1 /* 2 * Copyright (c) 1999, 2014, 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/init.hpp" 48 #include "runtime/java.hpp" 49 #include "runtime/javaCalls.hpp" 50 #include "runtime/mutexLocker.hpp" 51 #include "runtime/objectMonitor.hpp" 52 #include "runtime/osThread.hpp" 53 #include "runtime/perfMemory.hpp" 54 #include "runtime/sharedRuntime.hpp" 55 #include "runtime/statSampler.hpp" 56 #include "runtime/stubRoutines.hpp" 57 #include "runtime/thread.inline.hpp" 58 #include "runtime/threadCritical.hpp" 59 #include "runtime/timer.hpp" 60 #include "services/attachListener.hpp" 61 #include "services/memTracker.hpp" 62 #include "services/runtimeService.hpp" 63 #include "utilities/decoder.hpp" 64 #include "utilities/defaultStream.hpp" 65 #include "utilities/events.hpp" 66 #include "utilities/elfFile.hpp" 67 #include "utilities/growableArray.hpp" 68 #include "utilities/vmError.hpp" 69 70 // put OS-includes here 71 # include <sys/types.h> 72 # include <sys/mman.h> 73 # include <sys/stat.h> 74 # include <sys/select.h> 75 # include <pthread.h> 76 # include <signal.h> 77 # include <errno.h> 78 # include <dlfcn.h> 79 # include <stdio.h> 80 # include <unistd.h> 81 # include <sys/resource.h> 82 # include <pthread.h> 83 # include <sys/stat.h> 84 # include <sys/time.h> 85 # include <sys/times.h> 86 # include <sys/utsname.h> 87 # include <sys/socket.h> 88 # include <sys/wait.h> 89 # include <pwd.h> 90 # include <poll.h> 91 # include <semaphore.h> 92 # include <fcntl.h> 93 # include <string.h> 94 # include <syscall.h> 95 # include <sys/sysinfo.h> 96 # include <gnu/libc-version.h> 97 # include <sys/ipc.h> 98 # include <sys/shm.h> 99 # include <link.h> 100 # include <stdint.h> 101 # include <inttypes.h> 102 # include <sys/ioctl.h> 103 104 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 105 // getrusage() is prepared to handle the associated failure. 106 #ifndef RUSAGE_THREAD 107 #define RUSAGE_THREAD (1) /* only the calling thread */ 108 #endif 109 110 #define MAX_PATH (2 * K) 111 112 // for timer info max values which include all bits 113 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 114 115 #define LARGEPAGES_BIT (1 << 6) 116 //////////////////////////////////////////////////////////////////////////////// 117 // global variables 118 julong os::Linux::_physical_memory = 0; 119 120 address os::Linux::_initial_thread_stack_bottom = NULL; 121 uintptr_t os::Linux::_initial_thread_stack_size = 0; 122 123 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 124 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 125 Mutex* os::Linux::_createThread_lock = NULL; 126 pthread_t os::Linux::_main_thread; 127 int os::Linux::_page_size = -1; 128 const int os::Linux::_vm_default_page_size = (8 * K); 129 bool os::Linux::_is_floating_stack = false; 130 bool os::Linux::_is_NPTL = false; 131 bool os::Linux::_supports_fast_thread_cpu_time = false; 132 const char * os::Linux::_glibc_version = NULL; 133 const char * os::Linux::_libpthread_version = NULL; 134 pthread_condattr_t os::Linux::_condattr[1]; 135 136 static jlong initial_time_count=0; 137 138 static int clock_tics_per_sec = 100; 139 140 // For diagnostics to print a message once. see run_periodic_checks 141 static sigset_t check_signal_done; 142 static bool check_signals = true; 143 144 static pid_t _initial_pid = 0; 145 146 /* Signal number used to suspend/resume a thread */ 147 148 /* do not use any signal number less than SIGSEGV, see 4355769 */ 149 static int SR_signum = SIGUSR2; 150 sigset_t SR_sigset; 151 152 /* Used to protect dlsym() calls */ 153 static pthread_mutex_t dl_mutex; 154 155 // Declarations 156 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 157 158 #ifdef JAVASE_EMBEDDED 159 class MemNotifyThread: public Thread { 160 friend class VMStructs; 161 public: 162 virtual void run(); 163 164 private: 165 static MemNotifyThread* _memnotify_thread; 166 int _fd; 167 168 public: 169 170 // Constructor 171 MemNotifyThread(int fd); 172 173 // Tester 174 bool is_memnotify_thread() const { return true; } 175 176 // Printing 177 char* name() const { return (char*)"Linux MemNotify Thread"; } 178 179 // Returns the single instance of the MemNotifyThread 180 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; } 181 182 // Create and start the single instance of MemNotifyThread 183 static void start(); 184 }; 185 #endif // JAVASE_EMBEDDED 186 187 // utility functions 188 189 static int SR_initialize(); 190 191 julong os::available_memory() { 192 return Linux::available_memory(); 193 } 194 195 julong os::Linux::available_memory() { 196 // values in struct sysinfo are "unsigned long" 197 struct sysinfo si; 198 sysinfo(&si); 199 200 return (julong)si.freeram * si.mem_unit; 201 } 202 203 julong os::physical_memory() { 204 return Linux::physical_memory(); 205 } 206 207 //////////////////////////////////////////////////////////////////////////////// 208 // environment support 209 210 bool os::getenv(const char* name, char* buf, int len) { 211 const char* val = ::getenv(name); 212 if (val != NULL && strlen(val) < (size_t)len) { 213 strcpy(buf, val); 214 return true; 215 } 216 if (len > 0) buf[0] = 0; // return a null string 217 return false; 218 } 219 220 221 // Return true if user is running as root. 222 223 bool os::have_special_privileges() { 224 static bool init = false; 225 static bool privileges = false; 226 if (!init) { 227 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 228 init = true; 229 } 230 return privileges; 231 } 232 233 234 #ifndef SYS_gettid 235 // i386: 224, ia64: 1105, amd64: 186, sparc 143 236 #ifdef __ia64__ 237 #define SYS_gettid 1105 238 #elif __i386__ 239 #define SYS_gettid 224 240 #elif __amd64__ 241 #define SYS_gettid 186 242 #elif __sparc__ 243 #define SYS_gettid 143 244 #else 245 #error define gettid for the arch 246 #endif 247 #endif 248 249 // Cpu architecture string 250 #if defined(ZERO) 251 static char cpu_arch[] = ZERO_LIBARCH; 252 #elif defined(IA64) 253 static char cpu_arch[] = "ia64"; 254 #elif defined(IA32) 255 static char cpu_arch[] = "i386"; 256 #elif defined(AMD64) 257 static char cpu_arch[] = "amd64"; 258 #elif defined(ARM) 259 static char cpu_arch[] = "arm"; 260 #elif defined(PPC32) 261 static char cpu_arch[] = "ppc"; 262 #elif defined(PPC64) 263 static char cpu_arch[] = "ppc64"; 264 #elif defined(SPARC) 265 # ifdef _LP64 266 static char cpu_arch[] = "sparcv9"; 267 # else 268 static char cpu_arch[] = "sparc"; 269 # endif 270 #else 271 #error Add appropriate cpu_arch setting 272 #endif 273 274 275 // pid_t gettid() 276 // 277 // Returns the kernel thread id of the currently running thread. Kernel 278 // thread id is used to access /proc. 279 // 280 // (Note that getpid() on LinuxThreads returns kernel thread id too; but 281 // on NPTL, it returns the same pid for all threads, as required by POSIX.) 282 // 283 pid_t os::Linux::gettid() { 284 int rslt = syscall(SYS_gettid); 285 if (rslt == -1) { 286 // old kernel, no NPTL support 287 return getpid(); 288 } else { 289 return (pid_t)rslt; 290 } 291 } 292 293 // Most versions of linux have a bug where the number of processors are 294 // determined by looking at the /proc file system. In a chroot environment, 295 // the system call returns 1. This causes the VM to act as if it is 296 // a single processor and elide locking (see is_MP() call). 297 static bool unsafe_chroot_detected = false; 298 static const char *unstable_chroot_error = "/proc file system not found.\n" 299 "Java may be unstable running multithreaded in a chroot " 300 "environment on Linux when /proc filesystem is not mounted."; 301 302 void os::Linux::initialize_system_info() { 303 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 304 if (processor_count() == 1) { 305 pid_t pid = os::Linux::gettid(); 306 char fname[32]; 307 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 308 FILE *fp = fopen(fname, "r"); 309 if (fp == NULL) { 310 unsafe_chroot_detected = true; 311 } else { 312 fclose(fp); 313 } 314 } 315 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 316 assert(processor_count() > 0, "linux error"); 317 } 318 319 void os::init_system_properties_values() { 320 // char arch[12]; 321 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 322 323 // The next steps are taken in the product version: 324 // 325 // Obtain the JAVA_HOME value from the location of libjvm.so. 326 // This library should be located at: 327 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 328 // 329 // If "/jre/lib/" appears at the right place in the path, then we 330 // assume libjvm.so is installed in a JDK and we use this path. 331 // 332 // Otherwise exit with message: "Could not create the Java virtual machine." 333 // 334 // The following extra steps are taken in the debugging version: 335 // 336 // If "/jre/lib/" does NOT appear at the right place in the path 337 // instead of exit check for $JAVA_HOME environment variable. 338 // 339 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 340 // then we append a fake suffix "hotspot/libjvm.so" to this path so 341 // it looks like libjvm.so is installed there 342 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 343 // 344 // Otherwise exit. 345 // 346 // Important note: if the location of libjvm.so changes this 347 // code needs to be changed accordingly. 348 349 // The next few definitions allow the code to be verbatim: 350 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal) 351 #define getenv(n) ::getenv(n) 352 353 /* 354 * See ld(1): 355 * The linker uses the following search paths to locate required 356 * shared libraries: 357 * 1: ... 358 * ... 359 * 7: The default directories, normally /lib and /usr/lib. 360 */ 361 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 362 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 363 #else 364 #define DEFAULT_LIBPATH "/lib:/usr/lib" 365 #endif 366 367 #define EXTENSIONS_DIR "/lib/ext" 368 #define ENDORSED_DIR "/lib/endorsed" 369 #define REG_DIR "/usr/java/packages" 370 371 { 372 /* sysclasspath, java_home, dll_dir */ 373 { 374 char *home_path; 375 char *dll_path; 376 char *pslash; 377 char buf[MAXPATHLEN]; 378 os::jvm_path(buf, sizeof(buf)); 379 380 // Found the full path to libjvm.so. 381 // Now cut the path to <java_home>/jre if we can. 382 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 383 pslash = strrchr(buf, '/'); 384 if (pslash != NULL) 385 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 386 dll_path = malloc(strlen(buf) + 1); 387 if (dll_path == NULL) 388 return; 389 strcpy(dll_path, buf); 390 Arguments::set_dll_dir(dll_path); 391 392 if (pslash != NULL) { 393 pslash = strrchr(buf, '/'); 394 if (pslash != NULL) { 395 *pslash = '\0'; /* get rid of /<arch> */ 396 pslash = strrchr(buf, '/'); 397 if (pslash != NULL) 398 *pslash = '\0'; /* get rid of /lib */ 399 } 400 } 401 402 home_path = malloc(strlen(buf) + 1); 403 if (home_path == NULL) 404 return; 405 strcpy(home_path, buf); 406 Arguments::set_java_home(home_path); 407 408 if (!set_boot_path('/', ':')) 409 return; 410 } 411 412 /* 413 * Where to look for native libraries 414 * 415 * Note: Due to a legacy implementation, most of the library path 416 * is set in the launcher. This was to accomodate linking restrictions 417 * on legacy Linux implementations (which are no longer supported). 