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