464bool os::guard_memory(char* addr, size_t size) { 2465 return linux_mprotect(addr, size, PROT_NONE); 2466} 2467 2468bool os::unguard_memory(char* addr, size_t size) { 2469 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE|PROT_EXEC); 2470} 2471 2472// Large page support 2473 2474static size_t _large_page_size = 0; 2475 2476bool os::large_page_init() { 2477 if (!UseLargePages) return false; 2478 2479 if (LargePageSizeInBytes) { 2480 _large_page_size = LargePageSizeInBytes; 2481 } else { 2482 // large_page_size on Linux is used to round up heap size. x86 uses either 2483 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 2484 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 2485 // page as large as 256M. 2486 // 2487 // Here we try to figure out page size by parsing /proc/meminfo and looking 2488 // for a line with the following format: 2489 // Hugepagesize: 2048 kB 2490 // 2491 // If we can't determine the value (e.g. /proc is not mounted, or the text 2492 // format has been changed), we'll use the largest page size supported by 2493 // the processor. 2494 2495 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M); 2496 2497 FILE *fp = fopen("/proc/meminfo", "r"); 2498 if (fp) { 2499 while (!feof(fp)) { 2500 int x = 0; 2501 char buf[16]; 2502 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 2503 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 2504 _large_page_size = x * K; 2505 break; 2506 } 2507 } else { 2508 // skip to next line 2509 for (;;) { 2510 int ch = fgetc(fp); 2511 if (ch == EOF || ch == (int)'\n') break; 2512 } 2513 } 2514 } 2515 fclose(fp); 2516 } 2517 } 2518 2519 const size_t default_page_size = (size_t)Linux::page_size(); 2520 if (_large_page_size > default_page_size) { 2521 _page_sizes[0] = _large_page_size; 2522 _page_sizes[1] = default_page_size; 2523 _page_sizes[2] = 0; 2524 } 2525 2526 // Large page support is available on 2.6 or newer kernel, some vendors 2527 // (e.g. Redhat) have backported it to their 2.4 based distributions. 2528 // We optimistically assume the support is available. If later it turns out 2529 // not true, VM will automatically switch to use regular page size. 2530 return true; 2531} 2532 2533#ifndef SHM_HUGETLB 2534#define SHM_HUGETLB 04000 2535#endif 2536 2537char* os::reserve_memory_special(size_t bytes) { 2538 assert(UseLargePages, "only for large pages"); 2539 2540 key_t key = IPC_PRIVATE; 2541 char *addr; 2542 2543 bool warn_on_failure = UseLargePages && 2544 (!FLAG_IS_DEFAULT(UseLargePages) || 2545 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 2546 ); 2547 char msg[128]; 2548 2549 // Create a large shared memory region to attach to based on size. 2550 // Currently, size is the total size of the heap 2551 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 2552 if (shmid == -1) { 2553 // Possible reasons for shmget failure: 2554 // 1. shmmax is too small for Java heap. 2555 // > check shmmax value: cat /proc/sys/kernel/shmmax 2556 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 2557 // 2. not enough large page memory. 2558 // > check available large pages: cat /proc/meminfo 2559 // > increase amount of large pages: 2560 // echo new_value > /proc/sys/vm/nr_hugepages 2561 // Note 1: different Linux may use different name for this property, 2562 // e.g. on Redhat AS-3 it is "hugetlb_pool". 2563 // Note 2: it's possible there's enough physical memory available but 2564 // they are so fragmented after a long run that they can't 2565 // coalesce into large pages. Try to reserve large pages when 2566 // the system is still "fresh". 2567 if (warn_on_failure) { 2568 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 2569 warning(msg); 2570 } 2571 return NULL; 2572 } 2573 2574 // attach to the region 2575 addr = (char*)shmat(shmid, NULL, 0); 2576 int err = errno; 2577 2578 // Remove shmid. If shmat() is successful, the actual shared memory segment 2579 // will be deleted when it's detached by shmdt() or when the process 2580 // terminates. If shmat() is not successful this will remove the shared 2581 // segment immediately. 2582 shmctl(shmid, IPC_RMID, NULL); 2583 2584 if ((intptr_t)addr == -1) { 2585 if (warn_on_failure) { 2586 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 2587 warning(msg); 2588 } 2589 return NULL; 2590 } 2591 2592 return addr; 2593} 2594 2595bool os::release_memory_special(char* base, size_t bytes) { 2596 // detaching the SHM segment will also delete it, see reserve_memory_special() 2597 int rslt = shmdt(base); 2598 return rslt == 0; 2599} 2600 2601size_t os::large_page_size() { 2602 return _large_page_size; 2603} 2604 2605// Linux does not support anonymous mmap with large page memory. The only way 2606// to reserve large page memory without file backing is through SysV shared 2607// memory API. The entire memory region is committed and pinned upfront. 2608// Hopefully this will change in the future... 2609bool os::can_commit_large_page_memory() { 2610 return false; 2611} 2612 2613bool os::can_execute_large_page_memory() { 2614 return false; 2615} 2616 2617// Reserve memory at an arbitrary address, only if that area is 2618// available (and not reserved for something else). 2619 2620char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 2621 const int max_tries = 10; 2622 char* base[max_tries]; 2623 size_t size[max_tries]; 2624 const size_t gap = 0x000000; 2625 2626 // Assert only that the size is a multiple of the page size, since 2627 // that's all that mmap requires, and since that's all we really know 2628 // about at this low abstraction level. If we need higher alignment, 2629 // we can either pass an alignment to this method or verify alignment 2630 // in one of the methods further up the call chain. See bug 5044738. 2631 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 2632 2633 // Repeatedly allocate blocks until the block is allocated at the 2634 // right spot. Give up after max_tries. Note that reserve_memory() will 2635 // automatically update _highest_vm_reserved_address if the call is 2636 // successful. The variable tracks the highest memory address every reserved 2637 // by JVM. It is used to detect heap-stack collision if running with 2638 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 2639 // space than needed, it could confuse the collision detecting code. To 2640 // solve the problem, save current _highest_vm_reserved_address and 2641 // calculate the correct value before return. 2642 address old_highest = _highest_vm_reserved_address; 2643 2644 // Linux mmap allows caller to pass an address as hint; give it a try first, 2645 // if kernel honors the hint then we can return immediately. 2646 char * addr = anon_mmap(requested_addr, bytes, false); 2647 if (addr == requested_addr) { 2648 return requested_addr; 2649 } 2650 2651 if (addr != NULL) { 2652 // mmap() is successful but it fails to reserve at the requested address 2653 anon_munmap(addr, bytes); 2654 } 2655 2656 int i; 2657 for (i = 0; i < max_tries; ++i) { 2658 base[i] = reserve_memory(bytes); 2659 2660 if (base[i] != NULL) { 2661 // Is this the block we wanted? 2662 if (base[i] == requested_addr) { 2663 size[i] = bytes; 2664 break; 2665 } 2666 2667 // Does this overlap the block we wanted? Give back the overlapped 2668 // parts and try again. 2669 2670 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 2671 if (top_overlap >= 0 && top_overlap < bytes) { 2672 unmap_memory(base[i], top_overlap); 2673 base[i] += top_overlap; 2674 size[i] = bytes - top_overlap; 2675 } else { 2676 size_t bottom_overlap = base[i] + bytes - requested_addr; 2677 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 2678 unmap_memory(requested_addr, bottom_overlap); 2679 size[i] = bytes - bottom_overlap; 2680 } else { 2681 size[i] = bytes; 2682 } 2683 } 2684 } 2685 } 2686 2687 // Give back the unused reserved pieces. 2688 2689 for (int j = 0; j < i; ++j) { 2690 if (base[j] != NULL) { 2691 unmap_memory(base[j], size[j]); 2692 } 2693 } 2694 2695 if (i < max_tries) { 2696 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 2697 return requested_addr; 2698 } else { 2699 _highest_vm_reserved_address = old_highest; 2700 return NULL; 2701 } 2702} 2703 2704size_t os::read(int fd, void *buf, unsigned int nBytes) { 2705 return ::read(fd, buf, nBytes); 2706} 2707 2708// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 2709// Solaris uses poll(), linux uses park(). 2710// Poll() is likely a better choice, assuming that Thread.interrupt() 2711// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 2712// SIGSEGV, see 4355769. 2713 2714const int NANOSECS_PER_MILLISECS = 1000000; 2715 2716int os::sleep(Thread* thread, jlong millis, bool interruptible) { 2717 assert(thread == Thread::current(), "thread consistency check"); 2718 2719 ParkEvent * const slp = thread->_SleepEvent ; 2720 slp->reset() ; 2721 OrderAccess::fence() ; 2722 2723 if (interruptible) { 2724 jlong prevtime = javaTimeNanos(); 2725 2726 for (;;) { 2727 if (os::is_interrupted(thread, true)) { 2728 return OS_INTRPT; 2729 } 2730 2731 jlong newtime = javaTimeNanos(); 2732 2733 if (newtime - prevtime < 0) { 2734 // time moving backwards, should only happen if no monotonic clock 2735 // not a guarantee() because JVM should not abort on kernel/glibc bugs 2736 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 2737 } else { 2738 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 2739 } 2740 2741 if(millis <= 0) { 2742 return OS_OK; 2743 } 2744 2745 prevtime = newtime; 2746 2747 { 2748 assert(thread->is_Java_thread(), "sanity check"); 2749 JavaThread *jt = (JavaThread *) thread; 2750 ThreadBlockInVM tbivm(jt); 2751 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 2752 2753 jt->set_suspend_equivalent(); 2754 // cleared by handle_special_suspend_equivalent_condition() or 2755 // java_suspend_self() via check_and_wait_while_suspended() 2756 2757 slp->park(millis); 2758 2759 // were we externally suspended while we were waiting? 