os_linux.cpp revision 1887:828eafbd85cc
1/* 2 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. 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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25# define __STDC_FORMAT_MACROS 26 27// no precompiled headers 28#include "classfile/classLoader.hpp" 29#include "classfile/systemDictionary.hpp" 30#include "classfile/vmSymbols.hpp" 31#include "code/icBuffer.hpp" 32#include "code/vtableStubs.hpp" 33#include "compiler/compileBroker.hpp" 34#include "interpreter/interpreter.hpp" 35#include "jvm_linux.h" 36#include "memory/allocation.inline.hpp" 37#include "memory/filemap.hpp" 38#include "mutex_linux.inline.hpp" 39#include "oops/oop.inline.hpp" 40#include "os_share_linux.hpp" 41#include "prims/jniFastGetField.hpp" 42#include "prims/jvm.h" 43#include "prims/jvm_misc.hpp" 44#include "runtime/arguments.hpp" 45#include "runtime/extendedPC.hpp" 46#include "runtime/globals.hpp" 47#include "runtime/interfaceSupport.hpp" 48#include "runtime/java.hpp" 49#include "runtime/javaCalls.hpp" 50#include "runtime/mutexLocker.hpp" 51#include "runtime/objectMonitor.hpp" 52#include "runtime/osThread.hpp" 53#include "runtime/perfMemory.hpp" 54#include "runtime/sharedRuntime.hpp" 55#include "runtime/statSampler.hpp" 56#include "runtime/stubRoutines.hpp" 57#include "runtime/threadCritical.hpp" 58#include "runtime/timer.hpp" 59#include "services/attachListener.hpp" 60#include "services/runtimeService.hpp" 61#include "thread_linux.inline.hpp" 62#include "utilities/defaultStream.hpp" 63#include "utilities/events.hpp" 64#include "utilities/growableArray.hpp" 65#include "utilities/vmError.hpp" 66#ifdef TARGET_ARCH_x86 67# include "assembler_x86.inline.hpp" 68# include "nativeInst_x86.hpp" 69#endif 70#ifdef TARGET_ARCH_sparc 71# include "assembler_sparc.inline.hpp" 72# include "nativeInst_sparc.hpp" 73#endif 74#ifdef TARGET_ARCH_zero 75# include "assembler_zero.inline.hpp" 76# include "nativeInst_zero.hpp" 77#endif 78#ifdef COMPILER1 79#include "c1/c1_Runtime1.hpp" 80#endif 81#ifdef COMPILER2 82#include "opto/runtime.hpp" 83#endif 84 85// put OS-includes here 86# include <sys/types.h> 87# include <sys/mman.h> 88# include <sys/stat.h> 89# include <sys/select.h> 90# include <pthread.h> 91# include <signal.h> 92# include <errno.h> 93# include <dlfcn.h> 94# include <stdio.h> 95# include <unistd.h> 96# include <sys/resource.h> 97# include <pthread.h> 98# include <sys/stat.h> 99# include <sys/time.h> 100# include <sys/times.h> 101# include <sys/utsname.h> 102# include <sys/socket.h> 103# include <sys/wait.h> 104# include <pwd.h> 105# include <poll.h> 106# include <semaphore.h> 107# include <fcntl.h> 108# include <string.h> 109# include <syscall.h> 110# include <sys/sysinfo.h> 111# include <gnu/libc-version.h> 112# include <sys/ipc.h> 113# include <sys/shm.h> 114# include <link.h> 115# include <stdint.h> 116# include <inttypes.h> 117 118#define MAX_PATH (2 * K) 119 120// for timer info max values which include all bits 121#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 122#define SEC_IN_NANOSECS 1000000000LL 123 124//////////////////////////////////////////////////////////////////////////////// 125// global variables 126julong os::Linux::_physical_memory = 0; 127 128address os::Linux::_initial_thread_stack_bottom = NULL; 129uintptr_t os::Linux::_initial_thread_stack_size = 0; 130 131int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 132int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 133Mutex* os::Linux::_createThread_lock = NULL; 134pthread_t os::Linux::_main_thread; 135int os::Linux::_page_size = -1; 136bool os::Linux::_is_floating_stack = false; 137bool os::Linux::_is_NPTL = false; 138bool os::Linux::_supports_fast_thread_cpu_time = false; 139const char * os::Linux::_glibc_version = NULL; 140const char * os::Linux::_libpthread_version = NULL; 141 142static jlong initial_time_count=0; 143 144static int clock_tics_per_sec = 100; 145 146// For diagnostics to print a message once. see run_periodic_checks 147static sigset_t check_signal_done; 148static bool check_signals = true;; 149 150static pid_t _initial_pid = 0; 151 152/* Signal number used to suspend/resume a thread */ 153 154/* do not use any signal number less than SIGSEGV, see 4355769 */ 155static int SR_signum = SIGUSR2; 156sigset_t SR_sigset; 157 158/* Used to protect dlsym() calls */ 159static pthread_mutex_t dl_mutex; 160 161//////////////////////////////////////////////////////////////////////////////// 162// utility functions 163 164static int SR_initialize(); 165static int SR_finalize(); 166 167julong os::available_memory() { 168 return Linux::available_memory(); 169} 170 171julong os::Linux::available_memory() { 172 // values in struct sysinfo are "unsigned long" 173 struct sysinfo si; 174 sysinfo(&si); 175 176 return (julong)si.freeram * si.mem_unit; 177} 178 179julong os::physical_memory() { 180 return Linux::physical_memory(); 181} 182 183julong os::allocatable_physical_memory(julong size) { 184#ifdef _LP64 185 return size; 186#else 187 julong result = MIN2(size, (julong)3800*M); 188 if (!is_allocatable(result)) { 189 // See comments under solaris for alignment considerations 190 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); 191 result = MIN2(size, reasonable_size); 192 } 193 return result; 194#endif // _LP64 195} 196 197//////////////////////////////////////////////////////////////////////////////// 198// environment support 199 200bool os::getenv(const char* name, char* buf, int len) { 201 const char* val = ::getenv(name); 202 if (val != NULL && strlen(val) < (size_t)len) { 203 strcpy(buf, val); 204 return true; 205 } 206 if (len > 0) buf[0] = 0; // return a null string 207 return false; 208} 209 210 211// Return true if user is running as root. 212 213bool os::have_special_privileges() { 214 static bool init = false; 215 static bool privileges = false; 216 if (!init) { 217 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 218 init = true; 219 } 220 return privileges; 221} 222 223 224#ifndef SYS_gettid 225// i386: 224, ia64: 1105, amd64: 186, sparc 143 226#ifdef __ia64__ 227#define SYS_gettid 1105 228#elif __i386__ 229#define SYS_gettid 224 230#elif __amd64__ 231#define SYS_gettid 186 232#elif __sparc__ 233#define SYS_gettid 143 234#else 235#error define gettid for the arch 236#endif 237#endif 238 239// Cpu architecture string 240#if defined(ZERO) 241static char cpu_arch[] = ZERO_LIBARCH; 242#elif defined(IA64) 243static char cpu_arch[] = "ia64"; 244#elif defined(IA32) 245static char cpu_arch[] = "i386"; 246#elif defined(AMD64) 247static char cpu_arch[] = "amd64"; 248#elif defined(ARM) 249static char cpu_arch[] = "arm"; 250#elif defined(PPC) 251static char cpu_arch[] = "ppc"; 252#elif defined(SPARC) 253# ifdef _LP64 254static char cpu_arch[] = "sparcv9"; 255# else 256static char cpu_arch[] = "sparc"; 257# endif 258#else 259#error Add appropriate cpu_arch setting 260#endif 261 262 263// pid_t gettid() 264// 265// Returns the kernel thread id of the currently running thread. Kernel 266// thread id is used to access /proc. 267// 268// (Note that getpid() on LinuxThreads returns kernel thread id too; but 269// on NPTL, it returns the same pid for all threads, as required by POSIX.) 270// 271pid_t os::Linux::gettid() { 272 int rslt = syscall(SYS_gettid); 273 if (rslt == -1) { 274 // old kernel, no NPTL support 275 return getpid(); 276 } else { 277 return (pid_t)rslt; 278 } 279} 280 281// Most versions of linux have a bug where the number of processors are 282// determined by looking at the /proc file system. In a chroot environment, 283// the system call returns 1. This causes the VM to act as if it is 284// a single processor and elide locking (see is_MP() call). 285static bool unsafe_chroot_detected = false; 286static const char *unstable_chroot_error = "/proc file system not found.\n" 287 "Java may be unstable running multithreaded in a chroot " 288 "environment on Linux when /proc filesystem is not mounted."; 289 290void os::Linux::initialize_system_info() { 291 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 292 if (processor_count() == 1) { 293 pid_t pid = os::Linux::gettid(); 294 char fname[32]; 295 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 296 FILE *fp = fopen(fname, "r"); 297 if (fp == NULL) { 298 unsafe_chroot_detected = true; 299 } else { 300 fclose(fp); 301 } 302 } 303 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 304 assert(processor_count() > 0, "linux error"); 305} 306 307void os::init_system_properties_values() { 308// char arch[12]; 309// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 310 311 // The next steps are taken in the product version: 312 // 313 // Obtain the JAVA_HOME value from the location of libjvm[_g].so. 314 // This library should be located at: 315 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so. 316 // 317 // If "/jre/lib/" appears at the right place in the path, then we 318 // assume libjvm[_g].so is installed in a JDK and we use this path. 319 // 320 // Otherwise exit with message: "Could not create the Java virtual machine." 321 // 322 // The following extra steps are taken in the debugging version: 323 // 324 // If "/jre/lib/" does NOT appear at the right place in the path 325 // instead of exit check for $JAVA_HOME environment variable. 326 // 327 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 328 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so 329 // it looks like libjvm[_g].so is installed there 330 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so. 331 // 332 // Otherwise exit. 333 // 334 // Important note: if the location of libjvm.so changes this 335 // code needs to be changed accordingly. 336 337 // The next few definitions allow the code to be verbatim: 338#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n)) 339#define getenv(n) ::getenv(n) 340 341/* 342 * See ld(1): 343 * The linker uses the following search paths to locate required 344 * shared libraries: 345 * 1: ... 346 * ... 347 * 7: The default directories, normally /lib and /usr/lib. 348 */ 349#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 350#define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 351#else 352#define DEFAULT_LIBPATH "/lib:/usr/lib" 353#endif 354 355#define EXTENSIONS_DIR "/lib/ext" 356#define ENDORSED_DIR "/lib/endorsed" 357#define REG_DIR "/usr/java/packages" 358 359 { 360 /* sysclasspath, java_home, dll_dir */ 361 { 362 char *home_path; 363 char *dll_path; 364 char *pslash; 365 char buf[MAXPATHLEN]; 366 os::jvm_path(buf, sizeof(buf)); 367 368 // Found the full path to libjvm.so. 369 // Now cut the path to <java_home>/jre if we can. 370 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 371 pslash = strrchr(buf, '/'); 372 if (pslash != NULL) 373 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 374 dll_path = malloc(strlen(buf) + 1); 375 if (dll_path == NULL) 376 return; 377 strcpy(dll_path, buf); 378 Arguments::set_dll_dir(dll_path); 379 380 if (pslash != NULL) { 381 pslash = strrchr(buf, '/'); 382 if (pslash != NULL) { 383 *pslash = '\0'; /* get rid of /<arch> */ 384 pslash = strrchr(buf, '/'); 385 if (pslash != NULL) 386 *pslash = '\0'; /* get rid of /lib */ 387 } 388 } 389 390 home_path = malloc(strlen(buf) + 1); 391 if (home_path == NULL) 392 return; 393 strcpy(home_path, buf); 394 Arguments::set_java_home(home_path); 395 396 if (!set_boot_path('/', ':')) 397 return; 398 } 399 400 /* 401 * Where to look for native libraries 402 * 403 * Note: Due to a legacy implementation, most of the library path 404 * is set in the launcher. This was to accomodate linking restrictions 405 * on legacy Linux implementations (which are no longer supported). 406 * Eventually, all the library path setting will be done here. 407 * 408 * However, to prevent the proliferation of improperly built native 409 * libraries, the new path component /usr/java/packages is added here. 410 * Eventually, all the library path setting will be done here. 411 */ 412 { 413 char *ld_library_path; 414 415 /* 416 * Construct the invariant part of ld_library_path. Note that the 417 * space for the colon and the trailing null are provided by the 418 * nulls included by the sizeof operator (so actually we allocate 419 * a byte more than necessary). 420 */ 421 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 422 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 423 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 424 425 /* 426 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 427 * should always exist (until the legacy problem cited above is 428 * addressed). 429 */ 430 char *v = getenv("LD_LIBRARY_PATH"); 431 if (v != NULL) { 432 char *t = ld_library_path; 433 /* That's +1 for the colon and +1 for the trailing '\0' */ 434 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 435 sprintf(ld_library_path, "%s:%s", v, t); 436 } 437 Arguments::set_library_path(ld_library_path); 438 } 439 440 /* 441 * Extensions directories. 442 * 443 * Note that the space for the colon and the trailing null are provided 444 * by the nulls included by the sizeof operator (so actually one byte more 445 * than necessary is allocated). 446 */ 447 { 448 char *buf = malloc(strlen(Arguments::get_java_home()) + 449 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 450 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 451 Arguments::get_java_home()); 452 Arguments::set_ext_dirs(buf); 453 } 454 455 /* Endorsed standards default directory. */ 456 { 457 char * buf; 458 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 459 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 460 Arguments::set_endorsed_dirs(buf); 461 } 462 } 463 464#undef malloc 465#undef getenv 466#undef EXTENSIONS_DIR 467#undef ENDORSED_DIR 468 469 // Done 470 return; 471} 472 473//////////////////////////////////////////////////////////////////////////////// 474// breakpoint support 475 476void os::breakpoint() { 477 BREAKPOINT; 478} 479 480extern "C" void breakpoint() { 481 // use debugger to set breakpoint here 482} 483 484//////////////////////////////////////////////////////////////////////////////// 485// signal support 486 487debug_only(static bool signal_sets_initialized = false); 488static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 489 490bool os::Linux::is_sig_ignored(int sig) { 491 struct sigaction oact; 492 sigaction(sig, (struct sigaction*)NULL, &oact); 493 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 494 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 495 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 496 return true; 497 else 498 return false; 499} 500 501void os::Linux::signal_sets_init() { 502 // Should also have an assertion stating we are still single-threaded. 503 assert(!signal_sets_initialized, "Already initialized"); 504 // Fill in signals that are necessarily unblocked for all threads in 505 // the VM. Currently, we unblock the following signals: 506 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 507 // by -Xrs (=ReduceSignalUsage)); 508 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 509 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 510 // the dispositions or masks wrt these signals. 511 // Programs embedding the VM that want to use the above signals for their 512 // own purposes must, at this time, use the "-Xrs" option to prevent 513 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 514 // (See bug 4345157, and other related bugs). 515 // In reality, though, unblocking these signals is really a nop, since 516 // these signals are not blocked by default. 517 sigemptyset(&unblocked_sigs); 518 sigemptyset(&allowdebug_blocked_sigs); 519 sigaddset(&unblocked_sigs, SIGILL); 520 sigaddset(&unblocked_sigs, SIGSEGV); 521 sigaddset(&unblocked_sigs, SIGBUS); 522 sigaddset(&unblocked_sigs, SIGFPE); 523 sigaddset(&unblocked_sigs, SR_signum); 524 525 if (!ReduceSignalUsage) { 526 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 527 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 528 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 529 } 530 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 531 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 532 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 533 } 534 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 535 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 536 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 537 } 538 } 539 // Fill in signals that are blocked by all but the VM thread. 540 sigemptyset(&vm_sigs); 541 if (!ReduceSignalUsage) 542 sigaddset(&vm_sigs, BREAK_SIGNAL); 543 debug_only(signal_sets_initialized = true); 544 545} 546 547// These are signals that are unblocked while a thread is running Java. 548// (For some reason, they get blocked by default.) 549sigset_t* os::Linux::unblocked_signals() { 550 assert(signal_sets_initialized, "Not initialized"); 551 return &unblocked_sigs; 552} 553 554// These are the signals that are blocked while a (non-VM) thread is 555// running Java. Only the VM thread handles these signals. 556sigset_t* os::Linux::vm_signals() { 557 assert(signal_sets_initialized, "Not initialized"); 558 return &vm_sigs; 559} 560 561// These are signals that are blocked during cond_wait to allow debugger in 562sigset_t* os::Linux::allowdebug_blocked_signals() { 563 assert(signal_sets_initialized, "Not initialized"); 564 return &allowdebug_blocked_sigs; 565} 566 567void os::Linux::hotspot_sigmask(Thread* thread) { 568 569 //Save caller's signal mask before setting VM signal mask 570 sigset_t caller_sigmask; 571 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 572 573 OSThread* osthread = thread->osthread(); 574 osthread->set_caller_sigmask(caller_sigmask); 575 576 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 577 578 if (!ReduceSignalUsage) { 579 if (thread->is_VM_thread()) { 580 // Only the VM thread handles BREAK_SIGNAL ... 581 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 582 } else { 583 // ... all other threads block BREAK_SIGNAL 584 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 585 } 586 } 587} 588 589////////////////////////////////////////////////////////////////////////////// 590// detecting pthread library 591 592void os::Linux::libpthread_init() { 593 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 594 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 595 // generic name for earlier versions. 596 // Define macros here so we can build HotSpot on old systems. 597# ifndef _CS_GNU_LIBC_VERSION 598# define _CS_GNU_LIBC_VERSION 2 599# endif 600# ifndef _CS_GNU_LIBPTHREAD_VERSION 601# define _CS_GNU_LIBPTHREAD_VERSION 3 602# endif 603 604 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 605 if (n > 0) { 606 char *str = (char *)malloc(n); 607 confstr(_CS_GNU_LIBC_VERSION, str, n); 608 os::Linux::set_glibc_version(str); 609 } else { 610 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 611 static char _gnu_libc_version[32]; 612 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 613 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 614 os::Linux::set_glibc_version(_gnu_libc_version); 615 } 616 617 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 618 if (n > 0) { 619 char *str = (char *)malloc(n); 620 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 621 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 622 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 623 // is the case. LinuxThreads has a hard limit on max number of threads. 624 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 625 // On the other hand, NPTL does not have such a limit, sysconf() 626 // will return -1 and errno is not changed. Check if it is really NPTL. 627 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 628 strstr(str, "NPTL") && 629 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 630 free(str); 631 os::Linux::set_libpthread_version("linuxthreads"); 632 } else { 633 os::Linux::set_libpthread_version(str); 634 } 635 } else { 636 // glibc before 2.3.2 only has LinuxThreads. 637 os::Linux::set_libpthread_version("linuxthreads"); 638 } 639 640 if (strstr(libpthread_version(), "NPTL")) { 641 os::Linux::set_is_NPTL(); 642 } else { 643 os::Linux::set_is_LinuxThreads(); 644 } 645 646 // LinuxThreads have two flavors: floating-stack mode, which allows variable 647 // stack size; and fixed-stack mode. NPTL is always floating-stack. 648 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 649 os::Linux::set_is_floating_stack(); 650 } 651} 652 653///////////////////////////////////////////////////////////////////////////// 654// thread stack 655 656// Force Linux kernel to expand current thread stack. If "bottom" is close 657// to the stack guard, caller should block all signals. 658// 659// MAP_GROWSDOWN: 660// A special mmap() flag that is used to implement thread stacks. It tells 661// kernel that the memory region should extend downwards when needed. This 662// allows early versions of LinuxThreads to only mmap the first few pages 663// when creating a new thread. Linux kernel will automatically expand thread 664// stack as needed (on page faults). 665// 666// However, because the memory region of a MAP_GROWSDOWN stack can grow on 667// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 668// region, it's hard to tell if the fault is due to a legitimate stack 669// access or because of reading/writing non-exist memory (e.g. buffer 670// overrun). As a rule, if the fault happens below current stack pointer, 671// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 672// application (see Linux kernel fault.c). 673// 674// This Linux feature can cause SIGSEGV when VM bangs thread stack for 675// stack overflow detection. 676// 677// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 678// not use this flag. However, the stack of initial thread is not created 679// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 680// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 681// and then attach the thread to JVM. 682// 683// To get around the problem and allow stack banging on Linux, we need to 684// manually expand thread stack after receiving the SIGSEGV. 