1 /* 2 * Copyright (c) 1999, 2013, 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 // no precompiled headers 26 #include "classfile/classLoader.hpp" 27 #include "classfile/systemDictionary.hpp" 28 #include "classfile/vmSymbols.hpp" 29 #include "code/icBuffer.hpp" 30 #include "code/vtableStubs.hpp" 31 #include "compiler/compileBroker.hpp" 32 #include "interpreter/interpreter.hpp" 33 #include "jvm_linux.h" 34 #include "memory/allocation.inline.hpp" 35 #include "memory/filemap.hpp" 36 #include "mutex_linux.inline.hpp" 37 #include "oops/oop.inline.hpp" 38 #include "os_share_linux.hpp" 39 #include "prims/jniFastGetField.hpp" 40 #include "prims/jvm.h" 41 #include "prims/jvm_misc.hpp" 42 #include "runtime/arguments.hpp" 43 #include "runtime/extendedPC.hpp" 44 #include "runtime/globals.hpp" 45 #include "runtime/interfaceSupport.hpp" 46 #include "runtime/init.hpp" 47 #include "runtime/java.hpp" 48 #include "runtime/javaCalls.hpp" 49 #include "runtime/mutexLocker.hpp" 50 #include "runtime/objectMonitor.hpp" 51 #include "runtime/osThread.hpp" 52 #include "runtime/perfMemory.hpp" 53 #include "runtime/sharedRuntime.hpp" 54 #include "runtime/statSampler.hpp" 55 #include "runtime/stubRoutines.hpp" 56 #include "runtime/threadCritical.hpp" 57 #include "runtime/timer.hpp" 58 #include "services/attachListener.hpp" 59 #include "services/memTracker.hpp" 60 #include "services/runtimeService.hpp" 61 #include "thread_linux.inline.hpp" 62 #include "utilities/decoder.hpp" 63 #include "utilities/defaultStream.hpp" 64 #include "utilities/events.hpp" 65 #include "utilities/elfFile.hpp" 66 #include "utilities/growableArray.hpp" 67 #include "utilities/vmError.hpp" 68 #ifdef TARGET_ARCH_x86 69 # include "assembler_x86.inline.hpp" 70 # include "nativeInst_x86.hpp" 71 #endif 72 #ifdef TARGET_ARCH_sparc 73 # include "assembler_sparc.inline.hpp" 74 # include "nativeInst_sparc.hpp" 75 #endif 76 #ifdef TARGET_ARCH_zero 77 # include "assembler_zero.inline.hpp" 78 # include "nativeInst_zero.hpp" 79 #endif 80 #ifdef TARGET_ARCH_arm 81 # include "assembler_arm.inline.hpp" 82 # include "nativeInst_arm.hpp" 83 #endif 84 #ifdef TARGET_ARCH_ppc 85 # include "assembler_ppc.inline.hpp" 86 # include "nativeInst_ppc.hpp" 87 #endif 88 89 // put OS-includes here 90 # include <sys/types.h> 91 # include <sys/mman.h> 92 # include <sys/stat.h> 93 # include <sys/select.h> 94 # include <pthread.h> 95 # include <signal.h> 96 # include <errno.h> 97 # include <dlfcn.h> 98 # include <stdio.h> 99 # include <unistd.h> 100 # include <sys/resource.h> 101 # include <pthread.h> 102 # include <sys/stat.h> 103 # include <sys/time.h> 104 # include <sys/times.h> 105 # include <sys/utsname.h> 106 # include <sys/socket.h> 107 # include <sys/wait.h> 108 # include <pwd.h> 109 # include <poll.h> 110 # include <semaphore.h> 111 # include <fcntl.h> 112 # include <string.h> 113 # include <syscall.h> 114 # include <sys/sysinfo.h> 115 # include <gnu/libc-version.h> 116 # include <sys/ipc.h> 117 # include <sys/shm.h> 118 # include <link.h> 119 # include <stdint.h> 120 # include <inttypes.h> 121 # include <sys/ioctl.h> 122 123 #define MAX_PATH (2 * K) 124 125 // for timer info max values which include all bits 126 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 127 128 #define LARGEPAGES_BIT (1 << 6) 129 //////////////////////////////////////////////////////////////////////////////// 130 // global variables 131 julong os::Linux::_physical_memory = 0; 132 133 address os::Linux::_initial_thread_stack_bottom = NULL; 134 uintptr_t os::Linux::_initial_thread_stack_size = 0; 135 136 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 137 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 138 Mutex* os::Linux::_createThread_lock = NULL; 139 pthread_t os::Linux::_main_thread; 140 int os::Linux::_page_size = -1; 141 const int os::Linux::_vm_default_page_size = (8 * K); 142 bool os::Linux::_is_floating_stack = false; 143 bool os::Linux::_is_NPTL = false; 144 bool os::Linux::_supports_fast_thread_cpu_time = false; 145 const char * os::Linux::_glibc_version = NULL; 146 const char * os::Linux::_libpthread_version = NULL; 147 148 static jlong initial_time_count=0; 149 150 static int clock_tics_per_sec = 100; 151 152 // For diagnostics to print a message once. see run_periodic_checks 153 static sigset_t check_signal_done; 154 static bool check_signals = true;; 155 156 static pid_t _initial_pid = 0; 157 158 /* Signal number used to suspend/resume a thread */ 159 160 /* do not use any signal number less than SIGSEGV, see 4355769 */ 161 static int SR_signum = SIGUSR2; 162 sigset_t SR_sigset; 163 164 /* Used to protect dlsym() calls */ 165 static pthread_mutex_t dl_mutex; 166 167 // Declarations 168 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 169 170 #ifdef JAVASE_EMBEDDED 171 class MemNotifyThread: public Thread { 172 friend class VMStructs; 173 public: 174 virtual void run(); 175 176 private: 177 static MemNotifyThread* _memnotify_thread; 178 int _fd; 179 180 public: 181 182 // Constructor 183 MemNotifyThread(int fd); 184 185 // Tester 186 bool is_memnotify_thread() const { return true; } 187 188 // Printing 189 char* name() const { return (char*)"Linux MemNotify Thread"; } 190 191 // Returns the single instance of the MemNotifyThread 192 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; } 193 194 // Create and start the single instance of MemNotifyThread 195 static void start(); 196 }; 197 #endif // JAVASE_EMBEDDED 198 199 // utility functions 200 201 static int SR_initialize(); 202 static int SR_finalize(); 203 204 julong os::available_memory() { 205 return Linux::available_memory(); 206 } 207 208 julong os::Linux::available_memory() { 209 // values in struct sysinfo are "unsigned long" 210 struct sysinfo si; 211 sysinfo(&si); 212 213 return (julong)si.freeram * si.mem_unit; 214 } 215 216 julong os::physical_memory() { 217 return Linux::physical_memory(); 218 } 219 220 julong os::allocatable_physical_memory(julong size) { 221 #ifdef _LP64 222 return size; 223 #else 224 julong result = MIN2(size, (julong)3800*M); 225 if (!is_allocatable(result)) { 226 // See comments under solaris for alignment considerations 227 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size(); 228 result = MIN2(size, reasonable_size); 229 } 230 return result; 231 #endif // _LP64 232 } 233 234 //////////////////////////////////////////////////////////////////////////////// 235 // environment support 236 237 bool os::getenv(const char* name, char* buf, int len) { 238 const char* val = ::getenv(name); 239 if (val != NULL && strlen(val) < (size_t)len) { 240 strcpy(buf, val); 241 return true; 242 } 243 if (len > 0) buf[0] = 0; // return a null string 244 return false; 245 } 246 247 248 // Return true if user is running as root. 249 250 bool os::have_special_privileges() { 251 static bool init = false; 252 static bool privileges = false; 253 if (!init) { 254 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 255 init = true; 256 } 257 return privileges; 258 } 259 260 261 #ifndef SYS_gettid 262 // i386: 224, ia64: 1105, amd64: 186, sparc 143 263 #ifdef __ia64__ 264 #define SYS_gettid 1105 265 #elif __i386__ 266 #define SYS_gettid 224 267 #elif __amd64__ 268 #define SYS_gettid 186 269 #elif __sparc__ 270 #define SYS_gettid 143 271 #else 272 #error define gettid for the arch 273 #endif 274 #endif 275 276 // Cpu architecture string 277 #if defined(ZERO) 278 static char cpu_arch[] = ZERO_LIBARCH; 279 #elif defined(IA64) 280 static char cpu_arch[] = "ia64"; 281 #elif defined(IA32) 282 static char cpu_arch[] = "i386"; 283 #elif defined(AMD64) 284 static char cpu_arch[] = "amd64"; 285 #elif defined(ARM) 286 static char cpu_arch[] = "arm"; 287 #elif defined(PPC) 288 static char cpu_arch[] = "ppc"; 289 #elif defined(SPARC) 290 # ifdef _LP64 291 static char cpu_arch[] = "sparcv9"; 292 # else 293 static char cpu_arch[] = "sparc"; 294 # endif 295 #else 296 #error Add appropriate cpu_arch setting 297 #endif 298 299 300 // pid_t gettid() 301 // 302 // Returns the kernel thread id of the currently running thread. Kernel 303 // thread id is used to access /proc. 304 // 305 // (Note that getpid() on LinuxThreads returns kernel thread id too; but 306 // on NPTL, it returns the same pid for all threads, as required by POSIX.) 307 // 308 pid_t os::Linux::gettid() { 309 int rslt = syscall(SYS_gettid); 310 if (rslt == -1) { 311 // old kernel, no NPTL support 312 return getpid(); 313 } else { 314 return (pid_t)rslt; 315 } 316 } 317 318 // Most versions of linux have a bug where the number of processors are 319 // determined by looking at the /proc file system. In a chroot environment, 320 // the system call returns 1. This causes the VM to act as if it is 321 // a single processor and elide locking (see is_MP() call). 322 static bool unsafe_chroot_detected = false; 323 static const char *unstable_chroot_error = "/proc file system not found.\n" 324 "Java may be unstable running multithreaded in a chroot " 325 "environment on Linux when /proc filesystem is not mounted."; 326 327 void os::Linux::initialize_system_info() { 328 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 329 if (processor_count() == 1) { 330 pid_t pid = os::Linux::gettid(); 331 char fname[32]; 332 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 333 FILE *fp = fopen(fname, "r"); 334 if (fp == NULL) { 335 unsafe_chroot_detected = true; 336 } else { 337 fclose(fp); 338 } 339 } 340 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 341 assert(processor_count() > 0, "linux error"); 342 } 343 344 void os::init_system_properties_values() { 345 // char arch[12]; 346 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 347 348 // The next steps are taken in the product version: 349 // 350 // Obtain the JAVA_HOME value from the location of libjvm[_g].so. 351 // This library should be located at: 352 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so. 353 // 354 // If "/jre/lib/" appears at the right place in the path, then we 355 // assume libjvm[_g].so is installed in a JDK and we use this path. 356 // 357 // Otherwise exit with message: "Could not create the Java virtual machine." 358 // 359 // The following extra steps are taken in the debugging version: 360 // 361 // If "/jre/lib/" does NOT appear at the right place in the path 362 // instead of exit check for $JAVA_HOME environment variable. 363 // 364 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 365 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so 366 // it looks like libjvm[_g].so is installed there 367 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so. 368 // 369 // Otherwise exit. 370 // 371 // Important note: if the location of libjvm.so changes this 372 // code needs to be changed accordingly. 373 374 // The next few definitions allow the code to be verbatim: 375 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal) 376 #define getenv(n) ::getenv(n) 377 378 /* 379 * See ld(1): 380 * The linker uses the following search paths to locate required 381 * shared libraries: 382 * 1: ... 383 * ... 384 * 7: The default directories, normally /lib and /usr/lib. 385 */ 386 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 387 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 388 #else 389 #define DEFAULT_LIBPATH "/lib:/usr/lib" 390 #endif 391 392 #define EXTENSIONS_DIR "/lib/ext" 393 #define ENDORSED_DIR "/lib/endorsed" 394 #define REG_DIR "/usr/java/packages" 395 396 { 397 /* sysclasspath, java_home, dll_dir */ 398 { 399 char *home_path; 400 char *dll_path; 401 char *pslash; 402 char buf[MAXPATHLEN]; 403 os::jvm_path(buf, sizeof(buf)); 404 405 // Found the full path to libjvm.so. 406 // Now cut the path to <java_home>/jre if we can. 407 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 408 pslash = strrchr(buf, '/'); 409 if (pslash != NULL) 410 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 411 dll_path = malloc(strlen(buf) + 1); 412 if (dll_path == NULL) 413 return; 414 strcpy(dll_path, buf); 415 Arguments::set_dll_dir(dll_path); 416 417 if (pslash != NULL) { 418 pslash = strrchr(buf, '/'); 419 if (pslash != NULL) { 420 *pslash = '\0'; /* get rid of /<arch> */ 421 pslash = strrchr(buf, '/'); 422 if (pslash != NULL) 423 *pslash = '\0'; /* get rid of /lib */ 424 } 425 } 426 427 home_path = malloc(strlen(buf) + 1); 428 if (home_path == NULL) 429 return; 430 strcpy(home_path, buf); 431 Arguments::set_java_home(home_path); 432 433 if (!set_boot_path('/', ':')) 434 return; 435 } 436 437 /* 438 * Where to look for native libraries 439 * 440 * Note: Due to a legacy implementation, most of the library path 441 * is set in the launcher. This was to accomodate linking restrictions 442 * on legacy Linux implementations (which are no longer supported). 443 * Eventually, all the library path setting will be done here. 444 * 445 * However, to prevent the proliferation of improperly built native 446 * libraries, the new path component /usr/java/packages is added here. 447 * Eventually, all the library path setting will be done here. 448 */ 449 { 450 char *ld_library_path; 451 452 /* 453 * Construct the invariant part of ld_library_path. Note that the 454 * space for the colon and the trailing null are provided by the 455 * nulls included by the sizeof operator (so actually we allocate 456 * a byte more than necessary). 457 */ 458 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 459 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 460 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 461 462 /* 463 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 464 * should always exist (until the legacy problem cited above is 465 * addressed). 466 */ 467 char *v = getenv("LD_LIBRARY_PATH"); 468 if (v != NULL) { 469 char *t = ld_library_path; 470 /* That's +1 for the colon and +1 for the trailing '\0' */ 471 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 472 sprintf(ld_library_path, "%s:%s", v, t); 473 } 474 Arguments::set_library_path(ld_library_path); 475 } 476 477 /* 478 * Extensions directories. 479 * 480 * Note that the space for the colon and the trailing null are provided 481 * by the nulls included by the sizeof operator (so actually one byte more 482 * than necessary is allocated). 483 */ 484 { 485 char *buf = malloc(strlen(Arguments::get_java_home()) + 486 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 487 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 488 Arguments::get_java_home()); 489 Arguments::set_ext_dirs(buf); 490 } 491 492 /* Endorsed standards default directory. */ 493 { 494 char * buf; 495 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 496 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 497 Arguments::set_endorsed_dirs(buf); 498 } 499 } 500 501 #undef malloc 502 #undef getenv 503 #undef EXTENSIONS_DIR 504 #undef ENDORSED_DIR 505 506 // Done 507 return; 508 } 509 510 //////////////////////////////////////////////////////////////////////////////// 511 // breakpoint support 512 513 void os::breakpoint() { 514 BREAKPOINT; 515 } 516 517 extern "C" void breakpoint() { 518 // use debugger to set breakpoint here 519 } 520 521 //////////////////////////////////////////////////////////////////////////////// 522 // signal support 523 524 debug_only(static bool signal_sets_initialized = false); 525 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 526 527 bool os::Linux::is_sig_ignored(int sig) { 528 struct sigaction oact; 529 sigaction(sig, (struct sigaction*)NULL, &oact); 530 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 531 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 532 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 533 return true; 534 else 535 return false; 536 } 537 538 void os::Linux::signal_sets_init() { 539 // Should also have an assertion stating we are still single-threaded. 540 assert(!signal_sets_initialized, "Already initialized"); 541 // Fill in signals that are necessarily unblocked for all threads in 542 // the VM. Currently, we unblock the following signals: 543 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 544 // by -Xrs (=ReduceSignalUsage)); 545 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 546 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 547 // the dispositions or masks wrt these signals. 548 // Programs embedding the VM that want to use the above signals for their 549 // own purposes must, at this time, use the "-Xrs" option to prevent 550 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 551 // (See bug 4345157, and other related bugs). 552 // In reality, though, unblocking these signals is really a nop, since 553 // these signals are not blocked by default. 554 sigemptyset(&unblocked_sigs); 555 sigemptyset(&allowdebug_blocked_sigs); 556 sigaddset(&unblocked_sigs, SIGILL); 557 sigaddset(&unblocked_sigs, SIGSEGV); 558 sigaddset(&unblocked_sigs, SIGBUS); 559 sigaddset(&unblocked_sigs, SIGFPE); 560 sigaddset(&unblocked_sigs, SR_signum); 561 562 if (!ReduceSignalUsage) { 563 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 564 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 565 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 566 } 567 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 568 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 569 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 570 } 571 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 572 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 573 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 574 } 575 } 576 // Fill in signals that are blocked by all but the VM thread. 577 sigemptyset(&vm_sigs); 578 if (!ReduceSignalUsage) 579 sigaddset(&vm_sigs, BREAK_SIGNAL); 580 debug_only(signal_sets_initialized = true); 581 582 } 583 584 // These are signals that are unblocked while a thread is running Java. 585 // (For some reason, they get blocked by default.) 586 sigset_t* os::Linux::unblocked_signals() { 587 assert(signal_sets_initialized, "Not initialized"); 588 return &unblocked_sigs; 589 } 590 591 // These are the signals that are blocked while a (non-VM) thread is 592 // running Java. Only the VM thread handles these signals. 593 sigset_t* os::Linux::vm_signals() { 594 assert(signal_sets_initialized, "Not initialized"); 595 return &vm_sigs; 596 } 597 598 // These are signals that are blocked during cond_wait to allow debugger in 599 sigset_t* os::Linux::allowdebug_blocked_signals() { 600 assert(signal_sets_initialized, "Not initialized"); 601 return &allowdebug_blocked_sigs; 602 } 603 604 void os::Linux::hotspot_sigmask(Thread* thread) { 605 606 //Save caller's signal mask before setting VM signal mask 607 sigset_t caller_sigmask; 608 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 609 610 OSThread* osthread = thread->osthread(); 611 osthread->set_caller_sigmask(caller_sigmask); 612 613 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 614 615 if (!ReduceSignalUsage) { 616 if (thread->is_VM_thread()) { 617 // Only the VM thread handles BREAK_SIGNAL ... 618 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 619 } else { 620 // ... all other threads block BREAK_SIGNAL 621 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 622 } 623 } 624 } 625 626 ////////////////////////////////////////////////////////////////////////////// 627 // detecting pthread library 628 629 void os::Linux::libpthread_init() { 630 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 631 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 632 // generic name for earlier versions. 633 // Define macros here so we can build HotSpot on old systems. 634 # ifndef _CS_GNU_LIBC_VERSION 635 # define _CS_GNU_LIBC_VERSION 2 636 # endif 637 # ifndef _CS_GNU_LIBPTHREAD_VERSION 638 # define _CS_GNU_LIBPTHREAD_VERSION 3 639 # endif 640 641 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 642 if (n > 0) { 643 char *str = (char *)malloc(n, mtInternal); 644 confstr(_CS_GNU_LIBC_VERSION, str, n); 645 os::Linux::set_glibc_version(str); 646 } else { 647 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 648 static char _gnu_libc_version[32]; 649 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 650 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 651 os::Linux::set_glibc_version(_gnu_libc_version); 652 } 653 654 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 655 if (n > 0) { 656 char *str = (char *)malloc(n, mtInternal); 657 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 658 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 659 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 660 // is the case. LinuxThreads has a hard limit on max number of threads. 661 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 662 // On the other hand, NPTL does not have such a limit, sysconf() 663 // will return -1 and errno is not changed. Check if it is really NPTL. 664 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 665 strstr(str, "NPTL") && 666 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 667 free(str); 668 os::Linux::set_libpthread_version("linuxthreads"); 669 } else { 670 os::Linux::set_libpthread_version(str); 671 } 672 } else { 673 // glibc before 2.3.2 only has LinuxThreads. 674 os::Linux::set_libpthread_version("linuxthreads"); 675 } 676 677 if (strstr(libpthread_version(), "NPTL")) { 678 os::Linux::set_is_NPTL(); 679 } else { 680 os::Linux::set_is_LinuxThreads(); 681 } 682 683 // LinuxThreads have two flavors: floating-stack mode, which allows variable 684 // stack size; and fixed-stack mode. NPTL is always floating-stack. 685 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 686 os::Linux::set_is_floating_stack(); 687 } 688 } 689 690 ///////////////////////////////////////////////////////////////////////////// 691 // thread stack 692 693 // Force Linux kernel to expand current thread stack. If "bottom" is close 694 // to the stack guard, caller should block all signals. 695 // 696 // MAP_GROWSDOWN: 697 // A special mmap() flag that is used to implement thread stacks. It tells 698 // kernel that the memory region should extend downwards when needed. This 699 // allows early versions of LinuxThreads to only mmap the first few pages 700 // when creating a new thread. Linux kernel will automatically expand thread 701 // stack as needed (on page faults). 702 // 703 // However, because the memory region of a MAP_GROWSDOWN stack can grow on 704 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 705 // region, it's hard to tell if the fault is due to a legitimate stack 706 // access or because of reading/writing non-exist memory (e.g. buffer 707 // overrun). As a rule, if the fault happens below current stack pointer, 708 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the 709 // application (see Linux kernel fault.c). 710 // 711 // This Linux feature can cause SIGSEGV when VM bangs thread stack for 712 // stack overflow detection. 713 // 714 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 715 // not use this flag. However, the stack of initial thread is not created 716 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 717 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack 718 // and then attach the thread to JVM. 719 // 720 // To get around the problem and allow stack banging on Linux, we need to 721 // manually expand thread stack after receiving the SIGSEGV. 722 // 723 // There are two ways to expand thread stack to address "bottom", we used 724 // both of them in JVM before 1.5: 725 // 1. adjust stack pointer first so that it is below "bottom", and then 726 // touch "bottom" 727 // 2. mmap() the page in question 728 // 729 // Now alternate signal stack is gone, it's harder to use 2. For instance, 730 // if current sp is already near the lower end of page 101, and we need to 731 // call mmap() to map page 100, it is possible that part of the mmap() frame 732 // will be placed in page 100. When page 100 is mapped, it is zero-filled. 733 // That will destroy the mmap() frame and cause VM to crash. 734 // 735 // The following code works by adjusting sp first, then accessing the "bottom" 736 // page to force a page fault. Linux kernel will then automatically expand the 737 // stack mapping. 738 // 739 // _expand_stack_to() assumes its frame size is less than page size, which 740 // should always be true if the function is not inlined. 