2760 jt->check_and_wait_while_suspended(); 2761 } 2762 } 2763 } else { 2764 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 2765 jlong prevtime = javaTimeNanos(); 2766 2767 for (;;) { 2768 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 2769 // the 1st iteration ... 2770 jlong newtime = javaTimeNanos(); 2771 2772 if (newtime - prevtime < 0) { 2773 // time moving backwards, should only happen if no monotonic clock 2774 // not a guarantee() because JVM should not abort on kernel/glibc bugs 2775 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 2776 } else { 2777 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 2778 } 2779 2780 if(millis <= 0) break ; 2781 2782 prevtime = newtime; 2783 slp->park(millis); 2784 } 2785 return OS_OK ; 2786 } 2787} 2788 2789int os::naked_sleep() { 2790 // %% make the sleep time an integer flag. for now use 1 millisec. 2791 return os::sleep(Thread::current(), 1, false); 2792} 2793 2794// Sleep forever; naked call to OS-specific sleep; use with CAUTION 2795void os::infinite_sleep() { 2796 while (true) { // sleep forever ... 2797 ::sleep(100); // ... 100 seconds at a time 2798 } 2799} 2800 2801// Used to convert frequent JVM_Yield() to nops 2802bool os::dont_yield() { 2803 return DontYieldALot; 2804} 2805 2806void os::yield() { 2807 sched_yield(); 2808} 2809 2810os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 2811 2812void os::yield_all(int attempts) { 2813 // Yields to all threads, including threads with lower priorities 2814 // Threads on Linux are all with same priority. The Solaris style 2815 // os::yield_all() with nanosleep(1ms) is not necessary. 2816 sched_yield(); 2817} 2818 2819// Called from the tight loops to possibly influence time-sharing heuristics 2820void os::loop_breaker(int attempts) { 2821 os::yield_all(attempts); 2822} 2823 2824//////////////////////////////////////////////////////////////////////////////// 2825// thread priority support 2826 2827// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 2828// only supports dynamic priority, static priority must be zero. For real-time 2829// applications, Linux supports SCHED_RR which allows static priority (1-99). 2830// However, for large multi-threaded applications, SCHED_RR is not only slower 2831// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 2832// of 5 runs - Sep 2005). 2833// 2834// The following code actually changes the niceness of kernel-thread/LWP. It 2835// has an assumption that setpriority() only modifies one kernel-thread/LWP, 2836// not the entire user process, and user level threads are 1:1 mapped to kernel 2837// threads. It has always been the case, but could change in the future. For 2838// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 2839// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 2840 2841int os::java_to_os_priority[MaxPriority + 1] = { 2842 19, // 0 Entry should never be used 2843 2844 4, // 1 MinPriority 2845 3, // 2 2846 2, // 3 2847 2848 1, // 4 2849 0, // 5 NormPriority 2850 -1, // 6 2851 2852 -2, // 7 2853 -3, // 8 2854 -4, // 9 NearMaxPriority 2855 2856 -5 // 10 MaxPriority 2857}; 2858 2859static int prio_init() { 2860 if (ThreadPriorityPolicy == 1) { 2861 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 2862 // if effective uid is not root. Perhaps, a more elegant way of doing 2863 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 2864 if (geteuid() != 0) { 2865 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 2866 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 2867 } 2868 ThreadPriorityPolicy = 0; 2869 } 2870 } 2871 return 0; 2872} 2873 2874OSReturn os::set_native_priority(Thread* thread, int newpri) { 2875 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 2876 2877 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 2878 return (ret == 0) ? OS_OK : OS_ERR; 2879} 2880 2881OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 2882 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 2883 *priority_ptr = java_to_os_priority[NormPriority]; 2884 return OS_OK; 2885 } 2886 2887 errno = 0; 2888 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 2889 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 2890} 2891 2892// Hint to the underlying OS that a task switch would not be good. 2893// Void return because it's a hint and can fail. 2894void os::hint_no_preempt() {} 2895 2896//////////////////////////////////////////////////////////////////////////////// 2897// suspend/resume support 2898 2899// the low-level signal-based suspend/resume support is a remnant from the 2900// old VM-suspension that used to be for java-suspension, safepoints etc, 2901// within hotspot. Now there is a single use-case for this: 2902// - calling get_thread_pc() on the VMThread by the flat-profiler task 2903// that runs in the watcher thread. 2904// The remaining code is greatly simplified from the more general suspension 2905// code that used to be used. 2906// 2907// The protocol is quite simple: 2908// - suspend: 2909// - sends a signal to the target thread 2910// - polls the suspend state of the osthread using a yield loop 2911// - target thread signal handler (SR_handler) sets suspend state 2912// and blocks in sigsuspend until continued 2913// - resume: 2914// - sets target osthread state to continue 2915// - sends signal to end the sigsuspend loop in the SR_handler 2916// 2917// Note that the SR_lock plays no role in this suspend/resume protocol. 2918// 2919 2920static void resume_clear_context(OSThread *osthread) { 2921 osthread->set_ucontext(NULL); 2922 osthread->set_siginfo(NULL); 2923 2924 // notify the suspend action is completed, we have now resumed 2925 osthread->sr.clear_suspended(); 2926} 2927 2928static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 2929 osthread->set_ucontext(context); 2930 osthread->set_siginfo(siginfo); 2931} 2932 2933// 2934// Handler function invoked when a thread's execution is suspended or 2935// resumed. We have to be careful that only async-safe functions are 2936// called here (Note: most pthread functions are not async safe and 2937// should be avoided.) 2938// 2939// Note: sigwait() is a more natural fit than sigsuspend() from an 2940// interface point of view, but sigwait() prevents the signal hander 2941// from being run. libpthread would get very confused by not having 2942// its signal handlers run and prevents sigwait()'s use with the 2943// mutex granting granting signal. 2944// 2945// Currently only ever called on the VMThread 2946// 2947static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 2948 // Save and restore errno to avoid confusing native code with EINTR 2949 // after sigsuspend. 2950 int old_errno = errno; 2951 2952 Thread* thread = Thread::current(); 2953 OSThread* osthread = thread->osthread(); 2954 assert(thread->is_VM_thread(), "Must be VMThread"); 2955 // read current suspend action 2956 int action = osthread->sr.suspend_action(); 2957 if (action == SR_SUSPEND) { 2958 suspend_save_context(osthread, siginfo, context); 2959 2960 // Notify the suspend action is about to be completed. do_suspend() 2961 // waits until SR_SUSPENDED is set and then returns. We will wait 2962 // here for a resume signal and that completes the suspend-other 2963 // action. do_suspend/do_resume is always called as a pair from 2964 // the same thread - so there are no races 2965 2966 // notify the caller 2967 osthread->sr.set_suspended(); 2968 2969 sigset_t suspend_set; // signals for sigsuspend() 2970 2971 // get current set of blocked signals and unblock resume signal 2972 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 2973 sigdelset(&suspend_set, SR_signum); 2974 2975 // wait here until we are resumed 2976 do { 2977 sigsuspend(&suspend_set); 2978 // ignore all returns until we get a resume signal 2979 } while (osthread->sr.suspend_action() != SR_CONTINUE); 2980 2981 resume_clear_context(osthread); 2982 2983 } else { 2984 assert(action == SR_CONTINUE, "unexpected sr action"); 2985 // nothing special to do - just leave the handler 2986 } 2987 2988 errno = old_errno; 2989} 2990 2991 2992static int SR_initialize() { 2993 struct sigaction act; 2994 char *s; 2995 /* Get signal number to use for suspend/resume */ 2996 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 2997 int sig = ::strtol(s, 0, 10); 2998 if (sig > 0 || sig < _NSIG) { 2999 SR_signum = sig; 3000 } 3001 } 3002 3003 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3004 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3005 3006 sigemptyset(&SR_sigset); 3007 sigaddset(&SR_sigset, SR_signum); 3008 3009 /* Set up signal handler for suspend/resume */ 3010 act.sa_flags = SA_RESTART|SA_SIGINFO; 3011 act.sa_handler = (void (*)(int)) SR_handler; 3012 3013 // SR_signum is blocked by default. 