685// 686// There are two ways to expand thread stack to address "bottom", we used 687// both of them in JVM before 1.5: 688// 1. adjust stack pointer first so that it is below "bottom", and then 689// touch "bottom" 690// 2. mmap() the page in question 691// 692// Now alternate signal stack is gone, it's harder to use 2. For instance, 693// if current sp is already near the lower end of page 101, and we need to 694// call mmap() to map page 100, it is possible that part of the mmap() frame 695// will be placed in page 100. When page 100 is mapped, it is zero-filled. 696// That will destroy the mmap() frame and cause VM to crash. 697// 698// The following code works by adjusting sp first, then accessing the "bottom" 699// page to force a page fault. Linux kernel will then automatically expand the 700// stack mapping. 701// 702// _expand_stack_to() assumes its frame size is less than page size, which 703// should always be true if the function is not inlined. 704 705#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 706#define NOINLINE 707#else 708#define NOINLINE __attribute__ ((noinline)) 709#endif 710 711static void _expand_stack_to(address bottom) NOINLINE; 712 713static void _expand_stack_to(address bottom) { 714 address sp; 715 size_t size; 716 volatile char *p; 717 718 // Adjust bottom to point to the largest address within the same page, it 719 // gives us a one-page buffer if alloca() allocates slightly more memory. 720 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 721 bottom += os::Linux::page_size() - 1; 722 723 // sp might be slightly above current stack pointer; if that's the case, we 724 // will alloca() a little more space than necessary, which is OK. Don't use 725 // os::current_stack_pointer(), as its result can be slightly below current 726 // stack pointer, causing us to not alloca enough to reach "bottom". 727 sp = (address)&sp; 728 729 if (sp > bottom) { 730 size = sp - bottom; 731 p = (volatile char *)alloca(size); 732 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 733 p[0] = '\0'; 734 } 735} 736 737bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 738 assert(t!=NULL, "just checking"); 739 assert(t->osthread()->expanding_stack(), "expand should be set"); 740 assert(t->stack_base() != NULL, "stack_base was not initialized"); 741 742 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 743 sigset_t mask_all, old_sigset; 744 sigfillset(&mask_all); 745 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 746 _expand_stack_to(addr); 747 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 748 return true; 749 } 750 return false; 751} 752 753////////////////////////////////////////////////////////////////////////////// 754// create new thread 755 756static address highest_vm_reserved_address(); 757 758// check if it's safe to start a new thread 759static bool _thread_safety_check(Thread* thread) { 760 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 761 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 762 // Heap is mmap'ed at lower end of memory space. Thread stacks are 763 // allocated (MAP_FIXED) from high address space. Every thread stack 764 // occupies a fixed size slot (usually 2Mbytes, but user can change 765 // it to other values if they rebuild LinuxThreads). 766 // 767 // Problem with MAP_FIXED is that mmap() can still succeed even part of 768 // the memory region has already been mmap'ed. That means if we have too 769 // many threads and/or very large heap, eventually thread stack will 770 // collide with heap. 771 // 772 // Here we try to prevent heap/stack collision by comparing current 773 // stack bottom with the highest address that has been mmap'ed by JVM 774 // plus a safety margin for memory maps created by native code. 775 // 776 // This feature can be disabled by setting ThreadSafetyMargin to 0 777 // 778 if (ThreadSafetyMargin > 0) { 779 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 780 781 // not safe if our stack extends below the safety margin 782 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 783 } else { 784 return true; 785 } 786 } else { 787 // Floating stack LinuxThreads or NPTL: 788 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 789 // there's not enough space left, pthread_create() will fail. If we come 790 // here, that means enough space has been reserved for stack. 791 return true; 792 } 793} 794 795// Thread start routine for all newly created threads 796static void *java_start(Thread *thread) { 797 // Try to randomize the cache line index of hot stack frames. 798 // This helps when threads of the same stack traces evict each other's 799 // cache lines. The threads can be either from the same JVM instance, or 800 // from different JVM instances. The benefit is especially true for 801 // processors with hyperthreading technology. 802 static int counter = 0; 803 int pid = os::current_process_id(); 804 alloca(((pid ^ counter++) & 7) * 128); 805 806 ThreadLocalStorage::set_thread(thread); 807 808 OSThread* osthread = thread->osthread(); 809 Monitor* sync = osthread->startThread_lock(); 810 811 // non floating stack LinuxThreads needs extra check, see above 812 if (!_thread_safety_check(thread)) { 813 // notify parent thread 814 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 815 osthread->set_state(ZOMBIE); 816 sync->notify_all(); 817 return NULL; 818 } 819 820 // thread_id is kernel thread id (similar to Solaris LWP id) 821 osthread->set_thread_id(os::Linux::gettid()); 822 823 if (UseNUMA) { 824 int lgrp_id = os::numa_get_group_id(); 825 if (lgrp_id != -1) { 826 thread->set_lgrp_id(lgrp_id); 827 } 828 } 829 // initialize signal mask for this thread 830 os::Linux::hotspot_sigmask(thread); 831 832 // initialize floating point control register 833 os::Linux::init_thread_fpu_state(); 834 835 // handshaking with parent thread 836 { 837 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 838 839 // notify parent thread 840 osthread->set_state(INITIALIZED); 841 sync->notify_all(); 842 843 // wait until os::start_thread() 844 while (osthread->get_state() == INITIALIZED) { 845 sync->wait(Mutex::_no_safepoint_check_flag); 846 } 847 } 848 849 // call one more level start routine 850 thread->run(); 851 852 return 0; 853} 854 855bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 856 assert(thread->osthread() == NULL, "caller responsible"); 857 858 // Allocate the OSThread object 859 OSThread* osthread = new OSThread(NULL, NULL); 860 if (osthread == NULL) { 861 return false; 862 } 863 864 // set the correct thread state 865 osthread->set_thread_type(thr_type); 866 867 // Initial state is ALLOCATED but not INITIALIZED 868 osthread->set_state(ALLOCATED); 869 870 thread->set_osthread(osthread); 871 872 // init thread attributes 873 pthread_attr_t attr; 874 pthread_attr_init(&attr); 875 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 876 877 // stack size 878 if (os::Linux::supports_variable_stack_size()) { 879 // calculate stack size if it's not specified by caller 880 if (stack_size == 0) { 881 stack_size = os::Linux::default_stack_size(thr_type); 882 883 switch (thr_type) { 884 case os::java_thread: 885 // Java threads use ThreadStackSize which default value can be 886 // changed with the flag -Xss 887 assert (JavaThread::stack_size_at_create() > 0, "this should be set"); 888 stack_size = JavaThread::stack_size_at_create(); 889 break; 890 case os::compiler_thread: 891 if (CompilerThreadStackSize > 0) { 892 stack_size = (size_t)(CompilerThreadStackSize * K); 893 break; 894 } // else fall through: 895 // use VMThreadStackSize if CompilerThreadStackSize is not defined 896 case os::vm_thread: 897 case os::pgc_thread: 898 case os::cgc_thread: 899 case os::watcher_thread: 900 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 901 break; 902 } 903 } 904 905 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 906 pthread_attr_setstacksize(&attr, stack_size); 907 } else { 908 // let pthread_create() pick the default value. 909 } 910 911 // glibc guard page 912 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 913 914 ThreadState state; 915 916 { 917 // Serialize thread creation if we are running with fixed stack LinuxThreads 918 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 919 if (lock) { 920 os::Linux::createThread_lock()->lock_without_safepoint_check(); 921 } 922 923 pthread_t tid; 924 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 925 926 pthread_attr_destroy(&attr); 927 928 if (ret != 0) { 929 if (PrintMiscellaneous && (Verbose || WizardMode)) { 930 perror("pthread_create()"); 931 } 932 // Need to clean up stuff we've allocated so far 933 thread->set_osthread(NULL); 934 delete osthread; 935 if (lock) os::Linux::createThread_lock()->unlock(); 936 return false; 937 } 938 939 // Store pthread info into the OSThread 940 osthread->set_pthread_id(tid); 941 942 // Wait until child thread is either initialized or aborted 943 { 944 Monitor* sync_with_child = osthread->startThread_lock(); 945 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 946 while ((state = osthread->get_state()) == ALLOCATED) { 947 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 948 } 949 } 950 951 if (lock) { 952 os::Linux::createThread_lock()->unlock(); 953 } 954 } 955 956 // Aborted due to thread limit being reached 957 if (state == ZOMBIE) { 958 thread->set_osthread(NULL); 959 delete osthread; 960 return false; 961 } 962 963 // The thread is returned suspended (in state INITIALIZED), 964 // and is started higher up in the call chain 965 assert(state == INITIALIZED, "race condition"); 966 return true; 967} 968 969///////////////////////////////////////////////////////////////////////////// 970// attach existing thread 971 972// bootstrap the main thread 973bool os::create_main_thread(JavaThread* thread) { 974 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 975 return create_attached_thread(thread); 976} 977 978bool os::create_attached_thread(JavaThread* thread) { 979#ifdef ASSERT 980 thread->verify_not_published(); 981#endif 982 983 // Allocate the OSThread object 984 OSThread* osthread = new OSThread(NULL, NULL); 985 986 if (osthread == NULL) { 987 return false; 988 } 989 990 // Store pthread info into the OSThread 991 osthread->set_thread_id(os::Linux::gettid()); 992 osthread->set_pthread_id(::pthread_self()); 993 994 // initialize floating point control register 995 os::Linux::init_thread_fpu_state(); 996 997 // Initial thread state is RUNNABLE 998 osthread->set_state(RUNNABLE); 999 1000 thread->set_osthread(osthread); 1001 1002 if (UseNUMA) { 1003 int lgrp_id = os::numa_get_group_id(); 1004 if (lgrp_id != -1) { 1005 thread->set_lgrp_id(lgrp_id); 1006 } 1007 } 1008 1009 if (os::Linux::is_initial_thread()) { 1010 // If current thread is initial thread, its stack is mapped on demand, 1011 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 1012 // the entire stack region to avoid SEGV in stack banging. 1013 // It is also useful to get around the heap-stack-gap problem on SuSE 1014 // kernel (see 4821821 for details). We first expand stack to the top 1015 // of yellow zone, then enable stack yellow zone (order is significant, 1016 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1017 // is no gap between the last two virtual memory regions. 1018 1019 JavaThread *jt = (JavaThread *)thread; 1020 address addr = jt->stack_yellow_zone_base(); 1021 assert(addr != NULL, "initialization problem?"); 1022 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1023 1024 osthread->set_expanding_stack(); 1025 os::Linux::manually_expand_stack(jt, addr); 1026 osthread->clear_expanding_stack(); 1027 } 1028 1029 // initialize signal mask for this thread 1030 // and save the caller's signal mask 1031 os::Linux::hotspot_sigmask(thread); 1032 1033 return true; 1034} 1035 1036void os::pd_start_thread(Thread* thread) { 1037 OSThread * osthread = thread->osthread(); 1038 assert(osthread->get_state() != INITIALIZED, "just checking"); 1039 Monitor* sync_with_child = osthread->startThread_lock(); 1040 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1041 sync_with_child->notify(); 1042} 1043 1044// Free Linux resources related to the OSThread 1045void os::free_thread(OSThread* osthread) { 1046 assert(osthread != NULL, "osthread not set"); 1047 1048 if (Thread::current()->osthread() == osthread) { 1049 // Restore caller's signal mask 1050 sigset_t sigmask = osthread->caller_sigmask(); 1051 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1052 } 1053 1054 delete osthread; 1055} 1056 1057////////////////////////////////////////////////////////////////////////////// 1058// thread local storage 1059 1060int os::allocate_thread_local_storage() { 1061 pthread_key_t key; 1062 int rslt = pthread_key_create(&key, NULL); 1063 assert(rslt == 0, "cannot allocate thread local storage"); 1064 return (int)key; 1065} 1066 1067// Note: This is currently not used by VM, as we don't destroy TLS key 1068// on VM exit. 1069void os::free_thread_local_storage(int index) { 1070 int rslt = pthread_key_delete((pthread_key_t)index); 1071 assert(rslt == 0, "invalid index"); 1072} 1073 1074void os::thread_local_storage_at_put(int index, void* value) { 1075 int rslt = pthread_setspecific((pthread_key_t)index, value); 1076 assert(rslt == 0, "pthread_setspecific failed"); 1077} 1078 1079extern "C" Thread* get_thread() { 1080 return ThreadLocalStorage::thread(); 1081} 1082 1083////////////////////////////////////////////////////////////////////////////// 1084// initial thread 1085 1086// Check if current thread is the initial thread, similar to Solaris thr_main. 1087bool os::Linux::is_initial_thread(void) { 1088 char dummy; 1089 // If called before init complete, thread stack bottom will be null. 1090 // Can be called if fatal error occurs before initialization. 1091 if (initial_thread_stack_bottom() == NULL) return false; 1092 assert(initial_thread_stack_bottom() != NULL && 1093 initial_thread_stack_size() != 0, 1094 "os::init did not locate initial thread's stack region"); 1095 if ((address)&dummy >= initial_thread_stack_bottom() && 1096 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1097 return true; 1098 else return false; 1099} 1100 1101// Find the virtual memory area that contains addr 1102static bool find_vma(address addr, address* vma_low, address* vma_high) { 1103 FILE *fp = fopen("/proc/self/maps", "r"); 1104 if (fp) { 1105 address low, high; 1106 while (!feof(fp)) { 1107 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1108 if (low <= addr && addr < high) { 1109 if (vma_low) *vma_low = low; 1110 if (vma_high) *vma_high = high; 1111 fclose (fp); 1112 return true; 1113 } 1114 } 1115 for (;;) { 1116 int ch = fgetc(fp); 1117 if (ch == EOF || ch == (int)'\n') break; 1118 } 1119 } 1120 fclose(fp); 1121 } 1122 return false; 1123} 1124 1125// Locate initial thread stack. This special handling of initial thread stack 1126// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1127// bogus value for initial thread. 1128void os::Linux::capture_initial_stack(size_t max_size) { 1129 // stack size is the easy part, get it from RLIMIT_STACK 1130 size_t stack_size; 1131 struct rlimit rlim; 1132 getrlimit(RLIMIT_STACK, &rlim); 1133 stack_size = rlim.rlim_cur; 1134 1135 // 6308388: a bug in ld.so will relocate its own .data section to the 1136 // lower end of primordial stack; reduce ulimit -s value a little bit 1137 // so we won't install guard page on ld.so's data section. 1138 stack_size -= 2 * page_size(); 1139 1140 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1141 // 7.1, in both cases we will get 2G in return value. 1142 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1143 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1144 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1145 // in case other parts in glibc still assumes 2M max stack size. 1146 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1147#ifndef IA64 1148 if (stack_size > 2 * K * K) stack_size = 2 * K * K; 1149#else 1150 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1151 if (stack_size > 4 * K * K) stack_size = 4 * K * K; 1152#endif 1153 1154 // Try to figure out where the stack base (top) is. This is harder. 1155 // 1156 // When an application is started, glibc saves the initial stack pointer in 1157 // a global variable "__libc_stack_end", which is then used by system 1158 // libraries. __libc_stack_end should be pretty close to stack top. The 1159 // variable is available since the very early days. However, because it is 1160 // a private interface, it could disappear in the future. 1161 // 1162 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1163 // to __libc_stack_end, it is very close to stack top, but isn't the real 1164 // stack top. Note that /proc may not exist if VM is running as a chroot 1165 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1166 // /proc/<pid>/stat could change in the future (though unlikely). 1167 // 1168 // We try __libc_stack_end first. If that doesn't work, look for 1169 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1170 // as a hint, which should work well in most cases. 1171 1172 uintptr_t stack_start; 1173 1174 // try __libc_stack_end first 1175 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1176 if (p && *p) { 1177 stack_start = *p; 1178 } else { 1179 // see if we can get the start_stack field from /proc/self/stat 1180 FILE *fp; 1181 int pid; 1182 char state; 1183 int ppid; 1184 int pgrp; 1185 int session; 1186 int nr; 1187 int tpgrp; 1188 unsigned long flags; 1189 unsigned long minflt; 1190 unsigned long cminflt; 1191 unsigned long majflt; 1192 unsigned long cmajflt; 1193 unsigned long utime; 1194 unsigned long stime; 1195 long cutime; 1196 long cstime; 1197 long prio; 1198 long nice; 1199 long junk; 1200 long it_real; 1201 uintptr_t start; 1202 uintptr_t vsize; 1203 intptr_t rss; 1204 uintptr_t rsslim; 1205 uintptr_t scodes; 1206 uintptr_t ecode; 1207 int i; 1208 1209 // Figure what the primordial thread stack base is. Code is inspired 1210 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1211 // followed by command name surrounded by parentheses, state, etc. 1212 char stat[2048]; 1213 int statlen; 1214 1215 fp = fopen("/proc/self/stat", "r"); 1216 if (fp) { 1217 statlen = fread(stat, 1, 2047, fp); 1218 stat[statlen] = '\0'; 1219 fclose(fp); 1220 1221 // Skip pid and the command string. Note that we could be dealing with 1222 // weird command names, e.g. user could decide to rename java launcher 1223 // to "java 1.4.2 :)", then the stat file would look like 1224 // 1234 (java 1.4.2 :)) R ... ... 1225 // We don't really need to know the command string, just find the last 1226 // occurrence of ")" and then start parsing from there. See bug 4726580. 1227 char * s = strrchr(stat, ')'); 1228 1229 i = 0; 1230 if (s) { 1231 // Skip blank chars 1232 do s++; while (isspace(*s)); 1233 1234#define _UFM UINTX_FORMAT 1235#define _DFM INTX_FORMAT 1236 1237 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1238 /* 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 */ 1239 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM, 1240 &state, /* 3 %c */ 1241 &ppid, /* 4 %d */ 1242 &pgrp, /* 5 %d */ 1243 &session, /* 6 %d */ 1244 &nr, /* 7 %d */ 1245 &tpgrp, /* 8 %d */ 1246 &flags, /* 9 %lu */ 1247 &minflt, /* 10 %lu */ 1248 &cminflt, /* 11 %lu */ 1249 &majflt, /* 12 %lu */ 1250 &cmajflt, /* 13 %lu */ 1251 &utime, /* 14 %lu */ 1252 &stime, /* 15 %lu */ 1253 &cutime, /* 16 %ld */ 1254 &cstime, /* 17 %ld */ 1255 &prio, /* 18 %ld */ 1256 &nice, /* 19 %ld */ 1257 &junk, /* 20 %ld */ 1258 &it_real, /* 21 %ld */ 1259 &start, /* 22 UINTX_FORMAT */ 1260 &vsize, /* 23 UINTX_FORMAT */ 1261 &rss, /* 24 INTX_FORMAT */ 1262 &rsslim, /* 25 UINTX_FORMAT */ 1263 &scodes, /* 26 UINTX_FORMAT */ 1264 &ecode, /* 27 UINTX_FORMAT */ 1265 &stack_start); /* 28 UINTX_FORMAT */ 1266 } 1267 1268#undef _UFM 1269#undef _DFM 1270 1271 if (i != 28 - 2) { 1272 assert(false, "Bad conversion from /proc/self/stat"); 1273 // product mode - assume we are the initial thread, good luck in the 1274 // embedded case. 1275 warning("Can't detect initial thread stack location - bad conversion"); 1276 stack_start = (uintptr_t) &rlim; 1277 } 1278 } else { 1279 // For some reason we can't open /proc/self/stat (for example, running on 1280 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1281 // most cases, so don't abort: 1282 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1283 stack_start = (uintptr_t) &rlim; 1284 } 1285 } 1286 1287 // Now we have a pointer (stack_start) very close to the stack top, the 1288 // next thing to do is to figure out the exact location of stack top. We 1289 // can find out the virtual memory area that contains stack_start by 1290 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1291 // and its upper limit is the real stack top. (again, this would fail if 1292 // running inside chroot, because /proc may not exist.) 1293 1294 uintptr_t stack_top; 1295 address low, high; 1296 if (find_vma((address)stack_start, &low, &high)) { 1297 // success, "high" is the true stack top. (ignore "low", because initial 1298 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1299 stack_top = (uintptr_t)high; 1300 } else { 1301 // failed, likely because /proc/self/maps does not exist 1302 warning("Can't detect initial thread stack location - find_vma failed"); 1303 // best effort: stack_start is normally within a few pages below the real 1304 // stack top, use it as stack top, and reduce stack size so we won't put 1305 // guard page outside stack. 1306 stack_top = stack_start; 1307 stack_size -= 16 * page_size(); 1308 } 1309 1310 // stack_top could be partially down the page so align it 1311 stack_top = align_size_up(stack_top, page_size()); 1312 1313 if (max_size && stack_size > max_size) { 1314 _initial_thread_stack_size = max_size; 1315 } else { 1316 _initial_thread_stack_size = stack_size; 1317 } 1318 1319 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1320 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1321} 1322 1323//////////////////////////////////////////////////////////////////////////////// 1324// time support 1325 1326// Time since start-up in seconds to a fine granularity. 