741 742 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 743 #define NOINLINE 744 #else 745 #define NOINLINE __attribute__ ((noinline)) 746 #endif 747 748 static void _expand_stack_to(address bottom) NOINLINE; 749 750 static void _expand_stack_to(address bottom) { 751 address sp; 752 size_t size; 753 volatile char *p; 754 755 // Adjust bottom to point to the largest address within the same page, it 756 // gives us a one-page buffer if alloca() allocates slightly more memory. 757 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 758 bottom += os::Linux::page_size() - 1; 759 760 // sp might be slightly above current stack pointer; if that's the case, we 761 // will alloca() a little more space than necessary, which is OK. Don't use 762 // os::current_stack_pointer(), as its result can be slightly below current 763 // stack pointer, causing us to not alloca enough to reach "bottom". 764 sp = (address)&sp; 765 766 if (sp > bottom) { 767 size = sp - bottom; 768 p = (volatile char *)alloca(size); 769 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 770 p[0] = '\0'; 771 } 772 } 773 774 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 775 assert(t!=NULL, "just checking"); 776 assert(t->osthread()->expanding_stack(), "expand should be set"); 777 assert(t->stack_base() != NULL, "stack_base was not initialized"); 778 779 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 780 sigset_t mask_all, old_sigset; 781 sigfillset(&mask_all); 782 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 783 _expand_stack_to(addr); 784 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 785 return true; 786 } 787 return false; 788 } 789 790 ////////////////////////////////////////////////////////////////////////////// 791 // create new thread 792 793 static address highest_vm_reserved_address(); 794 795 // check if it's safe to start a new thread 796 static bool _thread_safety_check(Thread* thread) { 797 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 798 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 799 // Heap is mmap'ed at lower end of memory space. Thread stacks are 800 // allocated (MAP_FIXED) from high address space. Every thread stack 801 // occupies a fixed size slot (usually 2Mbytes, but user can change 802 // it to other values if they rebuild LinuxThreads). 803 // 804 // Problem with MAP_FIXED is that mmap() can still succeed even part of 805 // the memory region has already been mmap'ed. That means if we have too 806 // many threads and/or very large heap, eventually thread stack will 807 // collide with heap. 808 // 809 // Here we try to prevent heap/stack collision by comparing current 810 // stack bottom with the highest address that has been mmap'ed by JVM 811 // plus a safety margin for memory maps created by native code. 812 // 813 // This feature can be disabled by setting ThreadSafetyMargin to 0 814 // 815 if (ThreadSafetyMargin > 0) { 816 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 817 818 // not safe if our stack extends below the safety margin 819 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 820 } else { 821 return true; 822 } 823 } else { 824 // Floating stack LinuxThreads or NPTL: 825 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 826 // there's not enough space left, pthread_create() will fail. If we come 827 // here, that means enough space has been reserved for stack. 828 return true; 829 } 830 } 831 832 // Thread start routine for all newly created threads 833 static void *java_start(Thread *thread) { 834 // Try to randomize the cache line index of hot stack frames. 835 // This helps when threads of the same stack traces evict each other's 836 // cache lines. The threads can be either from the same JVM instance, or 837 // from different JVM instances. The benefit is especially true for 838 // processors with hyperthreading technology. 839 static int counter = 0; 840 int pid = os::current_process_id(); 841 alloca(((pid ^ counter++) & 7) * 128); 842 843 ThreadLocalStorage::set_thread(thread); 844 845 OSThread* osthread = thread->osthread(); 846 Monitor* sync = osthread->startThread_lock(); 847 848 // non floating stack LinuxThreads needs extra check, see above 849 if (!_thread_safety_check(thread)) { 850 // notify parent thread 851 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 852 osthread->set_state(ZOMBIE); 853 sync->notify_all(); 854 return NULL; 855 } 856 857 // thread_id is kernel thread id (similar to Solaris LWP id) 858 osthread->set_thread_id(os::Linux::gettid()); 859 860 if (UseNUMA) { 861 int lgrp_id = os::numa_get_group_id(); 862 if (lgrp_id != -1) { 863 thread->set_lgrp_id(lgrp_id); 864 } 865 } 866 // initialize signal mask for this thread 867 os::Linux::hotspot_sigmask(thread); 868 869 // initialize floating point control register 870 os::Linux::init_thread_fpu_state(); 871 872 // handshaking with parent thread 873 { 874 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 875 876 // notify parent thread 877 osthread->set_state(INITIALIZED); 878 sync->notify_all(); 879 880 // wait until os::start_thread() 881 while (osthread->get_state() == INITIALIZED) { 882 sync->wait(Mutex::_no_safepoint_check_flag); 883 } 884 } 885 886 // call one more level start routine 887 thread->run(); 888 889 return 0; 890 } 891 892 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 893 assert(thread->osthread() == NULL, "caller responsible"); 894 895 // Allocate the OSThread object 896 OSThread* osthread = new OSThread(NULL, NULL); 897 if (osthread == NULL) { 898 return false; 899 } 900 901 // set the correct thread state 902 osthread->set_thread_type(thr_type); 903 904 // Initial state is ALLOCATED but not INITIALIZED 905 osthread->set_state(ALLOCATED); 906 907 thread->set_osthread(osthread); 908 909 // init thread attributes 910 pthread_attr_t attr; 911 pthread_attr_init(&attr); 912 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 913 914 // stack size 915 if (os::Linux::supports_variable_stack_size()) { 916 // calculate stack size if it's not specified by caller 917 if (stack_size == 0) { 918 stack_size = os::Linux::default_stack_size(thr_type); 919 920 switch (thr_type) { 921 case os::java_thread: 922 // Java threads use ThreadStackSize which default value can be 923 // changed with the flag -Xss 924 assert (JavaThread::stack_size_at_create() > 0, "this should be set"); 925 stack_size = JavaThread::stack_size_at_create(); 926 break; 927 case os::compiler_thread: 928 if (CompilerThreadStackSize > 0) { 929 stack_size = (size_t)(CompilerThreadStackSize * K); 930 break; 931 } // else fall through: 932 // use VMThreadStackSize if CompilerThreadStackSize is not defined 933 case os::vm_thread: 934 case os::pgc_thread: 935 case os::cgc_thread: 936 case os::watcher_thread: 937 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 938 break; 939 } 940 } 941 942 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 943 pthread_attr_setstacksize(&attr, stack_size); 944 } else { 945 // let pthread_create() pick the default value. 946 } 947 948 // glibc guard page 949 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 950 951 ThreadState state; 952 953 { 954 // Serialize thread creation if we are running with fixed stack LinuxThreads 955 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 956 if (lock) { 957 os::Linux::createThread_lock()->lock_without_safepoint_check(); 958 } 959 960 pthread_t tid; 961 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 962 963 pthread_attr_destroy(&attr); 964 965 if (ret != 0) { 966 if (PrintMiscellaneous && (Verbose || WizardMode)) { 967 perror("pthread_create()"); 968 } 969 // Need to clean up stuff we've allocated so far 970 thread->set_osthread(NULL); 971 delete osthread; 972 if (lock) os::Linux::createThread_lock()->unlock(); 973 return false; 974 } 975 976 // Store pthread info into the OSThread 977 osthread->set_pthread_id(tid); 978 979 // Wait until child thread is either initialized or aborted 980 { 981 Monitor* sync_with_child = osthread->startThread_lock(); 982 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 983 while ((state = osthread->get_state()) == ALLOCATED) { 984 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 985 } 986 } 987 988 if (lock) { 989 os::Linux::createThread_lock()->unlock(); 990 } 991 } 992 993 // Aborted due to thread limit being reached 994 if (state == ZOMBIE) { 995 thread->set_osthread(NULL); 996 delete osthread; 997 return false; 998 } 999 1000 // The thread is returned suspended (in state INITIALIZED), 1001 // and is started higher up in the call chain 1002 assert(state == INITIALIZED, "race condition"); 1003 return true; 1004 } 1005 1006 ///////////////////////////////////////////////////////////////////////////// 1007 // attach existing thread 1008 1009 // bootstrap the main thread 1010 bool os::create_main_thread(JavaThread* thread) { 1011 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 1012 return create_attached_thread(thread); 1013 } 1014 1015 bool os::create_attached_thread(JavaThread* thread) { 1016 #ifdef ASSERT 1017 thread->verify_not_published(); 1018 #endif 1019 1020 // Allocate the OSThread object 1021 OSThread* osthread = new OSThread(NULL, NULL); 1022 1023 if (osthread == NULL) { 1024 return false; 1025 } 1026 1027 // Store pthread info into the OSThread 1028 osthread->set_thread_id(os::Linux::gettid()); 1029 osthread->set_pthread_id(::pthread_self()); 1030 1031 // initialize floating point control register 1032 os::Linux::init_thread_fpu_state(); 1033 1034 // Initial thread state is RUNNABLE 1035 osthread->set_state(RUNNABLE); 1036 1037 thread->set_osthread(osthread); 1038 1039 if (UseNUMA) { 1040 int lgrp_id = os::numa_get_group_id(); 1041 if (lgrp_id != -1) { 1042 thread->set_lgrp_id(lgrp_id); 1043 } 1044 } 1045 1046 if (os::Linux::is_initial_thread()) { 1047 // If current thread is initial thread, its stack is mapped on demand, 1048 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 1049 // the entire stack region to avoid SEGV in stack banging. 1050 // It is also useful to get around the heap-stack-gap problem on SuSE 1051 // kernel (see 4821821 for details). We first expand stack to the top 1052 // of yellow zone, then enable stack yellow zone (order is significant, 1053 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1054 // is no gap between the last two virtual memory regions. 1055 1056 JavaThread *jt = (JavaThread *)thread; 1057 address addr = jt->stack_yellow_zone_base(); 1058 assert(addr != NULL, "initialization problem?"); 1059 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1060 1061 osthread->set_expanding_stack(); 1062 os::Linux::manually_expand_stack(jt, addr); 1063 osthread->clear_expanding_stack(); 1064 } 1065 1066 // initialize signal mask for this thread 1067 // and save the caller's signal mask 1068 os::Linux::hotspot_sigmask(thread); 1069 1070 return true; 1071 } 1072 1073 void os::pd_start_thread(Thread* thread) { 1074 OSThread * osthread = thread->osthread(); 1075 assert(osthread->get_state() != INITIALIZED, "just checking"); 1076 Monitor* sync_with_child = osthread->startThread_lock(); 1077 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1078 sync_with_child->notify(); 1079 } 1080 1081 // Free Linux resources related to the OSThread 1082 void os::free_thread(OSThread* osthread) { 1083 assert(osthread != NULL, "osthread not set"); 1084 1085 if (Thread::current()->osthread() == osthread) { 1086 // Restore caller's signal mask 1087 sigset_t sigmask = osthread->caller_sigmask(); 1088 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1089 } 1090 1091 delete osthread; 1092 } 1093 1094 ////////////////////////////////////////////////////////////////////////////// 1095 // thread local storage 1096 1097 int os::allocate_thread_local_storage() { 1098 pthread_key_t key; 1099 int rslt = pthread_key_create(&key, NULL); 1100 assert(rslt == 0, "cannot allocate thread local storage"); 1101 return (int)key; 1102 } 1103 1104 // Note: This is currently not used by VM, as we don't destroy TLS key 1105 // on VM exit. 1106 void os::free_thread_local_storage(int index) { 1107 int rslt = pthread_key_delete((pthread_key_t)index); 1108 assert(rslt == 0, "invalid index"); 1109 } 1110 1111 void os::thread_local_storage_at_put(int index, void* value) { 1112 int rslt = pthread_setspecific((pthread_key_t)index, value); 1113 assert(rslt == 0, "pthread_setspecific failed"); 1114 } 1115 1116 extern "C" Thread* get_thread() { 1117 return ThreadLocalStorage::thread(); 1118 } 1119 1120 ////////////////////////////////////////////////////////////////////////////// 1121 // initial thread 1122 1123 // Check if current thread is the initial thread, similar to Solaris thr_main. 1124 bool os::Linux::is_initial_thread(void) { 1125 char dummy; 1126 // If called before init complete, thread stack bottom will be null. 1127 // Can be called if fatal error occurs before initialization. 1128 if (initial_thread_stack_bottom() == NULL) return false; 1129 assert(initial_thread_stack_bottom() != NULL && 1130 initial_thread_stack_size() != 0, 1131 "os::init did not locate initial thread's stack region"); 1132 if ((address)&dummy >= initial_thread_stack_bottom() && 1133 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1134 return true; 1135 else return false; 1136 } 1137 1138 // Find the virtual memory area that contains addr 1139 static bool find_vma(address addr, address* vma_low, address* vma_high) { 1140 FILE *fp = fopen("/proc/self/maps", "r"); 1141 if (fp) { 1142 address low, high; 1143 while (!feof(fp)) { 1144 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1145 if (low <= addr && addr < high) { 1146 if (vma_low) *vma_low = low; 1147 if (vma_high) *vma_high = high; 1148 fclose (fp); 1149 return true; 1150 } 1151 } 1152 for (;;) { 1153 int ch = fgetc(fp); 1154 if (ch == EOF || ch == (int)'\n') break; 1155 } 1156 } 1157 fclose(fp); 1158 } 1159 return false; 1160 } 1161 1162 // Locate initial thread stack. This special handling of initial thread stack 1163 // is needed because pthread_getattr_np() on most (all?) Linux distros returns 1164 // bogus value for initial thread. 1165 void os::Linux::capture_initial_stack(size_t max_size) { 1166 // stack size is the easy part, get it from RLIMIT_STACK 1167 size_t stack_size; 1168 struct rlimit rlim; 1169 getrlimit(RLIMIT_STACK, &rlim); 1170 stack_size = rlim.rlim_cur; 1171 1172 // 6308388: a bug in ld.so will relocate its own .data section to the 1173 // lower end of primordial stack; reduce ulimit -s value a little bit 1174 // so we won't install guard page on ld.so's data section. 1175 stack_size -= 2 * page_size(); 1176 1177 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1178 // 7.1, in both cases we will get 2G in return value. 1179 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1180 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1181 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1182 // in case other parts in glibc still assumes 2M max stack size. 1183 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1184 #ifndef IA64 1185 if (stack_size > 2 * K * K) stack_size = 2 * K * K; 1186 #else 1187 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1188 if (stack_size > 4 * K * K) stack_size = 4 * K * K; 1189 #endif 1190 1191 // Try to figure out where the stack base (top) is. This is harder. 1192 // 1193 // When an application is started, glibc saves the initial stack pointer in 1194 // a global variable "__libc_stack_end", which is then used by system 1195 // libraries. __libc_stack_end should be pretty close to stack top. The 1196 // variable is available since the very early days. However, because it is 1197 // a private interface, it could disappear in the future. 1198 // 1199 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1200 // to __libc_stack_end, it is very close to stack top, but isn't the real 1201 // stack top. Note that /proc may not exist if VM is running as a chroot 1202 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1203 // /proc/<pid>/stat could change in the future (though unlikely). 1204 // 1205 // We try __libc_stack_end first. If that doesn't work, look for 1206 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1207 // as a hint, which should work well in most cases. 1208 1209 uintptr_t stack_start; 1210 1211 // try __libc_stack_end first 1212 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1213 if (p && *p) { 1214 stack_start = *p; 1215 } else { 1216 // see if we can get the start_stack field from /proc/self/stat 1217 FILE *fp; 1218 int pid; 1219 char state; 1220 int ppid; 1221 int pgrp; 1222 int session; 1223 int nr; 1224 int tpgrp; 1225 unsigned long flags; 1226 unsigned long minflt; 1227 unsigned long cminflt; 1228 unsigned long majflt; 1229 unsigned long cmajflt; 1230 unsigned long utime; 1231 unsigned long stime; 1232 long cutime; 1233 long cstime; 1234 long prio; 1235 long nice; 1236 long junk; 1237 long it_real; 1238 uintptr_t start; 1239 uintptr_t vsize; 1240 intptr_t rss; 1241 uintptr_t rsslim; 1242 uintptr_t scodes; 1243 uintptr_t ecode; 1244 int i; 1245 1246 // Figure what the primordial thread stack base is. Code is inspired 1247 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1248 // followed by command name surrounded by parentheses, state, etc. 1249 char stat[2048]; 1250 int statlen; 1251 1252 fp = fopen("/proc/self/stat", "r"); 1253 if (fp) { 1254 statlen = fread(stat, 1, 2047, fp); 1255 stat[statlen] = '\0'; 1256 fclose(fp); 1257 1258 // Skip pid and the command string. Note that we could be dealing with 1259 // weird command names, e.g. user could decide to rename java launcher 1260 // to "java 1.4.2 :)", then the stat file would look like 1261 // 1234 (java 1.4.2 :)) R ... ... 1262 // We don't really need to know the command string, just find the last 1263 // occurrence of ")" and then start parsing from there. See bug 4726580. 1264 char * s = strrchr(stat, ')'); 1265 1266 i = 0; 1267 if (s) { 1268 // Skip blank chars 1269 do s++; while (isspace(*s)); 1270 1271 #define _UFM UINTX_FORMAT 1272 #define _DFM INTX_FORMAT 1273 1274 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1275 /* 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 */ 1276 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, 1277 &state, /* 3 %c */ 1278 &ppid, /* 4 %d */ 1279 &pgrp, /* 5 %d */ 1280 &session, /* 6 %d */ 1281 &nr, /* 7 %d */ 1282 &tpgrp, /* 8 %d */ 1283 &flags, /* 9 %lu */ 1284 &minflt, /* 10 %lu */ 1285 &cminflt, /* 11 %lu */ 1286 &majflt, /* 12 %lu */ 1287 &cmajflt, /* 13 %lu */ 1288 &utime, /* 14 %lu */ 1289 &stime, /* 15 %lu */ 1290 &cutime, /* 16 %ld */ 1291 &cstime, /* 17 %ld */ 1292 &prio, /* 18 %ld */ 1293 &nice, /* 19 %ld */ 1294 &junk, /* 20 %ld */ 1295 &it_real, /* 21 %ld */ 1296 &start, /* 22 UINTX_FORMAT */ 1297 &vsize, /* 23 UINTX_FORMAT */ 1298 &rss, /* 24 INTX_FORMAT */ 1299 &rsslim, /* 25 UINTX_FORMAT */ 1300 &scodes, /* 26 UINTX_FORMAT */ 1301 &ecode, /* 27 UINTX_FORMAT */ 1302 &stack_start); /* 28 UINTX_FORMAT */ 1303 } 1304 1305 #undef _UFM 1306 #undef _DFM 1307 1308 if (i != 28 - 2) { 1309 assert(false, "Bad conversion from /proc/self/stat"); 1310 // product mode - assume we are the initial thread, good luck in the 1311 // embedded case. 1312 warning("Can't detect initial thread stack location - bad conversion"); 1313 stack_start = (uintptr_t) &rlim; 1314 } 1315 } else { 1316 // For some reason we can't open /proc/self/stat (for example, running on 1317 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1318 // most cases, so don't abort: 1319 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1320 stack_start = (uintptr_t) &rlim; 1321 } 1322 } 1323 1324 // Now we have a pointer (stack_start) very close to the stack top, the 1325 // next thing to do is to figure out the exact location of stack top. We 1326 // can find out the virtual memory area that contains stack_start by 1327 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1328 // and its upper limit is the real stack top. (again, this would fail if 1329 // running inside chroot, because /proc may not exist.) 1330 1331 uintptr_t stack_top; 1332 address low, high; 1333 if (find_vma((address)stack_start, &low, &high)) { 1334 // success, "high" is the true stack top. (ignore "low", because initial 1335 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1336 stack_top = (uintptr_t)high; 1337 } else { 1338 // failed, likely because /proc/self/maps does not exist 1339 warning("Can't detect initial thread stack location - find_vma failed"); 1340 // best effort: stack_start is normally within a few pages below the real 1341 // stack top, use it as stack top, and reduce stack size so we won't put 1342 // guard page outside stack. 1343 stack_top = stack_start; 1344 stack_size -= 16 * page_size(); 1345 } 1346 1347 // stack_top could be partially down the page so align it 1348 stack_top = align_size_up(stack_top, page_size()); 1349 1350 if (max_size && stack_size > max_size) { 1351 _initial_thread_stack_size = max_size; 1352 } else { 1353 _initial_thread_stack_size = stack_size; 1354 } 1355 1356 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1357 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1358 } 1359 1360 //////////////////////////////////////////////////////////////////////////////// 1361 // time support 1362 1363 // Time since start-up in seconds to a fine granularity. 1364 // Used by VMSelfDestructTimer and the MemProfiler. 1365 double os::elapsedTime() { 1366 1367 return (double)(os::elapsed_counter()) * 0.000001; 1368 } 1369 1370 jlong os::elapsed_counter() { 1371 timeval time; 1372 int status = gettimeofday(&time, NULL); 1373 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count; 1374 } 1375 1376 jlong os::elapsed_frequency() { 1377 return (1000 * 1000); 1378 } 1379 1380 // For now, we say that linux does not support vtime. I have no idea 1381 // whether it can actually be made to (DLD, 9/13/05). 1382 1383 bool os::supports_vtime() { return false; } 1384 bool os::enable_vtime() { return false; } 1385 bool os::vtime_enabled() { return false; } 1386 double os::elapsedVTime() { 1387 // better than nothing, but not much 1388 return elapsedTime(); 1389 } 1390 1391 jlong os::javaTimeMillis() { 1392 timeval time; 1393 int status = gettimeofday(&time, NULL); 1394 assert(status != -1, "linux error"); 1395 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1396 } 1397 1398 #ifndef CLOCK_MONOTONIC 1399 #define CLOCK_MONOTONIC (1) 1400 #endif 1401 1402 void os::Linux::clock_init() { 1403 // we do dlopen's in this particular order due to bug in linux 1404 // dynamical loader (see 6348968) leading to crash on exit 1405 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1406 if (handle == NULL) { 1407 handle = dlopen("librt.so", RTLD_LAZY); 1408 } 1409 1410 if (handle) { 1411 int (*clock_getres_func)(clockid_t, struct timespec*) = 1412 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1413 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1414 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1415 if (clock_getres_func && clock_gettime_func) { 1416 // See if monotonic clock is supported by the kernel. Note that some 1417 // early implementations simply return kernel jiffies (updated every 1418 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1419 // for nano time (though the monotonic property is still nice to have). 1420 // It's fixed in newer kernels, however clock_getres() still returns 1421 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1422 // resolution for now. Hopefully as people move to new kernels, this 1423 // won't be a problem. 1424 struct timespec res; 1425 struct timespec tp; 1426 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1427 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1428 // yes, monotonic clock is supported 1429 _clock_gettime = clock_gettime_func; 1430 } else { 1431 // close librt if there is no monotonic clock 1432 dlclose(handle); 1433 } 1434 } 1435 } 1436 } 1437 1438 #ifndef SYS_clock_getres 1439 1440 #if defined(IA32) || defined(AMD64) 1441 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1442 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1443 #else 1444 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1445 #define sys_clock_getres(x,y) -1 1446 #endif 1447 1448 #else 1449 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1450 #endif 1451 1452 void os::Linux::fast_thread_clock_init() { 1453 if (!UseLinuxPosixThreadCPUClocks) { 1454 return; 1455 } 1456 clockid_t clockid; 1457 struct timespec tp; 1458 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1459 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1460 1461 // Switch to using fast clocks for thread cpu time if 1462 // the sys_clock_getres() returns 0 error code. 1463 // Note, that some kernels may support the current thread 1464 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1465 // returned by the pthread_getcpuclockid(). 