418 * Eventually, all the library path setting will be done here. 419 * 420 * However, to prevent the proliferation of improperly built native 421 * libraries, the new path component /usr/java/packages is added here. 422 * Eventually, all the library path setting will be done here. 423 */ 424 { 425 char *ld_library_path; 426 427 /* 428 * Construct the invariant part of ld_library_path. Note that the 429 * space for the colon and the trailing null are provided by the 430 * nulls included by the sizeof operator (so actually we allocate 431 * a byte more than necessary). 432 */ 433 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 434 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 435 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 436 437 /* 438 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 439 * should always exist (until the legacy problem cited above is 440 * addressed). 441 */ 442 char *v = getenv("LD_LIBRARY_PATH"); 443 if (v != NULL) { 444 char *t = ld_library_path; 445 /* That's +1 for the colon and +1 for the trailing '\0' */ 446 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 447 sprintf(ld_library_path, "%s:%s", v, t); 448 } 449 Arguments::set_library_path(ld_library_path); 450 } 451 452 /* 453 * Extensions directories. 454 * 455 * Note that the space for the colon and the trailing null are provided 456 * by the nulls included by the sizeof operator (so actually one byte more 457 * than necessary is allocated). 458 */ 459 { 460 char *buf = malloc(strlen(Arguments::get_java_home()) + 461 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 462 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 463 Arguments::get_java_home()); 464 Arguments::set_ext_dirs(buf); 465 } 466 467 /* Endorsed standards default directory. */ 468 { 469 char * buf; 470 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 471 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 472 Arguments::set_endorsed_dirs(buf); 473 } 474 } 475 476 #undef malloc 477 #undef getenv 478 #undef EXTENSIONS_DIR 479 #undef ENDORSED_DIR 480 481 // Done 482 return; 483 } 484 485 //////////////////////////////////////////////////////////////////////////////// 486 // breakpoint support 487 488 void os::breakpoint() { 489 BREAKPOINT; 490 } 491 492 extern "C" void breakpoint() { 493 // use debugger to set breakpoint here 494 } 495 496 //////////////////////////////////////////////////////////////////////////////// 497 // signal support 498 499 debug_only(static bool signal_sets_initialized = false); 500 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 501 502 bool os::Linux::is_sig_ignored(int sig) { 503 struct sigaction oact; 504 sigaction(sig, (struct sigaction*)NULL, &oact); 505 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 506 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 507 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 508 return true; 509 else 510 return false; 511 } 512 513 void os::Linux::signal_sets_init() { 514 // Should also have an assertion stating we are still single-threaded. 515 assert(!signal_sets_initialized, "Already initialized"); 516 // Fill in signals that are necessarily unblocked for all threads in 517 // the VM. Currently, we unblock the following signals: 518 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 519 // by -Xrs (=ReduceSignalUsage)); 520 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 521 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 522 // the dispositions or masks wrt these signals. 523 // Programs embedding the VM that want to use the above signals for their 524 // own purposes must, at this time, use the "-Xrs" option to prevent 525 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 526 // (See bug 4345157, and other related bugs). 527 // In reality, though, unblocking these signals is really a nop, since 528 // these signals are not blocked by default. 529 sigemptyset(&unblocked_sigs); 530 sigemptyset(&allowdebug_blocked_sigs); 531 sigaddset(&unblocked_sigs, SIGILL); 532 sigaddset(&unblocked_sigs, SIGSEGV); 533 sigaddset(&unblocked_sigs, SIGBUS); 534 sigaddset(&unblocked_sigs, SIGFPE); 535 #if defined(PPC64) 536 sigaddset(&unblocked_sigs, SIGTRAP); 537 #endif 538 sigaddset(&unblocked_sigs, SR_signum); 539 540 if (!ReduceSignalUsage) { 541 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 542 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 543 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 544 } 545 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 546 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 547 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 548 } 549 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 550 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 551 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 552 } 553 } 554 // Fill in signals that are blocked by all but the VM thread. 555 sigemptyset(&vm_sigs); 556 if (!ReduceSignalUsage) 557 sigaddset(&vm_sigs, BREAK_SIGNAL); 558 debug_only(signal_sets_initialized = true); 559 560 } 561 562 // These are signals that are unblocked while a thread is running Java. 563 // (For some reason, they get blocked by default.) 564 sigset_t* os::Linux::unblocked_signals() { 565 assert(signal_sets_initialized, "Not initialized"); 566 return &unblocked_sigs; 567 } 568 569 // These are the signals that are blocked while a (non-VM) thread is 570 // running Java. Only the VM thread handles these signals. 571 sigset_t* os::Linux::vm_signals() { 572 assert(signal_sets_initialized, "Not initialized"); 573 return &vm_sigs; 574 } 575 576 // These are signals that are blocked during cond_wait to allow debugger in 577 sigset_t* os::Linux::allowdebug_blocked_signals() { 578 assert(signal_sets_initialized, "Not initialized"); 579 return &allowdebug_blocked_sigs; 580 } 581 582 void os::Linux::hotspot_sigmask(Thread* thread) { 583 584 //Save caller's signal mask before setting VM signal mask 585 sigset_t caller_sigmask; 586 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 587 588 OSThread* osthread = thread->osthread(); 589 osthread->set_caller_sigmask(caller_sigmask); 590 591 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 592 593 if (!ReduceSignalUsage) { 594 if (thread->is_VM_thread()) { 595 // Only the VM thread handles BREAK_SIGNAL ... 596 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 597 } else { 598 // ... all other threads block BREAK_SIGNAL 599 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 600 } 601 } 602 } 603 604 ////////////////////////////////////////////////////////////////////////////// 605 // detecting pthread library 606 607 void os::Linux::libpthread_init() { 608 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 609 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 610 // generic name for earlier versions. 611 // Define macros here so we can build HotSpot on old systems. 612 # ifndef _CS_GNU_LIBC_VERSION 613 # define _CS_GNU_LIBC_VERSION 2 614 # endif 615 # ifndef _CS_GNU_LIBPTHREAD_VERSION 616 # define _CS_GNU_LIBPTHREAD_VERSION 3 617 # endif 618 619 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 620 if (n > 0) { 621 char *str = (char *)malloc(n, mtInternal); 622 confstr(_CS_GNU_LIBC_VERSION, str, n); 623 os::Linux::set_glibc_version(str); 624 } else { 625 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 626 static char _gnu_libc_version[32]; 627 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 628 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 629 os::Linux::set_glibc_version(_gnu_libc_version); 630 } 631 632 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 633 if (n > 0) { 634 char *str = (char *)malloc(n, mtInternal); 635 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 636 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 637 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 638 // is the case. LinuxThreads has a hard limit on max number of threads. 639 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 640 // On the other hand, NPTL does not have such a limit, sysconf() 641 // will return -1 and errno is not changed. Check if it is really NPTL. 642 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 643 strstr(str, "NPTL") && 644 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 645 free(str); 646 os::Linux::set_libpthread_version("linuxthreads"); 647 } else { 648 os::Linux::set_libpthread_version(str); 649 } 650 } else { 651 // glibc before 2.3.2 only has LinuxThreads. 652 os::Linux::set_libpthread_version("linuxthreads"); 653 } 654 655 if (strstr(libpthread_version(), "NPTL")) { 656 os::Linux::set_is_NPTL(); 657 } else { 658 os::Linux::set_is_LinuxThreads(); 659 } 660 661 // LinuxThreads have two flavors: floating-stack mode, which allows variable 662 // stack size; and fixed-stack mode. NPTL is always floating-stack. 663 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 664 os::Linux::set_is_floating_stack(); 665 } 666 } 667 668 ///////////////////////////////////////////////////////////////////////////// 669 // thread stack 670 671 // Force Linux kernel to expand current thread stack. If "bottom" is close 672 // to the stack guard, caller should block all signals. 673 // 674 // MAP_GROWSDOWN: 675 // A special mmap() flag that is used to implement thread stacks. It tells 676 // kernel that the memory region should extend downwards when needed. This 677 // allows early versions of LinuxThreads to only mmap the first few pages 678 // when creating a new thread. Linux kernel will automatically expand thread 679 // stack as needed (on page faults). 680 // 681 // However, because the memory region of a MAP_GROWSDOWN stack can grow on 682 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 683 // region, it's hard to tell if the fault is due to a legitimate stack 684 // access or because of reading/writing non-exist memory (e.g. buffer 685 // overrun). As a rule, if the fault happens below current stack pointer, 686 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the 687 // application (see Linux kernel fault.c). 