1327// Used by VMSelfDestructTimer and the MemProfiler. 1328double os::elapsedTime() { 1329 1330 return (double)(os::elapsed_counter()) * 0.000001; 1331} 1332 1333jlong os::elapsed_counter() { 1334 timeval time; 1335 int status = gettimeofday(&time, NULL); 1336 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1337} 1338 1339jlong os::elapsed_frequency() { 1340 return (1000 * 1000); 1341} 1342 1343// For now, we say that linux does not support vtime. I have no idea 1344// whether it can actually be made to (DLD, 9/13/05). 1345 1346bool os::supports_vtime() { return false; } 1347bool os::enable_vtime() { return false; } 1348bool os::vtime_enabled() { return false; } 1349double os::elapsedVTime() { 1350 // better than nothing, but not much 1351 return elapsedTime(); 1352} 1353 1354jlong os::javaTimeMillis() { 1355 timeval time; 1356 int status = gettimeofday(&time, NULL); 1357 assert(status != -1, "linux error"); 1358 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1359} 1360 1361#ifndef CLOCK_MONOTONIC 1362#define CLOCK_MONOTONIC (1) 1363#endif 1364 1365void os::Linux::clock_init() { 1366 // we do dlopen's in this particular order due to bug in linux 1367 // dynamical loader (see 6348968) leading to crash on exit 1368 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1369 if (handle == NULL) { 1370 handle = dlopen("librt.so", RTLD_LAZY); 1371 } 1372 1373 if (handle) { 1374 int (*clock_getres_func)(clockid_t, struct timespec*) = 1375 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1376 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1377 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1378 if (clock_getres_func && clock_gettime_func) { 1379 // See if monotonic clock is supported by the kernel. Note that some 1380 // early implementations simply return kernel jiffies (updated every 1381 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1382 // for nano time (though the monotonic property is still nice to have). 1383 // It's fixed in newer kernels, however clock_getres() still returns 1384 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1385 // resolution for now. Hopefully as people move to new kernels, this 1386 // won't be a problem. 1387 struct timespec res; 1388 struct timespec tp; 1389 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1390 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1391 // yes, monotonic clock is supported 1392 _clock_gettime = clock_gettime_func; 1393 } else { 1394 // close librt if there is no monotonic clock 1395 dlclose(handle); 1396 } 1397 } 1398 } 1399} 1400 1401#ifndef SYS_clock_getres 1402 1403#if defined(IA32) || defined(AMD64) 1404#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1405#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1406#else 1407#warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1408#define sys_clock_getres(x,y) -1 1409#endif 1410 1411#else 1412#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1413#endif 1414 1415void os::Linux::fast_thread_clock_init() { 1416 if (!UseLinuxPosixThreadCPUClocks) { 1417 return; 1418 } 1419 clockid_t clockid; 1420 struct timespec tp; 1421 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1422 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1423 1424 // Switch to using fast clocks for thread cpu time if 1425 // the sys_clock_getres() returns 0 error code. 1426 // Note, that some kernels may support the current thread 1427 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1428 // returned by the pthread_getcpuclockid(). 1429 // If the fast Posix clocks are supported then the sys_clock_getres() 1430 // must return at least tp.tv_sec == 0 which means a resolution 1431 // better than 1 sec. This is extra check for reliability. 1432 1433 if(pthread_getcpuclockid_func && 1434 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1435 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1436 1437 _supports_fast_thread_cpu_time = true; 1438 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1439 } 1440} 1441 1442jlong os::javaTimeNanos() { 1443 if (Linux::supports_monotonic_clock()) { 1444 struct timespec tp; 1445 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1446 assert(status == 0, "gettime error"); 1447 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1448 return result; 1449 } else { 1450 timeval time; 1451 int status = gettimeofday(&time, NULL); 1452 assert(status != -1, "linux error"); 1453 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1454 return 1000 * usecs; 1455 } 1456} 1457 1458void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1459 if (Linux::supports_monotonic_clock()) { 1460 info_ptr->max_value = ALL_64_BITS; 1461 1462 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1463 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1464 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1465 } else { 1466 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1467 info_ptr->max_value = ALL_64_BITS; 1468 1469 // gettimeofday is a real time clock so it skips 1470 info_ptr->may_skip_backward = true; 1471 info_ptr->may_skip_forward = true; 1472 } 1473 1474 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1475} 1476 1477// Return the real, user, and system times in seconds from an 1478// arbitrary fixed point in the past. 1479bool os::getTimesSecs(double* process_real_time, 1480 double* process_user_time, 1481 double* process_system_time) { 1482 struct tms ticks; 1483 clock_t real_ticks = times(&ticks); 1484 1485 if (real_ticks == (clock_t) (-1)) { 1486 return false; 1487 } else { 1488 double ticks_per_second = (double) clock_tics_per_sec; 1489 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1490 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1491 *process_real_time = ((double) real_ticks) / ticks_per_second; 1492 1493 return true; 1494 } 1495} 1496 1497 1498char * os::local_time_string(char *buf, size_t buflen) { 1499 struct tm t; 1500 time_t long_time; 1501 time(&long_time); 1502 localtime_r(&long_time, &t); 1503 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1504 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1505 t.tm_hour, t.tm_min, t.tm_sec); 1506 return buf; 1507} 1508 1509struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1510 return localtime_r(clock, res); 1511} 1512 1513//////////////////////////////////////////////////////////////////////////////// 1514// runtime exit support 1515 1516// Note: os::shutdown() might be called very early during initialization, or 1517// called from signal handler. Before adding something to os::shutdown(), make 1518// sure it is async-safe and can handle partially initialized VM. 1519void os::shutdown() { 1520 1521 // allow PerfMemory to attempt cleanup of any persistent resources 1522 perfMemory_exit(); 1523 1524 // needs to remove object in file system 1525 AttachListener::abort(); 1526 1527 // flush buffered output, finish log files 1528 ostream_abort(); 1529 1530 // Check for abort hook 1531 abort_hook_t abort_hook = Arguments::abort_hook(); 1532 if (abort_hook != NULL) { 1533 abort_hook(); 1534 } 1535 1536} 1537 1538// Note: os::abort() might be called very early during initialization, or 1539// called from signal handler. Before adding something to os::abort(), make 1540// sure it is async-safe and can handle partially initialized VM. 1541void os::abort(bool dump_core) { 1542 os::shutdown(); 1543 if (dump_core) { 1544#ifndef PRODUCT 1545 fdStream out(defaultStream::output_fd()); 1546 out.print_raw("Current thread is "); 1547 char buf[16]; 1548 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1549 out.print_raw_cr(buf); 1550 out.print_raw_cr("Dumping core ..."); 1551#endif 1552 ::abort(); // dump core 1553 } 1554 1555 ::exit(1); 1556} 1557 1558// Die immediately, no exit hook, no abort hook, no cleanup. 1559void os::die() { 1560 // _exit() on LinuxThreads only kills current thread 1561 ::abort(); 1562} 1563 1564// unused on linux for now. 1565void os::set_error_file(const char *logfile) {} 1566 1567 1568// This method is a copy of JDK's sysGetLastErrorString 1569// from src/solaris/hpi/src/system_md.c 1570 1571size_t os::lasterror(char *buf, size_t len) { 1572 1573 if (errno == 0) return 0; 1574 1575 const char *s = ::strerror(errno); 1576 size_t n = ::strlen(s); 1577 if (n >= len) { 1578 n = len - 1; 1579 } 1580 ::strncpy(buf, s, n); 1581 buf[n] = '\0'; 1582 return n; 1583} 1584 1585intx os::current_thread_id() { return (intx)pthread_self(); } 1586int os::current_process_id() { 1587 1588 // Under the old linux thread library, linux gives each thread 1589 // its own process id. Because of this each thread will return 1590 // a different pid if this method were to return the result 1591 // of getpid(2). Linux provides no api that returns the pid 1592 // of the launcher thread for the vm. This implementation 1593 // returns a unique pid, the pid of the launcher thread 1594 // that starts the vm 'process'. 1595 1596 // Under the NPTL, getpid() returns the same pid as the 1597 // launcher thread rather than a unique pid per thread. 1598 // Use gettid() if you want the old pre NPTL behaviour. 1599 1600 // if you are looking for the result of a call to getpid() that 1601 // returns a unique pid for the calling thread, then look at the 1602 // OSThread::thread_id() method in osThread_linux.hpp file 1603 1604 return (int)(_initial_pid ? _initial_pid : getpid()); 1605} 1606 1607// DLL functions 1608 1609const char* os::dll_file_extension() { return ".so"; } 1610 1611const char* os::get_temp_directory() { 1612 const char *prop = Arguments::get_property("java.io.tmpdir"); 1613 return prop == NULL ? "/tmp" : prop; 1614} 1615 1616static bool file_exists(const char* filename) { 1617 struct stat statbuf; 1618 if (filename == NULL || strlen(filename) == 0) { 1619 return false; 1620 } 1621 return os::stat(filename, &statbuf) == 0; 1622} 1623 1624void os::dll_build_name(char* buffer, size_t buflen, 1625 const char* pname, const char* fname) { 1626 // Copied from libhpi 1627 const size_t pnamelen = pname ? strlen(pname) : 0; 1628 1629 // Quietly truncate on buffer overflow. Should be an error. 1630 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1631 *buffer = '\0'; 1632 return; 1633 } 1634 1635 if (pnamelen == 0) { 1636 snprintf(buffer, buflen, "lib%s.so", fname); 1637 } else if (strchr(pname, *os::path_separator()) != NULL) { 1638 int n; 1639 char** pelements = split_path(pname, &n); 1640 for (int i = 0 ; i < n ; i++) { 1641 // Really shouldn't be NULL, but check can't hurt 1642 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1643 continue; // skip the empty path values 1644 } 1645 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1646 if (file_exists(buffer)) { 1647 break; 1648 } 1649 } 1650 // release the storage 1651 for (int i = 0 ; i < n ; i++) { 1652 if (pelements[i] != NULL) { 1653 FREE_C_HEAP_ARRAY(char, pelements[i]); 1654 } 1655 } 1656 if (pelements != NULL) { 1657 FREE_C_HEAP_ARRAY(char*, pelements); 1658 } 1659 } else { 1660 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1661 } 1662} 1663 1664const char* os::get_current_directory(char *buf, int buflen) { 1665 return getcwd(buf, buflen); 1666} 1667 1668// check if addr is inside libjvm[_g].so 1669bool os::address_is_in_vm(address addr) { 1670 static address libjvm_base_addr; 1671 Dl_info dlinfo; 1672 1673 if (libjvm_base_addr == NULL) { 1674 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); 1675 libjvm_base_addr = (address)dlinfo.dli_fbase; 1676 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1677 } 1678 1679 if (dladdr((void *)addr, &dlinfo)) { 1680 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1681 } 1682 1683 return false; 1684} 1685 1686bool os::dll_address_to_function_name(address addr, char *buf, 1687 int buflen, int *offset) { 1688 Dl_info dlinfo; 1689 1690 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { 1691 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1692 if (offset) *offset = addr - (address)dlinfo.dli_saddr; 1693 return true; 1694 } else { 1695 if (buf) buf[0] = '\0'; 1696 if (offset) *offset = -1; 1697 return false; 1698 } 1699} 1700 1701struct _address_to_library_name { 1702 address addr; // input : memory address 1703 size_t buflen; // size of fname 1704 char* fname; // output: library name 1705 address base; // library base addr 1706}; 1707 1708static int address_to_library_name_callback(struct dl_phdr_info *info, 1709 size_t size, void *data) { 1710 int i; 1711 bool found = false; 1712 address libbase = NULL; 1713 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1714 1715 // iterate through all loadable segments 1716 for (i = 0; i < info->dlpi_phnum; i++) { 1717 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1718 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1719 // base address of a library is the lowest address of its loaded 1720 // segments. 1721 if (libbase == NULL || libbase > segbase) { 1722 libbase = segbase; 1723 } 1724 // see if 'addr' is within current segment 1725 if (segbase <= d->addr && 1726 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1727 found = true; 1728 } 1729 } 1730 } 1731 1732 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1733 // so dll_address_to_library_name() can fall through to use dladdr() which 1734 // can figure out executable name from argv[0]. 1735 if (found && info->dlpi_name && info->dlpi_name[0]) { 1736 d->base = libbase; 1737 if (d->fname) { 1738 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1739 } 1740 return 1; 1741 } 1742 return 0; 1743} 1744 1745bool os::dll_address_to_library_name(address addr, char* buf, 1746 int buflen, int* offset) { 1747 Dl_info dlinfo; 1748 struct _address_to_library_name data; 1749 1750 // There is a bug in old glibc dladdr() implementation that it could resolve 1751 // to wrong library name if the .so file has a base address != NULL. Here 1752 // we iterate through the program headers of all loaded libraries to find 1753 // out which library 'addr' really belongs to. This workaround can be 1754 // removed once the minimum requirement for glibc is moved to 2.3.x. 1755 data.addr = addr; 1756 data.fname = buf; 1757 data.buflen = buflen; 1758 data.base = NULL; 1759 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1760 1761 if (rslt) { 1762 // buf already contains library name 1763 if (offset) *offset = addr - data.base; 1764 return true; 1765 } else if (dladdr((void*)addr, &dlinfo)){ 1766 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1767 if (offset) *offset = addr - (address)dlinfo.dli_fbase; 1768 return true; 1769 } else { 1770 if (buf) buf[0] = '\0'; 1771 if (offset) *offset = -1; 1772 return false; 1773 } 1774} 1775 1776 // Loads .dll/.so and 1777 // in case of error it checks if .dll/.so was built for the 1778 // same architecture as Hotspot is running on 1779 1780void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1781{ 1782 void * result= ::dlopen(filename, RTLD_LAZY); 1783 if (result != NULL) { 1784 // Successful loading 1785 return result; 1786 } 1787 1788 Elf32_Ehdr elf_head; 1789 1790 // Read system error message into ebuf 1791 // It may or may not be overwritten below 1792 ::strncpy(ebuf, ::dlerror(), ebuflen-1); 1793 ebuf[ebuflen-1]='\0'; 1794 int diag_msg_max_length=ebuflen-strlen(ebuf); 1795 char* diag_msg_buf=ebuf+strlen(ebuf); 1796 1797 if (diag_msg_max_length==0) { 1798 // No more space in ebuf for additional diagnostics message 1799 return NULL; 1800 } 1801 1802 1803 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1804 1805 if (file_descriptor < 0) { 1806 // Can't open library, report dlerror() message 1807 return NULL; 1808 } 1809 1810 bool failed_to_read_elf_head= 1811 (sizeof(elf_head)!= 1812 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1813 1814 ::close(file_descriptor); 1815 if (failed_to_read_elf_head) { 1816 // file i/o error - report dlerror() msg 1817 return NULL; 1818 } 1819 1820 typedef struct { 1821 Elf32_Half code; // Actual value as defined in elf.h 1822 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1823 char elf_class; // 32 or 64 bit 1824 char endianess; // MSB or LSB 1825 char* name; // String representation 1826 } arch_t; 1827 1828 #ifndef EM_486 1829 #define EM_486 6 /* Intel 80486 */ 1830 #endif 1831 1832 static const arch_t arch_array[]={ 1833 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1834 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1835 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1836 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1837 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1838 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1839 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1840 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1841 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1842 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1843 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1844 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1845 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1846 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1847 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1848 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1849 }; 1850 1851 #if (defined IA32) 1852 static Elf32_Half running_arch_code=EM_386; 1853 #elif (defined AMD64) 1854 static Elf32_Half running_arch_code=EM_X86_64; 1855 #elif (defined IA64) 1856 static Elf32_Half running_arch_code=EM_IA_64; 1857 #elif (defined __sparc) && (defined _LP64) 1858 static Elf32_Half running_arch_code=EM_SPARCV9; 1859 #elif (defined __sparc) && (!defined _LP64) 1860 static Elf32_Half running_arch_code=EM_SPARC; 1861 #elif (defined __powerpc64__) 1862 static Elf32_Half running_arch_code=EM_PPC64; 1863 #elif (defined __powerpc__) 1864 static Elf32_Half running_arch_code=EM_PPC; 1865 #elif (defined ARM) 1866 static Elf32_Half running_arch_code=EM_ARM; 1867 #elif (defined S390) 1868 static Elf32_Half running_arch_code=EM_S390; 1869 #elif (defined ALPHA) 1870 static Elf32_Half running_arch_code=EM_ALPHA; 1871 #elif (defined MIPSEL) 1872 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1873 #elif (defined PARISC) 1874 static Elf32_Half running_arch_code=EM_PARISC; 1875 #elif (defined MIPS) 1876 static Elf32_Half running_arch_code=EM_MIPS; 1877 #elif (defined M68K) 1878 static Elf32_Half running_arch_code=EM_68K; 1879 #else 1880 #error Method os::dll_load requires that one of following is defined:\ 1881 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 1882 #endif 1883 1884 // Identify compatability class for VM's architecture and library's architecture 1885 // Obtain string descriptions for architectures 1886 1887 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1888 int running_arch_index=-1; 1889 1890 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1891 if (running_arch_code == arch_array[i].code) { 1892 running_arch_index = i; 1893 } 1894 if (lib_arch.code == arch_array[i].code) { 1895 lib_arch.compat_class = arch_array[i].compat_class; 1896 lib_arch.name = arch_array[i].name; 1897 } 1898 } 1899 1900 assert(running_arch_index != -1, 1901 "Didn't find running architecture code (running_arch_code) in arch_array"); 1902 if (running_arch_index == -1) { 1903 // Even though running architecture detection failed 1904 // we may still continue with reporting dlerror() message 1905 return NULL; 1906 } 1907 1908 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1909 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1910 return NULL; 1911 } 1912 1913#ifndef S390 1914 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1915 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1916 return NULL; 1917 } 1918#endif // !S390 1919 1920 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1921 if ( lib_arch.name!=NULL ) { 1922 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1923 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1924 lib_arch.name, arch_array[running_arch_index].name); 1925 } else { 1926 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1927 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1928 lib_arch.code, 1929 arch_array[running_arch_index].name); 1930 } 1931 } 1932 1933 return NULL; 1934} 1935 1936/* 1937 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 1938 * chances are you might want to run the generated bits against glibc-2.0 1939 * libdl.so, so always use locking for any version of glibc. 1940 */ 1941void* os::dll_lookup(void* handle, const char* name) { 1942 pthread_mutex_lock(&dl_mutex); 1943 void* res = dlsym(handle, name); 1944 pthread_mutex_unlock(&dl_mutex); 1945 return res; 1946} 1947 1948 1949static bool _print_ascii_file(const char* filename, outputStream* st) { 1950 int fd = ::open(filename, O_RDONLY); 1951 if (fd == -1) { 1952 return false; 1953 } 1954 1955 char buf[32]; 1956 int bytes; 1957 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 1958 st->print_raw(buf, bytes); 1959 } 1960 1961 ::close(fd); 1962 1963 return true; 1964} 1965 1966void os::print_dll_info(outputStream *st) { 1967 st->print_cr("Dynamic libraries:"); 1968 1969 char fname[32]; 1970 pid_t pid = os::Linux::gettid(); 1971 1972 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1973 1974 if (!_print_ascii_file(fname, st)) { 1975 st->print("Can not get library information for pid = %d\n", pid); 1976 } 1977} 1978 1979 1980void os::print_os_info(outputStream* st) { 1981 st->print("OS:"); 1982 1983 // Try to identify popular distros. 1984 // Most Linux distributions have /etc/XXX-release file, which contains 1985 // the OS version string. Some have more than one /etc/XXX-release file 1986 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 1987 // so the order is important. 1988 if (!_print_ascii_file("/etc/mandrake-release", st) && 1989 !