1466 // If the fast Posix clocks are supported then the sys_clock_getres() 1467 // must return at least tp.tv_sec == 0 which means a resolution 1468 // better than 1 sec. This is extra check for reliability. 1469 1470 if(pthread_getcpuclockid_func && 1471 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1472 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1473 1474 _supports_fast_thread_cpu_time = true; 1475 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1476 } 1477 } 1478 1479 jlong os::javaTimeNanos() { 1480 if (Linux::supports_monotonic_clock()) { 1481 struct timespec tp; 1482 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1483 assert(status == 0, "gettime error"); 1484 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1485 return result; 1486 } else { 1487 timeval time; 1488 int status = gettimeofday(&time, NULL); 1489 assert(status != -1, "linux error"); 1490 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1491 return 1000 * usecs; 1492 } 1493 } 1494 1495 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1496 if (Linux::supports_monotonic_clock()) { 1497 info_ptr->max_value = ALL_64_BITS; 1498 1499 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1500 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1501 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1502 } else { 1503 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1504 info_ptr->max_value = ALL_64_BITS; 1505 1506 // gettimeofday is a real time clock so it skips 1507 info_ptr->may_skip_backward = true; 1508 info_ptr->may_skip_forward = true; 1509 } 1510 1511 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1512 } 1513 1514 // Return the real, user, and system times in seconds from an 1515 // arbitrary fixed point in the past. 1516 bool os::getTimesSecs(double* process_real_time, 1517 double* process_user_time, 1518 double* process_system_time) { 1519 struct tms ticks; 1520 clock_t real_ticks = times(&ticks); 1521 1522 if (real_ticks == (clock_t) (-1)) { 1523 return false; 1524 } else { 1525 double ticks_per_second = (double) clock_tics_per_sec; 1526 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1527 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1528 *process_real_time = ((double) real_ticks) / ticks_per_second; 1529 1530 return true; 1531 } 1532 } 1533 1534 1535 char * os::local_time_string(char *buf, size_t buflen) { 1536 struct tm t; 1537 time_t long_time; 1538 time(&long_time); 1539 localtime_r(&long_time, &t); 1540 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1541 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1542 t.tm_hour, t.tm_min, t.tm_sec); 1543 return buf; 1544 } 1545 1546 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1547 return localtime_r(clock, res); 1548 } 1549 1550 //////////////////////////////////////////////////////////////////////////////// 1551 // runtime exit support 1552 1553 // Note: os::shutdown() might be called very early during initialization, or 1554 // called from signal handler. Before adding something to os::shutdown(), make 1555 // sure it is async-safe and can handle partially initialized VM. 1556 void os::shutdown() { 1557 1558 // allow PerfMemory to attempt cleanup of any persistent resources 1559 perfMemory_exit(); 1560 1561 // needs to remove object in file system 1562 AttachListener::abort(); 1563 1564 // flush buffered output, finish log files 1565 ostream_abort(); 1566 1567 // Check for abort hook 1568 abort_hook_t abort_hook = Arguments::abort_hook(); 1569 if (abort_hook != NULL) { 1570 abort_hook(); 1571 } 1572 1573 } 1574 1575 // Note: os::abort() might be called very early during initialization, or 1576 // called from signal handler. Before adding something to os::abort(), make 1577 // sure it is async-safe and can handle partially initialized VM. 1578 void os::abort(bool dump_core) { 1579 os::shutdown(); 1580 if (dump_core) { 1581 #ifndef PRODUCT 1582 fdStream out(defaultStream::output_fd()); 1583 out.print_raw("Current thread is "); 1584 char buf[16]; 1585 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1586 out.print_raw_cr(buf); 1587 out.print_raw_cr("Dumping core ..."); 1588 #endif 1589 ::abort(); // dump core 1590 } 1591 1592 ::exit(1); 1593 } 1594 1595 // Die immediately, no exit hook, no abort hook, no cleanup. 1596 void os::die() { 1597 // _exit() on LinuxThreads only kills current thread 1598 ::abort(); 1599 } 1600 1601 // unused on linux for now. 1602 void os::set_error_file(const char *logfile) {} 1603 1604 1605 // This method is a copy of JDK's sysGetLastErrorString 1606 // from src/solaris/hpi/src/system_md.c 1607 1608 size_t os::lasterror(char *buf, size_t len) { 1609 1610 if (errno == 0) return 0; 1611 1612 const char *s = ::strerror(errno); 1613 size_t n = ::strlen(s); 1614 if (n >= len) { 1615 n = len - 1; 1616 } 1617 ::strncpy(buf, s, n); 1618 buf[n] = '\0'; 1619 return n; 1620 } 1621 1622 intx os::current_thread_id() { return (intx)pthread_self(); } 1623 int os::current_process_id() { 1624 1625 // Under the old linux thread library, linux gives each thread 1626 // its own process id. Because of this each thread will return 1627 // a different pid if this method were to return the result 1628 // of getpid(2). Linux provides no api that returns the pid 1629 // of the launcher thread for the vm. This implementation 1630 // returns a unique pid, the pid of the launcher thread 1631 // that starts the vm 'process'. 1632 1633 // Under the NPTL, getpid() returns the same pid as the 1634 // launcher thread rather than a unique pid per thread. 1635 // Use gettid() if you want the old pre NPTL behaviour. 1636 1637 // if you are looking for the result of a call to getpid() that 1638 // returns a unique pid for the calling thread, then look at the 1639 // OSThread::thread_id() method in osThread_linux.hpp file 1640 1641 return (int)(_initial_pid ? _initial_pid : getpid()); 1642 } 1643 1644 // DLL functions 1645 1646 const char* os::dll_file_extension() { return ".so"; } 1647 1648 // This must be hard coded because it's the system's temporary 1649 // directory not the java application's temp directory, ala java.io.tmpdir. 1650 const char* os::get_temp_directory() { return "/tmp"; } 1651 1652 static bool file_exists(const char* filename) { 1653 struct stat statbuf; 1654 if (filename == NULL || strlen(filename) == 0) { 1655 return false; 1656 } 1657 return os::stat(filename, &statbuf) == 0; 1658 } 1659 1660 void os::dll_build_name(char* buffer, size_t buflen, 1661 const char* pname, const char* fname) { 1662 // Copied from libhpi 1663 const size_t pnamelen = pname ? strlen(pname) : 0; 1664 1665 // Quietly truncate on buffer overflow. Should be an error. 1666 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1667 *buffer = '\0'; 1668 return; 1669 } 1670 1671 if (pnamelen == 0) { 1672 snprintf(buffer, buflen, "lib%s.so", fname); 1673 } else if (strchr(pname, *os::path_separator()) != NULL) { 1674 int n; 1675 char** pelements = split_path(pname, &n); 1676 for (int i = 0 ; i < n ; i++) { 1677 // Really shouldn't be NULL, but check can't hurt 1678 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1679 continue; // skip the empty path values 1680 } 1681 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1682 if (file_exists(buffer)) { 1683 break; 1684 } 1685 } 1686 // release the storage 1687 for (int i = 0 ; i < n ; i++) { 1688 if (pelements[i] != NULL) { 1689 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1690 } 1691 } 1692 if (pelements != NULL) { 1693 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1694 } 1695 } else { 1696 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1697 } 1698 } 1699 1700 const char* os::get_current_directory(char *buf, int buflen) { 1701 return getcwd(buf, buflen); 1702 } 1703 1704 // check if addr is inside libjvm[_g].so 1705 bool os::address_is_in_vm(address addr) { 1706 static address libjvm_base_addr; 1707 Dl_info dlinfo; 1708 1709 if (libjvm_base_addr == NULL) { 1710 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo); 1711 libjvm_base_addr = (address)dlinfo.dli_fbase; 1712 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1713 } 1714 1715 if (dladdr((void *)addr, &dlinfo)) { 1716 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1717 } 1718 1719 return false; 1720 } 1721 1722 bool os::dll_address_to_function_name(address addr, char *buf, 1723 int buflen, int *offset) { 1724 Dl_info dlinfo; 1725 1726 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) { 1727 if (buf != NULL) { 1728 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1729 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1730 } 1731 } 1732 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1733 return true; 1734 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) { 1735 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1736 buf, buflen, offset, dlinfo.dli_fname)) { 1737 return true; 1738 } 1739 } 1740 1741 if (buf != NULL) buf[0] = '\0'; 1742 if (offset != NULL) *offset = -1; 1743 return false; 1744 } 1745 1746 struct _address_to_library_name { 1747 address addr; // input : memory address 1748 size_t buflen; // size of fname 1749 char* fname; // output: library name 1750 address base; // library base addr 1751 }; 1752 1753 static int address_to_library_name_callback(struct dl_phdr_info *info, 1754 size_t size, void *data) { 1755 int i; 1756 bool found = false; 1757 address libbase = NULL; 1758 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1759 1760 // iterate through all loadable segments 1761 for (i = 0; i < info->dlpi_phnum; i++) { 1762 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1763 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1764 // base address of a library is the lowest address of its loaded 1765 // segments. 1766 if (libbase == NULL || libbase > segbase) { 1767 libbase = segbase; 1768 } 1769 // see if 'addr' is within current segment 1770 if (segbase <= d->addr && 1771 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1772 found = true; 1773 } 1774 } 1775 } 1776 1777 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1778 // so dll_address_to_library_name() can fall through to use dladdr() which 1779 // can figure out executable name from argv[0]. 1780 if (found && info->dlpi_name && info->dlpi_name[0]) { 1781 d->base = libbase; 1782 if (d->fname) { 1783 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1784 } 1785 return 1; 1786 } 1787 return 0; 1788 } 1789 1790 bool os::dll_address_to_library_name(address addr, char* buf, 1791 int buflen, int* offset) { 1792 Dl_info dlinfo; 1793 struct _address_to_library_name data; 1794 1795 // There is a bug in old glibc dladdr() implementation that it could resolve 1796 // to wrong library name if the .so file has a base address != NULL. Here 1797 // we iterate through the program headers of all loaded libraries to find 1798 // out which library 'addr' really belongs to. This workaround can be 1799 // removed once the minimum requirement for glibc is moved to 2.3.x. 1800 data.addr = addr; 1801 data.fname = buf; 1802 data.buflen = buflen; 1803 data.base = NULL; 1804 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1805 1806 if (rslt) { 1807 // buf already contains library name 1808 if (offset) *offset = addr - data.base; 1809 return true; 1810 } else if (dladdr((void*)addr, &dlinfo)){ 1811 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1812 if (offset) *offset = addr - (address)dlinfo.dli_fbase; 1813 return true; 1814 } else { 1815 if (buf) buf[0] = '\0'; 1816 if (offset) *offset = -1; 1817 return false; 1818 } 1819 } 1820 1821 // Loads .dll/.so and 1822 // in case of error it checks if .dll/.so was built for the 1823 // same architecture as Hotspot is running on 1824 1825 1826 // Remember the stack's state. The Linux dynamic linker will change 1827 // the stack to 'executable' at most once, so we must safepoint only once. 1828 bool os::Linux::_stack_is_executable = false; 1829 1830 // VM operation that loads a library. This is necessary if stack protection 1831 // of the Java stacks can be lost during loading the library. If we 1832 // do not stop the Java threads, they can stack overflow before the stacks 1833 // are protected again. 1834 class VM_LinuxDllLoad: public VM_Operation { 1835 private: 1836 const char *_filename; 1837 char *_ebuf; 1838 int _ebuflen; 1839 void *_lib; 1840 public: 1841 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1842 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1843 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1844 void doit() { 1845 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1846 os::Linux::_stack_is_executable = true; 1847 } 1848 void* loaded_library() { return _lib; } 1849 }; 1850 1851 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1852 { 1853 void * result = NULL; 1854 bool load_attempted = false; 1855 1856 // Check whether the library to load might change execution rights 1857 // of the stack. If they are changed, the protection of the stack 1858 // guard pages will be lost. We need a safepoint to fix this. 1859 // 1860 // See Linux man page execstack(8) for more info. 1861 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1862 ElfFile ef(filename); 1863 if (!ef.specifies_noexecstack()) { 1864 if (!is_init_completed()) { 1865 os::Linux::_stack_is_executable = true; 1866 // This is OK - No Java threads have been created yet, and hence no 1867 // stack guard pages to fix. 1868 // 1869 // This should happen only when you are building JDK7 using a very 1870 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1871 // 1872 // Dynamic loader will make all stacks executable after 1873 // this function returns, and will not do that again. 1874 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1875 } else { 1876 warning("You have loaded library %s which might have disabled stack guard. " 1877 "The VM will try to fix the stack guard now.\n" 1878 "It's highly recommended that you fix the library with " 1879 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1880 filename); 1881 1882 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1883 JavaThread *jt = JavaThread::current(); 1884 if (jt->thread_state() != _thread_in_native) { 1885 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1886 // that requires ExecStack. Cannot enter safe point. Let's give up. 1887 warning("Unable to fix stack guard. Giving up."); 1888 } else { 1889 if (!LoadExecStackDllInVMThread) { 1890 // This is for the case where the DLL has an static 1891 // constructor function that executes JNI code. We cannot 1892 // load such DLLs in the VMThread. 1893 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1894 } 1895 1896 ThreadInVMfromNative tiv(jt); 1897 debug_only(VMNativeEntryWrapper vew;) 1898 1899 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1900 VMThread::execute(&op); 1901 if (LoadExecStackDllInVMThread) { 1902 result = op.loaded_library(); 1903 } 1904 load_attempted = true; 1905 } 1906 } 1907 } 1908 } 1909 1910 if (!load_attempted) { 1911 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1912 } 1913 1914 if (result != NULL) { 1915 // Successful loading 1916 return result; 1917 } 1918 1919 Elf32_Ehdr elf_head; 1920 int diag_msg_max_length=ebuflen-strlen(ebuf); 1921 char* diag_msg_buf=ebuf+strlen(ebuf); 1922 1923 if (diag_msg_max_length==0) { 1924 // No more space in ebuf for additional diagnostics message 1925 return NULL; 1926 } 1927 1928 1929 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1930 1931 if (file_descriptor < 0) { 1932 // Can't open library, report dlerror() message 1933 return NULL; 1934 } 1935 1936 bool failed_to_read_elf_head= 1937 (sizeof(elf_head)!= 1938 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1939 1940 ::close(file_descriptor); 1941 if (failed_to_read_elf_head) { 1942 // file i/o error - report dlerror() msg 1943 return NULL; 1944 } 1945 1946 typedef struct { 1947 Elf32_Half code; // Actual value as defined in elf.h 1948 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1949 char elf_class; // 32 or 64 bit 1950 char endianess; // MSB or LSB 1951 char* name; // String representation 1952 } arch_t; 1953 1954 #ifndef EM_486 1955 #define EM_486 6 /* Intel 80486 */ 1956 #endif 1957 1958 static const arch_t arch_array[]={ 1959 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1960 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1961 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1962 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1963 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1964 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1965 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1966 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1967 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1968 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1969 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1970 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1971 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1972 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1973 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1974 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1975 }; 1976 1977 #if (defined IA32) 1978 static Elf32_Half running_arch_code=EM_386; 1979 #elif (defined AMD64) 1980 static Elf32_Half running_arch_code=EM_X86_64; 1981 #elif (defined IA64) 1982 static Elf32_Half running_arch_code=EM_IA_64; 1983 #elif (defined __sparc) && (defined _LP64) 1984 static Elf32_Half running_arch_code=EM_SPARCV9; 1985 #elif (defined __sparc) && (!defined _LP64) 1986 static Elf32_Half running_arch_code=EM_SPARC; 1987 #elif (defined __powerpc64__) 1988 static Elf32_Half running_arch_code=EM_PPC64; 1989 #elif (defined __powerpc__) 1990 static Elf32_Half running_arch_code=EM_PPC; 1991 #elif (defined ARM) 1992 static Elf32_Half running_arch_code=EM_ARM; 1993 #elif (defined S390) 1994 static Elf32_Half running_arch_code=EM_S390; 1995 #elif (defined ALPHA) 1996 static Elf32_Half running_arch_code=EM_ALPHA; 1997 #elif (defined MIPSEL) 1998 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1999 #elif (defined PARISC) 2000 static Elf32_Half running_arch_code=EM_PARISC; 2001 #elif (defined MIPS) 2002 static Elf32_Half running_arch_code=EM_MIPS; 2003 #elif (defined M68K) 2004 static Elf32_Half running_arch_code=EM_68K; 2005 #else 2006 #error Method os::dll_load requires that one of following is defined:\ 2007 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 2008 #endif 2009 2010 // Identify compatability class for VM's architecture and library's architecture 2011 // Obtain string descriptions for architectures 2012 2013 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 2014 int running_arch_index=-1; 2015 2016 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 2017 if (running_arch_code == arch_array[i].code) { 2018 running_arch_index = i; 2019 } 2020 if (lib_arch.code == arch_array[i].code) { 2021 lib_arch.compat_class = arch_array[i].compat_class; 2022 lib_arch.name = arch_array[i].name; 2023 } 2024 } 2025 2026 assert(running_arch_index != -1, 2027 "Didn't find running architecture code (running_arch_code) in arch_array"); 2028 if (running_arch_index == -1) { 2029 // Even though running architecture detection failed 2030 // we may still continue with reporting dlerror() message 2031 return NULL; 2032 } 2033 2034 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 2035 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 2036 return NULL; 2037 } 2038 2039 #ifndef S390 2040 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 2041 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 2042 return NULL; 2043 } 2044 #endif // !S390 2045 2046 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2047 if ( lib_arch.name!=NULL ) { 2048 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2049 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2050 lib_arch.name, arch_array[running_arch_index].name); 2051 } else { 2052 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2053 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2054 lib_arch.code, 2055 arch_array[running_arch_index].name); 2056 } 2057 } 2058 2059 return NULL; 2060 } 2061 2062 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) { 2063 void * result = ::dlopen(filename, RTLD_LAZY); 2064 if (result == NULL) { 2065 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2066 ebuf[ebuflen-1] = '\0'; 2067 } 2068 return result; 2069 } 2070 2071 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) { 2072 void * result = NULL; 2073 if (LoadExecStackDllInVMThread) { 2074 result = dlopen_helper(filename, ebuf, ebuflen); 2075 } 2076 2077 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2078 // library that requires an executable stack, or which does not have this 2079 // stack attribute set, dlopen changes the stack attribute to executable. The 2080 // read protection of the guard pages gets lost. 2081 // 2082 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2083 // may have been queued at the same time. 2084 2085 if (!_stack_is_executable) { 2086 JavaThread *jt = Threads::first(); 2087 2088 while (jt) { 2089 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2090 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2091 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2092 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2093 warning("Attempt to reguard stack yellow zone failed."); 2094 } 2095 } 2096 jt = jt->next(); 2097 } 2098 } 2099 2100 return result; 2101 } 2102 2103 /* 2104 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2105 * chances are you might want to run the generated bits against glibc-2.0 2106 * libdl.so, so always use locking for any version of glibc. 2107 */ 2108 void* os::dll_lookup(void* handle, const char* name) { 2109 pthread_mutex_lock(&dl_mutex); 2110 void* res = dlsym(handle, name); 2111 pthread_mutex_unlock(&dl_mutex); 2112 return res; 2113 } 2114 2115 2116 static bool _print_ascii_file(const char* filename, outputStream* st) { 2117 int fd = ::open(filename, O_RDONLY); 2118 if (fd == -1) { 2119 return false; 2120 } 2121 2122 char buf[32]; 2123 int bytes; 2124 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2125 st->print_raw(buf, bytes); 2126 } 2127 2128 ::close(fd); 2129 2130 return true; 2131 } 2132 2133 void os::print_dll_info(outputStream *st) { 2134 st->print_cr("Dynamic libraries:"); 2135 2136 char fname[32]; 2137 pid_t pid = os::Linux::gettid(); 2138 2139 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2140 2141 if (!_print_ascii_file(fname, st)) { 2142 st->print("Can not get library information for pid = %d\n", pid); 2143 } 2144 } 2145 2146 void os::print_os_info_brief(outputStream* st) { 2147 os::Linux::print_distro_info(st); 2148 2149 os::Posix::print_uname_info(st); 2150 2151 os::Linux::print_libversion_info(st); 2152 2153 } 2154 2155 void os::print_os_info(outputStream* st) { 2156 st->print("OS:"); 2157 2158 os::Linux::print_distro_info(st); 2159 2160 os::Posix::print_uname_info(st); 2161 2162 // Print warning if unsafe chroot environment detected 2163 if (unsafe_chroot_detected) { 2164 st->print("WARNING!! "); 2165 st->print_cr(unstable_chroot_error); 2166 } 2167 2168 os::Linux::print_libversion_info(st); 2169 2170 os::Posix::print_rlimit_info(st); 2171 2172 os::Posix::print_load_average(st); 2173 2174 os::Linux::print_full_memory_info(st); 2175 } 2176 2177 // Try to identify popular distros. 2178 // Most Linux distributions have /etc/XXX-release file, which contains 2179 // the OS version string. Some have more than one /etc/XXX-release file 2180 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.), 2181 // so the order is important. 2182 void os::Linux::print_distro_info(outputStream* st) { 2183 if (!_print_ascii_file("/etc/mandrake-release", st) && 2184 !