688 // 689 // This Linux feature can cause SIGSEGV when VM bangs thread stack for 690 // stack overflow detection. 691 // 692 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 693 // not use this flag. However, the stack of initial thread is not created 694 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 695 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack 696 // and then attach the thread to JVM. 697 // 698 // To get around the problem and allow stack banging on Linux, we need to 699 // manually expand thread stack after receiving the SIGSEGV. 700 // 701 // There are two ways to expand thread stack to address "bottom", we used 702 // both of them in JVM before 1.5: 703 // 1. adjust stack pointer first so that it is below "bottom", and then 704 // touch "bottom" 705 // 2. mmap() the page in question 706 // 707 // Now alternate signal stack is gone, it's harder to use 2. For instance, 708 // if current sp is already near the lower end of page 101, and we need to 709 // call mmap() to map page 100, it is possible that part of the mmap() frame 710 // will be placed in page 100. When page 100 is mapped, it is zero-filled. 711 // That will destroy the mmap() frame and cause VM to crash. 712 // 713 // The following code works by adjusting sp first, then accessing the "bottom" 714 // page to force a page fault. Linux kernel will then automatically expand the 715 // stack mapping. 716 // 717 // _expand_stack_to() assumes its frame size is less than page size, which 718 // should always be true if the function is not inlined. 719 720 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 721 #define NOINLINE 722 #else 723 #define NOINLINE __attribute__ ((noinline)) 724 #endif 725 726 static void _expand_stack_to(address bottom) NOINLINE; 727 728 static void _expand_stack_to(address bottom) { 729 address sp; 730 size_t size; 731 volatile char *p; 732 733 // Adjust bottom to point to the largest address within the same page, it 734 // gives us a one-page buffer if alloca() allocates slightly more memory. 735 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 736 bottom += os::Linux::page_size() - 1; 737 738 // sp might be slightly above current stack pointer; if that's the case, we 739 // will alloca() a little more space than necessary, which is OK. Don't use 740 // os::current_stack_pointer(), as its result can be slightly below current 741 // stack pointer, causing us to not alloca enough to reach "bottom". 742 sp = (address)&sp; 743 744 if (sp > bottom) { 745 size = sp - bottom; 746 p = (volatile char *)alloca(size); 747 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 748 p[0] = '\0'; 749 } 750 } 751 752 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 753 assert(t!=NULL, "just checking"); 754 assert(t->osthread()->expanding_stack(), "expand should be set"); 755 assert(t->stack_base() != NULL, "stack_base was not initialized"); 756 757 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 758 sigset_t mask_all, old_sigset; 759 sigfillset(&mask_all); 760 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 761 _expand_stack_to(addr); 762 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 763 return true; 764 } 765 return false; 766 } 767 768 ////////////////////////////////////////////////////////////////////////////// 769 // create new thread 770 771 static address highest_vm_reserved_address(); 772 773 // check if it's safe to start a new thread 774 static bool _thread_safety_check(Thread* thread) { 775 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 776 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 777 // Heap is mmap'ed at lower end of memory space. Thread stacks are 778 // allocated (MAP_FIXED) from high address space. Every thread stack 779 // occupies a fixed size slot (usually 2Mbytes, but user can change 780 // it to other values if they rebuild LinuxThreads). 781 // 782 // Problem with MAP_FIXED is that mmap() can still succeed even part of 783 // the memory region has already been mmap'ed. That means if we have too 784 // many threads and/or very large heap, eventually thread stack will 785 // collide with heap. 786 // 787 // Here we try to prevent heap/stack collision by comparing current 788 // stack bottom with the highest address that has been mmap'ed by JVM 789 // plus a safety margin for memory maps created by native code. 790 // 791 // This feature can be disabled by setting ThreadSafetyMargin to 0 792 // 793 if (ThreadSafetyMargin > 0) { 794 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 795 796 // not safe if our stack extends below the safety margin 797 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 798 } else { 799 return true; 800 } 801 } else { 802 // Floating stack LinuxThreads or NPTL: 803 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 804 // there's not enough space left, pthread_create() will fail. If we come 805 // here, that means enough space has been reserved for stack. 806 return true; 807 } 808 } 809 810 // Thread start routine for all newly created threads 811 static void *java_start(Thread *thread) { 812 // Try to randomize the cache line index of hot stack frames. 813 // This helps when threads of the same stack traces evict each other's 814 // cache lines. The threads can be either from the same JVM instance, or 815 // from different JVM instances. The benefit is especially true for 816 // processors with hyperthreading technology. 817 static int counter = 0; 818 int pid = os::current_process_id(); 819 alloca(((pid ^ counter++) & 7) * 128); 820 821 ThreadLocalStorage::set_thread(thread); 822 823 OSThread* osthread = thread->osthread(); 824 Monitor* sync = osthread->startThread_lock(); 825 826 // non floating stack LinuxThreads needs extra check, see above 827 if (!_thread_safety_check(thread)) { 828 // notify parent thread 829 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 830 osthread->set_state(ZOMBIE); 831 sync->notify_all(); 832 return NULL; 833 } 834 835 // thread_id is kernel thread id (similar to Solaris LWP id) 836 osthread->set_thread_id(os::Linux::gettid()); 837 838 if (UseNUMA) { 839 int lgrp_id = os::numa_get_group_id(); 840 if (lgrp_id != -1) { 841 thread->set_lgrp_id(lgrp_id); 842 } 843 } 844 // initialize signal mask for this thread 845 os::Linux::hotspot_sigmask(thread); 846 847 // initialize floating point control register 848 os::Linux::init_thread_fpu_state(); 849 850 // handshaking with parent thread 851 { 852 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 853 854 // notify parent thread 855 osthread->set_state(INITIALIZED); 856 sync->notify_all(); 857 858 // wait until os::start_thread() 859 while (osthread->get_state() == INITIALIZED) { 860 sync->wait(Mutex::_no_safepoint_check_flag); 861 } 862 } 863 864 // call one more level start routine 865 thread->run(); 866 867 return 0; 868 } 869 870 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 871 assert(thread->osthread() == NULL, "caller responsible"); 872 873 // Allocate the OSThread object 874 OSThread* osthread = new OSThread(NULL, NULL); 875 if (osthread == NULL) { 876 return false; 877 } 878 879 // set the correct thread state 880 osthread->set_thread_type(thr_type); 881 882 // Initial state is ALLOCATED but not INITIALIZED 883 osthread->set_state(ALLOCATED); 884 885 thread->set_osthread(osthread); 886 887 // init thread attributes 888 pthread_attr_t attr; 889 pthread_attr_init(&attr); 890 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 891 892 // stack size 893 if (os::Linux::supports_variable_stack_size()) { 894 // calculate stack size if it's not specified by caller 895 if (stack_size == 0) { 896 stack_size = os::Linux::default_stack_size(thr_type); 897 898 switch (thr_type) { 899 case os::java_thread: 900 // Java threads use ThreadStackSize which default value can be 901 // changed with the flag -Xss 902 assert (JavaThread::stack_size_at_create() > 0, "this should be set"); 903 stack_size = JavaThread::stack_size_at_create(); 904 break; 905 case os::compiler_thread: 906 if (CompilerThreadStackSize > 0) { 907 stack_size = (size_t)(CompilerThreadStackSize * K); 908 break; 909 } // else fall through: 910 // use VMThreadStackSize if CompilerThreadStackSize is not defined 911 case os::vm_thread: 912 case os::pgc_thread: 913 case os::cgc_thread: 914 case os::watcher_thread: 915 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 916 break; 917 } 918 } 919 920 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 921 pthread_attr_setstacksize(&attr, stack_size); 922 } else { 923 // let pthread_create() pick the default value. 924 } 925 926 // glibc guard page 927 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 928 929 ThreadState state; 930 931 { 932 // Serialize thread creation if we are running with fixed stack LinuxThreads 933 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 934 if (lock) { 935 os::Linux::createThread_lock()->lock_without_safepoint_check(); 936 } 937 938 pthread_t tid; 939 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 940 941 pthread_attr_destroy(&attr); 942 943 if (ret != 0) { 944 if (PrintMiscellaneous && (Verbose || WizardMode)) { 945 perror("pthread_create()"); 946 } 947 // Need to clean up stuff we've allocated so far 948 thread->set_osthread(NULL); 949 delete osthread; 950 if (lock) os::Linux::createThread_lock()->unlock(); 951 return false; 952 } 953 954 // Store pthread info into the OSThread 955 osthread->set_pthread_id(tid); 956 957 // Wait until child thread is either initialized or aborted 958 { 959 Monitor* sync_with_child = osthread->startThread_lock(); 960 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 961 while ((state = osthread->get_state()) == ALLOCATED) { 962 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 963 } 964 } 965 966 if (lock) { 967 os::Linux::createThread_lock()->unlock(); 968 } 969 } 970 971 // Aborted due to thread limit being reached 972 if (state == ZOMBIE) { 973 thread->set_osthread(NULL); 974 delete osthread; 975 return false; 976 } 977 978 // The thread is returned suspended (in state INITIALIZED), 979 // and is started higher up in the call chain 980 assert(state == INITIALIZED, "race condition"); 981 return true; 982 } 983 984 ///////////////////////////////////////////////////////////////////////////// 985 // attach existing thread 986 987 // bootstrap the main thread 988 bool os::create_main_thread(JavaThread* thread) { 989 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 990 return create_attached_thread(thread); 991 } 992 993 bool os::create_attached_thread(JavaThread* thread) { 994 #ifdef ASSERT 995 thread->verify_not_published(); 996 #endif 997 998 // Allocate the OSThread object 999 OSThread* osthread = new OSThread(NULL, NULL); 1000 1001 if (osthread == NULL) { 1002 return false; 1003 } 1004 1005 // Store pthread info into the OSThread 1006 osthread->set_thread_id(os::Linux::gettid()); 1007 osthread->set_pthread_id(::pthread_self()); 1008 1009 // initialize floating point control register 1010 os::Linux::init_thread_fpu_state(); 1011 1012 // Initial thread state is RUNNABLE 1013 osthread->set_state(RUNNABLE); 1014 1015 thread->set_osthread(osthread); 1016 1017 if (UseNUMA) { 1018 int lgrp_id = os::numa_get_group_id(); 1019 if (lgrp_id != -1) { 1020 thread->set_lgrp_id(lgrp_id); 1021 } 1022 } 1023 1024 if (os::Linux::is_initial_thread()) { 1025 // If current thread is initial thread, its stack is mapped on demand, 1026 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 1027 // the entire stack region to avoid SEGV in stack banging. 