3014 // 4528190 - We also need to block pthread restart signal (32 on all 3015 // supported Linux platforms). Note that LinuxThreads need to block 3016 // this signal for all threads to work properly. So we don't have 3017 // to use hard-coded signal number when setting up the mask. 3018 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 3019 3020 if (sigaction(SR_signum, &act, 0) == -1) { 3021 return -1; 3022 } 3023 3024 // Save signal flag 3025 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 3026 return 0; 3027} 3028 3029static int SR_finalize() { 3030 return 0; 3031} 3032 3033 3034// returns true on success and false on error - really an error is fatal 3035// but this seems the normal response to library errors 3036static bool do_suspend(OSThread* osthread) { 3037 // mark as suspended and send signal 3038 osthread->sr.set_suspend_action(SR_SUSPEND); 3039 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3040 assert_status(status == 0, status, "pthread_kill"); 3041 3042 // check status and wait until notified of suspension 3043 if (status == 0) { 3044 for (int i = 0; !osthread->sr.is_suspended(); i++) { 3045 os::yield_all(i); 3046 } 3047 osthread->sr.set_suspend_action(SR_NONE); 3048 return true; 3049 } 3050 else { 3051 osthread->sr.set_suspend_action(SR_NONE); 3052 return false; 3053 } 3054} 3055 3056static void do_resume(OSThread* osthread) { 3057 assert(osthread->sr.is_suspended(), "thread should be suspended"); 3058 osthread->sr.set_suspend_action(SR_CONTINUE); 3059 3060 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3061 assert_status(status == 0, status, "pthread_kill"); 3062 // check status and wait unit notified of resumption 3063 if (status == 0) { 3064 for (int i = 0; osthread->sr.is_suspended(); i++) { 3065 os::yield_all(i); 3066 } 3067 } 3068 osthread->sr.set_suspend_action(SR_NONE); 3069} 3070 3071//////////////////////////////////////////////////////////////////////////////// 3072// interrupt support 3073 3074void os::interrupt(Thread* thread) { 3075 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3076 "possibility of dangling Thread pointer"); 3077 3078 OSThread* osthread = thread->osthread(); 3079 3080 if (!osthread->interrupted()) { 3081 osthread->set_interrupted(true); 3082 // More than one thread can get here with the same value of osthread, 3083 // resulting in multiple notifications. We do, however, want the store 3084 // to interrupted() to be visible to other threads before we execute unpark(). 3085 OrderAccess::fence(); 3086 ParkEvent * const slp = thread->_SleepEvent ; 3087 if (slp != NULL) slp->unpark() ; 3088 } 3089 3090 // For JSR166. Unpark even if interrupt status already was set 3091 if (thread->is_Java_thread()) 3092 ((JavaThread*)thread)->parker()->unpark(); 3093 3094 ParkEvent * ev = thread->_ParkEvent ; 3095 if (ev != NULL) ev->unpark() ; 3096 3097} 3098 3099bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 3100 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3101 "possibility of dangling Thread pointer"); 3102 3103 OSThread* osthread = thread->osthread(); 3104 3105 bool interrupted = osthread->interrupted(); 3106 3107 if (interrupted && clear_interrupted) { 3108 osthread->set_interrupted(false); 3109 // consider thread->_SleepEvent->reset() ... optional optimization 3110 } 3111 3112 return interrupted; 3113} 3114 3115/////////////////////////////////////////////////////////////////////////////////// 3116// signal handling (except suspend/resume) 3117 3118// This routine may be used by user applications as a "hook" to catch signals. 3119// The user-defined signal handler must pass unrecognized signals to this 3120// routine, and if it returns true (non-zero), then the signal handler must 3121// return immediately. If the flag "abort_if_unrecognized" is true, then this 3122// routine will never retun false (zero), but instead will execute a VM panic 3123// routine kill the process. 3124// 3125// If this routine returns false, it is OK to call it again. This allows 3126// the user-defined signal handler to perform checks either before or after 3127// the VM performs its own checks. Naturally, the user code would be making 3128// a serious error if it tried to handle an exception (such as a null check 3129// or breakpoint) that the VM was generating for its own correct operation. 3130// 3131// This routine may recognize any of the following kinds of signals: 3132// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 3133// It should be consulted by handlers for any of those signals. 3134// 3135// The caller of this routine must pass in the three arguments supplied 3136// to the function referred to in the "sa_sigaction" (not the "sa_handler") 3137// field of the structure passed to sigaction(). This routine assumes that 3138// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 3139// 3140// Note that the VM will print warnings if it detects conflicting signal 3141// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 3142// 3143extern "C" int 3144JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 3145 void* ucontext, int abort_if_unrecognized); 3146 3147void signalHandler(int sig, siginfo_t* info, void* uc) { 3148 assert(info != NULL && uc != NULL, "it must be old kernel"); 3149 JVM_handle_linux_signal(sig, info, uc, true); 3150} 3151 3152 3153// This boolean allows users to forward their own non-matching signals 3154// to JVM_handle_linux_signal, harmlessly. 3155bool os::Linux::signal_handlers_are_installed = false; 3156 3157// For signal-chaining 3158struct sigaction os::Linux::sigact[MAXSIGNUM]; 3159unsigned int os::Linux::sigs = 0; 3160bool os::Linux::libjsig_is_loaded = false; 3161typedef struct sigaction *(*get_signal_t)(int); 3162get_signal_t os::Linux::get_signal_action = NULL; 3163 3164struct sigaction* os::Linux::get_chained_signal_action(int sig) { 3165 struct sigaction *actp = NULL; 3166 3167 if (libjsig_is_loaded) { 3168 // Retrieve the old signal handler from libjsig 3169 actp = (*get_signal_action)(sig); 3170 } 3171 if (actp == NULL) { 3172 // Retrieve the preinstalled signal handler from jvm 3173 actp = get_preinstalled_handler(sig); 3174 } 3175 3176 return actp; 3177} 3178 3179static bool call_chained_handler(struct sigaction *actp, int sig, 3180 siginfo_t *siginfo, void *context) { 3181 // Call the old signal handler 3182 if (actp->sa_handler == SIG_DFL) { 3183 // It's more reasonable to let jvm treat it as an unexpected exception 3184 // instead of taking the default action. 3185 return false; 3186 } else if (actp->sa_handler != SIG_IGN) { 3187 if ((actp->sa_flags & SA_NODEFER) == 0) { 3188 // automaticlly block the signal 3189 sigaddset(&(actp->sa_mask), sig); 3190 } 3191 3192 sa_handler_t hand; 3193 sa_sigaction_t sa; 3194 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 3195 // retrieve the chained handler 3196 if (siginfo_flag_set) { 3197 sa = actp->sa_sigaction; 3198 } else { 3199 hand = actp->sa_handler; 3200 } 3201 3202 if ((actp->sa_flags & SA_RESETHAND) != 0) { 3203 actp->sa_handler = SIG_DFL; 3204 } 3205 3206 // try to honor the signal mask 3207 sigset_t oset; 3208 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 3209 3210 // call into the chained handler 3211 if (siginfo_flag_set) { 3212 (*sa)(sig, siginfo, context); 3213 } else { 3214 (*hand)(sig); 3215 } 3216 3217 // restore the signal mask 3218 pthread_sigmask(SIG_SETMASK, &oset, 0); 3219 } 3220 // Tell jvm's signal handler the signal is taken care of. 3221 return true; 3222} 3223 3224bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 3225 bool chained = false; 3226 // signal-chaining 3227 if (UseSignalChaining) { 3228 struct sigaction *actp = get_chained_signal_action(sig); 3229 if (actp != NULL) { 3230 chained = call_chained_handler(actp, sig, siginfo, context); 3231 } 3232 } 3233 return chained; 3234} 3235 3236struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 3237 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 3238 return &sigact[sig]; 3239 } 3240 return NULL; 3241} 3242 3243void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 3244 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3245 sigact[sig] = oldAct; 3246 sigs |= (unsigned int)1 << sig; 3247} 3248 3249// for diagnostic 3250int os::Linux::sigflags[MAXSIGNUM]; 3251 3252int os::Linux::get_our_sigflags(int sig) { 3253 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3254 return sigflags[sig]; 3255} 3256 3257void os::Linux::set_our_sigflags(int sig, int flags) { 3258 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3259 sigflags[sig] = flags; 3260} 3261 3262void os::Linux::set_signal_handler(int sig, bool set_installed) { 3263 // Check for overwrite. 3264 struct sigaction oldAct; 3265 sigaction(sig, (struct sigaction*)NULL, &oldAct); 3266 3267 void* oldhand = oldAct.sa_sigaction 3268 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3269 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3270 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 3271 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 3272 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 3273 if (AllowUserSignalHandlers || !set_installed) { 3274 // Do not overwrite; user takes responsibility to forward to us. 3275 return; 3276 } else if (UseSignalChaining) { 3277 // save the old handler in jvm 3278 save_preinstalled_handler(sig, oldAct); 3279 // libjsig also interposes the sigaction() call below and saves the 3280 // old sigaction on it own. 3281 } else { 3282 fatal2("Encountered unexpected pre-existing sigaction handler %#lx for signal %d.", (long)oldhand, sig); 3283 } 3284 } 3285 3286 struct sigaction sigAct; 3287 sigfillset(&(sigAct.sa_mask)); 3288 sigAct.sa_handler = SIG_DFL; 3289 if (!set_installed) { 3290 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3291 } else { 3292 sigAct.sa_sigaction = signalHandler; 3293 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3294 } 3295 // Save flags, which are set by ours 3296 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3297 sigflags[sig] = sigAct.sa_flags; 3298 3299 int ret = sigaction(sig, &sigAct, &oldAct); 3300 assert(ret == 0, "check"); 3301 3302 void* oldhand2 = oldAct.sa_sigaction 3303 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3304 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3305 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 3306} 3307 3308// install signal handlers for signals that HotSpot needs to 3309// handle in order to support Java-level exception handling. 