_print_ascii_file("/etc/sun-release", st) && 1990 !_print_ascii_file("/etc/redhat-release", st) && 1991 !_print_ascii_file("/etc/SuSE-release", st) && 1992 !_print_ascii_file("/etc/turbolinux-release", st) && 1993 !_print_ascii_file("/etc/gentoo-release", st) && 1994 !_print_ascii_file("/etc/debian_version", st) && 1995 !_print_ascii_file("/etc/ltib-release", st) && 1996 !_print_ascii_file("/etc/angstrom-version", st)) { 1997 st->print("Linux"); 1998 } 1999 st->cr(); 2000 2001 // kernel 2002 st->print("uname:"); 2003 struct utsname name; 2004 uname(&name); 2005 st->print(name.sysname); st->print(" "); 2006 st->print(name.release); st->print(" "); 2007 st->print(name.version); st->print(" "); 2008 st->print(name.machine); 2009 st->cr(); 2010 2011 // Print warning if unsafe chroot environment detected 2012 if (unsafe_chroot_detected) { 2013 st->print("WARNING!! "); 2014 st->print_cr(unstable_chroot_error); 2015 } 2016 2017 // libc, pthread 2018 st->print("libc:"); 2019 st->print(os::Linux::glibc_version()); st->print(" "); 2020 st->print(os::Linux::libpthread_version()); st->print(" "); 2021 if (os::Linux::is_LinuxThreads()) { 2022 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2023 } 2024 st->cr(); 2025 2026 // rlimit 2027 st->print("rlimit:"); 2028 struct rlimit rlim; 2029 2030 st->print(" STACK "); 2031 getrlimit(RLIMIT_STACK, &rlim); 2032 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 2033 else st->print("%uk", rlim.rlim_cur >> 10); 2034 2035 st->print(", CORE "); 2036 getrlimit(RLIMIT_CORE, &rlim); 2037 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 2038 else st->print("%uk", rlim.rlim_cur >> 10); 2039 2040 st->print(", NPROC "); 2041 getrlimit(RLIMIT_NPROC, &rlim); 2042 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 2043 else st->print("%d", rlim.rlim_cur); 2044 2045 st->print(", NOFILE "); 2046 getrlimit(RLIMIT_NOFILE, &rlim); 2047 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 2048 else st->print("%d", rlim.rlim_cur); 2049 2050 st->print(", AS "); 2051 getrlimit(RLIMIT_AS, &rlim); 2052 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity"); 2053 else st->print("%uk", rlim.rlim_cur >> 10); 2054 st->cr(); 2055 2056 // load average 2057 st->print("load average:"); 2058 double loadavg[3]; 2059 os::loadavg(loadavg, 3); 2060 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]); 2061 st->cr(); 2062 2063 // meminfo 2064 st->print("\n/proc/meminfo:\n"); 2065 _print_ascii_file("/proc/meminfo", st); 2066 st->cr(); 2067} 2068 2069void os::print_memory_info(outputStream* st) { 2070 2071 st->print("Memory:"); 2072 st->print(" %dk page", os::vm_page_size()>>10); 2073 2074 // values in struct sysinfo are "unsigned long" 2075 struct sysinfo si; 2076 sysinfo(&si); 2077 2078 st->print(", physical " UINT64_FORMAT "k", 2079 os::physical_memory() >> 10); 2080 st->print("(" UINT64_FORMAT "k free)", 2081 os::available_memory() >> 10); 2082 st->print(", swap " UINT64_FORMAT "k", 2083 ((jlong)si.totalswap * si.mem_unit) >> 10); 2084 st->print("(" UINT64_FORMAT "k free)", 2085 ((jlong)si.freeswap * si.mem_unit) >> 10); 2086 st->cr(); 2087} 2088 2089// Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 2090// but they're the same for all the linux arch that we support 2091// and they're the same for solaris but there's no common place to put this. 2092const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 2093 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 2094 "ILL_COPROC", "ILL_BADSTK" }; 2095 2096const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 2097 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 2098 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 2099 2100const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 2101 2102const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 2103 2104void os::print_siginfo(outputStream* st, void* siginfo) { 2105 st->print("siginfo:"); 2106 2107 const int buflen = 100; 2108 char buf[buflen]; 2109 siginfo_t *si = (siginfo_t*)siginfo; 2110 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 2111 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 2112 st->print("si_errno=%s", buf); 2113 } else { 2114 st->print("si_errno=%d", si->si_errno); 2115 } 2116 const int c = si->si_code; 2117 assert(c > 0, "unexpected si_code"); 2118 switch (si->si_signo) { 2119 case SIGILL: 2120 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 2121 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2122 break; 2123 case SIGFPE: 2124 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 2125 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2126 break; 2127 case SIGSEGV: 2128 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 2129 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2130 break; 2131 case SIGBUS: 2132 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 2133 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2134 break; 2135 default: 2136 st->print(", si_code=%d", si->si_code); 2137 // no si_addr 2138 } 2139 2140 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2141 UseSharedSpaces) { 2142 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2143 if (mapinfo->is_in_shared_space(si->si_addr)) { 2144 st->print("\n\nError accessing class data sharing archive." \ 2145 " Mapped file inaccessible during execution, " \ 2146 " possible disk/network problem."); 2147 } 2148 } 2149 st->cr(); 2150} 2151 2152 2153static void print_signal_handler(outputStream* st, int sig, 2154 char* buf, size_t buflen); 2155 2156void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2157 st->print_cr("Signal Handlers:"); 2158 print_signal_handler(st, SIGSEGV, buf, buflen); 2159 print_signal_handler(st, SIGBUS , buf, buflen); 2160 print_signal_handler(st, SIGFPE , buf, buflen); 2161 print_signal_handler(st, SIGPIPE, buf, buflen); 2162 print_signal_handler(st, SIGXFSZ, buf, buflen); 2163 print_signal_handler(st, SIGILL , buf, buflen); 2164 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2165 print_signal_handler(st, SR_signum, buf, buflen); 2166 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2167 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2168 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2169 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2170} 2171 2172static char saved_jvm_path[MAXPATHLEN] = {0}; 2173 2174// Find the full path to the current module, libjvm.so or libjvm_g.so 2175void os::jvm_path(char *buf, jint buflen) { 2176 // Error checking. 2177 if (buflen < MAXPATHLEN) { 2178 assert(false, "must use a large-enough buffer"); 2179 buf[0] = '\0'; 2180 return; 2181 } 2182 // Lazy resolve the path to current module. 2183 if (saved_jvm_path[0] != 0) { 2184 strcpy(buf, saved_jvm_path); 2185 return; 2186 } 2187 2188 char dli_fname[MAXPATHLEN]; 2189 bool ret = dll_address_to_library_name( 2190 CAST_FROM_FN_PTR(address, os::jvm_path), 2191 dli_fname, sizeof(dli_fname), NULL); 2192 assert(ret != 0, "cannot locate libjvm"); 2193 char *rp = realpath(dli_fname, buf); 2194 if (rp == NULL) 2195 return; 2196 2197 if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) { 2198 // Support for the gamma launcher. Typical value for buf is 2199 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 2200 // the right place in the string, then assume we are installed in a JDK and 2201 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2202 // up the path so it looks like libjvm.so is installed there (append a 2203 // fake suffix hotspot/libjvm.so). 2204 const char *p = buf + strlen(buf) - 1; 2205 for (int count = 0; p > buf && count < 5; ++count) { 2206 for (--p; p > buf && *p != '/'; --p) 2207 /* empty */ ; 2208 } 2209 2210 if (strncmp(p, "/jre/lib/", 9) != 0) { 2211 // Look for JAVA_HOME in the environment. 2212 char* java_home_var = ::getenv("JAVA_HOME"); 2213 if (java_home_var != NULL && java_home_var[0] != 0) { 2214 char* jrelib_p; 2215 int len; 2216 2217 // Check the current module name "libjvm.so" or "libjvm_g.so". 2218 p = strrchr(buf, '/'); 2219 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2220 p = strstr(p, "_g") ? "_g" : ""; 2221 2222 rp = realpath(java_home_var, buf); 2223 if (rp == NULL) 2224 return; 2225 2226 // determine if this is a legacy image or modules image 2227 // modules image doesn't have "jre" subdirectory 2228 len = strlen(buf); 2229 jrelib_p = buf + len; 2230 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2231 if (0 != access(buf, F_OK)) { 2232 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2233 } 2234 2235 if (0 == access(buf, F_OK)) { 2236 // Use current module name "libjvm[_g].so" instead of 2237 // "libjvm"debug_only("_g")".so" since for fastdebug version 2238 // we should have "libjvm.so" but debug_only("_g") adds "_g"! 2239 len = strlen(buf); 2240 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p); 2241 } else { 2242 // Go back to path of .so 2243 rp = realpath(dli_fname, buf); 2244 if (rp == NULL) 2245 return; 2246 } 2247 } 2248 } 2249 } 2250 2251 strcpy(saved_jvm_path, buf); 2252} 2253 2254void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2255 // no prefix required, not even "_" 2256} 2257 2258void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2259 // no suffix required 2260} 2261 2262//////////////////////////////////////////////////////////////////////////////// 2263// sun.misc.Signal support 2264 2265static volatile jint sigint_count = 0; 2266 2267static void 2268UserHandler(int sig, void *siginfo, void *context) { 2269 // 4511530 - sem_post is serialized and handled by the manager thread. When 2270 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2271 // don't want to flood the manager thread with sem_post requests. 2272 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2273 return; 2274 2275 // Ctrl-C is pressed during error reporting, likely because the error 2276 // handler fails to abort. Let VM die immediately. 2277 if (sig == SIGINT && is_error_reported()) { 2278 os::die(); 2279 } 2280 2281 os::signal_notify(sig); 2282} 2283 2284void* os::user_handler() { 2285 return CAST_FROM_FN_PTR(void*, UserHandler); 2286} 2287 2288extern "C" { 2289 typedef void (*sa_handler_t)(int); 2290 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2291} 2292 2293void* os::signal(int signal_number, void* handler) { 2294 struct sigaction sigAct, oldSigAct; 2295 2296 sigfillset(&(sigAct.sa_mask)); 2297 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2298 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2299 2300 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2301 // -1 means registration failed 2302 return (void *)-1; 2303 } 2304 2305 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2306} 2307 2308void os::signal_raise(int signal_number) { 2309 ::raise(signal_number); 2310} 2311 2312/* 2313 * The following code is moved from os.cpp for making this 2314 * code platform specific, which it is by its very nature. 2315 */ 2316 2317// Will be modified when max signal is changed to be dynamic 2318int os::sigexitnum_pd() { 2319 return NSIG; 2320} 2321 2322// a counter for each possible signal value 2323static volatile jint pending_signals[NSIG+1] = { 0 }; 2324 2325// Linux(POSIX) specific hand shaking semaphore. 2326static sem_t sig_sem; 2327 2328void os::signal_init_pd() { 2329 // Initialize signal structures 2330 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2331 2332 // Initialize signal semaphore 2333 ::sem_init(&sig_sem, 0, 0); 2334} 2335 2336void os::signal_notify(int sig) { 2337 Atomic::inc(&pending_signals[sig]); 2338 ::sem_post(&sig_sem); 2339} 2340 2341static int check_pending_signals(bool wait) { 2342 Atomic::store(0, &sigint_count); 2343 for (;;) { 2344 for (int i = 0; i < NSIG + 1; i++) { 2345 jint n = pending_signals[i]; 2346 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2347 return i; 2348 } 2349 } 2350 if (!wait) { 2351 return -1; 2352 } 2353 JavaThread *thread = JavaThread::current(); 2354 ThreadBlockInVM tbivm(thread); 2355 2356 bool threadIsSuspended; 2357 do { 2358 thread->set_suspend_equivalent(); 2359 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2360 ::sem_wait(&sig_sem); 2361 2362 // were we externally suspended while we were waiting? 2363 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2364 if (threadIsSuspended) { 2365 // 2366 // The semaphore has been incremented, but while we were waiting 2367 // another thread suspended us. We don't want to continue running 2368 // while suspended because that would surprise the thread that 2369 // suspended us. 2370 // 2371 ::sem_post(&sig_sem); 2372 2373 thread->java_suspend_self(); 2374 } 2375 } while (threadIsSuspended); 2376 } 2377} 2378 2379int os::signal_lookup() { 2380 return check_pending_signals(false); 2381} 2382 2383int os::signal_wait() { 2384 return check_pending_signals(true); 2385} 2386 2387//////////////////////////////////////////////////////////////////////////////// 2388// Virtual Memory 2389 2390int os::vm_page_size() { 2391 // Seems redundant as all get out 2392 assert(os::Linux::page_size() != -1, "must call os::init"); 2393 return os::Linux::page_size(); 2394} 2395 2396// Solaris allocates memory by pages. 2397int os::vm_allocation_granularity() { 2398 assert(os::Linux::page_size() != -1, "must call os::init"); 2399 return os::Linux::page_size(); 2400} 2401 2402// Rationale behind this function: 2403// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2404// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2405// samples for JITted code. Here we create private executable mapping over the code cache 2406// and then we can use standard (well, almost, as mapping can change) way to provide 2407// info for the reporting script by storing timestamp and location of symbol 2408void linux_wrap_code(char* base, size_t size) { 2409 static volatile jint cnt = 0; 2410 2411 if (!UseOprofile) { 2412 return; 2413 } 2414 2415 char buf[PATH_MAX+1]; 2416 int num = Atomic::add(1, &cnt); 2417 2418 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2419 os::get_temp_directory(), os::current_process_id(), num); 2420 unlink(buf); 2421 2422 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2423 2424 if (fd != -1) { 2425 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2426 if (rv != (off_t)-1) { 2427 if (::write(fd, "", 1) == 1) { 2428 mmap(base, size, 2429 PROT_READ|PROT_WRITE|PROT_EXEC, 2430 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2431 } 2432 } 2433 ::close(fd); 2434 unlink(buf); 2435 } 2436} 2437 2438// NOTE: Linux kernel does not really reserve the pages for us. 2439// All it does is to check if there are enough free pages 2440// left at the time of mmap(). This could be a potential 2441// problem. 2442bool os::commit_memory(char* addr, size_t size, bool exec) { 2443 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2444 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2445 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2446 return res != (uintptr_t) MAP_FAILED; 2447} 2448 2449bool os::commit_memory(char* addr, size_t size, size_t alignment_hint, 2450 bool exec) { 2451 return commit_memory(addr, size, exec); 2452} 2453 2454void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) { } 2455 2456void os::free_memory(char *addr, size_t bytes) { 2457 ::mmap(addr, bytes, PROT_READ | PROT_WRITE, 2458 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2459} 2460 2461void os::numa_make_global(char *addr, size_t bytes) { 2462 Linux::numa_interleave_memory(addr, bytes); 2463} 2464 2465void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2466 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2467} 2468 2469bool os::numa_topology_changed() { return false; } 2470 2471size_t os::numa_get_groups_num() { 2472 int max_node = Linux::numa_max_node(); 2473 return max_node > 0 ? max_node + 1 : 1; 2474} 2475 2476int os::numa_get_group_id() { 2477 int cpu_id = Linux::sched_getcpu(); 2478 if (cpu_id != -1) { 2479 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2480 if (lgrp_id != -1) { 2481 return lgrp_id; 2482 } 2483 } 2484 return 0; 2485} 2486 2487size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2488 for (size_t i = 0; i < size; i++) { 2489 ids[i] = i; 2490 } 2491 return size; 2492} 2493 2494bool os::get_page_info(char *start, page_info* info) { 2495 return false; 2496} 2497 2498char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2499 return end; 2500} 2501 2502extern "C" void numa_warn(int number, char *where, ...) { } 2503extern "C" void numa_error(char *where) { } 2504 2505 2506// If we are running with libnuma version > 2, then we should 2507// be trying to use symbols with versions 1.1 2508// If we are running with earlier version, which did not have symbol versions, 2509// we should use the base version. 2510void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2511 void *f = dlvsym(handle, name, "libnuma_1.1"); 2512 if (f == NULL) { 2513 f = dlsym(handle, name); 2514 } 2515 return f; 2516} 2517 2518bool os::Linux::libnuma_init() { 2519 // sched_getcpu() should be in libc. 2520 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2521 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2522 2523 if (sched_getcpu() != -1) { // Does it work? 2524 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2525 if (handle != NULL) { 2526 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2527 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2528 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2529 libnuma_dlsym(handle, "numa_max_node"))); 2530 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2531 libnuma_dlsym(handle, "numa_available"))); 2532 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2533 libnuma_dlsym(handle, "numa_tonode_memory"))); 2534 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2535 libnuma_dlsym(handle, "numa_interleave_memory"))); 2536 2537 2538 if (numa_available() != -1) { 2539 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2540 // Create a cpu -> node mapping 2541 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true); 2542 rebuild_cpu_to_node_map(); 2543 return true; 2544 } 2545 } 2546 } 2547 return false; 2548} 2549 2550// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2551// The table is later used in get_node_by_cpu(). 2552void os::Linux::rebuild_cpu_to_node_map() { 2553 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2554 // in libnuma (possible values are starting from 16, 2555 // and continuing up with every other power of 2, but less 2556 // than the maximum number of CPUs supported by kernel), and 2557 // is a subject to change (in libnuma version 2 the requirements 2558 // are more reasonable) we'll just hardcode the number they use 2559 // in the library. 2560 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2561 2562 size_t cpu_num = os::active_processor_count(); 2563 size_t cpu_map_size = NCPUS / BitsPerCLong; 2564 size_t cpu_map_valid_size = 2565 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2566 2567 cpu_to_node()->clear(); 2568 cpu_to_node()->at_grow(cpu_num - 1); 2569 size_t node_num = numa_get_groups_num(); 2570 2571 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size); 2572 for (size_t i = 0; i < node_num; i++) { 2573 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2574 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2575 if (cpu_map[j] != 0) { 2576 for (size_t k = 0; k < BitsPerCLong; k++) { 2577 if (cpu_map[j] & (1UL << k)) { 2578 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2579 } 2580 } 2581 } 2582 } 2583 } 2584 } 2585 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2586} 2587 2588int os::Linux::get_node_by_cpu(int cpu_id) { 2589 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2590 return cpu_to_node()->at(cpu_id); 2591 } 2592 return -1; 2593} 2594 2595GrowableArray<int>* os::Linux::_cpu_to_node; 2596os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2597os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2598os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2599os::Linux::numa_available_func_t os::Linux::_numa_available; 2600os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2601os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2602unsigned long* os::Linux::_numa_all_nodes; 2603 2604bool os::uncommit_memory(char* addr, size_t size) { 2605 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2606 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2607 return res != (uintptr_t) MAP_FAILED; 2608} 2609 2610// Linux uses a growable mapping for the stack, and if the mapping for 2611// the stack guard pages is not removed when we detach a thread the 2612// stack cannot grow beyond the pages where the stack guard was 2613// mapped. If at some point later in the process the stack expands to 2614// that point, the Linux kernel cannot expand the stack any further 2615// because the guard pages are in the way, and a segfault occurs. 2616// 2617// However, it's essential not to split the stack region by unmapping 2618// a region (leaving a hole) that's already part of the stack mapping, 2619// so if the stack mapping has already grown beyond the guard pages at 2620// the time we create them, we have to truncate the stack mapping. 2621// So, we need to know the extent of the stack mapping when 2622// create_stack_guard_pages() is called. 2623 2624// Find the bounds of the stack mapping. Return true for success. 2625// 2626// We only need this for stacks that are growable: at the time of 2627// writing thread stacks don't use growable mappings (i.e. those 2628// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 2629// only applies to the main thread. 