1028 // It is also useful to get around the heap-stack-gap problem on SuSE 1029 // kernel (see 4821821 for details). We first expand stack to the top 1030 // of yellow zone, then enable stack yellow zone (order is significant, 1031 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1032 // is no gap between the last two virtual memory regions. 1033 1034 JavaThread *jt = (JavaThread *)thread; 1035 address addr = jt->stack_yellow_zone_base(); 1036 assert(addr != NULL, "initialization problem?"); 1037 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1038 1039 osthread->set_expanding_stack(); 1040 os::Linux::manually_expand_stack(jt, addr); 1041 osthread->clear_expanding_stack(); 1042 } 1043 1044 // initialize signal mask for this thread 1045 // and save the caller's signal mask 1046 os::Linux::hotspot_sigmask(thread); 1047 1048 return true; 1049 } 1050 1051 void os::pd_start_thread(Thread* thread) { 1052 OSThread * osthread = thread->osthread(); 1053 assert(osthread->get_state() != INITIALIZED, "just checking"); 1054 Monitor* sync_with_child = osthread->startThread_lock(); 1055 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1056 sync_with_child->notify(); 1057 } 1058 1059 // Free Linux resources related to the OSThread 1060 void os::free_thread(OSThread* osthread) { 1061 assert(osthread != NULL, "osthread not set"); 1062 1063 if (Thread::current()->osthread() == osthread) { 1064 // Restore caller's signal mask 1065 sigset_t sigmask = osthread->caller_sigmask(); 1066 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1067 } 1068 1069 delete osthread; 1070 } 1071 1072 ////////////////////////////////////////////////////////////////////////////// 1073 // thread local storage 1074 1075 int os::allocate_thread_local_storage() { 1076 pthread_key_t key; 1077 int rslt = pthread_key_create(&key, NULL); 1078 assert(rslt == 0, "cannot allocate thread local storage"); 1079 return (int)key; 1080 } 1081 1082 // Note: This is currently not used by VM, as we don't destroy TLS key 1083 // on VM exit. 1084 void os::free_thread_local_storage(int index) { 1085 int rslt = pthread_key_delete((pthread_key_t)index); 1086 assert(rslt == 0, "invalid index"); 1087 } 1088 1089 void os::thread_local_storage_at_put(int index, void* value) { 1090 int rslt = pthread_setspecific((pthread_key_t)index, value); 1091 assert(rslt == 0, "pthread_setspecific failed"); 1092 } 1093 1094 extern "C" Thread* get_thread() { 1095 return ThreadLocalStorage::thread(); 1096 } 1097 1098 ////////////////////////////////////////////////////////////////////////////// 1099 // initial thread 1100 1101 // Check if current thread is the initial thread, similar to Solaris thr_main. 1102 bool os::Linux::is_initial_thread(void) { 1103 char dummy; 1104 // If called before init complete, thread stack bottom will be null. 1105 // Can be called if fatal error occurs before initialization. 1106 if (initial_thread_stack_bottom() == NULL) return false; 1107 assert(initial_thread_stack_bottom() != NULL && 1108 initial_thread_stack_size() != 0, 1109 "os::init did not locate initial thread's stack region"); 1110 if ((address)&dummy >= initial_thread_stack_bottom() && 1111 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1112 return true; 1113 else return false; 1114 } 1115 1116 // Find the virtual memory area that contains addr 1117 static bool find_vma(address addr, address* vma_low, address* vma_high) { 1118 FILE *fp = fopen("/proc/self/maps", "r"); 1119 if (fp) { 1120 address low, high; 1121 while (!feof(fp)) { 1122 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1123 if (low <= addr && addr < high) { 1124 if (vma_low) *vma_low = low; 1125 if (vma_high) *vma_high = high; 1126 fclose (fp); 1127 return true; 1128 } 1129 } 1130 for (;;) { 1131 int ch = fgetc(fp); 1132 if (ch == EOF || ch == (int)'\n') break; 1133 } 1134 } 1135 fclose(fp); 1136 } 1137 return false; 1138 } 1139 1140 // Locate initial thread stack. This special handling of initial thread stack 1141 // is needed because pthread_getattr_np() on most (all?) Linux distros returns 1142 // bogus value for initial thread. 1143 void os::Linux::capture_initial_stack(size_t max_size) { 1144 // stack size is the easy part, get it from RLIMIT_STACK 1145 size_t stack_size; 1146 struct rlimit rlim; 1147 getrlimit(RLIMIT_STACK, &rlim); 1148 stack_size = rlim.rlim_cur; 1149 1150 // 6308388: a bug in ld.so will relocate its own .data section to the 1151 // lower end of primordial stack; reduce ulimit -s value a little bit 1152 // so we won't install guard page on ld.so's data section. 1153 stack_size -= 2 * page_size(); 1154 1155 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1156 // 7.1, in both cases we will get 2G in return value. 1157 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1158 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1159 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1160 // in case other parts in glibc still assumes 2M max stack size. 1161 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1162 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1163 if (stack_size > 2 * K * K IA64_ONLY(*2)) 1164 stack_size = 2 * K * K IA64_ONLY(*2); 1165 // Try to figure out where the stack base (top) is. This is harder. 1166 // 1167 // When an application is started, glibc saves the initial stack pointer in 1168 // a global variable "__libc_stack_end", which is then used by system 1169 // libraries. __libc_stack_end should be pretty close to stack top. The 1170 // variable is available since the very early days. However, because it is 1171 // a private interface, it could disappear in the future. 1172 // 1173 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1174 // to __libc_stack_end, it is very close to stack top, but isn't the real 1175 // stack top. Note that /proc may not exist if VM is running as a chroot 1176 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1177 // /proc/<pid>/stat could change in the future (though unlikely). 1178 // 1179 // We try __libc_stack_end first. If that doesn't work, look for 1180 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1181 // as a hint, which should work well in most cases. 1182 1183 uintptr_t stack_start; 1184 1185 // try __libc_stack_end first 1186 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1187 if (p && *p) { 1188 stack_start = *p; 1189 } else { 1190 // see if we can get the start_stack field from /proc/self/stat 1191 FILE *fp; 1192 int pid; 1193 char state; 1194 int ppid; 1195 int pgrp; 1196 int session; 1197 int nr; 1198 int tpgrp; 1199 unsigned long flags; 1200 unsigned long minflt; 1201 unsigned long cminflt; 1202 unsigned long majflt; 1203 unsigned long cmajflt; 1204 unsigned long utime; 1205 unsigned long stime; 1206 long cutime; 1207 long cstime; 1208 long prio; 1209 long nice; 1210 long junk; 1211 long it_real; 1212 uintptr_t start; 1213 uintptr_t vsize; 1214 intptr_t rss; 1215 uintptr_t rsslim; 1216 uintptr_t scodes; 1217 uintptr_t ecode; 1218 int i; 1219 1220 // Figure what the primordial thread stack base is. Code is inspired 1221 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1222 // followed by command name surrounded by parentheses, state, etc. 1223 char stat[2048]; 1224 int statlen; 1225 1226 fp = fopen("/proc/self/stat", "r"); 1227 if (fp) { 1228 statlen = fread(stat, 1, 2047, fp); 1229 stat[statlen] = '\0'; 1230 fclose(fp); 1231 1232 // Skip pid and the command string. Note that we could be dealing with 1233 // weird command names, e.g. user could decide to rename java launcher 1234 // to "java 1.4.2 :)", then the stat file would look like 1235 // 1234 (java 1.4.2 :)) R ... ... 1236 // We don't really need to know the command string, just find the last 1237 // occurrence of ")" and then start parsing from there. See bug 4726580. 1238 char * s = strrchr(stat, ')'); 1239 1240 i = 0; 1241 if (s) { 1242 // Skip blank chars 1243 do s++; while (isspace(*s)); 1244 1245 #define _UFM UINTX_FORMAT 1246 #define _DFM INTX_FORMAT 1247 1248 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1249 /* 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 */ 1250 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, 1251 &state, /* 3 %c */ 1252 &ppid, /* 4 %d */ 1253 &pgrp, /* 5 %d */ 1254 &session, /* 6 %d */ 1255 &nr, /* 7 %d */ 1256 &tpgrp, /* 8 %d */ 1257 &flags, /* 9 %lu */ 1258 &minflt, /* 10 %lu */ 1259 &cminflt, /* 11 %lu */ 1260 &majflt, /* 12 %lu */ 1261 &cmajflt, /* 13 %lu */ 1262 &utime, /* 14 %lu */ 1263 &stime, /* 15 %lu */ 1264 &cutime, /* 16 %ld */ 1265 &cstime, /* 17 %ld */ 1266 &prio, /* 18 %ld */ 1267 &nice, /* 19 %ld */ 1268 &junk, /* 20 %ld */ 1269 &it_real, /* 21 %ld */ 1270 &start, /* 22 UINTX_FORMAT */ 1271 &vsize, /* 23 UINTX_FORMAT */ 1272 &rss, /* 24 INTX_FORMAT */ 1273 &rsslim, /* 25 UINTX_FORMAT */ 1274 &scodes, /* 26 UINTX_FORMAT */ 1275 &ecode, /* 27 UINTX_FORMAT */ 1276 &stack_start); /* 28 UINTX_FORMAT */ 1277 } 1278 1279 #undef _UFM 1280 #undef _DFM 1281 1282 if (i != 28 - 2) { 1283 assert(false, "Bad conversion from /proc/self/stat"); 1284 // product mode - assume we are the initial thread, good luck in the 1285 // embedded case. 1286 warning("Can't detect initial thread stack location - bad conversion"); 1287 stack_start = (uintptr_t) &rlim; 1288 } 1289 } else { 1290 // For some reason we can't open /proc/self/stat (for example, running on 1291 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1292 // most cases, so don't abort: 1293 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1294 stack_start = (uintptr_t) &rlim; 1295 } 1296 } 1297 1298 // Now we have a pointer (stack_start) very close to the stack top, the 1299 // next thing to do is to figure out the exact location of stack top. We 1300 // can find out the virtual memory area that contains stack_start by 1301 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1302 // and its upper limit is the real stack top. (again, this would fail if 1303 // running inside chroot, because /proc may not exist.) 