3310 3311void os::Linux::install_signal_handlers() { 3312 if (!signal_handlers_are_installed) { 3313 signal_handlers_are_installed = true; 3314 3315 // signal-chaining 3316 typedef void (*signal_setting_t)(); 3317 signal_setting_t begin_signal_setting = NULL; 3318 signal_setting_t end_signal_setting = NULL; 3319 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3320 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 3321 if (begin_signal_setting != NULL) { 3322 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3323 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 3324 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 3325 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 3326 libjsig_is_loaded = true; 3327 assert(UseSignalChaining, "should enable signal-chaining"); 3328 } 3329 if (libjsig_is_loaded) { 3330 // Tell libjsig jvm is setting signal handlers 3331 (*begin_signal_setting)(); 3332 } 3333 3334 set_signal_handler(SIGSEGV, true); 3335 set_signal_handler(SIGPIPE, true); 3336 set_signal_handler(SIGBUS, true); 3337 set_signal_handler(SIGILL, true); 3338 set_signal_handler(SIGFPE, true); 3339 set_signal_handler(SIGXFSZ, true); 3340 3341 if (libjsig_is_loaded) { 3342 // Tell libjsig jvm finishes setting signal handlers 3343 (*end_signal_setting)(); 3344 } 3345 3346 // We don't activate signal checker if libjsig is in place, we trust ourselves 3347 // and if UserSignalHandler is installed all bets are off 3348 if (CheckJNICalls) { 3349 if (libjsig_is_loaded) { 3350 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 3351 check_signals = false; 3352 } 3353 if (AllowUserSignalHandlers) { 3354 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 3355 check_signals = false; 3356 } 3357 } 3358 } 3359} 3360 3361// This is the fastest way to get thread cpu time on Linux. 3362// Returns cpu time (user+sys) for any thread, not only for current. 3363// POSIX compliant clocks are implemented in the kernels 2.6.16+. 3364// It might work on 2.6.10+ with a special kernel/glibc patch. 3365// For reference, please, see IEEE Std 1003.1-2004: 3366// http://www.unix.org/single_unix_specification 3367 3368jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 3369 struct timespec tp; 3370 int rc = os::Linux::clock_gettime(clockid, &tp); 3371 assert(rc == 0, "clock_gettime is expected to return 0 code"); 3372 3373 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec; 3374} 3375 3376///// 3377// glibc on Linux platform uses non-documented flag 3378// to indicate, that some special sort of signal 3379// trampoline is used. 3380// We will never set this flag, and we should 3381// ignore this flag in our diagnostic 3382#ifdef SIGNIFICANT_SIGNAL_MASK 3383#undef SIGNIFICANT_SIGNAL_MASK 3384#endif 3385#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 3386 3387static const char* get_signal_handler_name(address handler, 3388 char* buf, int buflen) { 3389 int offset; 3390 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 3391 if (found) { 3392 // skip directory names 3393 const char *p1, *p2; 3394 p1 = buf; 3395 size_t len = strlen(os::file_separator()); 3396 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 3397 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 3398 } else { 3399 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 3400 } 3401 return buf; 3402} 3403 3404static void print_signal_handler(outputStream* st, int sig, 3405 char* buf, size_t buflen) { 3406 struct sigaction sa; 3407 3408 sigaction(sig, NULL, &sa); 3409 3410 // See comment for SIGNIFICANT_SIGNAL_MASK define 3411 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3412 3413 st->print("%s: ", os::exception_name(sig, buf, buflen)); 3414 3415 address handler = (sa.sa_flags & SA_SIGINFO) 3416 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 3417 : CAST_FROM_FN_PTR(address, sa.sa_handler); 3418 3419 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 3420 st->print("SIG_DFL"); 3421 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 3422 st->print("SIG_IGN"); 3423 } else { 3424 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 3425 } 3426 3427 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 3428 3429 address rh = VMError::get_resetted_sighandler(sig); 3430 // May be, handler was resetted by VMError? 3431 if(rh != NULL) { 3432 handler = rh; 3433 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 3434 } 3435 3436 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 3437 3438 // Check: is it our handler? 3439 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 3440 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 3441 // It is our signal handler 3442 // check for flags, reset system-used one! 3443 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 3444 st->print( 3445 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 3446 os::Linux::get_our_sigflags(sig)); 3447 } 3448 } 3449 st->cr(); 3450} 3451 3452 3453#define DO_SIGNAL_CHECK(sig) \ 3454 if (!sigismember(&check_signal_done, sig)) \ 3455 os::Linux::check_signal_handler(sig) 3456 3457// This method is a periodic task to check for misbehaving JNI applications 3458// under CheckJNI, we can add any periodic checks here 3459 3460void os::run_periodic_checks() { 3461 3462 if (check_signals == false) return; 3463 3464 // SEGV and BUS if overridden could potentially prevent 3465 // generation of hs*.log in the event of a crash, debugging 3466 // such a case can be very challenging, so we absolutely 3467 // check the following for a good measure: 3468 DO_SIGNAL_CHECK(SIGSEGV); 3469 DO_SIGNAL_CHECK(SIGILL); 3470 DO_SIGNAL_CHECK(SIGFPE); 3471 DO_SIGNAL_CHECK(SIGBUS); 3472 DO_SIGNAL_CHECK(SIGPIPE); 3473 DO_SIGNAL_CHECK(SIGXFSZ); 3474 3475 3476 // ReduceSignalUsage allows the user to override these handlers 3477 // see comments at the very top and jvm_solaris.h 3478 if (!ReduceSignalUsage) { 3479 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 3480 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 3481 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 3482 DO_SIGNAL_CHECK(BREAK_SIGNAL); 3483 } 3484 3485 DO_SIGNAL_CHECK(SR_signum); 3486 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 3487} 3488 3489typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 3490 3491static os_sigaction_t os_sigaction = NULL; 3492 3493void os::Linux::check_signal_handler(int sig) { 3494 char buf[O_BUFLEN]; 3495 address jvmHandler = NULL; 3496 3497 3498 struct sigaction act; 3499 if (os_sigaction == NULL) { 3500 // only trust the default sigaction, in case it has been interposed 3501 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 3502 if (os_sigaction == NULL) return; 3503 } 3504 3505 os_sigaction(sig, (struct sigaction*)NULL, &act); 3506 3507 3508 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3509 3510 address thisHandler = (act.sa_flags & SA_SIGINFO) 3511 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 3512 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 3513 3514 3515 switch(sig) { 3516 case SIGSEGV: 3517 case SIGBUS: 3518 case SIGFPE: 3519 case SIGPIPE: 3520 case SIGILL: 3521 case SIGXFSZ: 3522 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 3523 break; 3524 3525 case SHUTDOWN1_SIGNAL: 3526 case SHUTDOWN2_SIGNAL: 3527 case SHUTDOWN3_SIGNAL: 3528 case BREAK_SIGNAL: 3529 jvmHandler = (address)user_handler(); 3530 break; 3531 3532 case INTERRUPT_SIGNAL: 3533 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 3534 break; 3535 3536 default: 3537 if (sig == SR_signum) { 3538 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 3539 } else { 3540 return; 3541 } 3542 break; 3543 } 3544 3545 if (thisHandler != jvmHandler) { 3546 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 3547 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 3548 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 3549 // No need to check this sig any longer 3550 sigaddset(&check_signal_done, sig); 3551 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 3552 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 3553 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 3554 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 3555 // No need to check this sig any longer 3556 sigaddset(&check_signal_done, sig); 3557 } 3558 3559 // Dump all the signal 3560 if (sigismember(&check_signal_done, sig)) { 3561 print_signal_handlers(tty, buf, O_BUFLEN); 3562 } 3563} 3564 3565extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 3566 3567extern bool signal_name(int signo, char* buf, size_t len); 3568 3569const char* os::exception_name(int exception_code, char* buf, size_t size) { 3570 if (0 < exception_code && exception_code <= SIGRTMAX) { 3571 // signal 3572 if (!signal_name(exception_code, buf, size)) { 3573 jio_snprintf(buf, size, "SIG%d", exception_code); 3574 } 3575 return buf; 3576 } else { 3577 return NULL; 3578 } 3579} 3580 3581// this is called _before_ the most of global arguments have been parsed 3582void os::init(void) { 3583 char dummy; /* used to get a guess on initial stack address */ 3584// first_hrtime = gethrtime(); 3585 3586 // With LinuxThreads the JavaMain thread pid (primordial thread) 3587 // is different than the pid of the java launcher thread. 3588 // So, on Linux, the launcher thread pid is passed to the VM 3589 // via the sun.java.launcher.pid property. 3590 // Use this property instead of getpid() if it was correctly passed. 3591 // See bug 6351349. 