2630static bool 2631get_stack_bounds(uintptr_t *bottom, uintptr_t *top) 2632{ 2633 FILE *f = fopen("/proc/self/maps", "r"); 2634 if (f == NULL) 2635 return false; 2636 2637 while (!feof(f)) { 2638 size_t dummy; 2639 char *str = NULL; 2640 ssize_t len = getline(&str, &dummy, f); 2641 if (len == -1) { 2642 fclose(f); 2643 return false; 2644 } 2645 2646 if (len > 0 && str[len-1] == '\n') { 2647 str[len-1] = 0; 2648 len--; 2649 } 2650 2651 static const char *stack_str = "[stack]"; 2652 if (len > (ssize_t)strlen(stack_str) 2653 && (strcmp(str + len - strlen(stack_str), stack_str) == 0)) { 2654 if (sscanf(str, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) { 2655 uintptr_t sp = (uintptr_t)__builtin_frame_address(0); 2656 if (sp >= *bottom && sp <= *top) { 2657 free(str); 2658 fclose(f); 2659 return true; 2660 } 2661 } 2662 } 2663 free(str); 2664 } 2665 fclose(f); 2666 return false; 2667} 2668 2669// If the (growable) stack mapping already extends beyond the point 2670// where we're going to put our guard pages, truncate the mapping at 2671// that point by munmap()ping it. This ensures that when we later 2672// munmap() the guard pages we don't leave a hole in the stack 2673// mapping. This only affects the main/initial thread, but guard 2674// against future OS changes 2675bool os::create_stack_guard_pages(char* addr, size_t size) { 2676 uintptr_t stack_extent, stack_base; 2677 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2678 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2679 assert(os::Linux::is_initial_thread(), 2680 "growable stack in non-initial thread"); 2681 if (stack_extent < (uintptr_t)addr) 2682 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent); 2683 } 2684 2685 return os::commit_memory(addr, size); 2686} 2687 2688// If this is a growable mapping, remove the guard pages entirely by 2689// munmap()ping them. If not, just call uncommit_memory(). This only 2690// affects the main/initial thread, but guard against future OS changes 2691bool os::remove_stack_guard_pages(char* addr, size_t size) { 2692 uintptr_t stack_extent, stack_base; 2693 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2694 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2695 assert(os::Linux::is_initial_thread(), 2696 "growable stack in non-initial thread"); 2697 2698 return ::munmap(addr, size) == 0; 2699 } 2700 2701 return os::uncommit_memory(addr, size); 2702} 2703 2704static address _highest_vm_reserved_address = NULL; 2705 2706// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2707// at 'requested_addr'. If there are existing memory mappings at the same 2708// location, however, they will be overwritten. If 'fixed' is false, 2709// 'requested_addr' is only treated as a hint, the return value may or 2710// may not start from the requested address. Unlike Linux mmap(), this 2711// function returns NULL to indicate failure. 2712static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2713 char * addr; 2714 int flags; 2715 2716 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2717 if (fixed) { 2718 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2719 flags |= MAP_FIXED; 2720 } 2721 2722 // Map uncommitted pages PROT_READ and PROT_WRITE, change access 2723 // to PROT_EXEC if executable when we commit the page. 2724 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE, 2725 flags, -1, 0); 2726 2727 if (addr != MAP_FAILED) { 2728 // anon_mmap() should only get called during VM initialization, 2729 // don't need lock (actually we can skip locking even it can be called 2730 // from multiple threads, because _highest_vm_reserved_address is just a 2731 // hint about the upper limit of non-stack memory regions.) 2732 if ((address)addr + bytes > _highest_vm_reserved_address) { 2733 _highest_vm_reserved_address = (address)addr + bytes; 2734 } 2735 } 2736 2737 return addr == MAP_FAILED ? NULL : addr; 2738} 2739 2740// Don't update _highest_vm_reserved_address, because there might be memory 2741// regions above addr + size. If so, releasing a memory region only creates 2742// a hole in the address space, it doesn't help prevent heap-stack collision. 2743// 2744static int anon_munmap(char * addr, size_t size) { 2745 return ::munmap(addr, size) == 0; 2746} 2747 2748char* os::reserve_memory(size_t bytes, char* requested_addr, 2749 size_t alignment_hint) { 2750 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 2751} 2752 2753bool os::release_memory(char* addr, size_t size) { 2754 return anon_munmap(addr, size); 2755} 2756 2757static address highest_vm_reserved_address() { 2758 return _highest_vm_reserved_address; 2759} 2760 2761static bool linux_mprotect(char* addr, size_t size, int prot) { 2762 // Linux wants the mprotect address argument to be page aligned. 2763 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 2764 2765 // According to SUSv3, mprotect() should only be used with mappings 2766 // established by mmap(), and mmap() always maps whole pages. Unaligned 2767 // 'addr' likely indicates problem in the VM (e.g. trying to change 2768 // protection of malloc'ed or statically allocated memory). Check the 2769 // caller if you hit this assert. 2770 assert(addr == bottom, "sanity check"); 2771 2772 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 2773 return ::mprotect(bottom, size, prot) == 0; 2774} 2775 2776// Set protections specified 2777bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 2778 bool is_committed) { 2779 unsigned int p = 0; 2780 switch (prot) { 2781 case MEM_PROT_NONE: p = PROT_NONE; break; 2782 case MEM_PROT_READ: p = PROT_READ; break; 2783 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 2784 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 2785 default: 2786 ShouldNotReachHere(); 2787 } 2788 // is_committed is unused. 2789 return linux_mprotect(addr, bytes, p); 2790} 2791 2792bool os::guard_memory(char* addr, size_t size) { 2793 return linux_mprotect(addr, size, PROT_NONE); 2794} 2795 2796bool os::unguard_memory(char* addr, size_t size) { 2797 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 2798} 2799 2800// Large page support 2801 2802static size_t _large_page_size = 0; 2803 2804bool os::large_page_init() { 2805 if (!UseLargePages) return false; 2806 2807 if (LargePageSizeInBytes) { 2808 _large_page_size = LargePageSizeInBytes; 2809 } else { 2810 // large_page_size on Linux is used to round up heap size. x86 uses either 2811 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 2812 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 2813 // page as large as 256M. 2814 // 2815 // Here we try to figure out page size by parsing /proc/meminfo and looking 2816 // for a line with the following format: 2817 // Hugepagesize: 2048 kB 2818 // 2819 // If we can't determine the value (e.g. /proc is not mounted, or the text 2820 // format has been changed), we'll use the largest page size supported by 2821 // the processor. 2822 2823#ifndef ZERO 2824 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 2825 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 2826#endif // ZERO 2827 2828 FILE *fp = fopen("/proc/meminfo", "r"); 2829 if (fp) { 2830 while (!feof(fp)) { 2831 int x = 0; 2832 char buf[16]; 2833 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 2834 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 2835 _large_page_size = x * K; 2836 break; 2837 } 2838 } else { 2839 // skip to next line 2840 for (;;) { 2841 int ch = fgetc(fp); 2842 if (ch == EOF || ch == (int)'\n') break; 2843 } 2844 } 2845 } 2846 fclose(fp); 2847 } 2848 } 2849 2850 const size_t default_page_size = (size_t)Linux::page_size(); 2851 if (_large_page_size > default_page_size) { 2852 _page_sizes[0] = _large_page_size; 2853 _page_sizes[1] = default_page_size; 2854 _page_sizes[2] = 0; 2855 } 2856 2857 // Large page support is available on 2.6 or newer kernel, some vendors 2858 // (e.g. Redhat) have backported it to their 2.4 based distributions. 2859 // We optimistically assume the support is available. If later it turns out 2860 // not true, VM will automatically switch to use regular page size. 2861 return true; 2862} 2863 2864#ifndef SHM_HUGETLB 2865#define SHM_HUGETLB 04000 2866#endif 2867 2868char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) { 2869 // "exec" is passed in but not used. Creating the shared image for 2870 // the code cache doesn't have an SHM_X executable permission to check. 2871 assert(UseLargePages, "only for large pages"); 2872 2873 key_t key = IPC_PRIVATE; 2874 char *addr; 2875 2876 bool warn_on_failure = UseLargePages && 2877 (!FLAG_IS_DEFAULT(UseLargePages) || 2878 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 2879 ); 2880 char msg[128]; 2881 2882 // Create a large shared memory region to attach to based on size. 2883 // Currently, size is the total size of the heap 2884 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 2885 if (shmid == -1) { 2886 // Possible reasons for shmget failure: 2887 // 1. shmmax is too small for Java heap. 2888 // > check shmmax value: cat /proc/sys/kernel/shmmax 2889 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 2890 // 2. not enough large page memory. 2891 // > check available large pages: cat /proc/meminfo 2892 // > increase amount of large pages: 2893 // echo new_value > /proc/sys/vm/nr_hugepages 2894 // Note 1: different Linux may use different name for this property, 2895 // e.g. on Redhat AS-3 it is "hugetlb_pool". 2896 // Note 2: it's possible there's enough physical memory available but 2897 // they are so fragmented after a long run that they can't 2898 // coalesce into large pages. Try to reserve large pages when 2899 // the system is still "fresh". 2900 if (warn_on_failure) { 2901 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 2902 warning(msg); 2903 } 2904 return NULL; 2905 } 2906 2907 // attach to the region 2908 addr = (char*)shmat(shmid, req_addr, 0); 2909 int err = errno; 2910 2911 // Remove shmid. If shmat() is successful, the actual shared memory segment 2912 // will be deleted when it's detached by shmdt() or when the process 2913 // terminates. If shmat() is not successful this will remove the shared 2914 // segment immediately. 2915 shmctl(shmid, IPC_RMID, NULL); 2916 2917 if ((intptr_t)addr == -1) { 2918 if (warn_on_failure) { 2919 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 2920 warning(msg); 2921 } 2922 return NULL; 2923 } 2924 2925 return addr; 2926} 2927 2928bool os::release_memory_special(char* base, size_t bytes) { 2929 // detaching the SHM segment will also delete it, see reserve_memory_special() 2930 int rslt = shmdt(base); 2931 return rslt == 0; 2932} 2933 2934size_t os::large_page_size() { 2935 return _large_page_size; 2936} 2937 2938// Linux does not support anonymous mmap with large page memory. The only way 2939// to reserve large page memory without file backing is through SysV shared 2940// memory API. The entire memory region is committed and pinned upfront. 2941// Hopefully this will change in the future... 2942bool os::can_commit_large_page_memory() { 2943 return false; 2944} 2945 2946bool os::can_execute_large_page_memory() { 2947 return false; 2948} 2949 2950// Reserve memory at an arbitrary address, only if that area is 2951// available (and not reserved for something else). 2952 2953char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 2954 const int max_tries = 10; 2955 char* base[max_tries]; 2956 size_t size[max_tries]; 2957 const size_t gap = 0x000000; 2958 2959 // Assert only that the size is a multiple of the page size, since 2960 // that's all that mmap requires, and since that's all we really know 2961 // about at this low abstraction level. If we need higher alignment, 2962 // we can either pass an alignment to this method or verify alignment 2963 // in one of the methods further up the call chain. See bug 5044738. 2964 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 2965 2966 // Repeatedly allocate blocks until the block is allocated at the 2967 // right spot. Give up after max_tries. Note that reserve_memory() will 2968 // automatically update _highest_vm_reserved_address if the call is 2969 // successful. The variable tracks the highest memory address every reserved 2970 // by JVM. It is used to detect heap-stack collision if running with 2971 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 2972 // space than needed, it could confuse the collision detecting code. To 2973 // solve the problem, save current _highest_vm_reserved_address and 2974 // calculate the correct value before return. 2975 address old_highest = _highest_vm_reserved_address; 2976 2977 // Linux mmap allows caller to pass an address as hint; give it a try first, 2978 // if kernel honors the hint then we can return immediately. 2979 char * addr = anon_mmap(requested_addr, bytes, false); 2980 if (addr == requested_addr) { 2981 return requested_addr; 2982 } 2983 2984 if (addr != NULL) { 2985 // mmap() is successful but it fails to reserve at the requested address 2986 anon_munmap(addr, bytes); 2987 } 2988 2989 int i; 2990 for (i = 0; i < max_tries; ++i) { 2991 base[i] = reserve_memory(bytes); 2992 2993 if (base[i] != NULL) { 2994 // Is this the block we wanted? 2995 if (base[i] == requested_addr) { 2996 size[i] = bytes; 2997 break; 2998 } 2999 3000 // Does this overlap the block we wanted? Give back the overlapped 3001 // parts and try again. 3002 3003 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3004 if (top_overlap >= 0 && top_overlap < bytes) { 3005 unmap_memory(base[i], top_overlap); 3006 base[i] += top_overlap; 3007 size[i] = bytes - top_overlap; 3008 } else { 3009 size_t bottom_overlap = base[i] + bytes - requested_addr; 3010 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3011 unmap_memory(requested_addr, bottom_overlap); 3012 size[i] = bytes - bottom_overlap; 3013 } else { 3014 size[i] = bytes; 3015 } 3016 } 3017 } 3018 } 3019 3020 // Give back the unused reserved pieces. 3021 3022 for (int j = 0; j < i; ++j) { 3023 if (base[j] != NULL) { 3024 unmap_memory(base[j], size[j]); 3025 } 3026 } 3027 3028 if (i < max_tries) { 3029 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3030 return requested_addr; 3031 } else { 3032 _highest_vm_reserved_address = old_highest; 3033 return NULL; 3034 } 3035} 3036 3037size_t os::read(int fd, void *buf, unsigned int nBytes) { 3038 return ::read(fd, buf, nBytes); 3039} 3040 3041// TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 3042// Solaris uses poll(), linux uses park(). 3043// Poll() is likely a better choice, assuming that Thread.interrupt() 3044// generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 3045// SIGSEGV, see 4355769. 3046 3047const int NANOSECS_PER_MILLISECS = 1000000; 3048 3049int os::sleep(Thread* thread, jlong millis, bool interruptible) { 3050 assert(thread == Thread::current(), "thread consistency check"); 3051 3052 ParkEvent * const slp = thread->_SleepEvent ; 3053 slp->reset() ; 3054 OrderAccess::fence() ; 3055 3056 if (interruptible) { 3057 jlong prevtime = javaTimeNanos(); 3058 3059 for (;;) { 3060 if (os::is_interrupted(thread, true)) { 3061 return OS_INTRPT; 3062 } 3063 3064 jlong newtime = javaTimeNanos(); 3065 3066 if (newtime - prevtime < 0) { 3067 // time moving backwards, should only happen if no monotonic clock 3068 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3069 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3070 } else { 3071 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 3072 } 3073 3074 if(millis <= 0) { 3075 return OS_OK; 3076 } 3077 3078 prevtime = newtime; 3079 3080 { 3081 assert(thread->is_Java_thread(), "sanity check"); 3082 JavaThread *jt = (JavaThread *) thread; 3083 ThreadBlockInVM tbivm(jt); 3084 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 3085 3086 jt->set_suspend_equivalent(); 3087 // cleared by handle_special_suspend_equivalent_condition() or 3088 // java_suspend_self() via check_and_wait_while_suspended() 3089 3090 slp->park(millis); 3091 3092 // were we externally suspended while we were waiting? 3093 jt->check_and_wait_while_suspended(); 3094 } 3095 } 3096 } else { 3097 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 3098 jlong prevtime = javaTimeNanos(); 3099 3100 for (;;) { 3101 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 3102 // the 1st iteration ... 3103 jlong newtime = javaTimeNanos(); 3104 3105 if (newtime - prevtime < 0) { 3106 // time moving backwards, should only happen if no monotonic clock 3107 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3108 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3109 } else { 3110 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS; 3111 } 3112 3113 if(millis <= 0) break ; 3114 3115 prevtime = newtime; 3116 slp->park(millis); 3117 } 3118 return OS_OK ; 3119 } 3120} 3121 3122int os::naked_sleep() { 3123 // %% make the sleep time an integer flag. for now use 1 millisec. 3124 return os::sleep(Thread::current(), 1, false); 3125} 3126 3127// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3128void os::infinite_sleep() { 3129 while (true) { // sleep forever ... 3130 ::sleep(100); // ... 100 seconds at a time 3131 } 3132} 3133 3134// Used to convert frequent JVM_Yield() to nops 3135bool os::dont_yield() { 3136 return DontYieldALot; 3137} 3138 3139void os::yield() { 3140 sched_yield(); 3141} 3142 3143os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3144 3145void os::yield_all(int attempts) { 3146 // Yields to all threads, including threads with lower priorities 3147 // Threads on Linux are all with same priority. The Solaris style 3148 // os::yield_all() with nanosleep(1ms) is not necessary. 3149 sched_yield(); 3150} 3151 3152// Called from the tight loops to possibly influence time-sharing heuristics 3153void os::loop_breaker(int attempts) { 3154 os::yield_all(attempts); 3155} 3156 3157//////////////////////////////////////////////////////////////////////////////// 3158// thread priority support 3159 3160// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3161// only supports dynamic priority, static priority must be zero. For real-time 3162// applications, Linux supports SCHED_RR which allows static priority (1-99). 3163// However, for large multi-threaded applications, SCHED_RR is not only slower 3164// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3165// of 5 runs - Sep 2005). 3166// 3167// The following code actually changes the niceness of kernel-thread/LWP. It 3168// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3169// not the entire user process, and user level threads are 1:1 mapped to kernel 3170// threads. It has always been the case, but could change in the future. For 3171// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3172// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3173 3174int os::java_to_os_priority[MaxPriority + 1] = { 3175 19, // 0 Entry should never be used 3176 3177 4, // 1 MinPriority 3178 3, // 2 3179 2, // 3 3180 3181 1, // 4 3182 0, // 5 NormPriority 3183 -1, // 6 3184 3185 -2, // 7 3186 -3, // 8 3187 -4, // 9 NearMaxPriority 3188 3189 -5 // 10 MaxPriority 3190}; 3191 3192static int prio_init() { 3193 if (ThreadPriorityPolicy == 1) { 3194 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3195 // if effective uid is not root. Perhaps, a more elegant way of doing 3196 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3197 if (geteuid() != 0) { 3198 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3199 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3200 } 3201 ThreadPriorityPolicy = 0; 3202 } 3203 } 3204 return 0; 3205} 3206 3207OSReturn os::set_native_priority(Thread* thread, int newpri) { 3208 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3209 3210 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3211 return (ret == 0) ? OS_OK : OS_ERR; 3212} 3213 3214OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3215 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3216 *priority_ptr = java_to_os_priority[NormPriority]; 3217 return OS_OK; 3218 } 3219 3220 errno = 0; 3221 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3222 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3223} 3224 3225// Hint to the underlying OS that a task switch would not be good. 3226// Void return because it's a hint and can fail. 3227void os::hint_no_preempt() {} 3228 3229//////////////////////////////////////////////////////////////////////////////// 3230// suspend/resume support 3231 3232// the low-level signal-based suspend/resume support is a remnant from the 3233// old VM-suspension that used to be for java-suspension, safepoints etc, 3234// within hotspot. Now there is a single use-case for this: 3235// - calling get_thread_pc() on the VMThread by the flat-profiler task 3236// that runs in the watcher thread. 3237// The remaining code is greatly simplified from the more general suspension 3238// code that used to be used. 3239// 3240// The protocol is quite simple: 3241// - suspend: 3242// - sends a signal to the target thread 3243// - polls the suspend state of the osthread using a yield loop 3244// - target thread signal handler (SR_handler) sets suspend state 3245// and blocks in sigsuspend until continued 3246// - resume: 3247// - sets target osthread state to continue 3248// - sends signal to end the sigsuspend loop in the SR_handler 3249// 3250// Note that the SR_lock plays no role in this suspend/resume protocol. 3251// 3252 3253static void resume_clear_context(OSThread *osthread) { 3254 osthread->set_ucontext(NULL); 3255 osthread->set_siginfo(NULL); 3256 3257 // notify the suspend action is completed, we have now resumed 3258 osthread->sr.clear_suspended(); 3259} 3260 3261static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3262 osthread->set_ucontext(context); 3263 osthread->set_siginfo(siginfo); 3264} 3265 3266// 3267// Handler function invoked when a thread's execution is suspended or 3268// resumed. We have to be careful that only async-safe functions are 3269// called here (Note: most pthread functions are not async safe and 3270// should be avoided.) 3271// 3272// Note: sigwait() is a more natural fit than sigsuspend() from an 3273// interface point of view, but sigwait() prevents the signal hander 3274// from being run. libpthread would get very confused by not having 3275// its signal handlers run and prevents sigwait()'s use with the 3276// mutex granting granting signal. 