_print_ascii_file("/etc/sun-release", st) && 2185 !_print_ascii_file("/etc/redhat-release", st) && 2186 !_print_ascii_file("/etc/SuSE-release", st) && 2187 !_print_ascii_file("/etc/turbolinux-release", st) && 2188 !_print_ascii_file("/etc/gentoo-release", st) && 2189 !_print_ascii_file("/etc/debian_version", st) && 2190 !_print_ascii_file("/etc/ltib-release", st) && 2191 !_print_ascii_file("/etc/angstrom-version", st)) { 2192 st->print("Linux"); 2193 } 2194 st->cr(); 2195 } 2196 2197 void os::Linux::print_libversion_info(outputStream* st) { 2198 // libc, pthread 2199 st->print("libc:"); 2200 st->print(os::Linux::glibc_version()); st->print(" "); 2201 st->print(os::Linux::libpthread_version()); st->print(" "); 2202 if (os::Linux::is_LinuxThreads()) { 2203 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2204 } 2205 st->cr(); 2206 } 2207 2208 void os::Linux::print_full_memory_info(outputStream* st) { 2209 st->print("\n/proc/meminfo:\n"); 2210 _print_ascii_file("/proc/meminfo", st); 2211 st->cr(); 2212 } 2213 2214 void os::print_memory_info(outputStream* st) { 2215 2216 st->print("Memory:"); 2217 st->print(" %dk page", os::vm_page_size()>>10); 2218 2219 // values in struct sysinfo are "unsigned long" 2220 struct sysinfo si; 2221 sysinfo(&si); 2222 2223 st->print(", physical " UINT64_FORMAT "k", 2224 os::physical_memory() >> 10); 2225 st->print("(" UINT64_FORMAT "k free)", 2226 os::available_memory() >> 10); 2227 st->print(", swap " UINT64_FORMAT "k", 2228 ((jlong)si.totalswap * si.mem_unit) >> 10); 2229 st->print("(" UINT64_FORMAT "k free)", 2230 ((jlong)si.freeswap * si.mem_unit) >> 10); 2231 st->cr(); 2232 } 2233 2234 void os::pd_print_cpu_info(outputStream* st) { 2235 st->print("\n/proc/cpuinfo:\n"); 2236 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2237 st->print(" <Not Available>"); 2238 } 2239 st->cr(); 2240 } 2241 2242 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific 2243 // but they're the same for all the linux arch that we support 2244 // and they're the same for solaris but there's no common place to put this. 2245 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR", 2246 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG", 2247 "ILL_COPROC", "ILL_BADSTK" }; 2248 2249 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV", 2250 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES", 2251 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" }; 2252 2253 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" }; 2254 2255 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" }; 2256 2257 void os::print_siginfo(outputStream* st, void* siginfo) { 2258 st->print("siginfo:"); 2259 2260 const int buflen = 100; 2261 char buf[buflen]; 2262 siginfo_t *si = (siginfo_t*)siginfo; 2263 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen)); 2264 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) { 2265 st->print("si_errno=%s", buf); 2266 } else { 2267 st->print("si_errno=%d", si->si_errno); 2268 } 2269 const int c = si->si_code; 2270 assert(c > 0, "unexpected si_code"); 2271 switch (si->si_signo) { 2272 case SIGILL: 2273 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]); 2274 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2275 break; 2276 case SIGFPE: 2277 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]); 2278 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2279 break; 2280 case SIGSEGV: 2281 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]); 2282 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2283 break; 2284 case SIGBUS: 2285 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]); 2286 st->print(", si_addr=" PTR_FORMAT, si->si_addr); 2287 break; 2288 default: 2289 st->print(", si_code=%d", si->si_code); 2290 // no si_addr 2291 } 2292 2293 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2294 UseSharedSpaces) { 2295 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2296 if (mapinfo->is_in_shared_space(si->si_addr)) { 2297 st->print("\n\nError accessing class data sharing archive." \ 2298 " Mapped file inaccessible during execution, " \ 2299 " possible disk/network problem."); 2300 } 2301 } 2302 st->cr(); 2303 } 2304 2305 2306 static void print_signal_handler(outputStream* st, int sig, 2307 char* buf, size_t buflen); 2308 2309 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2310 st->print_cr("Signal Handlers:"); 2311 print_signal_handler(st, SIGSEGV, buf, buflen); 2312 print_signal_handler(st, SIGBUS , buf, buflen); 2313 print_signal_handler(st, SIGFPE , buf, buflen); 2314 print_signal_handler(st, SIGPIPE, buf, buflen); 2315 print_signal_handler(st, SIGXFSZ, buf, buflen); 2316 print_signal_handler(st, SIGILL , buf, buflen); 2317 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2318 print_signal_handler(st, SR_signum, buf, buflen); 2319 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2320 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2321 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2322 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2323 } 2324 2325 static char saved_jvm_path[MAXPATHLEN] = {0}; 2326 2327 // Find the full path to the current module, libjvm.so or libjvm_g.so 2328 void os::jvm_path(char *buf, jint buflen) { 2329 // Error checking. 2330 if (buflen < MAXPATHLEN) { 2331 assert(false, "must use a large-enough buffer"); 2332 buf[0] = '\0'; 2333 return; 2334 } 2335 // Lazy resolve the path to current module. 2336 if (saved_jvm_path[0] != 0) { 2337 strcpy(buf, saved_jvm_path); 2338 return; 2339 } 2340 2341 char dli_fname[MAXPATHLEN]; 2342 bool ret = dll_address_to_library_name( 2343 CAST_FROM_FN_PTR(address, os::jvm_path), 2344 dli_fname, sizeof(dli_fname), NULL); 2345 assert(ret != 0, "cannot locate libjvm"); 2346 char *rp = realpath(dli_fname, buf); 2347 if (rp == NULL) 2348 return; 2349 2350 if (Arguments::created_by_gamma_launcher()) { 2351 // Support for the gamma launcher. Typical value for buf is 2352 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at 2353 // the right place in the string, then assume we are installed in a JDK and 2354 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix 2355 // up the path so it looks like libjvm.so is installed there (append a 2356 // fake suffix hotspot/libjvm.so). 2357 const char *p = buf + strlen(buf) - 1; 2358 for (int count = 0; p > buf && count < 5; ++count) { 2359 for (--p; p > buf && *p != '/'; --p) 2360 /* empty */ ; 2361 } 2362 2363 if (strncmp(p, "/jre/lib/", 9) != 0) { 2364 // Look for JAVA_HOME in the environment. 2365 char* java_home_var = ::getenv("JAVA_HOME"); 2366 if (java_home_var != NULL && java_home_var[0] != 0) { 2367 char* jrelib_p; 2368 int len; 2369 2370 // Check the current module name "libjvm.so" or "libjvm_g.so". 2371 p = strrchr(buf, '/'); 2372 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2373 p = strstr(p, "_g") ? "_g" : ""; 2374 2375 rp = realpath(java_home_var, buf); 2376 if (rp == NULL) 2377 return; 2378 2379 // determine if this is a legacy image or modules image 2380 // modules image doesn't have "jre" subdirectory 2381 len = strlen(buf); 2382 jrelib_p = buf + len; 2383 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2384 if (0 != access(buf, F_OK)) { 2385 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2386 } 2387 2388 if (0 == access(buf, F_OK)) { 2389 // Use current module name "libjvm[_g].so" instead of 2390 // "libjvm"debug_only("_g")".so" since for fastdebug version 2391 // we should have "libjvm.so" but debug_only("_g") adds "_g"! 2392 len = strlen(buf); 2393 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p); 2394 } else { 2395 // Go back to path of .so 2396 rp = realpath(dli_fname, buf); 2397 if (rp == NULL) 2398 return; 2399 } 2400 } 2401 } 2402 } 2403 2404 strcpy(saved_jvm_path, buf); 2405 } 2406 2407 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2408 // no prefix required, not even "_" 2409 } 2410 2411 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2412 // no suffix required 2413 } 2414 2415 //////////////////////////////////////////////////////////////////////////////// 2416 // sun.misc.Signal support 2417 2418 static volatile jint sigint_count = 0; 2419 2420 static void 2421 UserHandler(int sig, void *siginfo, void *context) { 2422 // 4511530 - sem_post is serialized and handled by the manager thread. When 2423 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2424 // don't want to flood the manager thread with sem_post requests. 2425 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2426 return; 2427 2428 // Ctrl-C is pressed during error reporting, likely because the error 2429 // handler fails to abort. Let VM die immediately. 2430 if (sig == SIGINT && is_error_reported()) { 2431 os::die(); 2432 } 2433 2434 os::signal_notify(sig); 2435 } 2436 2437 void* os::user_handler() { 2438 return CAST_FROM_FN_PTR(void*, UserHandler); 2439 } 2440 2441 class Semaphore : public StackObj { 2442 public: 2443 Semaphore(); 2444 ~Semaphore(); 2445 void signal(); 2446 void wait(); 2447 bool trywait(); 2448 bool timedwait(unsigned int sec, int nsec); 2449 private: 2450 sem_t _semaphore; 2451 }; 2452 2453 2454 Semaphore::Semaphore() { 2455 sem_init(&_semaphore, 0, 0); 2456 } 2457 2458 Semaphore::~Semaphore() { 2459 sem_destroy(&_semaphore); 2460 } 2461 2462 void Semaphore::signal() { 2463 sem_post(&_semaphore); 2464 } 2465 2466 void Semaphore::wait() { 2467 sem_wait(&_semaphore); 2468 } 2469 2470 bool Semaphore::trywait() { 2471 return sem_trywait(&_semaphore) == 0; 2472 } 2473 2474 bool Semaphore::timedwait(unsigned int sec, int nsec) { 2475 struct timespec ts; 2476 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec); 2477 2478 while (1) { 2479 int result = sem_timedwait(&_semaphore, &ts); 2480 if (result == 0) { 2481 return true; 2482 } else if (errno == EINTR) { 2483 continue; 2484 } else if (errno == ETIMEDOUT) { 2485 return false; 2486 } else { 2487 return false; 2488 } 2489 } 2490 } 2491 2492 extern "C" { 2493 typedef void (*sa_handler_t)(int); 2494 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2495 } 2496 2497 void* os::signal(int signal_number, void* handler) { 2498 struct sigaction sigAct, oldSigAct; 2499 2500 sigfillset(&(sigAct.sa_mask)); 2501 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2502 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2503 2504 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2505 // -1 means registration failed 2506 return (void *)-1; 2507 } 2508 2509 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2510 } 2511 2512 void os::signal_raise(int signal_number) { 2513 ::raise(signal_number); 2514 } 2515 2516 /* 2517 * The following code is moved from os.cpp for making this 2518 * code platform specific, which it is by its very nature. 2519 */ 2520 2521 // Will be modified when max signal is changed to be dynamic 2522 int os::sigexitnum_pd() { 2523 return NSIG; 2524 } 2525 2526 // a counter for each possible signal value 2527 static volatile jint pending_signals[NSIG+1] = { 0 }; 2528 2529 // Linux(POSIX) specific hand shaking semaphore. 2530 static sem_t sig_sem; 2531 static Semaphore sr_semaphore; 2532 2533 void os::signal_init_pd() { 2534 // Initialize signal structures 2535 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2536 2537 // Initialize signal semaphore 2538 ::sem_init(&sig_sem, 0, 0); 2539 } 2540 2541 void os::signal_notify(int sig) { 2542 Atomic::inc(&pending_signals[sig]); 2543 ::sem_post(&sig_sem); 2544 } 2545 2546 static int check_pending_signals(bool wait) { 2547 Atomic::store(0, &sigint_count); 2548 for (;;) { 2549 for (int i = 0; i < NSIG + 1; i++) { 2550 jint n = pending_signals[i]; 2551 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2552 return i; 2553 } 2554 } 2555 if (!wait) { 2556 return -1; 2557 } 2558 JavaThread *thread = JavaThread::current(); 2559 ThreadBlockInVM tbivm(thread); 2560 2561 bool threadIsSuspended; 2562 do { 2563 thread->set_suspend_equivalent(); 2564 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2565 ::sem_wait(&sig_sem); 2566 2567 // were we externally suspended while we were waiting? 2568 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2569 if (threadIsSuspended) { 2570 // 2571 // The semaphore has been incremented, but while we were waiting 2572 // another thread suspended us. We don't want to continue running 2573 // while suspended because that would surprise the thread that 2574 // suspended us. 2575 // 2576 ::sem_post(&sig_sem); 2577 2578 thread->java_suspend_self(); 2579 } 2580 } while (threadIsSuspended); 2581 } 2582 } 2583 2584 int os::signal_lookup() { 2585 return check_pending_signals(false); 2586 } 2587 2588 int os::signal_wait() { 2589 return check_pending_signals(true); 2590 } 2591 2592 //////////////////////////////////////////////////////////////////////////////// 2593 // Virtual Memory 2594 2595 int os::vm_page_size() { 2596 // Seems redundant as all get out 2597 assert(os::Linux::page_size() != -1, "must call os::init"); 2598 return os::Linux::page_size(); 2599 } 2600 2601 // Solaris allocates memory by pages. 2602 int os::vm_allocation_granularity() { 2603 assert(os::Linux::page_size() != -1, "must call os::init"); 2604 return os::Linux::page_size(); 2605 } 2606 2607 // Rationale behind this function: 2608 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2609 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2610 // samples for JITted code. Here we create private executable mapping over the code cache 2611 // and then we can use standard (well, almost, as mapping can change) way to provide 2612 // info for the reporting script by storing timestamp and location of symbol 2613 void linux_wrap_code(char* base, size_t size) { 2614 static volatile jint cnt = 0; 2615 2616 if (!UseOprofile) { 2617 return; 2618 } 2619 2620 char buf[PATH_MAX+1]; 2621 int num = Atomic::add(1, &cnt); 2622 2623 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2624 os::get_temp_directory(), os::current_process_id(), num); 2625 unlink(buf); 2626 2627 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2628 2629 if (fd != -1) { 2630 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2631 if (rv != (off_t)-1) { 2632 if (::write(fd, "", 1) == 1) { 2633 mmap(base, size, 2634 PROT_READ|PROT_WRITE|PROT_EXEC, 2635 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2636 } 2637 } 2638 ::close(fd); 2639 unlink(buf); 2640 } 2641 } 2642 2643 // NOTE: Linux kernel does not really reserve the pages for us. 2644 // All it does is to check if there are enough free pages 2645 // left at the time of mmap(). This could be a potential 2646 // problem. 2647 bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2648 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2649 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2650 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2651 if (res != (uintptr_t) MAP_FAILED) { 2652 if (UseNUMAInterleaving) { 2653 numa_make_global(addr, size); 2654 } 2655 return true; 2656 } 2657 return false; 2658 } 2659 2660 // Define MAP_HUGETLB here so we can build HotSpot on old systems. 2661 #ifndef MAP_HUGETLB 2662 #define MAP_HUGETLB 0x40000 2663 #endif 2664 2665 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2666 #ifndef MADV_HUGEPAGE 2667 #define MADV_HUGEPAGE 14 2668 #endif 2669 2670 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2671 bool exec) { 2672 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) { 2673 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2674 uintptr_t res = 2675 (uintptr_t) ::mmap(addr, size, prot, 2676 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB, 2677 -1, 0); 2678 if (res != (uintptr_t) MAP_FAILED) { 2679 if (UseNUMAInterleaving) { 2680 numa_make_global(addr, size); 2681 } 2682 return true; 2683 } 2684 // Fall through and try to use small pages 2685 } 2686 2687 if (commit_memory(addr, size, exec)) { 2688 realign_memory(addr, size, alignment_hint); 2689 return true; 2690 } 2691 return false; 2692 } 2693 2694 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2695 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) { 2696 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2697 // be supported or the memory may already be backed by huge pages. 2698 ::madvise(addr, bytes, MADV_HUGEPAGE); 2699 } 2700 } 2701 2702 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2703 // This method works by doing an mmap over an existing mmaping and effectively discarding 2704 // the existing pages. However it won't work for SHM-based large pages that cannot be 2705 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2706 // small pages on top of the SHM segment. This method always works for small pages, so we 2707 // allow that in any case. 2708 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) { 2709 commit_memory(addr, bytes, alignment_hint, false); 2710 } 2711 } 2712 2713 void os::numa_make_global(char *addr, size_t bytes) { 2714 Linux::numa_interleave_memory(addr, bytes); 2715 } 2716 2717 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2718 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2719 } 2720 2721 bool os::numa_topology_changed() { return false; } 2722 2723 size_t os::numa_get_groups_num() { 2724 int max_node = Linux::numa_max_node(); 2725 return max_node > 0 ? max_node + 1 : 1; 2726 } 2727 2728 int os::numa_get_group_id() { 2729 int cpu_id = Linux::sched_getcpu(); 2730 if (cpu_id != -1) { 2731 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2732 if (lgrp_id != -1) { 2733 return lgrp_id; 2734 } 2735 } 2736 return 0; 2737 } 2738 2739 size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2740 for (size_t i = 0; i < size; i++) { 2741 ids[i] = i; 2742 } 2743 return size; 2744 } 2745 2746 bool os::get_page_info(char *start, page_info* info) { 2747 return false; 2748 } 2749 2750 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2751 return end; 2752 } 2753 2754 2755 int os::Linux::sched_getcpu_syscall(void) { 2756 unsigned int cpu; 2757 int retval = -1; 2758 2759 #if defined(IA32) 2760 # ifndef SYS_getcpu 2761 # define SYS_getcpu 318 2762 # endif 2763 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2764 #elif defined(AMD64) 2765 // Unfortunately we have to bring all these macros here from vsyscall.h 2766 // to be able to compile on old linuxes. 2767 # define __NR_vgetcpu 2 2768 # define VSYSCALL_START (-10UL << 20) 2769 # define VSYSCALL_SIZE 1024 2770 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2771 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2772 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2773 retval = vgetcpu(&cpu, NULL, NULL); 2774 #endif 2775 2776 return (retval == -1) ? retval : cpu; 2777 } 2778 2779 // Something to do with the numa-aware allocator needs these symbols 2780 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2781 extern "C" JNIEXPORT void numa_error(char *where) { } 2782 extern "C" JNIEXPORT int fork1() { return fork(); } 2783 2784 2785 // If we are running with libnuma version > 2, then we should 2786 // be trying to use symbols with versions 1.1 2787 // If we are running with earlier version, which did not have symbol versions, 2788 // we should use the base version. 2789 void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2790 void *f = dlvsym(handle, name, "libnuma_1.1"); 2791 if (f == NULL) { 2792 f = dlsym(handle, name); 2793 } 2794 return f; 2795 } 2796 2797 bool os::Linux::libnuma_init() { 2798 // sched_getcpu() should be in libc. 2799 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2800 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2801 2802 // If it's not, try a direct syscall. 2803 if (sched_getcpu() == -1) 2804 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall)); 2805 2806 if (sched_getcpu() != -1) { // Does it work? 2807 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2808 if (handle != NULL) { 2809 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2810 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2811 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2812 libnuma_dlsym(handle, "numa_max_node"))); 2813 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2814 libnuma_dlsym(handle, "numa_available"))); 2815 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2816 libnuma_dlsym(handle, "numa_tonode_memory"))); 2817 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2818 libnuma_dlsym(handle, "numa_interleave_memory"))); 2819 2820 2821 if (numa_available() != -1) { 2822 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2823 // Create a cpu -> node mapping 2824 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2825 rebuild_cpu_to_node_map(); 2826 return true; 2827 } 2828 } 2829 } 2830 return false; 2831 } 2832 2833 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2834 // The table is later used in get_node_by_cpu(). 2835 void os::Linux::rebuild_cpu_to_node_map() { 2836 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2837 // in libnuma (possible values are starting from 16, 2838 // and continuing up with every other power of 2, but less 2839 // than the maximum number of CPUs supported by kernel), and 2840 // is a subject to change (in libnuma version 2 the requirements 2841 // are more reasonable) we'll just hardcode the number they use 2842 // in the library. 2843 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2844 2845 size_t cpu_num = os::active_processor_count(); 2846 size_t cpu_map_size = NCPUS / BitsPerCLong; 2847 size_t cpu_map_valid_size = 2848 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2849 2850 cpu_to_node()->clear(); 2851 cpu_to_node()->at_grow(cpu_num - 1); 2852 size_t node_num = numa_get_groups_num(); 2853 2854 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2855 for (size_t i = 0; i < node_num; i++) { 2856 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2857 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2858 if (cpu_map[j] != 0) { 2859 for (size_t k = 0; k < BitsPerCLong; k++) { 2860 if (cpu_map[j] & (1UL << k)) { 2861 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2862 } 2863 } 2864 } 2865 } 2866 } 2867 } 2868 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2869 } 2870 2871 int os::Linux::get_node_by_cpu(int cpu_id) { 2872 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2873 return cpu_to_node()->at(cpu_id); 2874 } 2875 return -1; 2876 } 2877 2878 GrowableArray<int>* os::Linux::_cpu_to_node; 2879 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2880 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2881 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2882 os::Linux::numa_available_func_t os::Linux::_numa_available; 2883 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2884 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2885 unsigned long* os::Linux::_numa_all_nodes; 2886 2887 bool os::pd_uncommit_memory(char* addr, size_t size) { 2888 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2889 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2890 return res != (uintptr_t) MAP_FAILED; 2891 } 2892 2893 // Linux uses a growable mapping for the stack, and if the mapping for 2894 // the stack guard pages is not removed when we detach a thread the 2895 // stack cannot grow beyond the pages where the stack guard was 2896 // mapped. If at some point later in the process the stack expands to 2897 // that point, the Linux kernel cannot expand the stack any further 2898 // because the guard pages are in the way, and a segfault occurs. 2899 // 2900 // However, it's essential not to split the stack region by unmapping 2901 // a region (leaving a hole) that's already part of the stack mapping, 2902 // so if the stack mapping has already grown beyond the guard pages at 2903 // the time we create them, we have to truncate the stack mapping. 2904 // So, we need to know the extent of the stack mapping when 2905 // create_stack_guard_pages() is called. 2906 2907 // Find the bounds of the stack mapping. Return true for success. 2908 // 2909 // We only need this for stacks that are growable: at the time of 2910 // writing thread stacks don't use growable mappings (i.e. those 2911 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 2912 // only applies to the main thread. 