3592 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 3593 3594 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 3595 3596 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 3597 3598 init_random(1234567); 3599 3600 ThreadCritical::initialize(); 3601 3602 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 3603 if (Linux::page_size() == -1) { 3604 fatal1("os_linux.cpp: os::init: sysconf failed (%s)", strerror(errno)); 3605 } 3606 init_page_sizes((size_t) Linux::page_size()); 3607 3608 Linux::initialize_system_info(); 3609 3610 // main_thread points to the aboriginal thread 3611 Linux::_main_thread = pthread_self(); 3612 3613 Linux::clock_init(); 3614 initial_time_count = os::elapsed_counter(); 3615 pthread_mutex_init(&dl_mutex, NULL); 3616} 3617 3618// To install functions for atexit system call 3619extern "C" { 3620 static void perfMemory_exit_helper() { 3621 perfMemory_exit(); 3622 } 3623} 3624 3625// this is called _after_ the global arguments have been parsed 3626jint os::init_2(void) 3627{ 3628 Linux::fast_thread_clock_init(); 3629 3630 // Allocate a single page and mark it as readable for safepoint polling 3631 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3632 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 3633 3634 os::set_polling_page( polling_page ); 3635 3636#ifndef PRODUCT 3637 if(Verbose && PrintMiscellaneous) 3638 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 3639#endif 3640 3641 if (!UseMembar) { 3642 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3643 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); 3644 os::set_memory_serialize_page( mem_serialize_page ); 3645 3646#ifndef PRODUCT 3647 if(Verbose && PrintMiscellaneous) 3648 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 3649#endif 3650 } 3651 3652 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init()); 3653 3654 // initialize suspend/resume support - must do this before signal_sets_init() 3655 if (SR_initialize() != 0) { 3656 perror("SR_initialize failed"); 3657 return JNI_ERR; 3658 } 3659 3660 Linux::signal_sets_init(); 3661 Linux::install_signal_handlers(); 3662 3663 size_t threadStackSizeInBytes = ThreadStackSize * K; 3664 if (threadStackSizeInBytes != 0 && 3665 threadStackSizeInBytes < Linux::min_stack_allowed) { 3666 tty->print_cr("\nThe stack size specified is too small, " 3667 "Specify at least %dk", 3668 Linux::min_stack_allowed / K); 3669 return JNI_ERR; 3670 } 3671 3672 // Make the stack size a multiple of the page size so that 3673 // the yellow/red zones can be guarded. 3674 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 3675 vm_page_size())); 3676 3677 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 3678 3679 Linux::libpthread_init(); 3680 if (PrintMiscellaneous && (Verbose || WizardMode)) { 3681 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 3682 Linux::glibc_version(), Linux::libpthread_version(), 3683 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 3684 } 3685 3686 if (UseNUMA) { 3687 Linux::libnuma_init(); 3688 } 3689 3690 if (MaxFDLimit) { 3691 // set the number of file descriptors to max. print out error 3692 // if getrlimit/setrlimit fails but continue regardless. 3693 struct rlimit nbr_files; 3694 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 3695 if (status != 0) { 3696 if (PrintMiscellaneous && (Verbose || WizardMode)) 3697 perror("os::init_2 getrlimit failed"); 3698 } else { 3699 nbr_files.rlim_cur = nbr_files.rlim_max; 3700 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 3701 if (status != 0) { 3702 if (PrintMiscellaneous && (Verbose || WizardMode)) 3703 perror("os::init_2 setrlimit failed"); 3704 } 3705 } 3706 } 3707 3708 // Initialize lock used to serialize thread creation (see os::create_thread) 3709 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 3710 3711 // Initialize HPI. 3712 jint hpi_result = hpi::initialize(); 3713 if (hpi_result != JNI_OK) { 3714 tty->print_cr("There was an error trying to initialize the HPI library."); 3715 return hpi_result; 3716 } 3717 3718 // at-exit methods are called in the reverse order of their registration. 3719 // atexit functions are called on return from main or as a result of a 3720 // call to exit(3C). There can be only 32 of these functions registered 3721 // and atexit() does not set errno. 3722 3723 if (PerfAllowAtExitRegistration) { 3724 // only register atexit functions if PerfAllowAtExitRegistration is set. 3725 // atexit functions can be delayed until process exit time, which 3726 // can be problematic for embedded VM situations. Embedded VMs should 3727 // call DestroyJavaVM() to assure that VM resources are released. 3728 3729 // note: perfMemory_exit_helper atexit function may be removed in 3730 // the future if the appropriate cleanup code can be added to the 3731 // VM_Exit VMOperation's doit method. 3732 if (atexit(perfMemory_exit_helper) != 0) { 3733 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 3734 } 3735 } 3736 3737 // initialize thread priority policy 3738 prio_init(); 3739 3740 return JNI_OK; 3741} 3742 3743// Mark the polling page as unreadable 3744void os::make_polling_page_unreadable(void) { 3745 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 3746 fatal("Could not disable polling page"); 3747}; 3748 3749// Mark the polling page as readable 3750void os::make_polling_page_readable(void) { 3751 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 3752 fatal("Could not enable polling page"); 3753 } 3754}; 3755 3756int os::active_processor_count() { 3757 // Linux doesn't yet have a (official) notion of processor sets, 3758 // so just return the number of online processors. 3759 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 3760 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 3761 return online_cpus; 3762} 3763 3764bool os::distribute_processes(uint length, uint* distribution) { 3765 // Not yet implemented. 3766 return false; 3767} 3768 3769bool os::bind_to_processor(uint processor_id) { 3770 // Not yet implemented. 3771 return false; 3772} 3773 3774/// 3775 3776// Suspends the target using the signal mechanism and then grabs the PC before 3777// resuming the target. Used by the flat-profiler only 3778ExtendedPC os::get_thread_pc(Thread* thread) { 3779 // Make sure that it is called by the watcher for the VMThread 3780 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 3781 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 3782 3783 ExtendedPC epc; 3784 3785 OSThread* osthread = thread->osthread(); 3786 if (do_suspend(osthread)) { 3787 if (osthread->ucontext() != NULL) { 3788 epc = os::Linux::ucontext_get_pc(osthread->ucontext()); 3789 } else { 3790 // NULL context is unexpected, double-check this is the VMThread 3791 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 3792 } 3793 do_resume(osthread); 3794 } 3795 // failure means pthread_kill failed for some reason - arguably this is 3796 // a fatal problem, but such problems are ignored elsewhere 3797 3798 return epc; 3799} 3800 3801int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 3802{ 3803 if (is_NPTL()) { 3804 return pthread_cond_timedwait(_cond, _mutex, _abstime); 3805 } else { 3806#ifndef IA64 3807 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 3808 // word back to default 64bit precision if condvar is signaled. Java 3809 // wants 53bit precision. Save and restore current value. 3810 int fpu = get_fpu_control_word(); 3811#endif // IA64 3812 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 3813#ifndef IA64 3814 set_fpu_control_word(fpu); 3815#endif // IA64 3816 return status; 3817 } 3818} 3819 3820//////////////////////////////////////////////////////////////////////////////// 3821// debug support 3822 3823#ifndef PRODUCT 3824static address same_page(address x, address y) { 3825 int page_bits = -os::vm_page_size(); 3826 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) 3827 return x; 3828 else if (x > y) 3829 return (address)(intptr_t(y) | ~page_bits) + 1; 3830 else 3831 return (address)(intptr_t(y) & page_bits); 3832} 3833 3834bool os::find(address addr) { 3835 Dl_info dlinfo; 3836 memset(&dlinfo, 0, sizeof(dlinfo)); 3837 if (dladdr(addr, &dlinfo)) { 3838 tty->print(PTR_FORMAT ": ", addr); 3839 if (dlinfo.dli_sname != NULL) { 3840 tty->print("%s+%#x", dlinfo.dli_sname, 3841 addr - (intptr_t)dlinfo.dli_saddr); 3842 } else if (dlinfo.dli_fname) { 3843 tty->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 3844 } else { 3845 tty->print("<absolute address>"); 3846 } 3847 if (dlinfo.dli_fname) { 3848 tty->print(" in %s", dlinfo.dli_fname); 3849 } 3850 if (dlinfo.dli_fbase) { 3851 tty->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 3852 } 3853 tty->cr(); 3854 3855 if (Verbose) { 3856 // decode some bytes around the PC 3857 address begin = same_page(addr-40, addr); 3858 address end = same_page(addr+40, addr); 3859 address lowest = (address) dlinfo.dli_sname; 3860 if (!lowest) lowest = (address) dlinfo.dli_fbase; 3861 if (begin < lowest) begin = lowest; 3862 Dl_info dlinfo2; 3863 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr 3864 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 3865 end = (address) dlinfo2.dli_saddr; 3866 Disassembler::decode(begin, end); 3867 } 3868 return true; 3869 } 3870 return false; 3871} 3872 3873#endif 3874 3875//////////////////////////////////////////////////////////////////////////////// 3876// misc 3877 3878// This does not do anything on Linux. This is basically a hook for being 3879// able to use structured exception handling (thread-local exception filters) 3880// on, e.g., Win32. 