3277// 3278// Currently only ever called on the VMThread 3279// 3280static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3281 // Save and restore errno to avoid confusing native code with EINTR 3282 // after sigsuspend. 3283 int old_errno = errno; 3284 3285 Thread* thread = Thread::current(); 3286 OSThread* osthread = thread->osthread(); 3287 assert(thread->is_VM_thread(), "Must be VMThread"); 3288 // read current suspend action 3289 int action = osthread->sr.suspend_action(); 3290 if (action == SR_SUSPEND) { 3291 suspend_save_context(osthread, siginfo, context); 3292 3293 // Notify the suspend action is about to be completed. do_suspend() 3294 // waits until SR_SUSPENDED is set and then returns. We will wait 3295 // here for a resume signal and that completes the suspend-other 3296 // action. do_suspend/do_resume is always called as a pair from 3297 // the same thread - so there are no races 3298 3299 // notify the caller 3300 osthread->sr.set_suspended(); 3301 3302 sigset_t suspend_set; // signals for sigsuspend() 3303 3304 // get current set of blocked signals and unblock resume signal 3305 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3306 sigdelset(&suspend_set, SR_signum); 3307 3308 // wait here until we are resumed 3309 do { 3310 sigsuspend(&suspend_set); 3311 // ignore all returns until we get a resume signal 3312 } while (osthread->sr.suspend_action() != SR_CONTINUE); 3313 3314 resume_clear_context(osthread); 3315 3316 } else { 3317 assert(action == SR_CONTINUE, "unexpected sr action"); 3318 // nothing special to do - just leave the handler 3319 } 3320 3321 errno = old_errno; 3322} 3323 3324 3325static int SR_initialize() { 3326 struct sigaction act; 3327 char *s; 3328 /* Get signal number to use for suspend/resume */ 3329 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3330 int sig = ::strtol(s, 0, 10); 3331 if (sig > 0 || sig < _NSIG) { 3332 SR_signum = sig; 3333 } 3334 } 3335 3336 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3337 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3338 3339 sigemptyset(&SR_sigset); 3340 sigaddset(&SR_sigset, SR_signum); 3341 3342 /* Set up signal handler for suspend/resume */ 3343 act.sa_flags = SA_RESTART|SA_SIGINFO; 3344 act.sa_handler = (void (*)(int)) SR_handler; 3345 3346 // SR_signum is blocked by default. 3347 // 4528190 - We also need to block pthread restart signal (32 on all 3348 // supported Linux platforms). Note that LinuxThreads need to block 3349 // this signal for all threads to work properly. So we don't have 3350 // to use hard-coded signal number when setting up the mask. 3351 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 3352 3353 if (sigaction(SR_signum, &act, 0) == -1) { 3354 return -1; 3355 } 3356 3357 // Save signal flag 3358 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 3359 return 0; 3360} 3361 3362static int SR_finalize() { 3363 return 0; 3364} 3365 3366 3367// returns true on success and false on error - really an error is fatal 3368// but this seems the normal response to library errors 3369static bool do_suspend(OSThread* osthread) { 3370 // mark as suspended and send signal 3371 osthread->sr.set_suspend_action(SR_SUSPEND); 3372 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3373 assert_status(status == 0, status, "pthread_kill"); 3374 3375 // check status and wait until notified of suspension 3376 if (status == 0) { 3377 for (int i = 0; !osthread->sr.is_suspended(); i++) { 3378 os::yield_all(i); 3379 } 3380 osthread->sr.set_suspend_action(SR_NONE); 3381 return true; 3382 } 3383 else { 3384 osthread->sr.set_suspend_action(SR_NONE); 3385 return false; 3386 } 3387} 3388 3389static void do_resume(OSThread* osthread) { 3390 assert(osthread->sr.is_suspended(), "thread should be suspended"); 3391 osthread->sr.set_suspend_action(SR_CONTINUE); 3392 3393 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3394 assert_status(status == 0, status, "pthread_kill"); 3395 // check status and wait unit notified of resumption 3396 if (status == 0) { 3397 for (int i = 0; osthread->sr.is_suspended(); i++) { 3398 os::yield_all(i); 3399 } 3400 } 3401 osthread->sr.set_suspend_action(SR_NONE); 3402} 3403 3404//////////////////////////////////////////////////////////////////////////////// 3405// interrupt support 3406 3407void os::interrupt(Thread* thread) { 3408 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3409 "possibility of dangling Thread pointer"); 3410 3411 OSThread* osthread = thread->osthread(); 3412 3413 if (!osthread->interrupted()) { 3414 osthread->set_interrupted(true); 3415 // More than one thread can get here with the same value of osthread, 3416 // resulting in multiple notifications. We do, however, want the store 3417 // to interrupted() to be visible to other threads before we execute unpark(). 3418 OrderAccess::fence(); 3419 ParkEvent * const slp = thread->_SleepEvent ; 3420 if (slp != NULL) slp->unpark() ; 3421 } 3422 3423 // For JSR166. Unpark even if interrupt status already was set 3424 if (thread->is_Java_thread()) 3425 ((JavaThread*)thread)->parker()->unpark(); 3426 3427 ParkEvent * ev = thread->_ParkEvent ; 3428 if (ev != NULL) ev->unpark() ; 3429 3430} 3431 3432bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 3433 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3434 "possibility of dangling Thread pointer"); 3435 3436 OSThread* osthread = thread->osthread(); 3437 3438 bool interrupted = osthread->interrupted(); 3439 3440 if (interrupted && clear_interrupted) { 3441 osthread->set_interrupted(false); 3442 // consider thread->_SleepEvent->reset() ... optional optimization 3443 } 3444 3445 return interrupted; 3446} 3447 3448/////////////////////////////////////////////////////////////////////////////////// 3449// signal handling (except suspend/resume) 3450 3451// This routine may be used by user applications as a "hook" to catch signals. 3452// The user-defined signal handler must pass unrecognized signals to this 3453// routine, and if it returns true (non-zero), then the signal handler must 3454// return immediately. If the flag "abort_if_unrecognized" is true, then this 3455// routine will never retun false (zero), but instead will execute a VM panic 3456// routine kill the process. 3457// 3458// If this routine returns false, it is OK to call it again. This allows 3459// the user-defined signal handler to perform checks either before or after 3460// the VM performs its own checks. Naturally, the user code would be making 3461// a serious error if it tried to handle an exception (such as a null check 3462// or breakpoint) that the VM was generating for its own correct operation. 3463// 3464// This routine may recognize any of the following kinds of signals: 3465// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 3466// It should be consulted by handlers for any of those signals. 3467// 3468// The caller of this routine must pass in the three arguments supplied 3469// to the function referred to in the "sa_sigaction" (not the "sa_handler") 3470// field of the structure passed to sigaction(). This routine assumes that 3471// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 3472// 3473// Note that the VM will print warnings if it detects conflicting signal 3474// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 3475// 3476extern "C" int 3477JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 3478 void* ucontext, int abort_if_unrecognized); 3479 3480void signalHandler(int sig, siginfo_t* info, void* uc) { 3481 assert(info != NULL && uc != NULL, "it must be old kernel"); 3482 JVM_handle_linux_signal(sig, info, uc, true); 3483} 3484 3485 3486// This boolean allows users to forward their own non-matching signals 3487// to JVM_handle_linux_signal, harmlessly. 3488bool os::Linux::signal_handlers_are_installed = false; 3489 3490// For signal-chaining 3491struct sigaction os::Linux::sigact[MAXSIGNUM]; 3492unsigned int os::Linux::sigs = 0; 3493bool os::Linux::libjsig_is_loaded = false; 3494typedef struct sigaction *(*get_signal_t)(int); 3495get_signal_t os::Linux::get_signal_action = NULL; 3496 3497struct sigaction* os::Linux::get_chained_signal_action(int sig) { 3498 struct sigaction *actp = NULL; 3499 3500 if (libjsig_is_loaded) { 3501 // Retrieve the old signal handler from libjsig 3502 actp = (*get_signal_action)(sig); 3503 } 3504 if (actp == NULL) { 3505 // Retrieve the preinstalled signal handler from jvm 3506 actp = get_preinstalled_handler(sig); 3507 } 3508 3509 return actp; 3510} 3511 3512static bool call_chained_handler(struct sigaction *actp, int sig, 3513 siginfo_t *siginfo, void *context) { 3514 // Call the old signal handler 3515 if (actp->sa_handler == SIG_DFL) { 3516 // It's more reasonable to let jvm treat it as an unexpected exception 3517 // instead of taking the default action. 3518 return false; 3519 } else if (actp->sa_handler != SIG_IGN) { 3520 if ((actp->sa_flags & SA_NODEFER) == 0) { 3521 // automaticlly block the signal 3522 sigaddset(&(actp->sa_mask), sig); 3523 } 3524 3525 sa_handler_t hand; 3526 sa_sigaction_t sa; 3527 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 3528 // retrieve the chained handler 3529 if (siginfo_flag_set) { 3530 sa = actp->sa_sigaction; 3531 } else { 3532 hand = actp->sa_handler; 3533 } 3534 3535 if ((actp->sa_flags & SA_RESETHAND) != 0) { 3536 actp->sa_handler = SIG_DFL; 3537 } 3538 3539 // try to honor the signal mask 3540 sigset_t oset; 3541 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 3542 3543 // call into the chained handler 3544 if (siginfo_flag_set) { 3545 (*sa)(sig, siginfo, context); 3546 } else { 3547 (*hand)(sig); 3548 } 3549 3550 // restore the signal mask 3551 pthread_sigmask(SIG_SETMASK, &oset, 0); 3552 } 3553 // Tell jvm's signal handler the signal is taken care of. 3554 return true; 3555} 3556 3557bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 3558 bool chained = false; 3559 // signal-chaining 3560 if (UseSignalChaining) { 3561 struct sigaction *actp = get_chained_signal_action(sig); 3562 if (actp != NULL) { 3563 chained = call_chained_handler(actp, sig, siginfo, context); 3564 } 3565 } 3566 return chained; 3567} 3568 3569struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 3570 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 3571 return &sigact[sig]; 3572 } 3573 return NULL; 3574} 3575 3576void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 3577 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3578 sigact[sig] = oldAct; 3579 sigs |= (unsigned int)1 << sig; 3580} 3581 3582// for diagnostic 3583int os::Linux::sigflags[MAXSIGNUM]; 3584 3585int os::Linux::get_our_sigflags(int sig) { 3586 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3587 return sigflags[sig]; 3588} 3589 3590void os::Linux::set_our_sigflags(int sig, int flags) { 3591 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3592 sigflags[sig] = flags; 3593} 3594 3595void os::Linux::set_signal_handler(int sig, bool set_installed) { 3596 // Check for overwrite. 3597 struct sigaction oldAct; 3598 sigaction(sig, (struct sigaction*)NULL, &oldAct); 3599 3600 void* oldhand = oldAct.sa_sigaction 3601 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3602 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3603 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 3604 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 3605 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 3606 if (AllowUserSignalHandlers || !set_installed) { 3607 // Do not overwrite; user takes responsibility to forward to us. 3608 return; 3609 } else if (UseSignalChaining) { 3610 // save the old handler in jvm 3611 save_preinstalled_handler(sig, oldAct); 3612 // libjsig also interposes the sigaction() call below and saves the 3613 // old sigaction on it own. 3614 } else { 3615 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 3616 "%#lx for signal %d.", (long)oldhand, sig)); 3617 } 3618 } 3619 3620 struct sigaction sigAct; 3621 sigfillset(&(sigAct.sa_mask)); 3622 sigAct.sa_handler = SIG_DFL; 3623 if (!set_installed) { 3624 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3625 } else { 3626 sigAct.sa_sigaction = signalHandler; 3627 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 3628 } 3629 // Save flags, which are set by ours 3630 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 3631 sigflags[sig] = sigAct.sa_flags; 3632 3633 int ret = sigaction(sig, &sigAct, &oldAct); 3634 assert(ret == 0, "check"); 3635 3636 void* oldhand2 = oldAct.sa_sigaction 3637 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 3638 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 3639 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 3640} 3641 3642// install signal handlers for signals that HotSpot needs to 3643// handle in order to support Java-level exception handling. 3644 3645void os::Linux::install_signal_handlers() { 3646 if (!signal_handlers_are_installed) { 3647 signal_handlers_are_installed = true; 3648 3649 // signal-chaining 3650 typedef void (*signal_setting_t)(); 3651 signal_setting_t begin_signal_setting = NULL; 3652 signal_setting_t end_signal_setting = NULL; 3653 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3654 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 3655 if (begin_signal_setting != NULL) { 3656 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 3657 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 3658 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 3659 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 3660 libjsig_is_loaded = true; 3661 assert(UseSignalChaining, "should enable signal-chaining"); 3662 } 3663 if (libjsig_is_loaded) { 3664 // Tell libjsig jvm is setting signal handlers 3665 (*begin_signal_setting)(); 3666 } 3667 3668 set_signal_handler(SIGSEGV, true); 3669 set_signal_handler(SIGPIPE, true); 3670 set_signal_handler(SIGBUS, true); 3671 set_signal_handler(SIGILL, true); 3672 set_signal_handler(SIGFPE, true); 3673 set_signal_handler(SIGXFSZ, true); 3674 3675 if (libjsig_is_loaded) { 3676 // Tell libjsig jvm finishes setting signal handlers 3677 (*end_signal_setting)(); 3678 } 3679 3680 // We don't activate signal checker if libjsig is in place, we trust ourselves 3681 // and if UserSignalHandler is installed all bets are off 3682 if (CheckJNICalls) { 3683 if (libjsig_is_loaded) { 3684 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 3685 check_signals = false; 3686 } 3687 if (AllowUserSignalHandlers) { 3688 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 3689 check_signals = false; 3690 } 3691 } 3692 } 3693} 3694 3695// This is the fastest way to get thread cpu time on Linux. 3696// Returns cpu time (user+sys) for any thread, not only for current. 3697// POSIX compliant clocks are implemented in the kernels 2.6.16+. 3698// It might work on 2.6.10+ with a special kernel/glibc patch. 3699// For reference, please, see IEEE Std 1003.1-2004: 3700// http://www.unix.org/single_unix_specification 3701 3702jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 3703 struct timespec tp; 3704 int rc = os::Linux::clock_gettime(clockid, &tp); 3705 assert(rc == 0, "clock_gettime is expected to return 0 code"); 3706 3707 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec; 3708} 3709 3710///// 3711// glibc on Linux platform uses non-documented flag 3712// to indicate, that some special sort of signal 3713// trampoline is used. 3714// We will never set this flag, and we should 3715// ignore this flag in our diagnostic 3716#ifdef SIGNIFICANT_SIGNAL_MASK 3717#undef SIGNIFICANT_SIGNAL_MASK 3718#endif 3719#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 3720 3721static const char* get_signal_handler_name(address handler, 3722 char* buf, int buflen) { 3723 int offset; 3724 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 3725 if (found) { 3726 // skip directory names 3727 const char *p1, *p2; 3728 p1 = buf; 3729 size_t len = strlen(os::file_separator()); 3730 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 3731 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 3732 } else { 3733 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 3734 } 3735 return buf; 3736} 3737 3738static void print_signal_handler(outputStream* st, int sig, 3739 char* buf, size_t buflen) { 3740 struct sigaction sa; 3741 3742 sigaction(sig, NULL, &sa); 3743 3744 // See comment for SIGNIFICANT_SIGNAL_MASK define 3745 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3746 3747 st->print("%s: ", os::exception_name(sig, buf, buflen)); 3748 3749 address handler = (sa.sa_flags & SA_SIGINFO) 3750 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 3751 : CAST_FROM_FN_PTR(address, sa.sa_handler); 3752 3753 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 3754 st->print("SIG_DFL"); 3755 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 3756 st->print("SIG_IGN"); 3757 } else { 3758 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 3759 } 3760 3761 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 3762 3763 address rh = VMError::get_resetted_sighandler(sig); 3764 // May be, handler was resetted by VMError? 3765 if(rh != NULL) { 3766 handler = rh; 3767 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 3768 } 3769 3770 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 3771 3772 // Check: is it our handler? 3773 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 3774 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 3775 // It is our signal handler 3776 // check for flags, reset system-used one! 3777 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 3778 st->print( 3779 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 3780 os::Linux::get_our_sigflags(sig)); 3781 } 3782 } 3783 st->cr(); 3784} 3785 3786 3787#define DO_SIGNAL_CHECK(sig) \ 3788 if (!sigismember(&check_signal_done, sig)) \ 3789 os::Linux::check_signal_handler(sig) 3790 3791// This method is a periodic task to check for misbehaving JNI applications 3792// under CheckJNI, we can add any periodic checks here 3793 3794void os::run_periodic_checks() { 3795 3796 if (check_signals == false) return; 3797 3798 // SEGV and BUS if overridden could potentially prevent 3799 // generation of hs*.log in the event of a crash, debugging 3800 // such a case can be very challenging, so we absolutely 3801 // check the following for a good measure: 3802 DO_SIGNAL_CHECK(SIGSEGV); 3803 DO_SIGNAL_CHECK(SIGILL); 3804 DO_SIGNAL_CHECK(SIGFPE); 3805 DO_SIGNAL_CHECK(SIGBUS); 3806 DO_SIGNAL_CHECK(SIGPIPE); 3807 DO_SIGNAL_CHECK(SIGXFSZ); 3808 3809 3810 // ReduceSignalUsage allows the user to override these handlers 3811 // see comments at the very top and jvm_solaris.h 3812 if (!ReduceSignalUsage) { 3813 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 3814 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 3815 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 3816 DO_SIGNAL_CHECK(BREAK_SIGNAL); 3817 } 3818 3819 DO_SIGNAL_CHECK(SR_signum); 3820 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 3821} 3822 3823typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 3824 3825static os_sigaction_t os_sigaction = NULL; 3826 3827void os::Linux::check_signal_handler(int sig) { 3828 char buf[O_BUFLEN]; 3829 address jvmHandler = NULL; 3830 3831 3832 struct sigaction act; 3833 if (os_sigaction == NULL) { 3834 // only trust the default sigaction, in case it has been interposed 3835 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 3836 if (os_sigaction == NULL) return; 3837 } 3838 3839 os_sigaction(sig, (struct sigaction*)NULL, &act); 3840 3841 3842 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 3843 3844 address thisHandler = (act.sa_flags & SA_SIGINFO) 3845 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 3846 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 3847 3848 3849 switch(sig) { 3850 case SIGSEGV: 3851 case SIGBUS: 3852 case SIGFPE: 3853 case SIGPIPE: 3854 case SIGILL: 3855 case SIGXFSZ: 3856 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 3857 break; 3858 3859 case SHUTDOWN1_SIGNAL: 3860 case SHUTDOWN2_SIGNAL: 3861 case SHUTDOWN3_SIGNAL: 3862 case BREAK_SIGNAL: 3863 jvmHandler = (address)user_handler(); 3864 break; 3865 3866 case INTERRUPT_SIGNAL: 3867 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 3868 break; 3869 3870 default: 3871 if (sig == SR_signum) { 3872 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 3873 } else { 3874 return; 3875 } 3876 break; 3877 } 3878 3879 if (thisHandler != jvmHandler) { 3880 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 3881 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 3882 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 3883 // No need to check this sig any longer 3884 sigaddset(&check_signal_done, sig); 3885 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 3886 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 3887 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 3888 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 3889 // No need to check this sig any longer 3890 sigaddset(&check_signal_done, sig); 3891 } 3892 3893 // Dump all the signal 3894 if (sigismember(&check_signal_done, sig)) { 3895 print_signal_handlers(tty, buf, O_BUFLEN); 3896 } 3897} 3898 3899extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 3900 3901extern bool signal_name(int signo, char* buf, size_t len); 3902 3903const char* os::exception_name(int exception_code, char* buf, size_t size) { 3904 if (0 < exception_code && exception_code <= SIGRTMAX) { 3905 // signal 3906 if (!signal_name(exception_code, buf, size)) { 3907 jio_snprintf(buf, size, "SIG%d", exception_code); 3908 } 3909 return buf; 3910 } else { 3911 return NULL; 3912 } 3913} 3914 3915// this is called _before_ the most of global arguments have been parsed 3916void os::init(void) { 3917 char dummy; /* used to get a guess on initial stack address */ 3918// first_hrtime = gethrtime(); 3919 3920 // With LinuxThreads the JavaMain thread pid (primordial thread) 3921 // is different than the pid of the java launcher thread. 3922 // So, on Linux, the launcher thread pid is passed to the VM 3923 // via the sun.java.launcher.pid property. 