2913 2914 static 2915 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) { 2916 2917 char buf[128]; 2918 int fd, sz; 2919 2920 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) { 2921 return false; 2922 } 2923 2924 const char kw[] = "[stack]"; 2925 const int kwlen = sizeof(kw)-1; 2926 2927 // Address part of /proc/self/maps couldn't be more than 128 bytes 2928 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) { 2929 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) { 2930 // Extract addresses 2931 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) { 2932 uintptr_t sp = (uintptr_t) __builtin_frame_address(0); 2933 if (sp >= *bottom && sp <= *top) { 2934 ::close(fd); 2935 return true; 2936 } 2937 } 2938 } 2939 } 2940 2941 ::close(fd); 2942 return false; 2943 } 2944 2945 2946 // If the (growable) stack mapping already extends beyond the point 2947 // where we're going to put our guard pages, truncate the mapping at 2948 // that point by munmap()ping it. This ensures that when we later 2949 // munmap() the guard pages we don't leave a hole in the stack 2950 // mapping. This only affects the main/initial thread, but guard 2951 // against future OS changes 2952 bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 2953 uintptr_t stack_extent, stack_base; 2954 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2955 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2956 assert(os::Linux::is_initial_thread(), 2957 "growable stack in non-initial thread"); 2958 if (stack_extent < (uintptr_t)addr) 2959 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent); 2960 } 2961 2962 return os::commit_memory(addr, size); 2963 } 2964 2965 // If this is a growable mapping, remove the guard pages entirely by 2966 // munmap()ping them. If not, just call uncommit_memory(). This only 2967 // affects the main/initial thread, but guard against future OS changes 2968 bool os::remove_stack_guard_pages(char* addr, size_t size) { 2969 uintptr_t stack_extent, stack_base; 2970 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true); 2971 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) { 2972 assert(os::Linux::is_initial_thread(), 2973 "growable stack in non-initial thread"); 2974 2975 return ::munmap(addr, size) == 0; 2976 } 2977 2978 return os::uncommit_memory(addr, size); 2979 } 2980 2981 static address _highest_vm_reserved_address = NULL; 2982 2983 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2984 // at 'requested_addr'. If there are existing memory mappings at the same 2985 // location, however, they will be overwritten. If 'fixed' is false, 2986 // 'requested_addr' is only treated as a hint, the return value may or 2987 // may not start from the requested address. Unlike Linux mmap(), this 2988 // function returns NULL to indicate failure. 2989 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2990 char * addr; 2991 int flags; 2992 2993 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2994 if (fixed) { 2995 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2996 flags |= MAP_FIXED; 2997 } 2998 2999 // Map uncommitted pages PROT_READ and PROT_WRITE, change access 3000 // to PROT_EXEC if executable when we commit the page. 3001 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE, 3002 flags, -1, 0); 3003 3004 if (addr != MAP_FAILED) { 3005 // anon_mmap() should only get called during VM initialization, 3006 // don't need lock (actually we can skip locking even it can be called 3007 // from multiple threads, because _highest_vm_reserved_address is just a 3008 // hint about the upper limit of non-stack memory regions.) 3009 if ((address)addr + bytes > _highest_vm_reserved_address) { 3010 _highest_vm_reserved_address = (address)addr + bytes; 3011 } 3012 } 3013 3014 return addr == MAP_FAILED ? NULL : addr; 3015 } 3016 3017 // Don't update _highest_vm_reserved_address, because there might be memory 3018 // regions above addr + size. If so, releasing a memory region only creates 3019 // a hole in the address space, it doesn't help prevent heap-stack collision. 3020 // 3021 static int anon_munmap(char * addr, size_t size) { 3022 return ::munmap(addr, size) == 0; 3023 } 3024 3025 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3026 size_t alignment_hint) { 3027 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3028 } 3029 3030 bool os::pd_release_memory(char* addr, size_t size) { 3031 return anon_munmap(addr, size); 3032 } 3033 3034 static address highest_vm_reserved_address() { 3035 return _highest_vm_reserved_address; 3036 } 3037 3038 static bool linux_mprotect(char* addr, size_t size, int prot) { 3039 // Linux wants the mprotect address argument to be page aligned. 3040 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3041 3042 // According to SUSv3, mprotect() should only be used with mappings 3043 // established by mmap(), and mmap() always maps whole pages. Unaligned 3044 // 'addr' likely indicates problem in the VM (e.g. trying to change 3045 // protection of malloc'ed or statically allocated memory). Check the 3046 // caller if you hit this assert. 3047 assert(addr == bottom, "sanity check"); 3048 3049 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3050 return ::mprotect(bottom, size, prot) == 0; 3051 } 3052 3053 // Set protections specified 3054 bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3055 bool is_committed) { 3056 unsigned int p = 0; 3057 switch (prot) { 3058 case MEM_PROT_NONE: p = PROT_NONE; break; 3059 case MEM_PROT_READ: p = PROT_READ; break; 3060 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3061 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3062 default: 3063 ShouldNotReachHere(); 3064 } 3065 // is_committed is unused. 3066 return linux_mprotect(addr, bytes, p); 3067 } 3068 3069 bool os::guard_memory(char* addr, size_t size) { 3070 return linux_mprotect(addr, size, PROT_NONE); 3071 } 3072 3073 bool os::unguard_memory(char* addr, size_t size) { 3074 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3075 } 3076 3077 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3078 bool result = false; 3079 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE, 3080 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3081 -1, 0); 3082 3083 if (p != (void *) -1) { 3084 // We don't know if this really is a huge page or not. 3085 FILE *fp = fopen("/proc/self/maps", "r"); 3086 if (fp) { 3087 while (!feof(fp)) { 3088 char chars[257]; 3089 long x = 0; 3090 if (fgets(chars, sizeof(chars), fp)) { 3091 if (sscanf(chars, "%lx-%*x", &x) == 1 3092 && x == (long)p) { 3093 if (strstr (chars, "hugepage")) { 3094 result = true; 3095 break; 3096 } 3097 } 3098 } 3099 } 3100 fclose(fp); 3101 } 3102 munmap (p, page_size); 3103 if (result) 3104 return true; 3105 } 3106 3107 if (warn) { 3108 warning("HugeTLBFS is not supported by the operating system."); 3109 } 3110 3111 return result; 3112 } 3113 3114 /* 3115 * Set the coredump_filter bits to include largepages in core dump (bit 6) 3116 * 3117 * From the coredump_filter documentation: 3118 * 3119 * - (bit 0) anonymous private memory 3120 * - (bit 1) anonymous shared memory 3121 * - (bit 2) file-backed private memory 3122 * - (bit 3) file-backed shared memory 3123 * - (bit 4) ELF header pages in file-backed private memory areas (it is 3124 * effective only if the bit 2 is cleared) 3125 * - (bit 5) hugetlb private memory 3126 * - (bit 6) hugetlb shared memory 3127 */ 3128 static void set_coredump_filter(void) { 3129 FILE *f; 3130 long cdm; 3131 3132 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3133 return; 3134 } 3135 3136 if (fscanf(f, "%lx", &cdm) != 1) { 3137 fclose(f); 3138 return; 3139 } 3140 3141 rewind(f); 3142 3143 if ((cdm & LARGEPAGES_BIT) == 0) { 3144 cdm |= LARGEPAGES_BIT; 3145 fprintf(f, "%#lx", cdm); 3146 } 3147 3148 fclose(f); 3149 } 3150 3151 // Large page support 3152 3153 static size_t _large_page_size = 0; 3154 3155 void os::large_page_init() { 3156 if (!UseLargePages) { 3157 UseHugeTLBFS = false; 3158 UseSHM = false; 3159 return; 3160 } 3161 3162 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) { 3163 // If UseLargePages is specified on the command line try both methods, 3164 // if it's default, then try only HugeTLBFS. 3165 if (FLAG_IS_DEFAULT(UseLargePages)) { 3166 UseHugeTLBFS = true; 3167 } else { 3168 UseHugeTLBFS = UseSHM = true; 3169 } 3170 } 3171 3172 if (LargePageSizeInBytes) { 3173 _large_page_size = LargePageSizeInBytes; 3174 } else { 3175 // large_page_size on Linux is used to round up heap size. x86 uses either 3176 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3177 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3178 // page as large as 256M. 3179 // 3180 // Here we try to figure out page size by parsing /proc/meminfo and looking 3181 // for a line with the following format: 3182 // Hugepagesize: 2048 kB 3183 // 3184 // If we can't determine the value (e.g. /proc is not mounted, or the text 3185 // format has been changed), we'll use the largest page size supported by 3186 // the processor. 3187 3188 #ifndef ZERO 3189 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3190 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3191 #endif // ZERO 3192 3193 FILE *fp = fopen("/proc/meminfo", "r"); 3194 if (fp) { 3195 while (!feof(fp)) { 3196 int x = 0; 3197 char buf[16]; 3198 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3199 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3200 _large_page_size = x * K; 3201 break; 3202 } 3203 } else { 3204 // skip to next line 3205 for (;;) { 3206 int ch = fgetc(fp); 3207 if (ch == EOF || ch == (int)'\n') break; 3208 } 3209 } 3210 } 3211 fclose(fp); 3212 } 3213 } 3214 3215 // print a warning if any large page related flag is specified on command line 3216 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3217 3218 const size_t default_page_size = (size_t)Linux::page_size(); 3219 if (_large_page_size > default_page_size) { 3220 _page_sizes[0] = _large_page_size; 3221 _page_sizes[1] = default_page_size; 3222 _page_sizes[2] = 0; 3223 } 3224 UseHugeTLBFS = UseHugeTLBFS && 3225 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size); 3226 3227 if (UseHugeTLBFS) 3228 UseSHM = false; 3229 3230 UseLargePages = UseHugeTLBFS || UseSHM; 3231 3232 set_coredump_filter(); 3233 } 3234 3235 #ifndef SHM_HUGETLB 3236 #define SHM_HUGETLB 04000 3237 #endif 3238 3239 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) { 3240 // "exec" is passed in but not used. Creating the shared image for 3241 // the code cache doesn't have an SHM_X executable permission to check. 3242 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3243 3244 key_t key = IPC_PRIVATE; 3245 char *addr; 3246 3247 bool warn_on_failure = UseLargePages && 3248 (!FLAG_IS_DEFAULT(UseLargePages) || 3249 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 3250 ); 3251 char msg[128]; 3252 3253 // Create a large shared memory region to attach to based on size. 3254 // Currently, size is the total size of the heap 3255 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3256 if (shmid == -1) { 3257 // Possible reasons for shmget failure: 3258 // 1. shmmax is too small for Java heap. 3259 // > check shmmax value: cat /proc/sys/kernel/shmmax 3260 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3261 // 2. not enough large page memory. 3262 // > check available large pages: cat /proc/meminfo 3263 // > increase amount of large pages: 3264 // echo new_value > /proc/sys/vm/nr_hugepages 3265 // Note 1: different Linux may use different name for this property, 3266 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3267 // Note 2: it's possible there's enough physical memory available but 3268 // they are so fragmented after a long run that they can't 3269 // coalesce into large pages. Try to reserve large pages when 3270 // the system is still "fresh". 3271 if (warn_on_failure) { 3272 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3273 warning(msg); 3274 } 3275 return NULL; 3276 } 3277 3278 // attach to the region 3279 addr = (char*)shmat(shmid, req_addr, 0); 3280 int err = errno; 3281 3282 // Remove shmid. If shmat() is successful, the actual shared memory segment 3283 // will be deleted when it's detached by shmdt() or when the process 3284 // terminates. If shmat() is not successful this will remove the shared 3285 // segment immediately. 3286 shmctl(shmid, IPC_RMID, NULL); 3287 3288 if ((intptr_t)addr == -1) { 3289 if (warn_on_failure) { 3290 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3291 warning(msg); 3292 } 3293 return NULL; 3294 } 3295 3296 if ((addr != NULL) && UseNUMAInterleaving) { 3297 numa_make_global(addr, bytes); 3298 } 3299 3300 // The memory is committed 3301 address pc = CALLER_PC; 3302 MemTracker::record_virtual_memory_reserve((address)addr, bytes, pc); 3303 MemTracker::record_virtual_memory_commit((address)addr, bytes, pc); 3304 3305 return addr; 3306 } 3307 3308 bool os::release_memory_special(char* base, size_t bytes) { 3309 // detaching the SHM segment will also delete it, see reserve_memory_special() 3310 int rslt = shmdt(base); 3311 if (rslt == 0) { 3312 MemTracker::record_virtual_memory_uncommit((address)base, bytes); 3313 MemTracker::record_virtual_memory_release((address)base, bytes); 3314 return true; 3315 } else { 3316 return false; 3317 } 3318 } 3319 3320 size_t os::large_page_size() { 3321 return _large_page_size; 3322 } 3323 3324 // HugeTLBFS allows application to commit large page memory on demand; 3325 // with SysV SHM the entire memory region must be allocated as shared 3326 // memory. 3327 bool os::can_commit_large_page_memory() { 3328 return UseHugeTLBFS; 3329 } 3330 3331 bool os::can_execute_large_page_memory() { 3332 return UseHugeTLBFS; 3333 } 3334 3335 // Reserve memory at an arbitrary address, only if that area is 3336 // available (and not reserved for something else). 3337 3338 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3339 const int max_tries = 10; 3340 char* base[max_tries]; 3341 size_t size[max_tries]; 3342 const size_t gap = 0x000000; 3343 3344 // Assert only that the size is a multiple of the page size, since 3345 // that's all that mmap requires, and since that's all we really know 3346 // about at this low abstraction level. If we need higher alignment, 3347 // we can either pass an alignment to this method or verify alignment 3348 // in one of the methods further up the call chain. See bug 5044738. 3349 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3350 3351 // Repeatedly allocate blocks until the block is allocated at the 3352 // right spot. Give up after max_tries. Note that reserve_memory() will 3353 // automatically update _highest_vm_reserved_address if the call is 3354 // successful. The variable tracks the highest memory address every reserved 3355 // by JVM. It is used to detect heap-stack collision if running with 3356 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3357 // space than needed, it could confuse the collision detecting code. To 3358 // solve the problem, save current _highest_vm_reserved_address and 3359 // calculate the correct value before return. 3360 address old_highest = _highest_vm_reserved_address; 3361 3362 // Linux mmap allows caller to pass an address as hint; give it a try first, 3363 // if kernel honors the hint then we can return immediately. 3364 char * addr = anon_mmap(requested_addr, bytes, false); 3365 if (addr == requested_addr) { 3366 return requested_addr; 3367 } 3368 3369 if (addr != NULL) { 3370 // mmap() is successful but it fails to reserve at the requested address 3371 anon_munmap(addr, bytes); 3372 } 3373 3374 int i; 3375 for (i = 0; i < max_tries; ++i) { 3376 base[i] = reserve_memory(bytes); 3377 3378 if (base[i] != NULL) { 3379 // Is this the block we wanted? 3380 if (base[i] == requested_addr) { 3381 size[i] = bytes; 3382 break; 3383 } 3384 3385 // Does this overlap the block we wanted? Give back the overlapped 3386 // parts and try again. 3387 3388 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3389 if (top_overlap >= 0 && top_overlap < bytes) { 3390 unmap_memory(base[i], top_overlap); 3391 base[i] += top_overlap; 3392 size[i] = bytes - top_overlap; 3393 } else { 3394 size_t bottom_overlap = base[i] + bytes - requested_addr; 3395 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3396 unmap_memory(requested_addr, bottom_overlap); 3397 size[i] = bytes - bottom_overlap; 3398 } else { 3399 size[i] = bytes; 3400 } 3401 } 3402 } 3403 } 3404 3405 // Give back the unused reserved pieces. 3406 3407 for (int j = 0; j < i; ++j) { 3408 if (base[j] != NULL) { 3409 unmap_memory(base[j], size[j]); 3410 } 3411 } 3412 3413 if (i < max_tries) { 3414 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3415 return requested_addr; 3416 } else { 3417 _highest_vm_reserved_address = old_highest; 3418 return NULL; 3419 } 3420 } 3421 3422 size_t os::read(int fd, void *buf, unsigned int nBytes) { 3423 return ::read(fd, buf, nBytes); 3424 } 3425 3426 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation. 3427 // Solaris uses poll(), linux uses park(). 3428 // Poll() is likely a better choice, assuming that Thread.interrupt() 3429 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with 3430 // SIGSEGV, see 4355769. 3431 3432 int os::sleep(Thread* thread, jlong millis, bool interruptible) { 3433 assert(thread == Thread::current(), "thread consistency check"); 3434 3435 ParkEvent * const slp = thread->_SleepEvent ; 3436 slp->reset() ; 3437 OrderAccess::fence() ; 3438 3439 if (interruptible) { 3440 jlong prevtime = javaTimeNanos(); 3441 3442 for (;;) { 3443 if (os::is_interrupted(thread, true)) { 3444 return OS_INTRPT; 3445 } 3446 3447 jlong newtime = javaTimeNanos(); 3448 3449 if (newtime - prevtime < 0) { 3450 // time moving backwards, should only happen if no monotonic clock 3451 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3452 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3453 } else { 3454 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3455 } 3456 3457 if(millis <= 0) { 3458 return OS_OK; 3459 } 3460 3461 prevtime = newtime; 3462 3463 { 3464 assert(thread->is_Java_thread(), "sanity check"); 3465 JavaThread *jt = (JavaThread *) thread; 3466 ThreadBlockInVM tbivm(jt); 3467 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */); 3468 3469 jt->set_suspend_equivalent(); 3470 // cleared by handle_special_suspend_equivalent_condition() or 3471 // java_suspend_self() via check_and_wait_while_suspended() 3472 3473 slp->park(millis); 3474 3475 // were we externally suspended while we were waiting? 3476 jt->check_and_wait_while_suspended(); 3477 } 3478 } 3479 } else { 3480 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 3481 jlong prevtime = javaTimeNanos(); 3482 3483 for (;;) { 3484 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on 3485 // the 1st iteration ... 3486 jlong newtime = javaTimeNanos(); 3487 3488 if (newtime - prevtime < 0) { 3489 // time moving backwards, should only happen if no monotonic clock 3490 // not a guarantee() because JVM should not abort on kernel/glibc bugs 3491 assert(!Linux::supports_monotonic_clock(), "time moving backwards"); 3492 } else { 3493 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC; 3494 } 3495 3496 if(millis <= 0) break ; 3497 3498 prevtime = newtime; 3499 slp->park(millis); 3500 } 3501 return OS_OK ; 3502 } 3503 } 3504 3505 int os::naked_sleep() { 3506 // %% make the sleep time an integer flag. for now use 1 millisec. 3507 return os::sleep(Thread::current(), 1, false); 3508 } 3509 3510 // Sleep forever; naked call to OS-specific sleep; use with CAUTION 3511 void os::infinite_sleep() { 3512 while (true) { // sleep forever ... 3513 ::sleep(100); // ... 100 seconds at a time 3514 } 3515 } 3516 3517 // Used to convert frequent JVM_Yield() to nops 3518 bool os::dont_yield() { 3519 return DontYieldALot; 3520 } 3521 3522 void os::yield() { 3523 sched_yield(); 3524 } 3525 3526 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3527 3528 void os::yield_all(int attempts) { 3529 // Yields to all threads, including threads with lower priorities 3530 // Threads on Linux are all with same priority. The Solaris style 3531 // os::yield_all() with nanosleep(1ms) is not necessary. 3532 sched_yield(); 3533 } 3534 3535 // Called from the tight loops to possibly influence time-sharing heuristics 3536 void os::loop_breaker(int attempts) { 3537 os::yield_all(attempts); 3538 } 3539 3540 //////////////////////////////////////////////////////////////////////////////// 3541 // thread priority support 3542 3543 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3544 // only supports dynamic priority, static priority must be zero. For real-time 3545 // applications, Linux supports SCHED_RR which allows static priority (1-99). 3546 // However, for large multi-threaded applications, SCHED_RR is not only slower 3547 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3548 // of 5 runs - Sep 2005). 3549 // 3550 // The following code actually changes the niceness of kernel-thread/LWP. It 3551 // has an assumption that setpriority() only modifies one kernel-thread/LWP, 3552 // not the entire user process, and user level threads are 1:1 mapped to kernel 3553 // threads. It has always been the case, but could change in the future. For 3554 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3555 // It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3556 3557 int os::java_to_os_priority[CriticalPriority + 1] = { 3558 19, // 0 Entry should never be used 3559 3560 4, // 1 MinPriority 3561 3, // 2 3562 2, // 3 3563 3564 1, // 4 3565 0, // 5 NormPriority 3566 -1, // 6 3567 3568 -2, // 7 3569 -3, // 8 3570 -4, // 9 NearMaxPriority 3571 3572 -5, // 10 MaxPriority 3573 3574 -5 // 11 CriticalPriority 3575 }; 3576 3577 static int prio_init() { 3578 if (ThreadPriorityPolicy == 1) { 3579 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3580 // if effective uid is not root. Perhaps, a more elegant way of doing 3581 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3582 if (geteuid() != 0) { 3583 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3584 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3585 } 3586 ThreadPriorityPolicy = 0; 3587 } 3588 } 3589 if (UseCriticalJavaThreadPriority) { 3590 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3591 } 3592 return 0; 3593 } 3594 3595 OSReturn os::set_native_priority(Thread* thread, int newpri) { 3596 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3597 3598 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3599 return (ret == 0) ? OS_OK : OS_ERR; 3600 } 3601 3602 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3603 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3604 *priority_ptr = java_to_os_priority[NormPriority]; 3605 return OS_OK; 3606 } 3607 3608 errno = 0; 3609 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3610 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3611 } 3612 3613 // Hint to the underlying OS that a task switch would not be good. 3614 // Void return because it's a hint and can fail. 3615 void os::hint_no_preempt() {} 3616 3617 //////////////////////////////////////////////////////////////////////////////// 3618 // suspend/resume support 3619 3620 // the low-level signal-based suspend/resume support is a remnant from the 3621 // old VM-suspension that used to be for java-suspension, safepoints etc, 3622 // within hotspot. Now there is a single use-case for this: 3623 // - calling get_thread_pc() on the VMThread by the flat-profiler task 3624 // that runs in the watcher thread. 3625 // The remaining code is greatly simplified from the more general suspension 3626 // code that used to be used. 3627 // 3628 // The protocol is quite simple: 3629 // - suspend: 3630 // - sends a signal to the target thread 3631 // - polls the suspend state of the osthread using a yield loop 3632 // - target thread signal handler (SR_handler) sets suspend state 3633 // and blocks in sigsuspend until continued 3634 // - resume: 3635 // - sets target osthread state to continue 3636 // - sends signal to end the sigsuspend loop in the SR_handler 3637 // 3638 // Note that the SR_lock plays no role in this suspend/resume protocol. 3639 // 3640 3641 static void resume_clear_context(OSThread *osthread) { 3642 osthread->set_ucontext(NULL); 3643 osthread->set_siginfo(NULL); 3644 } 3645 3646 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3647 osthread->set_ucontext(context); 3648 osthread->set_siginfo(siginfo); 3649 } 3650 3651 // 3652 // Handler function invoked when a thread's execution is suspended or 3653 // resumed. We have to be careful that only async-safe functions are 3654 // called here (Note: most pthread functions are not async safe and 3655 // should be avoided.) 