3881void 3882os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 3883 JavaCallArguments* args, Thread* thread) { 3884 f(value, method, args, thread); 3885} 3886 3887void os::print_statistics() { 3888} 3889 3890int os::message_box(const char* title, const char* message) { 3891 int i; 3892 fdStream err(defaultStream::error_fd()); 3893 for (i = 0; i < 78; i++) err.print_raw("="); 3894 err.cr(); 3895 err.print_raw_cr(title); 3896 for (i = 0; i < 78; i++) err.print_raw("-"); 3897 err.cr(); 3898 err.print_raw_cr(message); 3899 for (i = 0; i < 78; i++) err.print_raw("="); 3900 err.cr(); 3901 3902 char buf[16]; 3903 // Prevent process from exiting upon "read error" without consuming all CPU 3904 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 3905 3906 return buf[0] == 'y' || buf[0] == 'Y'; 3907} 3908 3909int os::stat(const char *path, struct stat *sbuf) { 3910 char pathbuf[MAX_PATH]; 3911 if (strlen(path) > MAX_PATH - 1) { 3912 errno = ENAMETOOLONG; 3913 return -1; 3914 } 3915 hpi::native_path(strcpy(pathbuf, path)); 3916 return ::stat(pathbuf, sbuf); 3917} 3918 3919bool os::check_heap(bool force) { 3920 return true; 3921} 3922 3923int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 3924 return ::vsnprintf(buf, count, format, args); 3925} 3926 3927// Is a (classpath) directory empty? 3928bool os::dir_is_empty(const char* path) { 3929 DIR *dir = NULL; 3930 struct dirent *ptr; 3931 3932 dir = opendir(path); 3933 if (dir == NULL) return true; 3934 3935 /* Scan the directory */ 3936 bool result = true; 3937 char buf[sizeof(struct dirent) + MAX_PATH]; 3938 while (result && (ptr = ::readdir(dir)) != NULL) { 3939 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 3940 result = false; 3941 } 3942 } 3943 closedir(dir); 3944 return result; 3945} 3946 3947// create binary file, rewriting existing file if required 3948int os::create_binary_file(const char* path, bool rewrite_existing) { 3949 int oflags = O_WRONLY | O_CREAT; 3950 if (!rewrite_existing) { 3951 oflags |= O_EXCL; 3952 } 3953 return ::open64(path, oflags, S_IREAD | S_IWRITE); 3954} 3955 3956// return current position of file pointer 3957jlong os::current_file_offset(int fd) { 3958 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 3959} 3960 3961// move file pointer to the specified offset 3962jlong os::seek_to_file_offset(int fd, jlong offset) { 3963 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 3964} 3965 3966// Map a block of memory. 3967char* os::map_memory(int fd, const char* file_name, size_t file_offset, 3968 char *addr, size_t bytes, bool read_only, 3969 bool allow_exec) { 3970 int prot; 3971 int flags; 3972 3973 if (read_only) { 3974 prot = PROT_READ; 3975 flags = MAP_SHARED; 3976 } else { 3977 prot = PROT_READ | PROT_WRITE; 3978 flags = MAP_PRIVATE; 3979 } 3980 3981 if (allow_exec) { 3982 prot |= PROT_EXEC; 3983 } 3984 3985 if (addr != NULL) { 3986 flags |= MAP_FIXED; 3987 } 3988 3989 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 3990 fd, file_offset); 3991 if (mapped_address == MAP_FAILED) { 3992 return NULL; 3993 } 3994 return mapped_address; 3995} 3996 3997 3998// Remap a block of memory. 3999char* os::remap_memory(int fd, const char* file_name, size_t file_offset, 4000 char *addr, size_t bytes, bool read_only, 4001 bool allow_exec) { 4002 // same as map_memory() on this OS 4003 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 4004 allow_exec); 4005} 4006 4007 4008// Unmap a block of memory. 4009bool os::unmap_memory(char* addr, size_t bytes) { 4010 return munmap(addr, bytes) == 0; 4011} 4012 4013static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 4014 4015static clockid_t thread_cpu_clockid(Thread* thread) { 4016 pthread_t tid = thread->osthread()->pthread_id(); 4017 clockid_t clockid; 4018 4019 // Get thread clockid 4020 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 4021 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 4022 return clockid; 4023} 4024 4025// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 4026// are used by JVM M&M and JVMTI to get user+sys or user CPU time 4027// of a thread. 4028// 4029// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 4030// the fast estimate available on the platform. 4031 4032jlong os::current_thread_cpu_time() { 4033 if (os::Linux::supports_fast_thread_cpu_time()) { 4034 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4035 } else { 4036 // return user + sys since the cost is the same 4037 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 4038 } 4039} 4040 4041jlong os::thread_cpu_time(Thread* thread) { 4042 // consistent with what current_thread_cpu_time() returns 4043 if (os::Linux::supports_fast_thread_cpu_time()) { 4044 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4045 } else { 4046 return slow_thread_cpu_time(thread, true /* user + sys */); 4047 } 4048} 4049 4050jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 4051 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4052 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4053 } else { 4054 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 4055 } 4056} 4057 4058jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4059 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4060 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4061 } else { 4062 return slow_thread_cpu_time(thread, user_sys_cpu_time); 4063 } 4064} 4065 4066// 4067// -1 on error. 4068// 4069 4070static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4071 static bool proc_pid_cpu_avail = true; 4072 static bool proc_task_unchecked = true; 4073 static const char *proc_stat_path = "/proc/%d/stat"; 4074 pid_t tid = thread->osthread()->thread_id(); 4075 int i; 4076 char *s; 4077 char stat[2048]; 4078 int statlen; 4079 char proc_name[64]; 4080 int count; 4081 long sys_time, user_time; 4082 char string[64]; 4083 int idummy; 4084 long ldummy; 4085 FILE *fp; 4086 4087 // We first try accessing /proc/<pid>/cpu since this is faster to 4088 // process. If this file is not present (linux kernels 2.5 and above) 4089 // then we open /proc/<pid>/stat. 4090 if ( proc_pid_cpu_avail ) { 4091 sprintf(proc_name, "/proc/%d/cpu", tid); 4092 fp = fopen(proc_name, "r"); 4093 if ( fp != NULL ) { 4094 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time); 4095 fclose(fp); 4096 if ( count != 3 ) return -1; 4097 4098 if (user_sys_cpu_time) { 4099 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4100 } else { 4101 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4102 } 4103 } 4104 else proc_pid_cpu_avail = false; 4105 } 4106 4107 // The /proc/<tid>/stat aggregates per-process usage on 4108 // new Linux kernels 2.6+ where NPTL is supported. 4109 // The /proc/self/task/<tid>/stat still has the per-thread usage. 4110 // See bug 6328462. 4111 // There can be no directory /proc/self/task on kernels 2.4 with NPTL 4112 // and possibly in some other cases, so we check its availability. 4113 if (proc_task_unchecked && os::Linux::is_NPTL()) { 4114 // This is executed only once 4115 proc_task_unchecked = false; 4116 fp = fopen("/proc/self/task", "r"); 4117 if (fp != NULL) { 4118 proc_stat_path = "/proc/self/task/%d/stat"; 4119 fclose(fp); 4120 } 4121 } 4122 4123 sprintf(proc_name, proc_stat_path, tid); 4124 fp = fopen(proc_name, "r"); 4125 if ( fp == NULL ) return -1; 4126 statlen = fread(stat, 1, 2047, fp); 4127 stat[statlen] = '\0'; 4128 fclose(fp); 4129 4130 // Skip pid and the command string. Note that we could be dealing with 4131 // weird command names, e.g. user could decide to rename java launcher 4132 // to "java 1.4.2 :)", then the stat file would look like 4133 // 1234 (java 1.4.2 :)) R ... ... 4134 // We don't really need to know the command string, just find the last 4135 // occurrence of ")" and then start parsing from there. See bug 4726580. 4136 s = strrchr(stat, ')'); 4137 i = 0; 4138 if (s == NULL ) return -1; 4139 4140 // Skip blank chars 4141 do s++; while (isspace(*s)); 4142 4143 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 4144 &idummy, &idummy, &idummy, &idummy, &idummy, &idummy, 4145 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 4146 &user_time, &sys_time); 4147 if ( count != 13 ) return -1; 4148 if (user_sys_cpu_time) { 4149 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4150 } else { 4151 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4152 } 4153} 4154 4155void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4156 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4157 info_ptr->may_skip_backward = false; // elapsed time not wall time 4158 info_ptr->may_skip_forward = false; // elapsed time not wall time 4159 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4160} 4161 4162void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4163 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4164 info_ptr->may_skip_backward = false; // elapsed time not wall time 4165 info_ptr->may_skip_forward = false; // elapsed time not wall time 4166 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4167} 4168 4169bool os::is_thread_cpu_time_supported() { 4170 return true; 4171} 4172 4173// System loadavg support. Returns -1 if load average cannot be obtained. 4174// Linux doesn't yet have a (official) notion of processor sets, 4175// so just return the system wide load average. 4176int os::loadavg(double loadavg[], int nelem) { 4177 return ::getloadavg(loadavg, nelem); 4178} 4179 4180void os::pause() { 4181 char filename[MAX_PATH]; 4182 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 4183 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 4184 } else { 4185 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 4186 } 4187 4188 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 4189 if (fd != -1) { 4190 struct stat buf; 4191 close(fd); 4192 while (::stat(filename, &buf) == 0) { 4193 (void)::poll(NULL, 0, 100); 4194 } 4195 } else { 4196 jio_fprintf(stderr, 4197 "Could not open pause file '%s', continuing immediately.