3924 // Use this property instead of getpid() if it was correctly passed. 3925 // See bug 6351349. 3926 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 3927 3928 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 3929 3930 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 3931 3932 init_random(1234567); 3933 3934 ThreadCritical::initialize(); 3935 3936 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 3937 if (Linux::page_size() == -1) { 3938 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 3939 strerror(errno))); 3940 } 3941 init_page_sizes((size_t) Linux::page_size()); 3942 3943 Linux::initialize_system_info(); 3944 3945 // main_thread points to the aboriginal thread 3946 Linux::_main_thread = pthread_self(); 3947 3948 Linux::clock_init(); 3949 initial_time_count = os::elapsed_counter(); 3950 pthread_mutex_init(&dl_mutex, NULL); 3951} 3952 3953// To install functions for atexit system call 3954extern "C" { 3955 static void perfMemory_exit_helper() { 3956 perfMemory_exit(); 3957 } 3958} 3959 3960// this is called _after_ the global arguments have been parsed 3961jint os::init_2(void) 3962{ 3963 Linux::fast_thread_clock_init(); 3964 3965 // Allocate a single page and mark it as readable for safepoint polling 3966 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3967 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 3968 3969 os::set_polling_page( polling_page ); 3970 3971#ifndef PRODUCT 3972 if(Verbose && PrintMiscellaneous) 3973 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 3974#endif 3975 3976 if (!UseMembar) { 3977 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 3978 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); 3979 os::set_memory_serialize_page( mem_serialize_page ); 3980 3981#ifndef PRODUCT 3982 if(Verbose && PrintMiscellaneous) 3983 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 3984#endif 3985 } 3986 3987 FLAG_SET_DEFAULT(UseLargePages, os::large_page_init()); 3988 3989 // initialize suspend/resume support - must do this before signal_sets_init() 3990 if (SR_initialize() != 0) { 3991 perror("SR_initialize failed"); 3992 return JNI_ERR; 3993 } 3994 3995 Linux::signal_sets_init(); 3996 Linux::install_signal_handlers(); 3997 3998 // Check minimum allowable stack size for thread creation and to initialize 3999 // the java system classes, including StackOverflowError - depends on page 4000 // size. Add a page for compiler2 recursion in main thread. 4001 // Add in 2*BytesPerWord times page size to account for VM stack during 4002 // class initialization depending on 32 or 64 bit VM. 4003 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4004 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+ 4005 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size()); 4006 4007 size_t threadStackSizeInBytes = ThreadStackSize * K; 4008 if (threadStackSizeInBytes != 0 && 4009 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4010 tty->print_cr("\nThe stack size specified is too small, " 4011 "Specify at least %dk", 4012 os::Linux::min_stack_allowed/ K); 4013 return JNI_ERR; 4014 } 4015 4016 // Make the stack size a multiple of the page size so that 4017 // the yellow/red zones can be guarded. 4018 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4019 vm_page_size())); 4020 4021 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4022 4023 Linux::libpthread_init(); 4024 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4025 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4026 Linux::glibc_version(), Linux::libpthread_version(), 4027 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4028 } 4029 4030 if (UseNUMA) { 4031 if (!Linux::libnuma_init()) { 4032 UseNUMA = false; 4033 } else { 4034 if ((Linux::numa_max_node() < 1)) { 4035 // There's only one node(they start from 0), disable NUMA. 4036 UseNUMA = false; 4037 } 4038 } 4039 if (!UseNUMA && ForceNUMA) { 4040 UseNUMA = true; 4041 } 4042 } 4043 4044 if (MaxFDLimit) { 4045 // set the number of file descriptors to max. print out error 4046 // if getrlimit/setrlimit fails but continue regardless. 4047 struct rlimit nbr_files; 4048 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4049 if (status != 0) { 4050 if (PrintMiscellaneous && (Verbose || WizardMode)) 4051 perror("os::init_2 getrlimit failed"); 4052 } else { 4053 nbr_files.rlim_cur = nbr_files.rlim_max; 4054 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4055 if (status != 0) { 4056 if (PrintMiscellaneous && (Verbose || WizardMode)) 4057 perror("os::init_2 setrlimit failed"); 4058 } 4059 } 4060 } 4061 4062 // Initialize lock used to serialize thread creation (see os::create_thread) 4063 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4064 4065 // at-exit methods are called in the reverse order of their registration. 4066 // atexit functions are called on return from main or as a result of a 4067 // call to exit(3C). There can be only 32 of these functions registered 4068 // and atexit() does not set errno. 4069 4070 if (PerfAllowAtExitRegistration) { 4071 // only register atexit functions if PerfAllowAtExitRegistration is set. 4072 // atexit functions can be delayed until process exit time, which 4073 // can be problematic for embedded VM situations. Embedded VMs should 4074 // call DestroyJavaVM() to assure that VM resources are released. 4075 4076 // note: perfMemory_exit_helper atexit function may be removed in 4077 // the future if the appropriate cleanup code can be added to the 4078 // VM_Exit VMOperation's doit method. 4079 if (atexit(perfMemory_exit_helper) != 0) { 4080 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 4081 } 4082 } 4083 4084 // initialize thread priority policy 4085 prio_init(); 4086 4087 return JNI_OK; 4088} 4089 4090// this is called at the end of vm_initialization 4091void os::init_3(void) { } 4092 4093// Mark the polling page as unreadable 4094void os::make_polling_page_unreadable(void) { 4095 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4096 fatal("Could not disable polling page"); 4097}; 4098 4099// Mark the polling page as readable 4100void os::make_polling_page_readable(void) { 4101 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4102 fatal("Could not enable polling page"); 4103 } 4104}; 4105 4106int os::active_processor_count() { 4107 // Linux doesn't yet have a (official) notion of processor sets, 4108 // so just return the number of online processors. 4109 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4110 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4111 return online_cpus; 4112} 4113 4114bool os::distribute_processes(uint length, uint* distribution) { 4115 // Not yet implemented. 4116 return false; 4117} 4118 4119bool os::bind_to_processor(uint processor_id) { 4120 // Not yet implemented. 4121 return false; 4122} 4123 4124/// 4125 4126// Suspends the target using the signal mechanism and then grabs the PC before 4127// resuming the target. Used by the flat-profiler only 4128ExtendedPC os::get_thread_pc(Thread* thread) { 4129 // Make sure that it is called by the watcher for the VMThread 4130 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4131 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4132 4133 ExtendedPC epc; 4134 4135 OSThread* osthread = thread->osthread(); 4136 if (do_suspend(osthread)) { 4137 if (osthread->ucontext() != NULL) { 4138 epc = os::Linux::ucontext_get_pc(osthread->ucontext()); 4139 } else { 4140 // NULL context is unexpected, double-check this is the VMThread 4141 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4142 } 4143 do_resume(osthread); 4144 } 4145 // failure means pthread_kill failed for some reason - arguably this is 4146 // a fatal problem, but such problems are ignored elsewhere 4147 4148 return epc; 4149} 4150 4151int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4152{ 4153 if (is_NPTL()) { 4154 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4155 } else { 4156#ifndef IA64 4157 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4158 // word back to default 64bit precision if condvar is signaled. Java 4159 // wants 53bit precision. Save and restore current value. 4160 int fpu = get_fpu_control_word(); 4161#endif // IA64 4162 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4163#ifndef IA64 4164 set_fpu_control_word(fpu); 4165#endif // IA64 4166 return status; 4167 } 4168} 4169 4170//////////////////////////////////////////////////////////////////////////////// 4171// debug support 4172 4173static address same_page(address x, address y) { 4174 int page_bits = -os::vm_page_size(); 4175 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) 4176 return x; 4177 else if (x > y) 4178 return (address)(intptr_t(y) | ~page_bits) + 1; 4179 else 4180 return (address)(intptr_t(y) & page_bits); 4181} 4182 4183bool os::find(address addr, outputStream* st) { 4184 Dl_info dlinfo; 4185 memset(&dlinfo, 0, sizeof(dlinfo)); 4186 if (dladdr(addr, &dlinfo)) { 4187 st->print(PTR_FORMAT ": ", addr); 4188 if (dlinfo.dli_sname != NULL) { 4189 st->print("%s+%#x", dlinfo.dli_sname, 4190 addr - (intptr_t)dlinfo.dli_saddr); 4191 } else if (dlinfo.dli_fname) { 4192 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4193 } else { 4194 st->print("<absolute address>"); 4195 } 4196 if (dlinfo.dli_fname) { 4197 st->print(" in %s", dlinfo.dli_fname); 4198 } 4199 if (dlinfo.dli_fbase) { 4200 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4201 } 4202 st->cr(); 4203 4204 if (Verbose) { 4205 // decode some bytes around the PC 4206 address begin = same_page(addr-40, addr); 4207 address end = same_page(addr+40, addr); 4208 address lowest = (address) dlinfo.dli_sname; 4209 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4210 if (begin < lowest) begin = lowest; 4211 Dl_info dlinfo2; 4212 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr 4213 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 4214 end = (address) dlinfo2.dli_saddr; 4215 Disassembler::decode(begin, end, st); 4216 } 4217 return true; 4218 } 4219 return false; 4220} 4221 4222//////////////////////////////////////////////////////////////////////////////// 4223// misc 4224 4225// This does not do anything on Linux. This is basically a hook for being 4226// able to use structured exception handling (thread-local exception filters) 4227// on, e.g., Win32. 4228void 4229os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4230 JavaCallArguments* args, Thread* thread) { 4231 f(value, method, args, thread); 4232} 4233 4234void os::print_statistics() { 4235} 4236 4237int os::message_box(const char* title, const char* message) { 4238 int i; 4239 fdStream err(defaultStream::error_fd()); 4240 for (i = 0; i < 78; i++) err.print_raw("="); 4241 err.cr(); 4242 err.print_raw_cr(title); 4243 for (i = 0; i < 78; i++) err.print_raw("-"); 4244 err.cr(); 4245 err.print_raw_cr(message); 4246 for (i = 0; i < 78; i++) err.print_raw("="); 4247 err.cr(); 4248 4249 char buf[16]; 4250 // Prevent process from exiting upon "read error" without consuming all CPU 4251 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 4252 4253 return buf[0] == 'y' || buf[0] == 'Y'; 4254} 4255 4256int os::stat(const char *path, struct stat *sbuf) { 4257 char pathbuf[MAX_PATH]; 4258 if (strlen(path) > MAX_PATH - 1) { 4259 errno = ENAMETOOLONG; 4260 return -1; 4261 } 4262 os::native_path(strcpy(pathbuf, path)); 4263 return ::stat(pathbuf, sbuf); 4264} 4265 4266bool os::check_heap(bool force) { 4267 return true; 4268} 4269 4270int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 4271 return ::vsnprintf(buf, count, format, args); 4272} 4273 4274// Is a (classpath) directory empty? 4275bool os::dir_is_empty(const char* path) { 4276 DIR *dir = NULL; 4277 struct dirent *ptr; 4278 4279 dir = opendir(path); 4280 if (dir == NULL) return true; 4281 4282 /* Scan the directory */ 4283 bool result = true; 4284 char buf[sizeof(struct dirent) + MAX_PATH]; 4285 while (result && (ptr = ::readdir(dir)) != NULL) { 4286 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 4287 result = false; 4288 } 4289 } 4290 closedir(dir); 4291 return result; 4292} 4293 4294// This code originates from JDK's sysOpen and open64_w 4295// from src/solaris/hpi/src/system_md.c 4296 4297#ifndef O_DELETE 4298#define O_DELETE 0x10000 4299#endif 4300 4301// Open a file. Unlink the file immediately after open returns 4302// if the specified oflag has the O_DELETE flag set. 4303// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 4304 4305int os::open(const char *path, int oflag, int mode) { 4306 4307 if (strlen(path) > MAX_PATH - 1) { 4308 errno = ENAMETOOLONG; 4309 return -1; 4310 } 4311 int fd; 4312 int o_delete = (oflag & O_DELETE); 4313 oflag = oflag & ~O_DELETE; 4314 4315 fd = ::open64(path, oflag, mode); 4316 if (fd == -1) return -1; 4317 4318 //If the open succeeded, the file might still be a directory 4319 { 4320 struct stat64 buf64; 4321 int ret = ::fstat64(fd, &buf64); 4322 int st_mode = buf64.st_mode; 4323 4324 if (ret != -1) { 4325 if ((st_mode & S_IFMT) == S_IFDIR) { 4326 errno = EISDIR; 4327 ::close(fd); 4328 return -1; 4329 } 4330 } else { 4331 ::close(fd); 4332 return -1; 4333 } 4334 } 4335 4336 /* 4337 * All file descriptors that are opened in the JVM and not 4338 * specifically destined for a subprocess should have the 4339 * close-on-exec flag set. If we don't set it, then careless 3rd 4340 * party native code might fork and exec without closing all 4341 * appropriate file descriptors (e.g. as we do in closeDescriptors in 4342 * UNIXProcess.c), and this in turn might: 4343 * 4344 * - cause end-of-file to fail to be detected on some file 4345 * descriptors, resulting in mysterious hangs, or 4346 * 4347 * - might cause an fopen in the subprocess to fail on a system 4348 * suffering from bug 1085341. 4349 * 4350 * (Yes, the default setting of the close-on-exec flag is a Unix 4351 * design flaw) 4352 * 4353 * See: 4354 * 1085341: 32-bit stdio routines should support file descriptors >255 4355 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 4356 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 4357 */ 4358#ifdef FD_CLOEXEC 4359 { 4360 int flags = ::fcntl(fd, F_GETFD); 4361 if (flags != -1) 4362 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 4363 } 4364#endif 4365 4366 if (o_delete != 0) { 4367 ::unlink(path); 4368 } 4369 return fd; 4370} 4371 4372 4373// create binary file, rewriting existing file if required 4374int os::create_binary_file(const char* path, bool rewrite_existing) { 4375 int oflags = O_WRONLY | O_CREAT; 4376 if (!rewrite_existing) { 4377 oflags |= O_EXCL; 4378 } 4379 return ::open64(path, oflags, S_IREAD | S_IWRITE); 4380} 4381 4382// return current position of file pointer 4383jlong os::current_file_offset(int fd) { 4384 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 4385} 4386 4387// move file pointer to the specified offset 4388jlong os::seek_to_file_offset(int fd, jlong offset) { 4389 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 4390} 4391 4392// This code originates from JDK's sysAvailable 4393// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 4394 4395int os::available(int fd, jlong *bytes) { 4396 jlong cur, end; 4397 int mode; 4398 struct stat64 buf64; 4399 4400 if (::fstat64(fd, &buf64) >= 0) { 4401 mode = buf64.st_mode; 4402 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 4403 /* 4404 * XXX: is the following call interruptible? If so, this might 4405 * need to go through the INTERRUPT_IO() wrapper as for other 4406 * blocking, interruptible calls in this file. 4407 */ 4408 int n; 4409 if (::ioctl(fd, FIONREAD, &n) >= 0) { 4410 *bytes = n; 4411 return 1; 4412 } 4413 } 4414 } 4415 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 4416 return 0; 4417 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 4418 return 0; 4419 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 4420 return 0; 4421 } 4422 *bytes = end - cur; 4423 return 1; 4424} 4425 4426// Map a block of memory. 4427char* os::map_memory(int fd, const char* file_name, size_t file_offset, 4428 char *addr, size_t bytes, bool read_only, 4429 bool allow_exec) { 4430 int prot; 4431 int flags; 4432 4433 if (read_only) { 4434 prot = PROT_READ; 4435 flags = MAP_SHARED; 4436 } else { 4437 prot = PROT_READ | PROT_WRITE; 4438 flags = MAP_PRIVATE; 4439 } 4440 4441 if (allow_exec) { 4442 prot |= PROT_EXEC; 4443 } 4444 4445 if (addr != NULL) { 4446 flags |= MAP_FIXED; 4447 } 4448 4449 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 4450 fd, file_offset); 4451 if (mapped_address == MAP_FAILED) { 4452 return NULL; 4453 } 4454 return mapped_address; 4455} 4456 4457 4458// Remap a block of memory. 4459char* os::remap_memory(int fd, const char* file_name, size_t file_offset, 4460 char *addr, size_t bytes, bool read_only, 4461 bool allow_exec) { 4462 // same as map_memory() on this OS 4463 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 4464 allow_exec); 4465} 4466 4467 4468// Unmap a block of memory. 4469bool os::unmap_memory(char* addr, size_t bytes) { 4470 return munmap(addr, bytes) == 0; 4471} 4472 4473static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 4474 4475static clockid_t thread_cpu_clockid(Thread* thread) { 4476 pthread_t tid = thread->osthread()->pthread_id(); 4477 clockid_t clockid; 4478 4479 // Get thread clockid 4480 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 4481 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 4482 return clockid; 4483} 4484 4485// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 4486// are used by JVM M&M and JVMTI to get user+sys or user CPU time 4487// of a thread. 4488// 4489// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 4490// the fast estimate available on the platform. 4491 4492jlong os::current_thread_cpu_time() { 4493 if (os::Linux::supports_fast_thread_cpu_time()) { 4494 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4495 } else { 4496 // return user + sys since the cost is the same 4497 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 4498 } 4499} 4500 4501jlong os::thread_cpu_time(Thread* thread) { 4502 // consistent with what current_thread_cpu_time() returns 4503 if (os::Linux::supports_fast_thread_cpu_time()) { 4504 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4505 } else { 4506 return slow_thread_cpu_time(thread, true /* user + sys */); 4507 } 4508} 4509 4510jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 4511 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4512 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 4513 } else { 4514 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 4515 } 4516} 4517 4518jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4519 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 4520 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 4521 } else { 4522 return slow_thread_cpu_time(thread, user_sys_cpu_time); 4523 } 4524} 4525 4526// 4527// -1 on error. 4528// 4529 4530static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 4531 static bool proc_pid_cpu_avail = true; 4532 static bool proc_task_unchecked = true; 4533 static const char *proc_stat_path = "/proc/%d/stat"; 4534 pid_t tid = thread->osthread()->thread_id(); 4535 int i; 4536 char *s; 4537 char stat[2048]; 4538 int statlen; 4539 char proc_name[64]; 4540 int count; 4541 long sys_time, user_time; 4542 char string[64]; 4543 char cdummy; 4544 int idummy; 4545 long ldummy; 4546 FILE *fp; 4547 4548 // We first try accessing /proc/<pid>/cpu since this is faster to 4549 // process. If this file is not present (linux kernels 2.5 and above) 4550 // then we open /proc/<pid>/stat. 4551 if ( proc_pid_cpu_avail ) { 4552 sprintf(proc_name, "/proc/%d/cpu", tid); 4553 fp = fopen(proc_name, "r"); 4554 if ( fp != NULL ) { 4555 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time); 4556 fclose(fp); 4557 if ( count != 3 ) return -1; 4558 4559 if (user_sys_cpu_time) { 4560 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4561 } else { 4562 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4563 } 4564 } 4565 else proc_pid_cpu_avail = false; 4566 } 4567 4568 // The /proc/<tid>/stat aggregates per-process usage on 4569 // new Linux kernels 2.6+ where NPTL is supported. 4570 // The /proc/self/task/<tid>/stat still has the per-thread usage. 4571 // See bug 6328462. 4572 // There can be no directory /proc/self/task on kernels 2.4 with NPTL 4573 // and possibly in some other cases, so we check its availability. 4574 if (proc_task_unchecked && os::Linux::is_NPTL()) { 4575 // This is executed only once 4576 proc_task_unchecked = false; 4577 fp = fopen("/proc/self/task", "r"); 4578 if (fp != NULL) { 4579 proc_stat_path = "/proc/self/task/%d/stat"; 4580 fclose(fp); 4581 } 4582 } 4583 4584 sprintf(proc_name, proc_stat_path, tid); 4585 fp = fopen(proc_name, "r"); 4586 if ( fp == NULL ) return -1; 4587 statlen = fread(stat, 1, 2047, fp); 4588 stat[statlen] = '\0'; 4589 fclose(fp); 4590 4591 // Skip pid and the command string. Note that we could be dealing with 4592 // weird command names, e.g. user could decide to rename java launcher 4593 // to "java 1.4.2 :)", then the stat file would look like 4594 // 1234 (java 1.4.2 :)) R ... ... 4595 // We don't really need to know the command string, just find the last 4596 // occurrence of ")" and then start parsing from there. See bug 4726580. 4597 s = strrchr(stat, ')'); 4598 i = 0; 4599 if (s == NULL ) return -1; 4600 4601 // Skip blank chars 4602 do s++; while (isspace(*s)); 4603 4604 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 4605 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 4606 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 4607 &user_time, &sys_time); 4608 if ( count != 13 ) return -1; 4609 if (user_sys_cpu_time) { 4610 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 4611 } else { 4612 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 4613 } 4614} 4615 4616void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4617 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4618 info_ptr->may_skip_backward = false; // elapsed time not wall time 4619 info_ptr->may_skip_forward = false; // elapsed time not wall time 4620 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4621} 4622 4623void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 4624 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 4625 info_ptr->may_skip_backward = false; // elapsed time not wall time 4626 info_ptr->may_skip_forward = false; // elapsed time not wall time 4627 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 4628} 4629 4630bool os::is_thread_cpu_time_supported() { 4631 return true; 4632} 4633 4634// System loadavg support. Returns -1 if load average cannot be obtained. 