1304 1305 uintptr_t stack_top; 1306 address low, high; 1307 if (find_vma((address)stack_start, &low, &high)) { 1308 // success, "high" is the true stack top. (ignore "low", because initial 1309 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1310 stack_top = (uintptr_t)high; 1311 } else { 1312 // failed, likely because /proc/self/maps does not exist 1313 warning("Can't detect initial thread stack location - find_vma failed"); 1314 // best effort: stack_start is normally within a few pages below the real 1315 // stack top, use it as stack top, and reduce stack size so we won't put 1316 // guard page outside stack. 1317 stack_top = stack_start; 1318 stack_size -= 16 * page_size(); 1319 } 1320 1321 // stack_top could be partially down the page so align it 1322 stack_top = align_size_up(stack_top, page_size()); 1323 1324 if (max_size && stack_size > max_size) { 1325 _initial_thread_stack_size = max_size; 1326 } else { 1327 _initial_thread_stack_size = stack_size; 1328 } 1329 1330 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1331 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1332 } 1333 1334 //////////////////////////////////////////////////////////////////////////////// 1335 // time support 1336 1337 // Time since start-up in seconds to a fine granularity. 1338 // Used by VMSelfDestructTimer and the MemProfiler. 1339 double os::elapsedTime() { 1340 1341 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1342 } 1343 1344 jlong os::elapsed_counter() { 1345 return javaTimeNanos() - initial_time_count; 1346 } 1347 1348 jlong os::elapsed_frequency() { 1349 return NANOSECS_PER_SEC; // nanosecond resolution 1350 } 1351 1352 bool os::supports_vtime() { return true; } 1353 bool os::enable_vtime() { return false; } 1354 bool os::vtime_enabled() { return false; } 1355 1356 double os::elapsedVTime() { 1357 struct rusage usage; 1358 int retval = getrusage(RUSAGE_THREAD, &usage); 1359 if (retval == 0) { 1360 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000); 1361 } else { 1362 // better than nothing, but not much 1363 return elapsedTime(); 1364 } 1365 } 1366 1367 jlong os::javaTimeMillis() { 1368 timeval time; 1369 int status = gettimeofday(&time, NULL); 1370 assert(status != -1, "linux error"); 1371 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1372 } 1373 1374 #ifndef CLOCK_MONOTONIC 1375 #define CLOCK_MONOTONIC (1) 1376 #endif 1377 1378 void os::Linux::clock_init() { 1379 // we do dlopen's in this particular order due to bug in linux 1380 // dynamical loader (see 6348968) leading to crash on exit 1381 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1382 if (handle == NULL) { 1383 handle = dlopen("librt.so", RTLD_LAZY); 1384 } 1385 1386 if (handle) { 1387 int (*clock_getres_func)(clockid_t, struct timespec*) = 1388 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1389 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1390 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1391 if (clock_getres_func && clock_gettime_func) { 1392 // See if monotonic clock is supported by the kernel. Note that some 1393 // early implementations simply return kernel jiffies (updated every 1394 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1395 // for nano time (though the monotonic property is still nice to have). 1396 // It's fixed in newer kernels, however clock_getres() still returns 1397 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1398 // resolution for now. Hopefully as people move to new kernels, this 1399 // won't be a problem. 1400 struct timespec res; 1401 struct timespec tp; 1402 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1403 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1404 // yes, monotonic clock is supported 1405 _clock_gettime = clock_gettime_func; 1406 return; 1407 } else { 1408 // close librt if there is no monotonic clock 1409 dlclose(handle); 1410 } 1411 } 1412 } 1413 warning("No monotonic clock was available - timed services may " \ 1414 "be adversely affected if the time-of-day clock changes"); 1415 } 1416 1417 #ifndef SYS_clock_getres 1418 1419 #if defined(IA32) || defined(AMD64) 1420 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1421 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1422 #else 1423 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1424 #define sys_clock_getres(x,y) -1 1425 #endif 1426 1427 #else 1428 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1429 #endif 1430 1431 void os::Linux::fast_thread_clock_init() { 1432 if (!UseLinuxPosixThreadCPUClocks) { 1433 return; 1434 } 1435 clockid_t clockid; 1436 struct timespec tp; 1437 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1438 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1439 1440 // Switch to using fast clocks for thread cpu time if 1441 // the sys_clock_getres() returns 0 error code. 1442 // Note, that some kernels may support the current thread 1443 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1444 // returned by the pthread_getcpuclockid(). 1445 // If the fast Posix clocks are supported then the sys_clock_getres() 1446 // must return at least tp.tv_sec == 0 which means a resolution 1447 // better than 1 sec. This is extra check for reliability. 1448 1449 if(pthread_getcpuclockid_func && 1450 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1451 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1452 1453 _supports_fast_thread_cpu_time = true; 1454 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1455 } 1456 } 1457 1458 jlong os::javaTimeNanos() { 1459 if (os::supports_monotonic_clock()) { 1460 struct timespec tp; 1461 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1462 assert(status == 0, "gettime error"); 1463 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1464 return result; 1465 } else { 1466 timeval time; 1467 int status = gettimeofday(&time, NULL); 1468 assert(status != -1, "linux error"); 1469 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1470 return 1000 * usecs; 1471 } 1472 } 1473 1474 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1475 if (os::supports_monotonic_clock()) { 1476 info_ptr->max_value = ALL_64_BITS; 1477 1478 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1479 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1480 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1481 } else { 1482 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1483 info_ptr->max_value = ALL_64_BITS; 1484 1485 // gettimeofday is a real time clock so it skips 1486 info_ptr->may_skip_backward = true; 1487 info_ptr->may_skip_forward = true; 1488 } 1489 1490 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1491 } 1492 1493 // Return the real, user, and system times in seconds from an 1494 // arbitrary fixed point in the past. 1495 bool os::getTimesSecs(double* process_real_time, 1496 double* process_user_time, 1497 double* process_system_time) { 1498 struct tms ticks; 1499 clock_t real_ticks = times(&ticks); 1500 1501 if (real_ticks == (clock_t) (-1)) { 1502 return false; 1503 } else { 1504 double ticks_per_second = (double) clock_tics_per_sec; 1505 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1506 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1507 *process_real_time = ((double) real_ticks) / ticks_per_second; 1508 1509 return true; 1510 } 1511 } 1512 1513 1514 char * os::local_time_string(char *buf, size_t buflen) { 1515 struct tm t; 1516 time_t long_time; 1517 time(&long_time); 1518 localtime_r(&long_time, &t); 1519 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1520 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1521 t.tm_hour, t.tm_min, t.tm_sec); 1522 return buf; 1523 } 1524 1525 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1526 return localtime_r(clock, res); 1527 } 1528 1529 //////////////////////////////////////////////////////////////////////////////// 1530 // runtime exit support 1531 1532 // Note: os::shutdown() might be called very early during initialization, or 1533 // called from signal handler. Before adding something to os::shutdown(), make 1534 // sure it is async-safe and can handle partially initialized VM. 1535 void os::shutdown() { 1536 1537 // allow PerfMemory to attempt cleanup of any persistent resources 1538 perfMemory_exit(); 1539 1540 // needs to remove object in file system 1541 AttachListener::abort(); 1542 1543 // flush buffered output, finish log files 1544 ostream_abort(); 1545 1546 // Check for abort hook 1547 abort_hook_t abort_hook = Arguments::abort_hook(); 1548 if (abort_hook != NULL) { 1549 abort_hook(); 1550 } 1551 1552 } 1553 1554 // Note: os::abort() might be called very early during initialization, or 1555 // called from signal handler. Before adding something to os::abort(), make 1556 // sure it is async-safe and can handle partially initialized VM. 1557 void os::abort(bool dump_core) { 1558 os::shutdown(); 1559 if (dump_core) { 1560 #ifndef PRODUCT 1561 fdStream out(defaultStream::output_fd()); 1562 out.print_raw("Current thread is "); 1563 char buf[16]; 1564 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1565 out.print_raw_cr(buf); 1566 out.print_raw_cr("Dumping core ..."); 1567 #endif 1568 ::abort(); // dump core 1569 } 1570 1571 ::exit(1); 1572 } 1573 1574 // Die immediately, no exit hook, no abort hook, no cleanup. 1575 void os::die() { 1576 // _exit() on LinuxThreads only kills current thread 1577 ::abort(); 1578 } 1579 1580 // unused on linux for now. 1581 void os::set_error_file(const char *logfile) {} 1582 1583 1584 // This method is a copy of JDK's sysGetLastErrorString 1585 // from src/solaris/hpi/src/system_md.c 1586 1587 size_t os::lasterror(char *buf, size_t len) { 1588 1589 if (errno == 0) return 0; 1590 1591 const char *s = ::strerror(errno); 1592 size_t n = ::strlen(s); 1593 if (n >= len) { 1594 n = len - 1; 1595 } 1596 ::strncpy(buf, s, n); 1597 buf[n] = '\0'; 1598 return n; 1599 } 1600 1601 intx os::current_thread_id() { return (intx)pthread_self(); } 1602 int os::current_process_id() { 1603 1604 // Under the old linux thread library, linux gives each thread 1605 // its own process id. Because of this each thread will return 1606 // a different pid if this method were to return the result 1607 // of getpid(2). Linux provides no api that returns the pid 1608 // of the launcher thread for the vm. This implementation 1609 // returns a unique pid, the pid of the launcher thread 1610 // that starts the vm 'process'. 