\n", filename); 4198 } 4199} 4200 4201extern "C" { 4202 4203/** 4204 * NOTE: the following code is to keep the green threads code 4205 * in the libjava.so happy. Once the green threads is removed, 4206 * these code will no longer be needed. 4207 */ 4208int 4209jdk_waitpid(pid_t pid, int* status, int options) { 4210 return waitpid(pid, status, options); 4211} 4212 4213int 4214fork1() { 4215 return fork(); 4216} 4217 4218int 4219jdk_sem_init(sem_t *sem, int pshared, unsigned int value) { 4220 return sem_init(sem, pshared, value); 4221} 4222 4223int 4224jdk_sem_post(sem_t *sem) { 4225 return sem_post(sem); 4226} 4227 4228int 4229jdk_sem_wait(sem_t *sem) { 4230 return sem_wait(sem); 4231} 4232 4233int 4234jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) { 4235 return pthread_sigmask(how , newmask, oldmask); 4236} 4237 4238} 4239 4240// Refer to the comments in os_solaris.cpp park-unpark. 4241// 4242// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 4243// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 4244// For specifics regarding the bug see GLIBC BUGID 261237 : 4245// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 4246// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 4247// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 4248// is used. (The simple C test-case provided in the GLIBC bug report manifests the 4249// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 4250// and monitorenter when we're using 1-0 locking. All those operations may result in 4251// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 4252// of libpthread avoids the problem, but isn't practical. 4253// 4254// Possible remedies: 4255// 4256// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 4257// This is palliative and probabilistic, however. If the thread is preempted 4258// between the call to compute_abstime() and pthread_cond_timedwait(), more 4259// than the minimum period may have passed, and the abstime may be stale (in the 4260// past) resultin in a hang. Using this technique reduces the odds of a hang 4261// but the JVM is still vulnerable, particularly on heavily loaded systems. 4262// 4263// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 4264// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 4265// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 4266// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 4267// thread. 4268// 4269// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 4270// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 4271// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 4272// This also works well. In fact it avoids kernel-level scalability impediments 4273// on certain platforms that don't handle lots of active pthread_cond_timedwait() 4274// timers in a graceful fashion. 4275// 4276// 4. When the abstime value is in the past it appears that control returns 4277// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 4278// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 4279// can avoid the problem by reinitializing the condvar -- by cond_destroy() 4280// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 4281// It may be possible to avoid reinitialization by checking the return 4282// value from pthread_cond_timedwait(). In addition to reinitializing the 4283// condvar we must establish the invariant that cond_signal() is only called 4284// within critical sections protected by the adjunct mutex. This prevents 4285// cond_signal() from "seeing" a condvar that's in the midst of being 4286// reinitialized or that is corrupt. Sadly, this invariant obviates the 4287// desirable signal-after-unlock optimization that avoids futile context switching. 4288// 4289// I'm also concerned that some versions of NTPL might allocate an auxilliary 4290// structure when a condvar is used or initialized. cond_destroy() would 4291// release the helper structure. Our reinitialize-after-timedwait fix 4292// put excessive stress on malloc/free and locks protecting the c-heap. 4293// 4294// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 4295// It may be possible to refine (4) by checking the kernel and NTPL verisons 4296// and only enabling the work-around for vulnerable environments. 4297 4298// utility to compute the abstime argument to timedwait: 4299// millis is the relative timeout time 4300// abstime will be the absolute timeout time 4301// TODO: replace compute_abstime() with unpackTime() 4302 4303static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 4304 if (millis < 0) millis = 0; 4305 struct timeval now; 4306 int status = gettimeofday(&now, NULL); 4307 assert(status == 0, "gettimeofday"); 4308 jlong seconds = millis / 1000; 4309 millis %= 1000; 4310 if (seconds > 50000000) { // see man cond_timedwait(3T) 4311 seconds = 50000000; 4312 } 4313 abstime->tv_sec = now.tv_sec + seconds; 4314 long usec = now.tv_usec + millis * 1000; 4315 if (usec >= 1000000) { 4316 abstime->tv_sec += 1; 4317 usec -= 1000000; 4318 } 4319 abstime->tv_nsec = usec * 1000; 4320 return abstime; 4321} 4322 4323 4324// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 4325// Conceptually TryPark() should be equivalent to park(0). 4326 4327int os::PlatformEvent::TryPark() { 4328 for (;;) { 4329 const int v = _Event ; 4330 guarantee ((v == 0) || (v == 1), "invariant") ; 4331 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 4332 } 4333} 4334 4335void os::PlatformEvent::park() { // AKA "down()" 4336 // Invariant: Only the thread associated with the Event/PlatformEvent 4337 // may call park(). 4338 // TODO: assert that _Assoc != NULL or _Assoc == Self 4339 int v ; 4340 for (;;) { 4341 v = _Event ; 4342 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4343 } 4344 guarantee (v >= 0, "invariant") ; 4345 if (v == 0) { 4346 // Do this the hard way by blocking ... 4347 int status = pthread_mutex_lock(_mutex); 4348 assert_status(status == 0, status, "mutex_lock"); 4349 guarantee (_nParked == 0, "invariant") ; 4350 ++ _nParked ; 4351 while (_Event < 0) { 4352 status = pthread_cond_wait(_cond, _mutex); 4353 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 4354 // Treat this the same as if the wait was interrupted 4355 if (status == ETIME) { status = EINTR; } 4356 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 4357 } 4358 -- _nParked ; 4359 4360 // In theory we could move the ST of 0 into _Event past the unlock(), 4361 // but then we'd need a MEMBAR after the ST. 4362 _Event = 0 ; 4363 status = pthread_mutex_unlock(_mutex); 4364 assert_status(status == 0, status, "mutex_unlock"); 4365 } 4366 guarantee (_Event >= 0, "invariant") ; 4367} 4368 4369int os::PlatformEvent::park(jlong millis) { 4370 guarantee (_nParked == 0, "invariant") ; 4371 4372 int v ; 4373 for (;;) { 4374 v = _Event ; 4375 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4376 } 4377 guarantee (v >= 0, "invariant") ; 4378 if (v != 0) return OS_OK ; 4379 4380 // We do this the hard way, by blocking the thread. 4381 // Consider enforcing a minimum timeout value. 4382 struct timespec abst; 4383 compute_abstime(&abst, millis); 4384 4385 int ret = OS_TIMEOUT; 4386 int status = pthread_mutex_lock(_mutex); 4387 assert_status(status == 0, status, "mutex_lock"); 4388 guarantee (_nParked == 0, "invariant") ; 4389 ++_nParked ; 4390 4391 // Object.wait(timo) will return because of 4392 // (a) notification 4393 // (b) timeout 4394 // (c) thread.interrupt 4395 // 4396 // Thread.interrupt and object.notify{All} both call Event::set. 4397 // That is, we treat thread.interrupt as a special case of notification. 4398 // The underlying Solaris implementation, cond_timedwait, admits 4399 // spurious/premature wakeups, but the JLS/JVM spec prevents the 4400 // JVM from making those visible to Java code. As such, we must 4401 // filter out spurious wakeups. We assume all ETIME returns are valid. 4402 // 4403 // TODO: properly differentiate simultaneous notify+interrupt. 4404 // In that case, we should propagate the notify to another waiter. 4405 4406 while (_Event < 0) { 4407 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 4408 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4409 pthread_cond_destroy (_cond); 4410 pthread_cond_init (_cond, NULL) ; 4411 } 4412 assert_status(status == 0 || status == EINTR || 4413 status == ETIME || status == ETIMEDOUT, 4414 status, "cond_timedwait"); 4415 if (!FilterSpuriousWakeups) break ; // previous semantics 4416 if (status == ETIME || status == ETIMEDOUT) break ; 4417 // We consume and ignore EINTR and spurious wakeups. 4418 } 4419 --_nParked ; 4420 if (_Event >= 0) { 4421 ret = OS_OK; 4422 } 4423 _Event = 0 ; 4424 status = pthread_mutex_unlock(_mutex); 4425 assert_status(status == 0, status, "mutex_unlock"); 4426 assert (_nParked == 0, "invariant") ; 4427 return ret; 4428} 4429 4430void os::PlatformEvent::unpark() { 4431 int v, AnyWaiters ; 4432 for (;;) { 4433 v = _Event ; 4434 if (v > 0) { 4435 // The LD of _Event could have reordered or be satisfied 4436 // by a read-aside from this processor's write buffer. 4437 // To avoid problems execute a barrier and then 4438 // ratify the value. 4439 OrderAccess::fence() ; 4440 if (_Event == v) return ; 4441 continue ; 4442 } 4443 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ; 4444 } 4445 if (v < 0) { 4446 // Wait for the thread associated with the event to vacate 4447 int status = pthread_mutex_lock(_mutex); 4448 assert_status(status == 0, status, "mutex_lock"); 4449 AnyWaiters = _nParked ; 4450 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ; 4451 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 4452 AnyWaiters = 0 ; 4453 pthread_cond_signal (_cond); 4454 } 4455 status = pthread_mutex_unlock(_mutex); 4456 assert_status(status == 0, status, "mutex_unlock"); 4457 if (AnyWaiters != 0) { 4458 status = pthread_cond_signal(_cond); 4459 assert_status(status == 0, status, "cond_signal"); 4460 } 4461 } 4462 4463 // Note that we signal() _after dropping the lock for "immortal" Events. 