4635// Linux doesn't yet have a (official) notion of processor sets, 4636// so just return the system wide load average. 4637int os::loadavg(double loadavg[], int nelem) { 4638 return ::getloadavg(loadavg, nelem); 4639} 4640 4641void os::pause() { 4642 char filename[MAX_PATH]; 4643 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 4644 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 4645 } else { 4646 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 4647 } 4648 4649 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 4650 if (fd != -1) { 4651 struct stat buf; 4652 ::close(fd); 4653 while (::stat(filename, &buf) == 0) { 4654 (void)::poll(NULL, 0, 100); 4655 } 4656 } else { 4657 jio_fprintf(stderr, 4658 "Could not open pause file '%s', continuing immediately.\n", filename); 4659 } 4660} 4661 4662extern "C" { 4663 4664/** 4665 * NOTE: the following code is to keep the green threads code 4666 * in the libjava.so happy. Once the green threads is removed, 4667 * these code will no longer be needed. 4668 */ 4669int 4670jdk_waitpid(pid_t pid, int* status, int options) { 4671 return waitpid(pid, status, options); 4672} 4673 4674int 4675fork1() { 4676 return fork(); 4677} 4678 4679int 4680jdk_sem_init(sem_t *sem, int pshared, unsigned int value) { 4681 return sem_init(sem, pshared, value); 4682} 4683 4684int 4685jdk_sem_post(sem_t *sem) { 4686 return sem_post(sem); 4687} 4688 4689int 4690jdk_sem_wait(sem_t *sem) { 4691 return sem_wait(sem); 4692} 4693 4694int 4695jdk_pthread_sigmask(int how , const sigset_t* newmask, sigset_t* oldmask) { 4696 return pthread_sigmask(how , newmask, oldmask); 4697} 4698 4699} 4700 4701// Refer to the comments in os_solaris.cpp park-unpark. 4702// 4703// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 4704// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 4705// For specifics regarding the bug see GLIBC BUGID 261237 : 4706// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 4707// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 4708// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 4709// is used. (The simple C test-case provided in the GLIBC bug report manifests the 4710// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 4711// and monitorenter when we're using 1-0 locking. All those operations may result in 4712// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 4713// of libpthread avoids the problem, but isn't practical. 4714// 4715// Possible remedies: 4716// 4717// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 4718// This is palliative and probabilistic, however. If the thread is preempted 4719// between the call to compute_abstime() and pthread_cond_timedwait(), more 4720// than the minimum period may have passed, and the abstime may be stale (in the 4721// past) resultin in a hang. Using this technique reduces the odds of a hang 4722// but the JVM is still vulnerable, particularly on heavily loaded systems. 4723// 4724// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 4725// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 4726// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 4727// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 4728// thread. 4729// 4730// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 4731// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 4732// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 4733// This also works well. In fact it avoids kernel-level scalability impediments 4734// on certain platforms that don't handle lots of active pthread_cond_timedwait() 4735// timers in a graceful fashion. 4736// 4737// 4. When the abstime value is in the past it appears that control returns 4738// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 4739// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 4740// can avoid the problem by reinitializing the condvar -- by cond_destroy() 4741// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 4742// It may be possible to avoid reinitialization by checking the return 4743// value from pthread_cond_timedwait(). In addition to reinitializing the 4744// condvar we must establish the invariant that cond_signal() is only called 4745// within critical sections protected by the adjunct mutex. This prevents 4746// cond_signal() from "seeing" a condvar that's in the midst of being 4747// reinitialized or that is corrupt. Sadly, this invariant obviates the 4748// desirable signal-after-unlock optimization that avoids futile context switching. 4749// 4750// I'm also concerned that some versions of NTPL might allocate an auxilliary 4751// structure when a condvar is used or initialized. cond_destroy() would 4752// release the helper structure. Our reinitialize-after-timedwait fix 4753// put excessive stress on malloc/free and locks protecting the c-heap. 4754// 4755// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 4756// It may be possible to refine (4) by checking the kernel and NTPL verisons 4757// and only enabling the work-around for vulnerable environments. 4758 4759// utility to compute the abstime argument to timedwait: 4760// millis is the relative timeout time 4761// abstime will be the absolute timeout time 4762// TODO: replace compute_abstime() with unpackTime() 4763 4764static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 4765 if (millis < 0) millis = 0; 4766 struct timeval now; 4767 int status = gettimeofday(&now, NULL); 4768 assert(status == 0, "gettimeofday"); 4769 jlong seconds = millis / 1000; 4770 millis %= 1000; 4771 if (seconds > 50000000) { // see man cond_timedwait(3T) 4772 seconds = 50000000; 4773 } 4774 abstime->tv_sec = now.tv_sec + seconds; 4775 long usec = now.tv_usec + millis * 1000; 4776 if (usec >= 1000000) { 4777 abstime->tv_sec += 1; 4778 usec -= 1000000; 4779 } 4780 abstime->tv_nsec = usec * 1000; 4781 return abstime; 4782} 4783 4784 4785// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 4786// Conceptually TryPark() should be equivalent to park(0). 4787 4788int os::PlatformEvent::TryPark() { 4789 for (;;) { 4790 const int v = _Event ; 4791 guarantee ((v == 0) || (v == 1), "invariant") ; 4792 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 4793 } 4794} 4795 4796void os::PlatformEvent::park() { // AKA "down()" 4797 // Invariant: Only the thread associated with the Event/PlatformEvent 4798 // may call park(). 4799 // TODO: assert that _Assoc != NULL or _Assoc == Self 4800 int v ; 4801 for (;;) { 4802 v = _Event ; 4803 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4804 } 4805 guarantee (v >= 0, "invariant") ; 4806 if (v == 0) { 4807 // Do this the hard way by blocking ... 4808 int status = pthread_mutex_lock(_mutex); 4809 assert_status(status == 0, status, "mutex_lock"); 4810 guarantee (_nParked == 0, "invariant") ; 4811 ++ _nParked ; 4812 while (_Event < 0) { 4813 status = pthread_cond_wait(_cond, _mutex); 4814 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 4815 // Treat this the same as if the wait was interrupted 4816 if (status == ETIME) { status = EINTR; } 4817 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 4818 } 4819 -- _nParked ; 4820 4821 // In theory we could move the ST of 0 into _Event past the unlock(), 4822 // but then we'd need a MEMBAR after the ST. 4823 _Event = 0 ; 4824 status = pthread_mutex_unlock(_mutex); 4825 assert_status(status == 0, status, "mutex_unlock"); 4826 } 4827 guarantee (_Event >= 0, "invariant") ; 4828} 4829 4830int os::PlatformEvent::park(jlong millis) { 4831 guarantee (_nParked == 0, "invariant") ; 4832 4833 int v ; 4834 for (;;) { 4835 v = _Event ; 4836 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 4837 } 4838 guarantee (v >= 0, "invariant") ; 4839 if (v != 0) return OS_OK ; 4840 4841 // We do this the hard way, by blocking the thread. 4842 // Consider enforcing a minimum timeout value. 4843 struct timespec abst; 4844 compute_abstime(&abst, millis); 4845 4846 int ret = OS_TIMEOUT; 4847 int status = pthread_mutex_lock(_mutex); 4848 assert_status(status == 0, status, "mutex_lock"); 4849 guarantee (_nParked == 0, "invariant") ; 4850 ++_nParked ; 4851 4852 // Object.wait(timo) will return because of 4853 // (a) notification 4854 // (b) timeout 4855 // (c) thread.interrupt 4856 // 4857 // Thread.interrupt and object.notify{All} both call Event::set. 4858 // That is, we treat thread.interrupt as a special case of notification. 4859 // The underlying Solaris implementation, cond_timedwait, admits 4860 // spurious/premature wakeups, but the JLS/JVM spec prevents the 4861 // JVM from making those visible to Java code. As such, we must 4862 // filter out spurious wakeups. We assume all ETIME returns are valid. 4863 // 4864 // TODO: properly differentiate simultaneous notify+interrupt. 4865 // In that case, we should propagate the notify to another waiter. 4866 4867 while (_Event < 0) { 4868 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 4869 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 4870 pthread_cond_destroy (_cond); 4871 pthread_cond_init (_cond, NULL) ; 4872 } 4873 assert_status(status == 0 || status == EINTR || 4874 status == ETIME || status == ETIMEDOUT, 4875 status, "cond_timedwait"); 4876 if (!FilterSpuriousWakeups) break ; // previous semantics 4877 if (status == ETIME || status == ETIMEDOUT) break ; 4878 // We consume and ignore EINTR and spurious wakeups. 4879 } 4880 --_nParked ; 4881 if (_Event >= 0) { 4882 ret = OS_OK; 4883 } 4884 _Event = 0 ; 4885 status = pthread_mutex_unlock(_mutex); 4886 assert_status(status == 0, status, "mutex_unlock"); 4887 assert (_nParked == 0, "invariant") ; 4888 return ret; 4889} 4890 4891void os::PlatformEvent::unpark() { 4892 int v, AnyWaiters ; 4893 for (;;) { 4894 v = _Event ; 4895 if (v > 0) { 4896 // The LD of _Event could have reordered or be satisfied 4897 // by a read-aside from this processor's write buffer. 4898 // To avoid problems execute a barrier and then 4899 // ratify the value. 4900 OrderAccess::fence() ; 4901 if (_Event == v) return ; 4902 continue ; 4903 } 4904 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ; 4905 } 4906 if (v < 0) { 4907 // Wait for the thread associated with the event to vacate 4908 int status = pthread_mutex_lock(_mutex); 4909 assert_status(status == 0, status, "mutex_lock"); 4910 AnyWaiters = _nParked ; 4911 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ; 4912 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 4913 AnyWaiters = 0 ; 4914 pthread_cond_signal (_cond); 4915 } 4916 status = pthread_mutex_unlock(_mutex); 4917 assert_status(status == 0, status, "mutex_unlock"); 4918 if (AnyWaiters != 0) { 4919 status = pthread_cond_signal(_cond); 4920 assert_status(status == 0, status, "cond_signal"); 4921 } 4922 } 4923 4924 // Note that we signal() _after dropping the lock for "immortal" Events. 4925 // This is safe and avoids a common class of futile wakeups. In rare 4926 // circumstances this can cause a thread to return prematurely from 4927 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 4928 // simply re-test the condition and re-park itself. 4929} 4930 4931 4932// JSR166 4933// ------------------------------------------------------- 4934 4935/* 4936 * The solaris and linux implementations of park/unpark are fairly 4937 * conservative for now, but can be improved. They currently use a 4938 * mutex/condvar pair, plus a a count. 4939 * Park decrements count if > 0, else does a condvar wait. Unpark 4940 * sets count to 1 and signals condvar. Only one thread ever waits 4941 * on the condvar. Contention seen when trying to park implies that someone 4942 * is unparking you, so don't wait. And spurious returns are fine, so there 4943 * is no need to track notifications. 4944 */ 4945 4946 4947#define NANOSECS_PER_SEC 1000000000 4948#define NANOSECS_PER_MILLISEC 1000000 4949#define MAX_SECS 100000000 4950/* 4951 * This code is common to linux and solaris and will be moved to a 4952 * common place in dolphin. 4953 * 4954 * The passed in time value is either a relative time in nanoseconds 4955 * or an absolute time in milliseconds. Either way it has to be unpacked 4956 * into suitable seconds and nanoseconds components and stored in the 4957 * given timespec structure. 4958 * Given time is a 64-bit value and the time_t used in the timespec is only 4959 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 4960 * overflow if times way in the future are given. Further on Solaris versions 4961 * prior to 10 there is a restriction (see cond_timedwait) that the specified 4962 * number of seconds, in abstime, is less than current_time + 100,000,000. 4963 * As it will be 28 years before "now + 100000000" will overflow we can 4964 * ignore overflow and just impose a hard-limit on seconds using the value 4965 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 4966 * years from "now". 4967 */ 4968 4969static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 4970 assert (time > 0, "convertTime"); 4971 4972 struct timeval now; 4973 int status = gettimeofday(&now, NULL); 4974 assert(status == 0, "gettimeofday"); 4975 4976 time_t max_secs = now.tv_sec + MAX_SECS; 4977 4978 if (isAbsolute) { 4979 jlong secs = time / 1000; 4980 if (secs > max_secs) { 4981 absTime->tv_sec = max_secs; 4982 } 4983 else { 4984 absTime->tv_sec = secs; 4985 } 4986 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 4987 } 4988 else { 4989 jlong secs = time / NANOSECS_PER_SEC; 4990 if (secs >= MAX_SECS) { 4991 absTime->tv_sec = max_secs; 4992 absTime->tv_nsec = 0; 4993 } 4994 else { 4995 absTime->tv_sec = now.tv_sec + secs; 4996 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 4997 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 4998 absTime->tv_nsec -= NANOSECS_PER_SEC; 4999 ++absTime->tv_sec; // note: this must be <= max_secs 5000 } 5001 } 5002 } 5003 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5004 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5005 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5006 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5007} 5008 5009void Parker::park(bool isAbsolute, jlong time) { 5010 // Optional fast-path check: 5011 // Return immediately if a permit is available. 5012 if (_counter > 0) { 5013 _counter = 0 ; 5014 OrderAccess::fence(); 5015 return ; 5016 } 5017 5018 Thread* thread = Thread::current(); 5019 assert(thread->is_Java_thread(), "Must be JavaThread"); 5020 JavaThread *jt = (JavaThread *)thread; 5021 5022 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5023 // Check interrupt before trying to wait 5024 if (Thread::is_interrupted(thread, false)) { 5025 return; 5026 } 5027 5028 // Next, demultiplex/decode time arguments 5029 timespec absTime; 5030 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5031 return; 5032 } 5033 if (time > 0) { 5034 unpackTime(&absTime, isAbsolute, time); 5035 } 5036 5037 5038 // Enter safepoint region 5039 // Beware of deadlocks such as 6317397. 5040 // The per-thread Parker:: mutex is a classic leaf-lock. 5041 // In particular a thread must never block on the Threads_lock while 5042 // holding the Parker:: mutex. If safepoints are pending both the 5043 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5044 ThreadBlockInVM tbivm(jt); 5045 5046 // Don't wait if cannot get lock since interference arises from 5047 // unblocking. Also. check interrupt before trying wait 5048 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5049 return; 5050 } 5051 5052 int status ; 5053 if (_counter > 0) { // no wait needed 5054 _counter = 0; 5055 status = pthread_mutex_unlock(_mutex); 5056 assert (status == 0, "invariant") ; 5057 OrderAccess::fence(); 5058 return; 5059 } 5060 5061#ifdef ASSERT 5062 // Don't catch signals while blocked; let the running threads have the signals. 5063 // (This allows a debugger to break into the running thread.) 5064 sigset_t oldsigs; 5065 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5066 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5067#endif 5068 5069 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5070 jt->set_suspend_equivalent(); 5071 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5072 5073 if (time == 0) { 5074 status = pthread_cond_wait (_cond, _mutex) ; 5075 } else { 5076 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 5077 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5078 pthread_cond_destroy (_cond) ; 5079 pthread_cond_init (_cond, NULL); 5080 } 5081 } 5082 assert_status(status == 0 || status == EINTR || 5083 status == ETIME || status == ETIMEDOUT, 5084 status, "cond_timedwait"); 5085 5086#ifdef ASSERT 5087 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5088#endif 5089 5090 _counter = 0 ; 5091 status = pthread_mutex_unlock(_mutex) ; 5092 assert_status(status == 0, status, "invariant") ; 5093 // If externally suspended while waiting, re-suspend 5094 if (jt->handle_special_suspend_equivalent_condition()) { 5095 jt->java_suspend_self(); 5096 } 5097 5098 OrderAccess::fence(); 5099} 5100 5101void Parker::unpark() { 5102 int s, status ; 5103 status = pthread_mutex_lock(_mutex); 5104 assert (status == 0, "invariant") ; 5105 s = _counter; 5106 _counter = 1; 5107 if (s < 1) { 5108 if (WorkAroundNPTLTimedWaitHang) { 5109 status = pthread_cond_signal (_cond) ; 5110 assert (status == 0, "invariant") ; 5111 status = pthread_mutex_unlock(_mutex); 5112 assert (status == 0, "invariant") ; 5113 } else { 5114 status = pthread_mutex_unlock(_mutex); 5115 assert (status == 0, "invariant") ; 5116 status = pthread_cond_signal (_cond) ; 5117 assert (status == 0, "invariant") ; 5118 } 5119 } else { 5120 pthread_mutex_unlock(_mutex); 5121 assert (status == 0, "invariant") ; 5122 } 5123} 5124 5125 5126extern char** environ; 5127 5128#ifndef __NR_fork 5129#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5130#endif 5131 5132#ifndef __NR_execve 5133#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5134#endif 5135 5136// Run the specified command in a separate process. Return its exit value, 5137// or -1 on failure (e.g. can't fork a new process). 5138// Unlike system(), this function can be called from signal handler. It 5139// doesn't block SIGINT et al. 5140int os::fork_and_exec(char* cmd) { 5141 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5142 5143 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5144 // pthread_atfork handlers and reset pthread library. All we need is a 5145 // separate process to execve. Make a direct syscall to fork process. 5146 // On IA64 there's no fork syscall, we have to use fork() and hope for 5147 // the best... 5148 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5149 IA64_ONLY(fork();) 5150 5151 if (pid < 0) { 5152 // fork failed 5153 return -1; 5154 5155 } else if (pid == 0) { 5156 // child process 5157 5158 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5159 // first to kill every thread on the thread list. Because this list is 5160 // not reset by fork() (see notes above), execve() will instead kill 5161 // every thread in the parent process. We know this is the only thread 5162 // in the new process, so make a system call directly. 5163 // IA64 should use normal execve() from glibc to match the glibc fork() 5164 // above. 5165 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5166 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5167 5168 // execve failed 5169 _exit(-1); 5170 5171 } else { 5172 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5173 // care about the actual exit code, for now. 5174 5175 int status; 5176 5177 // Wait for the child process to exit. This returns immediately if 5178 // the child has already exited. */ 5179 while (waitpid(pid, &status, 0) < 0) { 5180 switch (errno) { 5181 case ECHILD: return 0; 5182 case EINTR: break; 5183 default: return -1; 5184 } 5185 } 5186 5187 if (WIFEXITED(status)) { 5188 // The child exited normally; get its exit code. 5189 return WEXITSTATUS(status); 5190 } else if (WIFSIGNALED(status)) { 5191 // The child exited because of a signal 5192 // The best value to return is 0x80 + signal number, 5193 // because that is what all Unix shells do, and because 5194 // it allows callers to distinguish between process exit and 5195 // process death by signal. 5196 return 0x80 + WTERMSIG(status); 5197 } else { 5198 // Unknown exit code; pass it through 5199 return status; 5200 } 5201 } 5202} 5203 5204// is_headless_jre() 5205// 5206// Test for the existence of libmawt in motif21 or xawt directories 5207// in order to report if we are running in a headless jre 5208// 5209bool os::is_headless_jre() { 5210 struct stat statbuf; 5211 char buf[MAXPATHLEN]; 5212 char libmawtpath[MAXPATHLEN]; 5213 const char *xawtstr = "/xawt/libmawt.so"; 5214 const char *motifstr = "/motif21/libmawt.so"; 5215 char *p; 5216 5217 // Get path to libjvm.so 5218 os::jvm_path(buf, sizeof(buf)); 5219 5220 // Get rid of libjvm.so 5221 p = strrchr(buf, '/'); 5222 if (p == NULL) return false; 5223 else *p = '\0'; 5224 5225 // Get rid of client or server 5226 p = strrchr(buf, '/'); 5227 if (p == NULL) return false; 5228 else *p = '\0'; 5229 5230 // check xawt/libmawt.so 5231 strcpy(libmawtpath, buf); 5232 strcat(libmawtpath, xawtstr); 5233 if (::stat(libmawtpath, &statbuf) == 0) return false; 5234 5235 // check motif21/libmawt.so 5236 strcpy(libmawtpath, buf); 5237 strcat(libmawtpath, motifstr); 5238 if (::stat(libmawtpath, &statbuf) == 0) return false; 5239 5240 return true; 5241} 5242 5243