1611 1612 // Under the NPTL, getpid() returns the same pid as the 1613 // launcher thread rather than a unique pid per thread. 1614 // Use gettid() if you want the old pre NPTL behaviour. 1615 1616 // if you are looking for the result of a call to getpid() that 1617 // returns a unique pid for the calling thread, then look at the 1618 // OSThread::thread_id() method in osThread_linux.hpp file 1619 1620 return (int)(_initial_pid ? _initial_pid : getpid()); 1621 } 1622 1623 // DLL functions 1624 1625 const char* os::dll_file_extension() { return ".so"; } 1626 1627 // This must be hard coded because it's the system's temporary 1628 // directory not the java application's temp directory, ala java.io.tmpdir. 1629 const char* os::get_temp_directory() { return "/tmp"; } 1630 1631 static bool file_exists(const char* filename) { 1632 struct stat statbuf; 1633 if (filename == NULL || strlen(filename) == 0) { 1634 return false; 1635 } 1636 return os::stat(filename, &statbuf) == 0; 1637 } 1638 1639 bool os::dll_build_name(char* buffer, size_t buflen, 1640 const char* pname, const char* fname) { 1641 bool retval = false; 1642 // Copied from libhpi 1643 const size_t pnamelen = pname ? strlen(pname) : 0; 1644 1645 // Return error on buffer overflow. 1646 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1647 return retval; 1648 } 1649 1650 if (pnamelen == 0) { 1651 snprintf(buffer, buflen, "lib%s.so", fname); 1652 retval = true; 1653 } else if (strchr(pname, *os::path_separator()) != NULL) { 1654 int n; 1655 char** pelements = split_path(pname, &n); 1656 if (pelements == NULL) { 1657 return false; 1658 } 1659 for (int i = 0 ; i < n ; i++) { 1660 // Really shouldn't be NULL, but check can't hurt 1661 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1662 continue; // skip the empty path values 1663 } 1664 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1665 if (file_exists(buffer)) { 1666 retval = true; 1667 break; 1668 } 1669 } 1670 // release the storage 1671 for (int i = 0 ; i < n ; i++) { 1672 if (pelements[i] != NULL) { 1673 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1674 } 1675 } 1676 if (pelements != NULL) { 1677 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1678 } 1679 } else { 1680 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1681 retval = true; 1682 } 1683 return retval; 1684 } 1685 1686 // check if addr is inside libjvm.so 1687 bool os::address_is_in_vm(address addr) { 1688 static address libjvm_base_addr; 1689 Dl_info dlinfo; 1690 1691 if (libjvm_base_addr == NULL) { 1692 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1693 libjvm_base_addr = (address)dlinfo.dli_fbase; 1694 } 1695 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1696 } 1697 1698 if (dladdr((void *)addr, &dlinfo) != 0) { 1699 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1700 } 1701 1702 return false; 1703 } 1704 1705 bool os::dll_address_to_function_name(address addr, char *buf, 1706 int buflen, int *offset) { 1707 // buf is not optional, but offset is optional 1708 assert(buf != NULL, "sanity check"); 1709 1710 Dl_info dlinfo; 1711 1712 if (dladdr((void*)addr, &dlinfo) != 0) { 1713 // see if we have a matching symbol 1714 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1715 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1716 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1717 } 1718 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1719 return true; 1720 } 1721 // no matching symbol so try for just file info 1722 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1723 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1724 buf, buflen, offset, dlinfo.dli_fname)) { 1725 return true; 1726 } 1727 } 1728 } 1729 1730 buf[0] = '\0'; 1731 if (offset != NULL) *offset = -1; 1732 return false; 1733 } 1734 1735 struct _address_to_library_name { 1736 address addr; // input : memory address 1737 size_t buflen; // size of fname 1738 char* fname; // output: library name 1739 address base; // library base addr 1740 }; 1741 1742 static int address_to_library_name_callback(struct dl_phdr_info *info, 1743 size_t size, void *data) { 1744 int i; 1745 bool found = false; 1746 address libbase = NULL; 1747 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1748 1749 // iterate through all loadable segments 1750 for (i = 0; i < info->dlpi_phnum; i++) { 1751 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1752 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1753 // base address of a library is the lowest address of its loaded 1754 // segments. 1755 if (libbase == NULL || libbase > segbase) { 1756 libbase = segbase; 1757 } 1758 // see if 'addr' is within current segment 1759 if (segbase <= d->addr && 1760 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1761 found = true; 1762 } 1763 } 1764 } 1765 1766 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1767 // so dll_address_to_library_name() can fall through to use dladdr() which 1768 // can figure out executable name from argv[0]. 1769 if (found && info->dlpi_name && info->dlpi_name[0]) { 1770 d->base = libbase; 1771 if (d->fname) { 1772 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1773 } 1774 return 1; 1775 } 1776 return 0; 1777 } 1778 1779 bool os::dll_address_to_library_name(address addr, char* buf, 1780 int buflen, int* offset) { 1781 // buf is not optional, but offset is optional 1782 assert(buf != NULL, "sanity check"); 1783 1784 Dl_info dlinfo; 1785 struct _address_to_library_name data; 1786 1787 // There is a bug in old glibc dladdr() implementation that it could resolve 1788 // to wrong library name if the .so file has a base address != NULL. Here 1789 // we iterate through the program headers of all loaded libraries to find 1790 // out which library 'addr' really belongs to. This workaround can be 1791 // removed once the minimum requirement for glibc is moved to 2.3.x. 1792 data.addr = addr; 1793 data.fname = buf; 1794 data.buflen = buflen; 1795 data.base = NULL; 1796 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1797 1798 if (rslt) { 1799 // buf already contains library name 1800 if (offset) *offset = addr - data.base; 1801 return true; 1802 } 1803 if (dladdr((void*)addr, &dlinfo) != 0) { 1804 if (dlinfo.dli_fname != NULL) { 1805 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1806 } 1807 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1808 *offset = addr - (address)dlinfo.dli_fbase; 1809 } 1810 return true; 1811 } 1812 1813 buf[0] = '\0'; 1814 if (offset) *offset = -1; 1815 return false; 1816 } 1817 1818 // Loads .dll/.so and 1819 // in case of error it checks if .dll/.so was built for the 1820 // same architecture as Hotspot is running on 1821 1822 1823 // Remember the stack's state. The Linux dynamic linker will change 1824 // the stack to 'executable' at most once, so we must safepoint only once. 1825 bool os::Linux::_stack_is_executable = false; 1826 1827 // VM operation that loads a library. This is necessary if stack protection 1828 // of the Java stacks can be lost during loading the library. If we 1829 // do not stop the Java threads, they can stack overflow before the stacks 1830 // are protected again. 1831 class VM_LinuxDllLoad: public VM_Operation { 1832 private: 1833 const char *_filename; 1834 char *_ebuf; 1835 int _ebuflen; 1836 void *_lib; 1837 public: 1838 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1839 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1840 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1841 void doit() { 1842 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1843 os::Linux::_stack_is_executable = true; 1844 } 1845 void* loaded_library() { return _lib; } 1846 }; 1847 1848 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1849 { 1850 void * result = NULL; 1851 bool load_attempted = false; 1852 1853 // Check whether the library to load might change execution rights 1854 // of the stack. If they are changed, the protection of the stack 1855 // guard pages will be lost. We need a safepoint to fix this. 1856 // 1857 // See Linux man page execstack(8) for more info. 1858 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1859 ElfFile ef(filename); 1860 if (!ef.specifies_noexecstack()) { 1861 if (!is_init_completed()) { 1862 os::Linux::_stack_is_executable = true; 1863 // This is OK - No Java threads have been created yet, and hence no 1864 // stack guard pages to fix. 1865 // 1866 // This should happen only when you are building JDK7 using a very 1867 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1868 // 1869 // Dynamic loader will make all stacks executable after 1870 // this function returns, and will not do that again. 1871 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1872 } else { 1873 warning("You have loaded library %s which might have disabled stack guard. " 1874 "The VM will try to fix the stack guard now.\n" 1875 "It's highly recommended that you fix the library with " 1876 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1877 filename); 1878 1879 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1880 JavaThread *jt = JavaThread::current(); 1881 if (jt->thread_state() != _thread_in_native) { 1882 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1883 // that requires ExecStack. Cannot enter safe point. Let's give up. 1884 warning("Unable to fix stack guard. Giving up."); 1885 } else { 1886 if (!LoadExecStackDllInVMThread) { 1887 // This is for the case where the DLL has an static 1888 // constructor function that executes JNI code. We cannot 1889 // load such DLLs in the VMThread. 