4464 // This is safe and avoids a common class of futile wakeups. In rare 4465 // circumstances this can cause a thread to return prematurely from 4466 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 4467 // simply re-test the condition and re-park itself. 4468} 4469 4470 4471// JSR166 4472// ------------------------------------------------------- 4473 4474/* 4475 * The solaris and linux implementations of park/unpark are fairly 4476 * conservative for now, but can be improved. They currently use a 4477 * mutex/condvar pair, plus a a count. 4478 * Park decrements count if > 0, else does a condvar wait. Unpark 4479 * sets count to 1 and signals condvar. Only one thread ever waits 4480 * on the condvar. Contention seen when trying to park implies that someone 4481 * is unparking you, so don't wait. And spurious returns are fine, so there 4482 * is no need to track notifications. 4483 */ 4484 4485 4486#define NANOSECS_PER_SEC 1000000000 4487#define NANOSECS_PER_MILLISEC 1000000 4488#define MAX_SECS 100000000 4489/* 4490 * This code is common to linux and solaris and will be moved to a 4491 * common place in dolphin. 4492 * 4493 * The passed in time value is either a relative time in nanoseconds 4494 * or an absolute time in milliseconds. Either way it has to be unpacked 4495 * into suitable seconds and nanoseconds components and stored in the 4496 * given timespec structure. 4497 * Given time is a 64-bit value and the time_t used in the timespec is only 4498 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 4499 * overflow if times way in the future are given. Further on Solaris versions 4500 * prior to 10 there is a restriction (see cond_timedwait) that the specified 4501 * number of seconds, in abstime, is less than current_time + 100,000,000. 4502 * As it will be 28 years before "now + 100000000" will overflow we can 4503 * ignore overflow and just impose a hard-limit on seconds using the value 4504 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 4505 * years from "now". 4506 */ 4507 4508static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 4509 assert (time > 0, "convertTime"); 4510 4511 struct timeval now; 4512 int status = gettimeofday(&now, NULL); 4513 assert(status == 0, "gettimeofday"); 4514 4515 time_t max_secs = now.tv_sec + MAX_SECS; 4516 4517 if (isAbsolute) { 4518 jlong secs = time / 1000; 4519 if (secs > max_secs) { 4520 absTime->tv_sec = max_secs; 4521 } 4522 else { 4523 absTime->tv_sec = secs; 4524 } 4525 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 4526 } 4527 else { 4528 jlong secs = time / NANOSECS_PER_SEC; 4529 if (secs >= MAX_SECS) { 4530 absTime->tv_sec = max_secs; 4531 absTime->tv_nsec = 0; 4532 } 4533 else { 4534 absTime->tv_sec = now.tv_sec + secs; 4535 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 4536 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 4537 absTime->tv_nsec -= NANOSECS_PER_SEC; 4538 ++absTime->tv_sec; // note: this must be <= max_secs 4539 } 4540 } 4541 } 4542 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 4543 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 4544 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 4545 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 4546} 4547 4548void Parker::park(bool isAbsolute, jlong time) { 4549 // Optional fast-path check: 4550 // Return immediately if a permit is available. 4551 if (_counter > 0) { 4552 _counter = 0 ; 4553 return ; 4554 } 4555 4556 Thread* thread = Thread::current(); 4557 assert(thread->is_Java_thread(), "Must be JavaThread"); 4558 JavaThread *jt = (JavaThread *)thread; 4559 4560 // Optional optimization -- avoid state transitions if there's an interrupt pending. 4561 // Check interrupt before trying to wait 4562 if (Thread::is_interrupted(thread, false)) { 4563 return; 4564 } 4565 4566 // Next, demultiplex/decode time arguments 4567 timespec absTime; 4568 if (time < 0) { // don't wait at all 4569 return; 4570 } 4571 if (time > 0) { 4572 unpackTime(&absTime, isAbsolute, time); 4573 } 4574 4575 4576 // Enter safepoint region 4577 // Beware of deadlocks such as 6317397. 4578 // The per-thread Parker:: mutex is a classic leaf-lock. 4579 // In particular a thread must never block on the Threads_lock while 4580 // holding the Parker:: mutex. If safepoints are pending both the 4581 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 4582 ThreadBlockInVM tbivm(jt); 4583 4584 // Don't wait if cannot get lock since interference arises from 4585 // unblocking. Also. check interrupt before trying wait 4586 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 4587 return; 4588 } 4589 4590 int status ; 4591 if (_counter > 0) { // no wait needed 4592 _counter = 0; 4593 status = pthread_mutex_unlock(_mutex); 4594 assert (status == 0, "invariant") ; 4595 return; 4596 } 4597 4598#ifdef ASSERT 4599 // Don't catch signals while blocked; let the running threads have the signals. 4600 // (This allows a debugger to break into the running thread.) 4601 sigset_t oldsigs; 4602 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 4603 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 4604#endif 4605 4606 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 4607 jt->set_suspend_equivalent(); 4608 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 4609 4610 if (time == 0) { 4611 status = pthread_cond_wait (_cond, _mutex) ; 4612 } else { 4613 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 4614 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4615 pthread_cond_destroy (_cond) ; 4616 pthread_cond_init (_cond, NULL); 4617 } 4618 } 4619 assert_status(status == 0 || status == EINTR || 4620 status == ETIME || status == ETIMEDOUT, 4621 status, "cond_timedwait"); 4622 4623#ifdef ASSERT 4624 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 4625#endif 4626 4627 _counter = 0 ; 4628 status = pthread_mutex_unlock(_mutex) ; 4629 assert_status(status == 0, status, "invariant") ; 4630 // If externally suspended while waiting, re-suspend 4631 if (jt->handle_special_suspend_equivalent_condition()) { 4632 jt->java_suspend_self(); 4633 } 4634 4635} 4636 4637void Parker::unpark() { 4638 int s, status ; 4639 status = pthread_mutex_lock(_mutex); 4640 assert (status == 0, "invariant") ; 4641 s = _counter; 4642 _counter = 1; 4643 if (s < 1) { 4644 if (WorkAroundNPTLTimedWaitHang) { 4645 status = pthread_cond_signal (_cond) ; 4646 assert (status == 0, "invariant") ; 4647 status = pthread_mutex_unlock(_mutex); 4648 assert (status == 0, "invariant") ; 4649 } else { 4650 status = pthread_mutex_unlock(_mutex); 4651 assert (status == 0, "invariant") ; 4652 status = pthread_cond_signal (_cond) ; 4653 assert (status == 0, "invariant") ; 4654 } 4655 } else { 4656 pthread_mutex_unlock(_mutex); 4657 assert (status == 0, "invariant") ; 4658 } 4659} 4660 4661 4662extern char** environ; 4663 4664#ifndef __NR_fork 4665#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 4666#endif 4667 4668#ifndef __NR_execve 4669#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 4670#endif 4671 4672// Run the specified command in a separate process. Return its exit value, 4673// or -1 on failure (e.g. can't fork a new process). 4674// Unlike system(), this function can be called from signal handler. It 4675// doesn't block SIGINT et al. 4676int os::fork_and_exec(char* cmd) { 4677 const char * argv[4] = {"sh", "-c", cmd, NULL}; 4678 4679 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 4680 // pthread_atfork handlers and reset pthread library. All we need is a 4681 // separate process to execve. Make a direct syscall to fork process. 4682 // On IA64 there's no fork syscall, we have to use fork() and hope for 4683 // the best... 4684 pid_t pid = NOT_IA64(syscall(__NR_fork);) 4685 IA64_ONLY(fork();) 4686 4687 if (pid < 0) { 4688 // fork failed 4689 return -1; 4690 4691 } else if (pid == 0) { 4692 // child process 4693 4694 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 4695 // first to kill every thread on the thread list. Because this list is 4696 // not reset by fork() (see notes above), execve() will instead kill 4697 // every thread in the parent process. We know this is the only thread 4698 // in the new process, so make a system call directly. 4699 // IA64 should use normal execve() from glibc to match the glibc fork() 4700 // above. 4701 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 4702 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 4703 4704 // execve failed 4705 _exit(-1); 4706 4707 } else { 4708 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 4709 // care about the actual exit code, for now. 4710 4711 int status; 4712 4713 // Wait for the child process to exit. This returns immediately if 4714 // the child has already exited. */ 4715 while (waitpid(pid, &status, 0) < 0) { 4716 switch (errno) { 4717 case ECHILD: return 0; 4718 case EINTR: break; 4719 default: return -1; 4720 } 4721 } 4722 4723 if (WIFEXITED(status)) { 4724 // The child exited normally; get its exit code. 4725 return WEXITSTATUS(status); 4726 } else if (WIFSIGNALED(status)) { 4727 // The child exited because of a signal 4728 // The best value to return is 0x80 + signal number, 4729 // because that is what all Unix shells do, and because 4730 // it allows callers to distinguish between process exit and 4731 // process death by signal. 4732 return 0x80 + WTERMSIG(status); 4733 } else { 4734 // Unknown exit code; pass it through 4735 return status; 4736 } 4737 } 4738} 4739