3656 // 3657 // Note: sigwait() is a more natural fit than sigsuspend() from an 3658 // interface point of view, but sigwait() prevents the signal hander 3659 // from being run. libpthread would get very confused by not having 3660 // its signal handlers run and prevents sigwait()'s use with the 3661 // mutex granting granting signal. 3662 // 3663 // Currently only ever called on the VMThread and JavaThreads (PC sampling) 3664 // 3665 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3666 // Save and restore errno to avoid confusing native code with EINTR 3667 // after sigsuspend. 3668 int old_errno = errno; 3669 3670 Thread* thread = Thread::current(); 3671 OSThread* osthread = thread->osthread(); 3672 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3673 3674 os::SuspendResume::State current = osthread->sr.state(); 3675 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3676 suspend_save_context(osthread, siginfo, context); 3677 3678 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 3679 os::SuspendResume::State state = osthread->sr.suspended(); 3680 if (state == os::SuspendResume::SR_SUSPENDED) { 3681 sigset_t suspend_set; // signals for sigsuspend() 3682 3683 // get current set of blocked signals and unblock resume signal 3684 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3685 sigdelset(&suspend_set, SR_signum); 3686 3687 sr_semaphore.signal(); 3688 // wait here until we are resumed 3689 while (1) { 3690 sigsuspend(&suspend_set); 3691 3692 os::SuspendResume::State result = osthread->sr.running(); 3693 if (result == os::SuspendResume::SR_RUNNING) { 3694 sr_semaphore.signal(); 3695 break; 3696 } 3697 } 3698 3699 } else if (state == os::SuspendResume::SR_RUNNING) { 3700 // request was cancelled, continue 3701 } else { 3702 ShouldNotReachHere(); 3703 } 3704 3705 resume_clear_context(osthread); 3706 } else if (current == os::SuspendResume::SR_RUNNING) { 3707 // request was cancelled, continue 3708 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 3709 // ignore 3710 } else { 3711 // ignore 3712 } 3713 3714 errno = old_errno; 3715 } 3716 3717 3718 static int SR_initialize() { 3719 struct sigaction act; 3720 char *s; 3721 /* Get signal number to use for suspend/resume */ 3722 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3723 int sig = ::strtol(s, 0, 10); 3724 if (sig > 0 || sig < _NSIG) { 3725 SR_signum = sig; 3726 } 3727 } 3728 3729 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3730 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3731 3732 sigemptyset(&SR_sigset); 3733 sigaddset(&SR_sigset, SR_signum); 3734 3735 /* Set up signal handler for suspend/resume */ 3736 act.sa_flags = SA_RESTART|SA_SIGINFO; 3737 act.sa_handler = (void (*)(int)) SR_handler; 3738 3739 // SR_signum is blocked by default. 3740 // 4528190 - We also need to block pthread restart signal (32 on all 3741 // supported Linux platforms). Note that LinuxThreads need to block 3742 // this signal for all threads to work properly. So we don't have 3743 // to use hard-coded signal number when setting up the mask. 3744 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 3745 3746 if (sigaction(SR_signum, &act, 0) == -1) { 3747 return -1; 3748 } 3749 3750 // Save signal flag 3751 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 3752 return 0; 3753 } 3754 3755 static int SR_finalize() { 3756 return 0; 3757 } 3758 3759 static int sr_notify(OSThread* osthread) { 3760 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3761 assert_status(status == 0, status, "pthread_kill"); 3762 return status; 3763 } 3764 3765 // "Randomly" selected value for how long we want to spin 3766 // before bailing out on suspending a thread, also how often 3767 // we send a signal to a thread we want to resume 3768 static const int RANDOMLY_LARGE_INTEGER = 1000000; 3769 static const int RANDOMLY_LARGE_INTEGER2 = 100; 3770 3771 // returns true on success and false on error - really an error is fatal 3772 // but this seems the normal response to library errors 3773 static bool do_suspend(OSThread* osthread) { 3774 assert(osthread->sr.is_running(), "thread should be running"); 3775 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 3776 3777 // mark as suspended and send signal 3778 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 3779 // failed to switch, state wasn't running? 3780 ShouldNotReachHere(); 3781 return false; 3782 } 3783 3784 if (sr_notify(osthread) != 0) { 3785 ShouldNotReachHere(); 3786 } 3787 3788 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 3789 while (true) { 3790 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 3791 break; 3792 } else { 3793 // timeout 3794 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 3795 if (cancelled == os::SuspendResume::SR_RUNNING) { 3796 return false; 3797 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 3798 // make sure that we consume the signal on the semaphore as well 3799 sr_semaphore.wait(); 3800 break; 3801 } else { 3802 ShouldNotReachHere(); 3803 return false; 3804 } 3805 } 3806 } 3807 3808 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 3809 return true; 3810 } 3811 3812 static void do_resume(OSThread* osthread) { 3813 assert(osthread->sr.is_suspended(), "thread should be suspended"); 3814 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 3815 3816 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 3817 // failed to switch to WAKEUP_REQUEST 3818 ShouldNotReachHere(); 3819 return; 3820 } 3821 3822 while (true) { 3823 if (sr_notify(osthread) == 0) { 3824 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 3825 if (osthread->sr.is_running()) { 3826 return; 3827 } 3828 } 3829 } else { 3830 ShouldNotReachHere(); 3831 } 3832 } 3833 3834 guarantee(osthread->sr.is_running(), "Must be running!"); 3835 } 3836 3837 //////////////////////////////////////////////////////////////////////////////// 3838 // interrupt support 3839 3840 void os::interrupt(Thread* thread) { 3841 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3842 "possibility of dangling Thread pointer"); 3843 3844 OSThread* osthread = thread->osthread(); 3845 3846 if (!osthread->interrupted()) { 3847 osthread->set_interrupted(true); 3848 // More than one thread can get here with the same value of osthread, 3849 // resulting in multiple notifications. We do, however, want the store 3850 // to interrupted() to be visible to other threads before we execute unpark(). 3851 OrderAccess::fence(); 3852 ParkEvent * const slp = thread->_SleepEvent ; 3853 if (slp != NULL) slp->unpark() ; 3854 } 3855 3856 // For JSR166. Unpark even if interrupt status already was set 3857 if (thread->is_Java_thread()) 3858 ((JavaThread*)thread)->parker()->unpark(); 3859 3860 ParkEvent * ev = thread->_ParkEvent ; 3861 if (ev != NULL) ev->unpark() ; 3862 3863 } 3864 3865 bool os::is_interrupted(Thread* thread, bool clear_interrupted) { 3866 assert(Thread::current() == thread || Threads_lock->owned_by_self(), 3867 "possibility of dangling Thread pointer"); 3868 3869 OSThread* osthread = thread->osthread(); 3870 3871 bool interrupted = osthread->interrupted(); 3872 3873 if (interrupted && clear_interrupted) { 3874 osthread->set_interrupted(false); 3875 // consider thread->_SleepEvent->reset() ... optional optimization 3876 } 3877 3878 return interrupted; 3879 } 3880 3881 /////////////////////////////////////////////////////////////////////////////////// 3882 // signal handling (except suspend/resume) 3883 3884 // This routine may be used by user applications as a "hook" to catch signals. 3885 // The user-defined signal handler must pass unrecognized signals to this 3886 // routine, and if it returns true (non-zero), then the signal handler must 3887 // return immediately. If the flag "abort_if_unrecognized" is true, then this 3888 // routine will never retun false (zero), but instead will execute a VM panic 3889 // routine kill the process. 3890 // 3891 // If this routine returns false, it is OK to call it again. This allows 3892 // the user-defined signal handler to perform checks either before or after 3893 // the VM performs its own checks. Naturally, the user code would be making 3894 // a serious error if it tried to handle an exception (such as a null check 3895 // or breakpoint) that the VM was generating for its own correct operation. 3896 // 3897 // This routine may recognize any of the following kinds of signals: 3898 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 3899 // It should be consulted by handlers for any of those signals. 3900 // 3901 // The caller of this routine must pass in the three arguments supplied 3902 // to the function referred to in the "sa_sigaction" (not the "sa_handler") 3903 // field of the structure passed to sigaction(). This routine assumes that 3904 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 3905 // 3906 // Note that the VM will print warnings if it detects conflicting signal 3907 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 3908 // 3909 extern "C" JNIEXPORT int 3910 JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 3911 void* ucontext, int abort_if_unrecognized); 3912 3913 void signalHandler(int sig, siginfo_t* info, void* uc) { 3914 assert(info != NULL && uc != NULL, "it must be old kernel"); 3915 JVM_handle_linux_signal(sig, info, uc, true); 3916 } 3917 3918 3919 // This boolean allows users to forward their own non-matching signals 3920 // to JVM_handle_linux_signal, harmlessly. 3921 bool os::Linux::signal_handlers_are_installed = false; 3922 3923 // For signal-chaining 3924 struct sigaction os::Linux::sigact[MAXSIGNUM]; 3925 unsigned int os::Linux::sigs = 0; 3926 bool os::Linux::libjsig_is_loaded = false; 3927 typedef struct sigaction *(*get_signal_t)(int); 3928 get_signal_t os::Linux::get_signal_action = NULL; 3929 3930 struct sigaction* os::Linux::get_chained_signal_action(int sig) { 3931 struct sigaction *actp = NULL; 3932 3933 if (libjsig_is_loaded) { 3934 // Retrieve the old signal handler from libjsig 3935 actp = (*get_signal_action)(sig); 3936 } 3937 if (actp == NULL) { 3938 // Retrieve the preinstalled signal handler from jvm 3939 actp = get_preinstalled_handler(sig); 3940 } 3941 3942 return actp; 3943 } 3944 3945 static bool call_chained_handler(struct sigaction *actp, int sig, 3946 siginfo_t *siginfo, void *context) { 3947 // Call the old signal handler 3948 if (actp->sa_handler == SIG_DFL) { 3949 // It's more reasonable to let jvm treat it as an unexpected exception 3950 // instead of taking the default action. 3951 return false; 3952 } else if (actp->sa_handler != SIG_IGN) { 3953 if ((actp->sa_flags & SA_NODEFER) == 0) { 3954 // automaticlly block the signal 3955 sigaddset(&(actp->sa_mask), sig); 3956 } 3957 3958 sa_handler_t hand; 3959 sa_sigaction_t sa; 3960 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 3961 // retrieve the chained handler 3962 if (siginfo_flag_set) { 3963 sa = actp->sa_sigaction; 3964 } else { 3965 hand = actp->sa_handler; 3966 } 3967 3968 if ((actp->sa_flags & SA_RESETHAND) != 0) { 3969 actp->sa_handler = SIG_DFL; 3970 } 3971 3972 // try to honor the signal mask 3973 sigset_t oset; 3974 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 3975 3976 // call into the chained handler 3977 if (siginfo_flag_set) { 3978 (*sa)(sig, siginfo, context); 3979 } else { 3980 (*hand)(sig); 3981 } 3982 3983 // restore the signal mask 3984 pthread_sigmask(SIG_SETMASK, &oset, 0); 3985 } 3986 // Tell jvm's signal handler the signal is taken care of. 3987 return true; 3988 } 3989 3990 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 3991 bool chained = false; 3992 // signal-chaining 3993 if (UseSignalChaining) { 3994 struct sigaction *actp = get_chained_signal_action(sig); 3995 if (actp != NULL) { 3996 chained = call_chained_handler(actp, sig, siginfo, context); 3997 } 3998 } 3999 return chained; 4000 } 4001 4002 struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4003 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 4004 return &sigact[sig]; 4005 } 4006 return NULL; 4007 } 4008 4009 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4010 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4011 sigact[sig] = oldAct; 4012 sigs |= (unsigned int)1 << sig; 4013 } 4014 4015 // for diagnostic 4016 int os::Linux::sigflags[MAXSIGNUM]; 4017 4018 int os::Linux::get_our_sigflags(int sig) { 4019 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4020 return sigflags[sig]; 4021 } 4022 4023 void os::Linux::set_our_sigflags(int sig, int flags) { 4024 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4025 sigflags[sig] = flags; 4026 } 4027 4028 void os::Linux::set_signal_handler(int sig, bool set_installed) { 4029 // Check for overwrite. 4030 struct sigaction oldAct; 4031 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4032 4033 void* oldhand = oldAct.sa_sigaction 4034 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4035 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4036 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4037 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4038 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4039 if (AllowUserSignalHandlers || !set_installed) { 4040 // Do not overwrite; user takes responsibility to forward to us. 4041 return; 4042 } else if (UseSignalChaining) { 4043 // save the old handler in jvm 4044 save_preinstalled_handler(sig, oldAct); 4045 // libjsig also interposes the sigaction() call below and saves the 4046 // old sigaction on it own. 4047 } else { 4048 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4049 "%#lx for signal %d.", (long)oldhand, sig)); 4050 } 4051 } 4052 4053 struct sigaction sigAct; 4054 sigfillset(&(sigAct.sa_mask)); 4055 sigAct.sa_handler = SIG_DFL; 4056 if (!set_installed) { 4057 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4058 } else { 4059 sigAct.sa_sigaction = signalHandler; 4060 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4061 } 4062 // Save flags, which are set by ours 4063 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4064 sigflags[sig] = sigAct.sa_flags; 4065 4066 int ret = sigaction(sig, &sigAct, &oldAct); 4067 assert(ret == 0, "check"); 4068 4069 void* oldhand2 = oldAct.sa_sigaction 4070 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4071 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4072 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4073 } 4074 4075 // install signal handlers for signals that HotSpot needs to 4076 // handle in order to support Java-level exception handling. 4077 4078 void os::Linux::install_signal_handlers() { 4079 if (!signal_handlers_are_installed) { 4080 signal_handlers_are_installed = true; 4081 4082 // signal-chaining 4083 typedef void (*signal_setting_t)(); 4084 signal_setting_t begin_signal_setting = NULL; 4085 signal_setting_t end_signal_setting = NULL; 4086 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4087 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4088 if (begin_signal_setting != NULL) { 4089 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4090 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4091 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4092 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4093 libjsig_is_loaded = true; 4094 assert(UseSignalChaining, "should enable signal-chaining"); 4095 } 4096 if (libjsig_is_loaded) { 4097 // Tell libjsig jvm is setting signal handlers 4098 (*begin_signal_setting)(); 4099 } 4100 4101 set_signal_handler(SIGSEGV, true); 4102 set_signal_handler(SIGPIPE, true); 4103 set_signal_handler(SIGBUS, true); 4104 set_signal_handler(SIGILL, true); 4105 set_signal_handler(SIGFPE, true); 4106 set_signal_handler(SIGXFSZ, true); 4107 4108 if (libjsig_is_loaded) { 4109 // Tell libjsig jvm finishes setting signal handlers 4110 (*end_signal_setting)(); 4111 } 4112 4113 // We don't activate signal checker if libjsig is in place, we trust ourselves 4114 // and if UserSignalHandler is installed all bets are off. 4115 // Log that signal checking is off only if -verbose:jni is specified. 4116 if (CheckJNICalls) { 4117 if (libjsig_is_loaded) { 4118 if (PrintJNIResolving) { 4119 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4120 } 4121 check_signals = false; 4122 } 4123 if (AllowUserSignalHandlers) { 4124 if (PrintJNIResolving) { 4125 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4126 } 4127 check_signals = false; 4128 } 4129 } 4130 } 4131 } 4132 4133 // This is the fastest way to get thread cpu time on Linux. 4134 // Returns cpu time (user+sys) for any thread, not only for current. 4135 // POSIX compliant clocks are implemented in the kernels 2.6.16+. 4136 // It might work on 2.6.10+ with a special kernel/glibc patch. 4137 // For reference, please, see IEEE Std 1003.1-2004: 4138 // http://www.unix.org/single_unix_specification 4139 4140 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4141 struct timespec tp; 4142 int rc = os::Linux::clock_gettime(clockid, &tp); 4143 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4144 4145 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4146 } 4147 4148 ///// 4149 // glibc on Linux platform uses non-documented flag 4150 // to indicate, that some special sort of signal 4151 // trampoline is used. 4152 // We will never set this flag, and we should 4153 // ignore this flag in our diagnostic 4154 #ifdef SIGNIFICANT_SIGNAL_MASK 4155 #undef SIGNIFICANT_SIGNAL_MASK 4156 #endif 4157 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4158 4159 static const char* get_signal_handler_name(address handler, 4160 char* buf, int buflen) { 4161 int offset; 4162 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4163 if (found) { 4164 // skip directory names 4165 const char *p1, *p2; 4166 p1 = buf; 4167 size_t len = strlen(os::file_separator()); 4168 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4169 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4170 } else { 4171 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4172 } 4173 return buf; 4174 } 4175 4176 static void print_signal_handler(outputStream* st, int sig, 4177 char* buf, size_t buflen) { 4178 struct sigaction sa; 4179 4180 sigaction(sig, NULL, &sa); 4181 4182 // See comment for SIGNIFICANT_SIGNAL_MASK define 4183 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4184 4185 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4186 4187 address handler = (sa.sa_flags & SA_SIGINFO) 4188 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4189 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4190 4191 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4192 st->print("SIG_DFL"); 4193 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4194 st->print("SIG_IGN"); 4195 } else { 4196 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4197 } 4198 4199 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask); 4200 4201 address rh = VMError::get_resetted_sighandler(sig); 4202 // May be, handler was resetted by VMError? 4203 if(rh != NULL) { 4204 handler = rh; 4205 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4206 } 4207 4208 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags); 4209 4210 // Check: is it our handler? 4211 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4212 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4213 // It is our signal handler 4214 // check for flags, reset system-used one! 4215 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4216 st->print( 4217 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4218 os::Linux::get_our_sigflags(sig)); 4219 } 4220 } 4221 st->cr(); 4222 } 4223 4224 4225 #define DO_SIGNAL_CHECK(sig) \ 4226 if (!sigismember(&check_signal_done, sig)) \ 4227 os::Linux::check_signal_handler(sig) 4228 4229 // This method is a periodic task to check for misbehaving JNI applications 4230 // under CheckJNI, we can add any periodic checks here 4231 4232 void os::run_periodic_checks() { 4233 4234 if (check_signals == false) return; 4235 4236 // SEGV and BUS if overridden could potentially prevent 4237 // generation of hs*.log in the event of a crash, debugging 4238 // such a case can be very challenging, so we absolutely 4239 // check the following for a good measure: 4240 DO_SIGNAL_CHECK(SIGSEGV); 4241 DO_SIGNAL_CHECK(SIGILL); 4242 DO_SIGNAL_CHECK(SIGFPE); 4243 DO_SIGNAL_CHECK(SIGBUS); 4244 DO_SIGNAL_CHECK(SIGPIPE); 4245 DO_SIGNAL_CHECK(SIGXFSZ); 4246 4247 4248 // ReduceSignalUsage allows the user to override these handlers 4249 // see comments at the very top and jvm_solaris.h 4250 if (!ReduceSignalUsage) { 4251 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4252 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4253 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4254 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4255 } 4256 4257 DO_SIGNAL_CHECK(SR_signum); 4258 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4259 } 4260 4261 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4262 4263 static os_sigaction_t os_sigaction = NULL; 4264 4265 void os::Linux::check_signal_handler(int sig) { 4266 char buf[O_BUFLEN]; 4267 address jvmHandler = NULL; 4268 4269 4270 struct sigaction act; 4271 if (os_sigaction == NULL) { 4272 // only trust the default sigaction, in case it has been interposed 4273 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4274 if (os_sigaction == NULL) return; 4275 } 4276 4277 os_sigaction(sig, (struct sigaction*)NULL, &act); 4278 4279 4280 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4281 4282 address thisHandler = (act.sa_flags & SA_SIGINFO) 4283 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4284 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 4285 4286 4287 switch(sig) { 4288 case SIGSEGV: 4289 case SIGBUS: 4290 case SIGFPE: 4291 case SIGPIPE: 4292 case SIGILL: 4293 case SIGXFSZ: 4294 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4295 break; 4296 4297 case SHUTDOWN1_SIGNAL: 4298 case SHUTDOWN2_SIGNAL: 4299 case SHUTDOWN3_SIGNAL: 4300 case BREAK_SIGNAL: 4301 jvmHandler = (address)user_handler(); 4302 break; 4303 4304 case INTERRUPT_SIGNAL: 4305 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4306 break; 4307 4308 default: 4309 if (sig == SR_signum) { 4310 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4311 } else { 4312 return; 4313 } 4314 break; 4315 } 4316 4317 if (thisHandler != jvmHandler) { 4318 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4319 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4320 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4321 // No need to check this sig any longer 4322 sigaddset(&check_signal_done, sig); 4323 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4324 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4325 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4326 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4327 // No need to check this sig any longer 4328 sigaddset(&check_signal_done, sig); 4329 } 4330 4331 // Dump all the signal 4332 if (sigismember(&check_signal_done, sig)) { 4333 print_signal_handlers(tty, buf, O_BUFLEN); 4334 } 4335 } 4336 4337 extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 4338 4339 extern bool signal_name(int signo, char* buf, size_t len); 4340 4341 const char* os::exception_name(int exception_code, char* buf, size_t size) { 4342 if (0 < exception_code && exception_code <= SIGRTMAX) { 4343 // signal 4344 if (!signal_name(exception_code, buf, size)) { 4345 jio_snprintf(buf, size, "SIG%d", exception_code); 4346 } 4347 return buf; 4348 } else { 4349 return NULL; 4350 } 4351 } 4352 4353 // this is called _before_ the most of global arguments have been parsed 4354 void os::init(void) { 4355 char dummy; /* used to get a guess on initial stack address */ 4356 // first_hrtime = gethrtime(); 4357 4358 // With LinuxThreads the JavaMain thread pid (primordial thread) 4359 // is different than the pid of the java launcher thread. 4360 // So, on Linux, the launcher thread pid is passed to the VM 4361 // via the sun.java.launcher.pid property. 4362 // Use this property instead of getpid() if it was correctly passed. 4363 // See bug 6351349. 