1890 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1891 } 1892 1893 ThreadInVMfromNative tiv(jt); 1894 debug_only(VMNativeEntryWrapper vew;) 1895 1896 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1897 VMThread::execute(&op); 1898 if (LoadExecStackDllInVMThread) { 1899 result = op.loaded_library(); 1900 } 1901 load_attempted = true; 1902 } 1903 } 1904 } 1905 } 1906 1907 if (!load_attempted) { 1908 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1909 } 1910 1911 if (result != NULL) { 1912 // Successful loading 1913 return result; 1914 } 1915 1916 Elf32_Ehdr elf_head; 1917 int diag_msg_max_length=ebuflen-strlen(ebuf); 1918 char* diag_msg_buf=ebuf+strlen(ebuf); 1919 1920 if (diag_msg_max_length==0) { 1921 // No more space in ebuf for additional diagnostics message 1922 return NULL; 1923 } 1924 1925 1926 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1927 1928 if (file_descriptor < 0) { 1929 // Can't open library, report dlerror() message 1930 return NULL; 1931 } 1932 1933 bool failed_to_read_elf_head= 1934 (sizeof(elf_head)!= 1935 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1936 1937 ::close(file_descriptor); 1938 if (failed_to_read_elf_head) { 1939 // file i/o error - report dlerror() msg 1940 return NULL; 1941 } 1942 1943 typedef struct { 1944 Elf32_Half code; // Actual value as defined in elf.h 1945 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1946 char elf_class; // 32 or 64 bit 1947 char endianess; // MSB or LSB 1948 char* name; // String representation 1949 } arch_t; 1950 1951 #ifndef EM_486 1952 #define EM_486 6 /* Intel 80486 */ 1953 #endif 1954 1955 static const arch_t arch_array[]={ 1956 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1957 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1958 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1959 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1960 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1961 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1962 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1963 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1964 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1965 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1966 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1967 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1968 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1969 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1970 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1971 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1972 }; 1973 1974 #if (defined IA32) 1975 static Elf32_Half running_arch_code=EM_386; 1976 #elif (defined AMD64) 1977 static Elf32_Half running_arch_code=EM_X86_64; 1978 #elif (defined IA64) 1979 static Elf32_Half running_arch_code=EM_IA_64; 1980 #elif (defined __sparc) && (defined _LP64) 1981 static Elf32_Half running_arch_code=EM_SPARCV9; 1982 #elif (defined __sparc) && (!defined _LP64) 1983 static Elf32_Half running_arch_code=EM_SPARC; 1984 #elif (defined __powerpc64__) 1985 static Elf32_Half running_arch_code=EM_PPC64; 1986 #elif (defined __powerpc__) 1987 static Elf32_Half running_arch_code=EM_PPC; 1988 #elif (defined ARM) 1989 static Elf32_Half running_arch_code=EM_ARM; 1990 #elif (defined S390) 1991 static Elf32_Half running_arch_code=EM_S390; 1992 #elif (defined ALPHA) 1993 static Elf32_Half running_arch_code=EM_ALPHA; 1994 #elif (defined MIPSEL) 1995 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1996 #elif (defined PARISC) 1997 static Elf32_Half running_arch_code=EM_PARISC; 1998 #elif (defined MIPS) 1999 static Elf32_Half running_arch_code=EM_MIPS; 2000 #elif (defined M68K) 2001 static Elf32_Half running_arch_code=EM_68K; 2002 #else 2003 #error Method os::dll_load requires that one of following is defined:\ 2004 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 2005 #endif 2006 2007 // Identify compatability class for VM's architecture and library's architecture 2008 // Obtain string descriptions for architectures 2009 2010 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 2011 int running_arch_index=-1; 2012 2013 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 2014 if (running_arch_code == arch_array[i].code) { 2015 running_arch_index = i; 2016 } 2017 if (lib_arch.code == arch_array[i].code) { 2018 lib_arch.compat_class = arch_array[i].compat_class; 2019 lib_arch.name = arch_array[i].name; 2020 } 2021 } 2022 2023 assert(running_arch_index != -1, 2024 "Didn't find running architecture code (running_arch_code) in arch_array"); 2025 if (running_arch_index == -1) { 2026 // Even though running architecture detection failed 2027 // we may still continue with reporting dlerror() message 2028 return NULL; 2029 } 2030 2031 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 2032 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 2033 return NULL; 2034 } 2035 2036 #ifndef S390 2037 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 2038 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 2039 return NULL; 2040 } 2041 #endif // !S390 2042 2043 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2044 if ( lib_arch.name!=NULL ) { 2045 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2046 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2047 lib_arch.name, arch_array[running_arch_index].name); 2048 } else { 2049 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2050 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2051 lib_arch.code, 2052 arch_array[running_arch_index].name); 2053 } 2054 } 2055 2056 return NULL; 2057 } 2058 2059 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) { 2060 void * result = ::dlopen(filename, RTLD_LAZY); 2061 if (result == NULL) { 2062 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2063 ebuf[ebuflen-1] = '\0'; 2064 } 2065 return result; 2066 } 2067 2068 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) { 2069 void * result = NULL; 2070 if (LoadExecStackDllInVMThread) { 2071 result = dlopen_helper(filename, ebuf, ebuflen); 2072 } 2073 2074 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2075 // library that requires an executable stack, or which does not have this 2076 // stack attribute set, dlopen changes the stack attribute to executable. The 2077 // read protection of the guard pages gets lost. 2078 // 2079 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2080 // may have been queued at the same time. 2081 2082 if (!_stack_is_executable) { 2083 JavaThread *jt = Threads::first(); 2084 2085 while (jt) { 2086 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2087 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2088 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2089 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2090 warning("Attempt to reguard stack yellow zone failed."); 2091 } 2092 } 2093 jt = jt->next(); 2094 } 2095 } 2096 2097 return result; 2098 } 2099 2100 /* 2101 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2102 * chances are you might want to run the generated bits against glibc-2.0 2103 * libdl.so, so always use locking for any version of glibc. 2104 */ 2105 void* os::dll_lookup(void* handle, const char* name) { 2106 pthread_mutex_lock(&dl_mutex); 2107 void* res = dlsym(handle, name); 2108 pthread_mutex_unlock(&dl_mutex); 2109 return res; 2110 } 2111 2112 void* os::get_default_process_handle() { 2113 return (void*)::dlopen(NULL, RTLD_LAZY); 2114 } 2115 2116 static bool _print_ascii_file(const char* filename, outputStream* st) { 2117 int fd = ::open(filename, O_RDONLY); 2118 if (fd == -1) { 2119 return false; 2120 } 2121 2122 char buf[32]; 2123 int bytes; 2124 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2125 st->print_raw(buf, bytes); 2126 } 2127 2128 ::close(fd); 2129 2130 return true; 2131 } 2132 2133 void os::print_dll_info(outputStream *st) { 2134 st->print_cr("Dynamic libraries:"); 2135 2136 char fname[32]; 2137 pid_t pid = os::Linux::gettid(); 2138 2139 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2140 2141 if (!_print_ascii_file(fname, st)) { 2142 st->print("Can not get library information for pid = %d\n", pid); 2143 } 2144 } 2145 2146 void os::print_os_info_brief(outputStream* st) { 2147 os::Linux::print_distro_info(st); 2148 2149 os::Posix::print_uname_info(st); 2150 2151 os::Linux::print_libversion_info(st); 2152 2153 } 2154 2155 void os::print_os_info(outputStream* st) { 2156 st->print("OS:"); 2157 2158 os::Linux::print_distro_info(st); 2159 2160 os::Posix::print_uname_info(st); 2161 2162 // Print warning if unsafe chroot environment detected 2163 if (unsafe_chroot_detected) { 2164 st->print("WARNING!! "); 2165 st->print_cr(unstable_chroot_error); 2166 } 2167 2168 os::Linux::print_libversion_info(st); 2169 2170 os::Posix::print_rlimit_info(st); 2171 2172 os::Posix::print_load_average(st); 2173 2174 os::Linux::print_full_memory_info(st); 2175 } 2176 2177 // Try to identify popular distros. 2178 // Most Linux distributions have a /etc/XXX-release file, which contains 2179 // the OS version string. Newer Linux distributions have a /etc/lsb-release 2180 // file that also contains the OS version string. Some have more than one 2181 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2182 // /etc/redhat-release.), so the order is important. 2183 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2184 // their own specific XXX-release file as well as a redhat-release file. 2185 // Because of this the XXX-release file needs to be searched for before the 2186 // redhat-release file. 2187 // Since Red Hat has a lsb-release file that is not very descriptive the 2188 // search for redhat-release needs to be before lsb-release. 2189 // Since the lsb-release file is the new standard it needs to be searched 2190 // before the older style release files. 2191 // Searching system-release (Red Hat) and os-release (other Linuxes) are a 2192 // next to last resort. The os-release file is a new standard that contains 2193 // distribution information and the system-release file seems to be an old 2194 // standard that has been replaced by the lsb-release and os-release files. 2195 // Searching for the debian_version file is the last resort. It contains 2196 // an informative string like "6.0.6" or "wheezy/sid". Because of this 2197 // "Debian " is printed before the contents of the debian_version file. 2198 void os::Linux::print_distro_info(outputStream* st) { 2199 if (!_print_ascii_file("/etc/oracle-release", st) && 2200 !_print_ascii_file("/etc/mandriva-release", st) && 2201 !_print_ascii_file("/etc/mandrake-release", st) && 2202 !_print_ascii_file("/etc/sun-release", st) && 2203 !_print_ascii_file("/etc/redhat-release", st) && 2204 !_print_ascii_file("/etc/lsb-release", st) && 2205 !_print_ascii_file("/etc/SuSE-release", st) && 2206 !_print_ascii_file("/etc/turbolinux-release", st) && 2207 !_print_ascii_file("/etc/gentoo-release", st) && 2208 !_print_ascii_file("/etc/ltib-release", st) && 2209 !_print_ascii_file("/etc/angstrom-version", st) && 2210 !