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