4364 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4365 4366 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4367 4368 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4369 4370 init_random(1234567); 4371 4372 ThreadCritical::initialize(); 4373 4374 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4375 if (Linux::page_size() == -1) { 4376 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4377 strerror(errno))); 4378 } 4379 init_page_sizes((size_t) Linux::page_size()); 4380 4381 Linux::initialize_system_info(); 4382 4383 // main_thread points to the aboriginal thread 4384 Linux::_main_thread = pthread_self(); 4385 4386 Linux::clock_init(); 4387 initial_time_count = os::elapsed_counter(); 4388 pthread_mutex_init(&dl_mutex, NULL); 4389 4390 // If the pagesize of the VM is greater than 8K determine the appropriate 4391 // number of initial guard pages. The user can change this with the 4392 // command line arguments, if needed. 4393 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4394 StackYellowPages = 1; 4395 StackRedPages = 1; 4396 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4397 } 4398 } 4399 4400 // To install functions for atexit system call 4401 extern "C" { 4402 static void perfMemory_exit_helper() { 4403 perfMemory_exit(); 4404 } 4405 } 4406 4407 // this is called _after_ the global arguments have been parsed 4408 jint os::init_2(void) 4409 { 4410 Linux::fast_thread_clock_init(); 4411 4412 // Allocate a single page and mark it as readable for safepoint polling 4413 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4414 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 4415 4416 os::set_polling_page( polling_page ); 4417 4418 #ifndef PRODUCT 4419 if(Verbose && PrintMiscellaneous) 4420 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 4421 #endif 4422 4423 if (!UseMembar) { 4424 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4425 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page"); 4426 os::set_memory_serialize_page( mem_serialize_page ); 4427 4428 #ifndef PRODUCT 4429 if(Verbose && PrintMiscellaneous) 4430 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 4431 #endif 4432 } 4433 4434 os::large_page_init(); 4435 4436 // initialize suspend/resume support - must do this before signal_sets_init() 4437 if (SR_initialize() != 0) { 4438 perror("SR_initialize failed"); 4439 return JNI_ERR; 4440 } 4441 4442 Linux::signal_sets_init(); 4443 Linux::install_signal_handlers(); 4444 4445 // Check minimum allowable stack size for thread creation and to initialize 4446 // the java system classes, including StackOverflowError - depends on page 4447 // size. Add a page for compiler2 recursion in main thread. 4448 // Add in 2*BytesPerWord times page size to account for VM stack during 4449 // class initialization depending on 32 or 64 bit VM. 4450 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4451 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4452 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4453 4454 size_t threadStackSizeInBytes = ThreadStackSize * K; 4455 if (threadStackSizeInBytes != 0 && 4456 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4457 tty->print_cr("\nThe stack size specified is too small, " 4458 "Specify at least %dk", 4459 os::Linux::min_stack_allowed/ K); 4460 return JNI_ERR; 4461 } 4462 4463 // Make the stack size a multiple of the page size so that 4464 // the yellow/red zones can be guarded. 4465 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4466 vm_page_size())); 4467 4468 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4469 4470 Linux::libpthread_init(); 4471 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4472 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4473 Linux::glibc_version(), Linux::libpthread_version(), 4474 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4475 } 4476 4477 if (UseNUMA) { 4478 if (!Linux::libnuma_init()) { 4479 UseNUMA = false; 4480 } else { 4481 if ((Linux::numa_max_node() < 1)) { 4482 // There's only one node(they start from 0), disable NUMA. 4483 UseNUMA = false; 4484 } 4485 } 4486 // With SHM large pages we cannot uncommit a page, so there's not way 4487 // we can make the adaptive lgrp chunk resizing work. If the user specified 4488 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and 4489 // disable adaptive resizing. 4490 if (UseNUMA && UseLargePages && UseSHM) { 4491 if (!FLAG_IS_DEFAULT(UseNUMA)) { 4492 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) { 4493 UseLargePages = false; 4494 } else { 4495 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing"); 4496 UseAdaptiveSizePolicy = false; 4497 UseAdaptiveNUMAChunkSizing = false; 4498 } 4499 } else { 4500 UseNUMA = false; 4501 } 4502 } 4503 if (!UseNUMA && ForceNUMA) { 4504 UseNUMA = true; 4505 } 4506 } 4507 4508 if (MaxFDLimit) { 4509 // set the number of file descriptors to max. print out error 4510 // if getrlimit/setrlimit fails but continue regardless. 4511 struct rlimit nbr_files; 4512 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4513 if (status != 0) { 4514 if (PrintMiscellaneous && (Verbose || WizardMode)) 4515 perror("os::init_2 getrlimit failed"); 4516 } else { 4517 nbr_files.rlim_cur = nbr_files.rlim_max; 4518 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4519 if (status != 0) { 4520 if (PrintMiscellaneous && (Verbose || WizardMode)) 4521 perror("os::init_2 setrlimit failed"); 4522 } 4523 } 4524 } 4525 4526 // Initialize lock used to serialize thread creation (see os::create_thread) 4527 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4528 4529 // at-exit methods are called in the reverse order of their registration. 4530 // atexit functions are called on return from main or as a result of a 4531 // call to exit(3C). There can be only 32 of these functions registered 4532 // and atexit() does not set errno. 4533 4534 if (PerfAllowAtExitRegistration) { 4535 // only register atexit functions if PerfAllowAtExitRegistration is set. 4536 // atexit functions can be delayed until process exit time, which 4537 // can be problematic for embedded VM situations. Embedded VMs should 4538 // call DestroyJavaVM() to assure that VM resources are released. 4539 4540 // note: perfMemory_exit_helper atexit function may be removed in 4541 // the future if the appropriate cleanup code can be added to the 4542 // VM_Exit VMOperation's doit method. 4543 if (atexit(perfMemory_exit_helper) != 0) { 4544 warning("os::init2 atexit(perfMemory_exit_helper) failed"); 4545 } 4546 } 4547 4548 // initialize thread priority policy 4549 prio_init(); 4550 4551 return JNI_OK; 4552 } 4553 4554 // this is called at the end of vm_initialization 4555 void os::init_3(void) 4556 { 4557 #ifdef JAVASE_EMBEDDED 4558 // Start the MemNotifyThread 4559 if (LowMemoryProtection) { 4560 MemNotifyThread::start(); 4561 } 4562 return; 4563 #endif 4564 } 4565 4566 // Mark the polling page as unreadable 4567 void os::make_polling_page_unreadable(void) { 4568 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4569 fatal("Could not disable polling page"); 4570 }; 4571 4572 // Mark the polling page as readable 4573 void os::make_polling_page_readable(void) { 4574 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4575 fatal("Could not enable polling page"); 4576 } 4577 }; 4578 4579 int os::active_processor_count() { 4580 // Linux doesn't yet have a (official) notion of processor sets, 4581 // so just return the number of online processors. 4582 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4583 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4584 return online_cpus; 4585 } 4586 4587 void os::set_native_thread_name(const char *name) { 4588 // Not yet implemented. 4589 return; 4590 } 4591 4592 bool os::distribute_processes(uint length, uint* distribution) { 4593 // Not yet implemented. 4594 return false; 4595 } 4596 4597 bool os::bind_to_processor(uint processor_id) { 4598 // Not yet implemented. 4599 return false; 4600 } 4601 4602 /// 4603 4604 void os::SuspendedThreadTask::internal_do_task() { 4605 if (do_suspend(_thread->osthread())) { 4606 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4607 do_task(context); 4608 do_resume(_thread->osthread()); 4609 } 4610 } 4611 4612 class PcFetcher : public os::SuspendedThreadTask { 4613 public: 4614 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4615 ExtendedPC result(); 4616 protected: 4617 void do_task(const os::SuspendedThreadTaskContext& context); 4618 private: 4619 ExtendedPC _epc; 4620 }; 4621 4622 ExtendedPC PcFetcher::result() { 4623 guarantee(is_done(), "task is not done yet."); 4624 return _epc; 4625 } 4626 4627 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4628 Thread* thread = context.thread(); 4629 OSThread* osthread = thread->osthread(); 4630 if (osthread->ucontext() != NULL) { 4631 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4632 } else { 4633 // NULL context is unexpected, double-check this is the VMThread 4634 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4635 } 4636 } 4637 4638 // Suspends the target using the signal mechanism and then grabs the PC before 4639 // resuming the target. Used by the flat-profiler only 4640 ExtendedPC os::get_thread_pc(Thread* thread) { 4641 // Make sure that it is called by the watcher for the VMThread 4642 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4643 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4644 4645 PcFetcher fetcher(thread); 4646 fetcher.run(); 4647 return fetcher.result(); 4648 } 4649 4650 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4651 { 4652 if (is_NPTL()) { 4653 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4654 } else { 4655 #ifndef IA64 4656 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4657 // word back to default 64bit precision if condvar is signaled. Java 4658 // wants 53bit precision. Save and restore current value. 4659 int fpu = get_fpu_control_word(); 4660 #endif // IA64 4661 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4662 #ifndef IA64 4663 set_fpu_control_word(fpu); 4664 #endif // IA64 4665 return status; 4666 } 4667 } 4668 4669 //////////////////////////////////////////////////////////////////////////////// 4670 // debug support 4671 4672 static address same_page(address x, address y) { 4673 int page_bits = -os::vm_page_size(); 4674 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits)) 4675 return x; 4676 else if (x > y) 4677 return (address)(intptr_t(y) | ~page_bits) + 1; 4678 else 4679 return (address)(intptr_t(y) & page_bits); 4680 } 4681 4682 bool os::find(address addr, outputStream* st) { 4683 Dl_info dlinfo; 4684 memset(&dlinfo, 0, sizeof(dlinfo)); 4685 if (dladdr(addr, &dlinfo)) { 4686 st->print(PTR_FORMAT ": ", addr); 4687 if (dlinfo.dli_sname != NULL) { 4688 st->print("%s+%#x", dlinfo.dli_sname, 4689 addr - (intptr_t)dlinfo.dli_saddr); 4690 } else if (dlinfo.dli_fname) { 4691 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4692 } else { 4693 st->print("<absolute address>"); 4694 } 4695 if (dlinfo.dli_fname) { 4696 st->print(" in %s", dlinfo.dli_fname); 4697 } 4698 if (dlinfo.dli_fbase) { 4699 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4700 } 4701 st->cr(); 4702 4703 if (Verbose) { 4704 // decode some bytes around the PC 4705 address begin = same_page(addr-40, addr); 4706 address end = same_page(addr+40, addr); 4707 address lowest = (address) dlinfo.dli_sname; 4708 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4709 if (begin < lowest) begin = lowest; 4710 Dl_info dlinfo2; 4711 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr 4712 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 4713 end = (address) dlinfo2.dli_saddr; 4714 Disassembler::decode(begin, end, st); 4715 } 4716 return true; 4717 } 4718 return false; 4719 } 4720 4721 //////////////////////////////////////////////////////////////////////////////// 4722 // misc 4723 4724 // This does not do anything on Linux. This is basically a hook for being 4725 // able to use structured exception handling (thread-local exception filters) 4726 // on, e.g., Win32. 4727 void 4728 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4729 JavaCallArguments* args, Thread* thread) { 4730 f(value, method, args, thread); 4731 } 4732 4733 void os::print_statistics() { 4734 } 4735 4736 int os::message_box(const char* title, const char* message) { 4737 int i; 4738 fdStream err(defaultStream::error_fd()); 4739 for (i = 0; i < 78; i++) err.print_raw("="); 4740 err.cr(); 4741 err.print_raw_cr(title); 4742 for (i = 0; i < 78; i++) err.print_raw("-"); 4743 err.cr(); 4744 err.print_raw_cr(message); 4745 for (i = 0; i < 78; i++) err.print_raw("="); 4746 err.cr(); 4747 4748 char buf[16]; 4749 // Prevent process from exiting upon "read error" without consuming all CPU 4750 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 4751 4752 return buf[0] == 'y' || buf[0] == 'Y'; 4753 } 4754 4755 int os::stat(const char *path, struct stat *sbuf) { 4756 char pathbuf[MAX_PATH]; 4757 if (strlen(path) > MAX_PATH - 1) { 4758 errno = ENAMETOOLONG; 4759 return -1; 4760 } 4761 os::native_path(strcpy(pathbuf, path)); 4762 return ::stat(pathbuf, sbuf); 4763 } 4764 4765 bool os::check_heap(bool force) { 4766 return true; 4767 } 4768 4769 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 4770 return ::vsnprintf(buf, count, format, args); 4771 } 4772 4773 // Is a (classpath) directory empty? 4774 bool os::dir_is_empty(const char* path) { 4775 DIR *dir = NULL; 4776 struct dirent *ptr; 4777 4778 dir = opendir(path); 4779 if (dir == NULL) return true; 4780 4781 /* Scan the directory */ 4782 bool result = true; 4783 char buf[sizeof(struct dirent) + MAX_PATH]; 4784 while (result && (ptr = ::readdir(dir)) != NULL) { 4785 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 4786 result = false; 4787 } 4788 } 4789 closedir(dir); 4790 return result; 4791 } 4792 4793 // This code originates from JDK's sysOpen and open64_w 4794 // from src/solaris/hpi/src/system_md.c 4795 4796 #ifndef O_DELETE 4797 #define O_DELETE 0x10000 4798 #endif 4799 4800 // Open a file. Unlink the file immediately after open returns 4801 // if the specified oflag has the O_DELETE flag set. 4802 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 4803 4804 int os::open(const char *path, int oflag, int mode) { 4805 4806 if (strlen(path) > MAX_PATH - 1) { 4807 errno = ENAMETOOLONG; 4808 return -1; 4809 } 4810 int fd; 4811 int o_delete = (oflag & O_DELETE); 4812 oflag = oflag & ~O_DELETE; 4813 4814 fd = ::open64(path, oflag, mode); 4815 if (fd == -1) return -1; 4816 4817 //If the open succeeded, the file might still be a directory 4818 { 4819 struct stat64 buf64; 4820 int ret = ::fstat64(fd, &buf64); 4821 int st_mode = buf64.st_mode; 4822 4823 if (ret != -1) { 4824 if ((st_mode & S_IFMT) == S_IFDIR) { 4825 errno = EISDIR; 4826 ::close(fd); 4827 return -1; 4828 } 4829 } else { 4830 ::close(fd); 4831 return -1; 4832 } 4833 } 4834 4835 /* 4836 * All file descriptors that are opened in the JVM and not 4837 * specifically destined for a subprocess should have the 4838 * close-on-exec flag set. If we don't set it, then careless 3rd 4839 * party native code might fork and exec without closing all 4840 * appropriate file descriptors (e.g. as we do in closeDescriptors in 4841 * UNIXProcess.c), and this in turn might: 4842 * 4843 * - cause end-of-file to fail to be detected on some file 4844 * descriptors, resulting in mysterious hangs, or 4845 * 4846 * - might cause an fopen in the subprocess to fail on a system 4847 * suffering from bug 1085341. 4848 * 4849 * (Yes, the default setting of the close-on-exec flag is a Unix 4850 * design flaw) 4851 * 4852 * See: 4853 * 1085341: 32-bit stdio routines should support file descriptors >255 4854 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 4855 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 4856 */ 4857 #ifdef FD_CLOEXEC 4858 { 4859 int flags = ::fcntl(fd, F_GETFD); 4860 if (flags != -1) 4861 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 4862 } 4863 #endif 4864 4865 if (o_delete != 0) { 4866 ::unlink(path); 4867 } 4868 return fd; 4869 } 4870 4871 4872 // create binary file, rewriting existing file if required 4873 int os::create_binary_file(const char* path, bool rewrite_existing) { 4874 int oflags = O_WRONLY | O_CREAT; 4875 if (!rewrite_existing) { 4876 oflags |= O_EXCL; 4877 } 4878 return ::open64(path, oflags, S_IREAD | S_IWRITE); 4879 } 4880 4881 // return current position of file pointer 4882 jlong os::current_file_offset(int fd) { 4883 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 4884 } 4885 4886 // move file pointer to the specified offset 4887 jlong os::seek_to_file_offset(int fd, jlong offset) { 4888 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 4889 } 4890 4891 // This code originates from JDK's sysAvailable 4892 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c 4893 4894 int os::available(int fd, jlong *bytes) { 4895 jlong cur, end; 4896 int mode; 4897 struct stat64 buf64; 4898 4899 if (::fstat64(fd, &buf64) >= 0) { 4900 mode = buf64.st_mode; 4901 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 4902 /* 4903 * XXX: is the following call interruptible? If so, this might 4904 * need to go through the INTERRUPT_IO() wrapper as for other 4905 * blocking, interruptible calls in this file. 4906 */ 4907 int n; 4908 if (::ioctl(fd, FIONREAD, &n) >= 0) { 4909 *bytes = n; 4910 return 1; 4911 } 4912 } 4913 } 4914 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 4915 return 0; 4916 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 4917 return 0; 4918 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 4919 return 0; 4920 } 4921 *bytes = end - cur; 4922 return 1; 4923 } 4924 4925 int os::socket_available(int fd, jint *pbytes) { 4926 // Linux doc says EINTR not returned, unlike Solaris 4927 int ret = ::ioctl(fd, FIONREAD, pbytes); 4928 4929 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 4930 // is expected to return 0 on failure and 1 on success to the jdk. 4931 return (ret < 0) ? 0 : 1; 4932 } 4933 4934 // Map a block of memory. 4935 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 4936 char *addr, size_t bytes, bool read_only, 4937 bool allow_exec) { 4938 int prot; 4939 int flags = MAP_PRIVATE; 4940 4941 if (read_only) { 4942 prot = PROT_READ; 4943 } else { 4944 prot = PROT_READ | PROT_WRITE; 4945 } 4946 4947 if (allow_exec) { 4948 prot |= PROT_EXEC; 4949 } 4950 4951 if (addr != NULL) { 4952 flags |= MAP_FIXED; 4953 } 4954 4955 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 4956 fd, file_offset); 4957 if (mapped_address == MAP_FAILED) { 4958 return NULL; 4959 } 4960 return mapped_address; 4961 } 4962 4963 4964 // Remap a block of memory. 4965 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 4966 char *addr, size_t bytes, bool read_only, 4967 bool allow_exec) { 4968 // same as map_memory() on this OS 4969 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 4970 allow_exec); 4971 } 4972 4973 4974 // Unmap a block of memory. 4975 bool os::pd_unmap_memory(char* addr, size_t bytes) { 4976 return munmap(addr, bytes) == 0; 4977 } 4978 4979 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 4980 4981 static clockid_t thread_cpu_clockid(Thread* thread) { 4982 pthread_t tid = thread->osthread()->pthread_id(); 4983 clockid_t clockid; 4984 4985 // Get thread clockid 4986 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 4987 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 4988 return clockid; 4989 } 4990 4991 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 4992 // are used by JVM M&M and JVMTI to get user+sys or user CPU time 4993 // of a thread. 4994 // 4995 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns 4996 // the fast estimate available on the platform. 4997 4998 jlong os::current_thread_cpu_time() { 4999 if (os::Linux::supports_fast_thread_cpu_time()) { 5000 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5001 } else { 5002 // return user + sys since the cost is the same 5003 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5004 } 5005 } 5006 5007 jlong os::thread_cpu_time(Thread* thread) { 5008 // consistent with what current_thread_cpu_time() returns 5009 if (os::Linux::supports_fast_thread_cpu_time()) { 5010 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5011 } else { 5012 return slow_thread_cpu_time(thread, true /* user + sys */); 5013 } 5014 } 5015 5016 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5017 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5018 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5019 } else { 5020 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5021 } 5022 } 5023 5024 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5025 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5026 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5027 } else { 5028 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5029 } 5030 } 5031 5032 // 5033 // -1 on error. 5034 // 5035 5036 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5037 static bool proc_pid_cpu_avail = true; 5038 static bool proc_task_unchecked = true; 5039 static const char *proc_stat_path = "/proc/%d/stat"; 5040 pid_t tid = thread->osthread()->thread_id(); 5041 int i; 5042 char *s; 5043 char stat[2048]; 5044 int statlen; 5045 char proc_name[64]; 5046 int count; 5047 long sys_time, user_time; 5048 char string[64]; 5049 char cdummy; 5050 int idummy; 5051 long ldummy; 5052 FILE *fp; 5053 5054 // We first try accessing /proc/<pid>/cpu since this is faster to 5055 // process. If this file is not present (linux kernels 2.5 and above) 5056 // then we open /proc/<pid>/stat. 5057 if ( proc_pid_cpu_avail ) { 5058 sprintf(proc_name, "/proc/%d/cpu", tid); 5059 fp = fopen(proc_name, "r"); 5060 if ( fp != NULL ) { 5061 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time); 5062 fclose(fp); 5063 if ( count != 3 ) return -1; 5064 5065 if (user_sys_cpu_time) { 5066 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5067 } else { 5068 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5069 } 5070 } 5071 else proc_pid_cpu_avail = false; 5072 } 5073 5074 // The /proc/<tid>/stat aggregates per-process usage on 5075 // new Linux kernels 2.6+ where NPTL is supported. 5076 // The /proc/self/task/<tid>/stat still has the per-thread usage. 5077 // See bug 6328462. 5078 // There can be no directory /proc/self/task on kernels 2.4 with NPTL 5079 // and possibly in some other cases, so we check its availability. 5080 if (proc_task_unchecked && os::Linux::is_NPTL()) { 5081 // This is executed only once 5082 proc_task_unchecked = false; 5083 fp = fopen("/proc/self/task", "r"); 5084 if (fp != NULL) { 5085 proc_stat_path = "/proc/self/task/%d/stat"; 5086 fclose(fp); 5087 } 5088 } 5089 5090 sprintf(proc_name, proc_stat_path, tid); 5091 fp = fopen(proc_name, "r"); 5092 if ( fp == NULL ) return -1; 5093 statlen = fread(stat, 1, 2047, fp); 5094 stat[statlen] = '\0'; 5095 fclose(fp); 5096 5097 // Skip pid and the command string. Note that we could be dealing with 5098 // weird command names, e.g. user could decide to rename java launcher 5099 // to "java 1.4.2 :)", then the stat file would look like 5100 // 1234 (java 1.4.2 :)) R ... ... 5101 // We don't really need to know the command string, just find the last 5102 // occurrence of ")" and then start parsing from there. See bug 4726580. 5103 s = strrchr(stat, ')'); 5104 i = 0; 5105 if (s == NULL ) return -1; 5106 5107 // Skip blank chars 5108 do s++; while (isspace(*s)); 5109 5110 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5111 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5112 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5113 &user_time, &sys_time); 5114 if ( count != 13 ) return -1; 5115 if (user_sys_cpu_time) { 5116 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5117 } else { 5118 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5119 } 5120 } 5121 5122 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5123 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5124 info_ptr->may_skip_backward = false; // elapsed time not wall time 5125 info_ptr->may_skip_forward = false; // elapsed time not wall time 5126 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5127 } 5128 5129 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5130 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5131 info_ptr->may_skip_backward = false; // elapsed time not wall time 5132 info_ptr->may_skip_forward = false; // elapsed time not wall time 5133 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5134 } 5135 5136 bool os::is_thread_cpu_time_supported() { 5137 return true; 5138 } 5139 5140 // System loadavg support. Returns -1 if load average cannot be obtained. 5141 // Linux doesn't yet have a (official) notion of processor sets, 5142 // so just return the system wide load average. 5143 int os::loadavg(double loadavg[], int nelem) { 5144 return ::getloadavg(loadavg, nelem); 5145 } 5146 5147 void os::pause() { 5148 char filename[MAX_PATH]; 5149 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5150 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5151 } else { 5152 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5153 } 5154 5155 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5156 if (fd != -1) { 5157 struct stat buf; 5158 ::close(fd); 5159 while (::stat(filename, &buf) == 0) { 5160 (void)::poll(NULL, 0, 100); 5161 } 5162 } else { 5163 jio_fprintf(stderr, 5164 "Could not open pause file '%s', continuing immediately.\n", filename); 5165 } 5166 } 5167 5168 5169 // Refer to the comments in os_solaris.cpp park-unpark. 5170 // 5171 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5172 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5173 // For specifics regarding the bug see GLIBC BUGID 261237 : 5174 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5175 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5176 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5177 // is used. (The simple C test-case provided in the GLIBC bug report manifests the 5178 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5179 // and monitorenter when we're using 1-0 locking. All those operations may result in 5180 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5181 // of libpthread avoids the problem, but isn't practical. 5182 // 5183 // Possible remedies: 5184 // 5185 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5186 // This is palliative and probabilistic, however. If the thread is preempted 5187 // between the call to compute_abstime() and pthread_cond_timedwait(), more 5188 // than the minimum period may have passed, and the abstime may be stale (in the 5189 // past) resultin in a hang. Using this technique reduces the odds of a hang 5190 // but the JVM is still vulnerable, particularly on heavily loaded systems. 5191 // 5192 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5193 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5194 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5195 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5196 // thread. 5197 // 5198 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5199 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5200 // a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5201 // This also works well. In fact it avoids kernel-level scalability impediments 5202 // on certain platforms that don't handle lots of active pthread_cond_timedwait() 5203 // timers in a graceful fashion. 5204 // 5205 // 4. When the abstime value is in the past it appears that control returns 5206 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5207 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5208 // can avoid the problem by reinitializing the condvar -- by cond_destroy() 5209 // followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5210 // It may be possible to avoid reinitialization by checking the return 5211 // value from pthread_cond_timedwait(). In addition to reinitializing the 5212 // condvar we must establish the invariant that cond_signal() is only called 5213 // within critical sections protected by the adjunct mutex. This prevents 5214 // cond_signal() from "seeing" a condvar that's in the midst of being 5215 // reinitialized or that is corrupt. Sadly, this invariant obviates the 5216 // desirable signal-after-unlock optimization that avoids futile context switching. 5217 // 5218 // I'm also concerned that some versions of NTPL might allocate an auxilliary 5219 // structure when a condvar is used or initialized. cond_destroy() would 5220 // release the helper structure. Our reinitialize-after-timedwait fix 5221 // put excessive stress on malloc/free and locks protecting the c-heap. 5222 // 5223 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5224 // It may be possible to refine (4) by checking the kernel and NTPL verisons 5225 // and only enabling the work-around for vulnerable environments. 5226 5227 // utility to compute the abstime argument to timedwait: 5228 // millis is the relative timeout time 5229 // abstime will be the absolute timeout time 5230 // TODO: replace compute_abstime() with unpackTime() 5231 5232 static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5233 if (millis < 0) millis = 0; 5234 struct timeval now; 5235 int status = gettimeofday(&now, NULL); 5236 assert(status == 0, "gettimeofday"); 5237 jlong seconds = millis / 1000; 5238 millis %= 1000; 5239 if (seconds > 50000000) { // see man cond_timedwait(3T) 5240 seconds = 50000000; 5241 } 5242 abstime->tv_sec = now.tv_sec + seconds; 5243 long usec = now.tv_usec + millis * 1000; 5244 if (usec >= 1000000) { 5245 abstime->tv_sec += 1; 5246 usec -= 1000000; 5247 } 5248 abstime->tv_nsec = usec * 1000; 5249 return abstime; 5250 } 5251 5252 5253 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 5254 // Conceptually TryPark() should be equivalent to park(0). 5255 5256 int os::PlatformEvent::TryPark() { 5257 for (;;) { 5258 const int v = _Event ; 5259 guarantee ((v == 0) || (v == 1), "invariant") ; 5260 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 5261 } 5262 } 5263 5264 void os::PlatformEvent::park() { // AKA "down()" 5265 // Invariant: Only the thread associated with the Event/PlatformEvent 5266 // may call park(). 5267 // TODO: assert that _Assoc != NULL or _Assoc == Self 5268 int v ; 5269 for (;;) { 5270 v = _Event ; 5271 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5272 } 5273 guarantee (v >= 0, "invariant") ; 5274 if (v == 0) { 5275 // Do this the hard way by blocking ... 5276 int status = pthread_mutex_lock(_mutex); 5277 assert_status(status == 0, status, "mutex_lock"); 5278 guarantee (_nParked == 0, "invariant") ; 5279 ++ _nParked ; 5280 while (_Event < 0) { 5281 status = pthread_cond_wait(_cond, _mutex); 5282 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5283 // Treat this the same as if the wait was interrupted 5284 if (status == ETIME) { status = EINTR; } 5285 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5286 } 5287 -- _nParked ; 5288 5289 _Event = 0 ; 5290 status = pthread_mutex_unlock(_mutex); 5291 assert_status(status == 0, status, "mutex_unlock"); 5292 // Paranoia to ensure our locked and lock-free paths interact 5293 // correctly with each other. 5294 OrderAccess::fence(); 5295 } 5296 guarantee (_Event >= 0, "invariant") ; 5297 } 5298 5299 int os::PlatformEvent::park(jlong millis) { 5300 guarantee (_nParked == 0, "invariant") ; 5301 5302 int v ; 5303 for (;;) { 5304 v = _Event ; 5305 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5306 } 5307 guarantee (v >= 0, "invariant") ; 5308 if (v != 0) return OS_OK ; 5309 5310 // We do this the hard way, by blocking the thread. 5311 // Consider enforcing a minimum timeout value. 5312 struct timespec abst; 5313 compute_abstime(&abst, millis); 5314 5315 int ret = OS_TIMEOUT; 5316 int status = pthread_mutex_lock(_mutex); 5317 assert_status(status == 0, status, "mutex_lock"); 5318 guarantee (_nParked == 0, "invariant") ; 5319 ++_nParked ; 5320 5321 // Object.wait(timo) will return because of 5322 // (a) notification 5323 // (b) timeout 5324 // (c) thread.interrupt 5325 // 5326 // Thread.interrupt and object.notify{All} both call Event::set. 5327 // That is, we treat thread.interrupt as a special case of notification. 5328 // The underlying Solaris implementation, cond_timedwait, admits 5329 // spurious/premature wakeups, but the JLS/JVM spec prevents the 5330 // JVM from making those visible to Java code. As such, we must 5331 // filter out spurious wakeups. We assume all ETIME returns are valid. 5332 // 5333 // TODO: properly differentiate simultaneous notify+interrupt. 5334 // In that case, we should propagate the notify to another waiter. 5335 5336 while (_Event < 0) { 5337 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5338 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5339 pthread_cond_destroy (_cond); 5340 pthread_cond_init (_cond, NULL) ; 5341 } 5342 assert_status(status == 0 || status == EINTR || 5343 status == ETIME || status == ETIMEDOUT, 5344 status, "cond_timedwait"); 5345 if (!FilterSpuriousWakeups) break ; // previous semantics 5346 if (status == ETIME || status == ETIMEDOUT) break ; 5347 // We consume and ignore EINTR and spurious wakeups. 5348 } 5349 --_nParked ; 5350 if (_Event >= 0) { 5351 ret = OS_OK; 5352 } 5353 _Event = 0 ; 5354 status = pthread_mutex_unlock(_mutex); 5355 assert_status(status == 0, status, "mutex_unlock"); 5356 assert (_nParked == 0, "invariant") ; 5357 // Paranoia to ensure our locked and lock-free paths interact 5358 // correctly with each other. 5359 OrderAccess::fence(); 5360 return ret; 5361 } 5362 5363 void os::PlatformEvent::unpark() { 5364 // Transitions for _Event: 5365 // 0 :=> 1 5366 // 1 :=> 1 5367 // -1 :=> either 0 or 1; must signal target thread 5368 // That is, we can safely transition _Event from -1 to either 5369 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back 5370 // unpark() calls. 5371 // See also: "Semaphores in Plan 9" by Mullender & Cox 5372 // 5373 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5374 // that it will take two back-to-back park() calls for the owning 5375 // thread to block. This has the benefit of forcing a spurious return 5376 // from the first park() call after an unpark() call which will help 5377 // shake out uses of park() and unpark() without condition variables. 5378 5379 if (Atomic::xchg(1, &_Event) >= 0) return; 5380 5381 // Wait for the thread associated with the event to vacate 5382 int status = pthread_mutex_lock(_mutex); 5383 assert_status(status == 0, status, "mutex_lock"); 5384 int AnyWaiters = _nParked; 5385 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5386 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5387 AnyWaiters = 0; 5388 pthread_cond_signal(_cond); 5389 } 5390 status = pthread_mutex_unlock(_mutex); 5391 assert_status(status == 0, status, "mutex_unlock"); 5392 if (AnyWaiters != 0) { 5393 status = pthread_cond_signal(_cond); 5394 assert_status(status == 0, status, "cond_signal"); 5395 } 5396 5397 // Note that we signal() _after dropping the lock for "immortal" Events. 5398 // This is safe and avoids a common class of futile wakeups. In rare 5399 // circumstances this can cause a thread to return prematurely from 5400 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5401 // simply re-test the condition and re-park itself. 5402 } 5403 5404 5405 // JSR166 5406 // ------------------------------------------------------- 5407 5408 /* 5409 * The solaris and linux implementations of park/unpark are fairly 5410 * conservative for now, but can be improved. They currently use a 5411 * mutex/condvar pair, plus a a count. 5412 * Park decrements count if > 0, else does a condvar wait. Unpark 5413 * sets count to 1 and signals condvar. Only one thread ever waits 5414 * on the condvar. Contention seen when trying to park implies that someone 5415 * is unparking you, so don't wait. And spurious returns are fine, so there 5416 * is no need to track notifications. 5417 */ 5418 5419 #define MAX_SECS 100000000 5420 /* 5421 * This code is common to linux and solaris and will be moved to a 5422 * common place in dolphin. 5423 * 5424 * The passed in time value is either a relative time in nanoseconds 5425 * or an absolute time in milliseconds. Either way it has to be unpacked 5426 * into suitable seconds and nanoseconds components and stored in the 5427 * given timespec structure. 5428 * Given time is a 64-bit value and the time_t used in the timespec is only 5429 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 5430 * overflow if times way in the future are given. Further on Solaris versions 5431 * prior to 10 there is a restriction (see cond_timedwait) that the specified 5432 * number of seconds, in abstime, is less than current_time + 100,000,000. 5433 * As it will be 28 years before "now + 100000000" will overflow we can 5434 * ignore overflow and just impose a hard-limit on seconds using the value 5435 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 5436 * years from "now". 5437 */ 5438 5439 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5440 assert (time > 0, "convertTime"); 5441 5442 struct timeval now; 5443 int status = gettimeofday(&now, NULL); 5444 assert(status == 0, "gettimeofday"); 5445 5446 time_t max_secs = now.tv_sec + MAX_SECS; 5447 5448 if (isAbsolute) { 5449 jlong secs = time / 1000; 5450 if (secs > max_secs) { 5451 absTime->tv_sec = max_secs; 5452 } 5453 else { 5454 absTime->tv_sec = secs; 5455 } 5456 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5457 } 5458 else { 5459 jlong secs = time / NANOSECS_PER_SEC; 5460 if (secs >= MAX_SECS) { 5461 absTime->tv_sec = max_secs; 5462 absTime->tv_nsec = 0; 5463 } 5464 else { 5465 absTime->tv_sec = now.tv_sec + secs; 5466 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5467 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5468 absTime->tv_nsec -= NANOSECS_PER_SEC; 5469 ++absTime->tv_sec; // note: this must be <= max_secs 5470 } 5471 } 5472 } 5473 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5474 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5475 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5476 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5477 } 5478 5479 void Parker::park(bool isAbsolute, jlong time) { 5480 // Ideally we'd do something useful while spinning, such 5481 // as calling unpackTime(). 5482 5483 // Optional fast-path check: 5484 // Return immediately if a permit is available. 5485 // We depend on Atomic::xchg() having full barrier semantics 5486 // since we are doing a lock-free update to _counter. 5487 if (Atomic::xchg(0, &_counter) > 0) return; 5488 5489 Thread* thread = Thread::current(); 5490 assert(thread->is_Java_thread(), "Must be JavaThread"); 5491 JavaThread *jt = (JavaThread *)thread; 5492 5493 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5494 // Check interrupt before trying to wait 5495 if (Thread::is_interrupted(thread, false)) { 5496 return; 5497 } 5498 5499 // Next, demultiplex/decode time arguments 5500 timespec absTime; 5501 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5502 return; 5503 } 5504 if (time > 0) { 5505 unpackTime(&absTime, isAbsolute, time); 5506 } 5507 5508 5509 // Enter safepoint region 5510 // Beware of deadlocks such as 6317397. 5511 // The per-thread Parker:: mutex is a classic leaf-lock. 5512 // In particular a thread must never block on the Threads_lock while 5513 // holding the Parker:: mutex. If safepoints are pending both the 5514 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5515 ThreadBlockInVM tbivm(jt); 5516 5517 // Don't wait if cannot get lock since interference arises from 5518 // unblocking. Also. check interrupt before trying wait 5519 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5520 return; 5521 } 5522 5523 int status ; 5524 if (_counter > 0) { // no wait needed 5525 _counter = 0; 5526 status = pthread_mutex_unlock(_mutex); 5527 assert (status == 0, "invariant") ; 5528 // Paranoia to ensure our locked and lock-free paths interact 5529 // correctly with each other and Java-level accesses. 5530 OrderAccess::fence(); 5531 return; 5532 } 5533 5534 #ifdef ASSERT 5535 // Don't catch signals while blocked; let the running threads have the signals. 5536 // (This allows a debugger to break into the running thread.) 5537 sigset_t oldsigs; 5538 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5539 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5540 #endif 5541 5542 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5543 jt->set_suspend_equivalent(); 5544 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5545 5546 if (time == 0) { 5547 status = pthread_cond_wait (_cond, _mutex) ; 5548 } else { 5549 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ; 5550 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5551 pthread_cond_destroy (_cond) ; 5552 pthread_cond_init (_cond, NULL); 5553 } 5554 } 5555 assert_status(status == 0 || status == EINTR || 5556 status == ETIME || status == ETIMEDOUT, 5557 status, "cond_timedwait"); 5558 5559 #ifdef ASSERT 5560 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5561 #endif 5562 5563 _counter = 0 ; 5564 status = pthread_mutex_unlock(_mutex) ; 5565 assert_status(status == 0, status, "invariant") ; 5566 // Paranoia to ensure our locked and lock-free paths interact 5567 // correctly with each other and Java-level accesses. 5568 OrderAccess::fence(); 5569 5570 // If externally suspended while waiting, re-suspend 5571 if (jt->handle_special_suspend_equivalent_condition()) { 5572 jt->java_suspend_self(); 5573 } 5574 } 5575 5576 void Parker::unpark() { 5577 int s, status ; 5578 status = pthread_mutex_lock(_mutex); 5579 assert (status == 0, "invariant") ; 5580 s = _counter; 5581 _counter = 1; 5582 if (s < 1) { 5583 if (WorkAroundNPTLTimedWaitHang) { 5584 status = pthread_cond_signal (_cond) ; 5585 assert (status == 0, "invariant") ; 5586 status = pthread_mutex_unlock(_mutex); 5587 assert (status == 0, "invariant") ; 5588 } else { 5589 status = pthread_mutex_unlock(_mutex); 5590 assert (status == 0, "invariant") ; 5591 status = pthread_cond_signal (_cond) ; 5592 assert (status == 0, "invariant") ; 5593 } 5594 } else { 5595 pthread_mutex_unlock(_mutex); 5596 assert (status == 0, "invariant") ; 5597 } 5598 } 5599 5600 5601 extern char** environ; 5602 5603 #ifndef __NR_fork 5604 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5605 #endif 5606 5607 #ifndef __NR_execve 5608 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5609 #endif 5610 5611 // Run the specified command in a separate process. Return its exit value, 5612 // or -1 on failure (e.g. can't fork a new process). 5613 // Unlike system(), this function can be called from signal handler. It 5614 // doesn't block SIGINT et al. 5615 int os::fork_and_exec(char* cmd) { 5616 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5617 5618 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5619 // pthread_atfork handlers and reset pthread library. All we need is a 5620 // separate process to execve. Make a direct syscall to fork process. 5621 // On IA64 there's no fork syscall, we have to use fork() and hope for 5622 // the best... 5623 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5624 IA64_ONLY(fork();) 5625 5626 if (pid < 0) { 5627 // fork failed 5628 return -1; 5629 5630 } else if (pid == 0) { 5631 // child process 5632 5633 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5634 // first to kill every thread on the thread list. Because this list is 5635 // not reset by fork() (see notes above), execve() will instead kill 5636 // every thread in the parent process. We know this is the only thread 5637 // in the new process, so make a system call directly. 5638 // IA64 should use normal execve() from glibc to match the glibc fork() 5639 // above. 5640 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5641 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5642 5643 // execve failed 5644 _exit(-1); 5645 5646 } else { 5647 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5648 // care about the actual exit code, for now. 5649 5650 int status; 5651 5652 // Wait for the child process to exit. This returns immediately if 5653 // the child has already exited. */ 5654 while (waitpid(pid, &status, 0) < 0) { 5655 switch (errno) { 5656 case ECHILD: return 0; 5657 case EINTR: break; 5658 default: return -1; 5659 } 5660 } 5661 5662 if (WIFEXITED(status)) { 5663 // The child exited normally; get its exit code. 5664 return WEXITSTATUS(status); 5665 } else if (WIFSIGNALED(status)) { 5666 // The child exited because of a signal 5667 // The best value to return is 0x80 + signal number, 5668 // because that is what all Unix shells do, and because 5669 // it allows callers to distinguish between process exit and 5670 // process death by signal. 5671 return 0x80 + WTERMSIG(status); 5672 } else { 5673 // Unknown exit code; pass it through 5674 return status; 5675 } 5676 } 5677 } 5678 5679 // is_headless_jre() 5680 // 5681 // Test for the existence of xawt/libmawt.so or libawt_xawt.so 5682 // in order to report if we are running in a headless jre 5683 // 5684 // Since JDK8 xawt/libmawt.so was moved into the same directory 5685 // as libawt.so, and renamed libawt_xawt.so 5686 // 5687 bool os::is_headless_jre() { 5688 struct stat statbuf; 5689 char buf[MAXPATHLEN]; 5690 char libmawtpath[MAXPATHLEN]; 5691 const char *xawtstr = "/xawt/libmawt.so"; 5692 const char *new_xawtstr = "/libawt_xawt.so"; 5693 char *p; 5694 5695 // Get path to libjvm.so 5696 os::jvm_path(buf, sizeof(buf)); 5697 5698 // Get rid of libjvm.so 5699 p = strrchr(buf, '/'); 5700 if (p == NULL) return false; 5701 else *p = '\0'; 5702 5703 // Get rid of client or server 5704 p = strrchr(buf, '/'); 5705 if (p == NULL) return false; 5706 else *p = '\0'; 5707 5708 // check xawt/libmawt.so 5709 strcpy(libmawtpath, buf); 5710 strcat(libmawtpath, xawtstr); 5711 if (::stat(libmawtpath, &statbuf) == 0) return false; 5712 5713 // check libawt_xawt.so 5714 strcpy(libmawtpath, buf); 5715 strcat(libmawtpath, new_xawtstr); 5716 if (::stat(libmawtpath, &statbuf) == 0) return false; 5717 5718 return true; 5719 } 5720 5721 // Get the default path to the core file 5722 // Returns the length of the string 5723 int os::get_core_path(char* buffer, size_t bufferSize) { 5724 const char* p = get_current_directory(buffer, bufferSize); 5725 5726 if (p == NULL) { 5727 assert(p != NULL, "failed to get current directory"); 5728 return 0; 5729 } 5730 5731 return strlen(buffer); 5732 } 5733 5734 #ifdef JAVASE_EMBEDDED 5735 // 5736 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 5737 // 5738 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 5739 5740 // ctor 5741 // 5742 MemNotifyThread::MemNotifyThread(int fd): Thread() { 5743 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 5744 _fd = fd; 5745 5746 if (os::create_thread(this, os::os_thread)) { 5747 _memnotify_thread = this; 5748 os::set_priority(this, NearMaxPriority); 5749 os::start_thread(this); 5750 } 5751 } 5752 5753 // Where all the work gets done 5754 // 5755 void MemNotifyThread::run() { 5756 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 5757 5758 // Set up the select arguments 5759 fd_set rfds; 5760 if (_fd != -1) { 5761 FD_ZERO(&rfds); 5762 FD_SET(_fd, &rfds); 5763 } 5764 5765 // Now wait for the mem_notify device to wake up 5766 while (1) { 5767 // Wait for the mem_notify device to signal us.. 5768 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 5769 if (rc == -1) { 5770 perror("select!\n"); 5771 break; 5772 } else if (rc) { 5773 //ssize_t free_before = os::available_memory(); 5774 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 5775 5776 // The kernel is telling us there is not much memory left... 5777 // try to do something about that 5778 5779 // If we are not already in a GC, try one. 5780 if (!Universe::heap()->is_gc_active()) { 5781 Universe::heap()->collect(GCCause::_allocation_failure); 5782 5783 //ssize_t free_after = os::available_memory(); 5784 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 5785 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 5786 } 5787 // We might want to do something like the following if we find the GC's are not helping... 5788 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 5789 } 5790 } 5791 } 5792 5793 // 5794 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 5795 // 5796 void MemNotifyThread::start() { 5797 int fd; 5798 fd = open ("/dev/mem_notify", O_RDONLY, 0); 5799 if (fd < 0) { 5800 return; 5801 } 5802 5803 if (memnotify_thread() == NULL) { 5804 new MemNotifyThread(fd); 5805 } 5806 } 5807 5808 #endif // JAVASE_EMBEDDED