_print_ascii_file("/etc/system-release", st) && 2211 !_print_ascii_file("/etc/os-release", st)) { 2212 2213 if (file_exists("/etc/debian_version")) { 2214 st->print("Debian "); 2215 _print_ascii_file("/etc/debian_version", st); 2216 } else { 2217 st->print("Linux"); 2218 } 2219 } 2220 st->cr(); 2221 } 2222 2223 void os::Linux::print_libversion_info(outputStream* st) { 2224 // libc, pthread 2225 st->print("libc:"); 2226 st->print(os::Linux::glibc_version()); st->print(" "); 2227 st->print(os::Linux::libpthread_version()); st->print(" "); 2228 if (os::Linux::is_LinuxThreads()) { 2229 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2230 } 2231 st->cr(); 2232 } 2233 2234 void os::Linux::print_full_memory_info(outputStream* st) { 2235 st->print("\n/proc/meminfo:\n"); 2236 _print_ascii_file("/proc/meminfo", st); 2237 st->cr(); 2238 } 2239 2240 void os::print_memory_info(outputStream* st) { 2241 2242 st->print("Memory:"); 2243 st->print(" %dk page", os::vm_page_size()>>10); 2244 2245 // values in struct sysinfo are "unsigned long" 2246 struct sysinfo si; 2247 sysinfo(&si); 2248 2249 st->print(", physical " UINT64_FORMAT "k", 2250 os::physical_memory() >> 10); 2251 st->print("(" UINT64_FORMAT "k free)", 2252 os::available_memory() >> 10); 2253 st->print(", swap " UINT64_FORMAT "k", 2254 ((jlong)si.totalswap * si.mem_unit) >> 10); 2255 st->print("(" UINT64_FORMAT "k free)", 2256 ((jlong)si.freeswap * si.mem_unit) >> 10); 2257 st->cr(); 2258 } 2259 2260 void os::pd_print_cpu_info(outputStream* st) { 2261 st->print("\n/proc/cpuinfo:\n"); 2262 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2263 st->print(" <Not Available>"); 2264 } 2265 st->cr(); 2266 } 2267 2268 void os::print_siginfo(outputStream* st, void* siginfo) { 2269 const siginfo_t* si = (const siginfo_t*)siginfo; 2270 2271 os::Posix::print_siginfo_brief(st, si); 2272 2273 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2274 UseSharedSpaces) { 2275 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2276 if (mapinfo->is_in_shared_space(si->si_addr)) { 2277 st->print("\n\nError accessing class data sharing archive." \ 2278 " Mapped file inaccessible during execution, " \ 2279 " possible disk/network problem."); 2280 } 2281 } 2282 st->cr(); 2283 } 2284 2285 2286 static void print_signal_handler(outputStream* st, int sig, 2287 char* buf, size_t buflen); 2288 2289 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2290 st->print_cr("Signal Handlers:"); 2291 print_signal_handler(st, SIGSEGV, buf, buflen); 2292 print_signal_handler(st, SIGBUS , buf, buflen); 2293 print_signal_handler(st, SIGFPE , buf, buflen); 2294 print_signal_handler(st, SIGPIPE, buf, buflen); 2295 print_signal_handler(st, SIGXFSZ, buf, buflen); 2296 print_signal_handler(st, SIGILL , buf, buflen); 2297 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2298 print_signal_handler(st, SR_signum, buf, buflen); 2299 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2300 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2301 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2302 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2303 #if defined(PPC64) 2304 print_signal_handler(st, SIGTRAP, buf, buflen); 2305 #endif 2306 } 2307 2308 static char saved_jvm_path[MAXPATHLEN] = {0}; 2309 2310 // Find the full path to the current module, libjvm.so 2311 void os::jvm_path(char *buf, jint buflen) { 2312 // Error checking. 2313 if (buflen < MAXPATHLEN) { 2314 assert(false, "must use a large-enough buffer"); 2315 buf[0] = '\0'; 2316 return; 2317 } 2318 // Lazy resolve the path to current module. 2319 if (saved_jvm_path[0] != 0) { 2320 strcpy(buf, saved_jvm_path); 2321 return; 2322 } 2323 2324 char dli_fname[MAXPATHLEN]; 2325 bool ret = dll_address_to_library_name( 2326 CAST_FROM_FN_PTR(address, os::jvm_path), 2327 dli_fname, sizeof(dli_fname), NULL); 2328 assert(ret, "cannot locate libjvm"); 2329 char *rp = NULL; 2330 if (ret && dli_fname[0] != '\0') { 2331 rp = realpath(dli_fname, buf); 2332 } 2333 if (rp == NULL) 2334 return; 2335 2336 if (Arguments::sun_java_launcher_is_altjvm()) { 2337 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2338 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2339 // If "/jre/lib/" appears at the right place in the string, then 2340 // assume we are installed in a JDK and we're done. Otherwise, check 2341 // for a JAVA_HOME environment variable and fix up the path so it 2342 // looks like libjvm.so is installed there (append a fake suffix 2343 // hotspot/libjvm.so). 2344 const char *p = buf + strlen(buf) - 1; 2345 for (int count = 0; p > buf && count < 5; ++count) { 2346 for (--p; p > buf && *p != '/'; --p) 2347 /* empty */ ; 2348 } 2349 2350 if (strncmp(p, "/jre/lib/", 9) != 0) { 2351 // Look for JAVA_HOME in the environment. 2352 char* java_home_var = ::getenv("JAVA_HOME"); 2353 if (java_home_var != NULL && java_home_var[0] != 0) { 2354 char* jrelib_p; 2355 int len; 2356 2357 // Check the current module name "libjvm.so". 2358 p = strrchr(buf, '/'); 2359 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2360 2361 rp = realpath(java_home_var, buf); 2362 if (rp == NULL) 2363 return; 2364 2365 // determine if this is a legacy image or modules image 2366 // modules image doesn't have "jre" subdirectory 2367 len = strlen(buf); 2368 jrelib_p = buf + len; 2369 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2370 if (0 != access(buf, F_OK)) { 2371 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2372 } 2373 2374 if (0 == access(buf, F_OK)) { 2375 // Use current module name "libjvm.so" 2376 len = strlen(buf); 2377 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2378 } else { 2379 // Go back to path of .so 2380 rp = realpath(dli_fname, buf); 2381 if (rp == NULL) 2382 return; 2383 } 2384 } 2385 } 2386 } 2387 2388 strcpy(saved_jvm_path, buf); 2389 } 2390 2391 void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2392 // no prefix required, not even "_" 2393 } 2394 2395 void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2396 // no suffix required 2397 } 2398 2399 //////////////////////////////////////////////////////////////////////////////// 2400 // sun.misc.Signal support 2401 2402 static volatile jint sigint_count = 0; 2403 2404 static void 2405 UserHandler(int sig, void *siginfo, void *context) { 2406 // 4511530 - sem_post is serialized and handled by the manager thread. When 2407 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2408 // don't want to flood the manager thread with sem_post requests. 2409 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2410 return; 2411 2412 // Ctrl-C is pressed during error reporting, likely because the error 2413 // handler fails to abort. Let VM die immediately. 2414 if (sig == SIGINT && is_error_reported()) { 2415 os::die(); 2416 } 2417 2418 os::signal_notify(sig); 2419 } 2420 2421 void* os::user_handler() { 2422 return CAST_FROM_FN_PTR(void*, UserHandler); 2423 } 2424 2425 class Semaphore : public StackObj { 2426 public: 2427 Semaphore(); 2428 ~Semaphore(); 2429 void signal(); 2430 void wait(); 2431 bool trywait(); 2432 bool timedwait(unsigned int sec, int nsec); 2433 private: 2434 sem_t _semaphore; 2435 }; 2436 2437 2438 Semaphore::Semaphore() { 2439 sem_init(&_semaphore, 0, 0); 2440 } 2441 2442 Semaphore::~Semaphore() { 2443 sem_destroy(&_semaphore); 2444 } 2445 2446 void Semaphore::signal() { 2447 sem_post(&_semaphore); 2448 } 2449 2450 void Semaphore::wait() { 2451 sem_wait(&_semaphore); 2452 } 2453 2454 bool Semaphore::trywait() { 2455 return sem_trywait(&_semaphore) == 0; 2456 } 2457 2458 bool Semaphore::timedwait(unsigned int sec, int nsec) { 2459 struct timespec ts; 2460 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec); 2461 2462 while (1) { 2463 int result = sem_timedwait(&_semaphore, &ts); 2464 if (result == 0) { 2465 return true; 2466 } else if (errno == EINTR) { 2467 continue; 2468 } else if (errno == ETIMEDOUT) { 2469 return false; 2470 } else { 2471 return false; 2472 } 2473 } 2474 } 2475 2476 extern "C" { 2477 typedef void (*sa_handler_t)(int); 2478 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2479 } 2480 2481 void* os::signal(int signal_number, void* handler) { 2482 struct sigaction sigAct, oldSigAct; 2483 2484 sigfillset(&(sigAct.sa_mask)); 2485 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2486 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2487 2488 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2489 // -1 means registration failed 2490 return (void *)-1; 2491 } 2492 2493 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2494 } 2495 2496 void os::signal_raise(int signal_number) { 2497 ::raise(signal_number); 2498 } 2499 2500 /* 2501 * The following code is moved from os.cpp for making this 2502 * code platform specific, which it is by its very nature. 2503 */ 2504 2505 // Will be modified when max signal is changed to be dynamic 2506 int os::sigexitnum_pd() { 2507 return NSIG; 2508 } 2509 2510 // a counter for each possible signal value 2511 static volatile jint pending_signals[NSIG+1] = { 0 }; 2512 2513 // Linux(POSIX) specific hand shaking semaphore. 2514 static sem_t sig_sem; 2515 static Semaphore sr_semaphore; 2516 2517 void os::signal_init_pd() { 2518 // Initialize signal structures 2519 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2520 2521 // Initialize signal semaphore 2522 ::sem_init(&sig_sem, 0, 0); 2523 } 2524 2525 void os::signal_notify(int sig) { 2526 Atomic::inc(&pending_signals[sig]); 2527 ::sem_post(&sig_sem); 2528 } 2529 2530 static int check_pending_signals(bool wait) { 2531 Atomic::store(0, &sigint_count); 2532 for (;;) { 2533 for (int i = 0; i < NSIG + 1; i++) { 2534 jint n = pending_signals[i]; 2535 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2536 return i; 2537 } 2538 } 2539 if (!wait) { 2540 return -1; 2541 } 2542 JavaThread *thread = JavaThread::current(); 2543 ThreadBlockInVM tbivm(thread); 2544 2545 bool threadIsSuspended; 2546 do { 2547 thread->set_suspend_equivalent(); 2548 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2549 ::sem_wait(&sig_sem); 2550 2551 // were we externally suspended while we were waiting? 2552 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2553 if (threadIsSuspended) { 2554 // 2555 // The semaphore has been incremented, but while we were waiting 2556 // another thread suspended us. We don't want to continue running 2557 // while suspended because that would surprise the thread that 2558 // suspended us. 2559 // 2560 ::sem_post(&sig_sem); 2561 2562 thread->java_suspend_self(); 2563 } 2564 } while (threadIsSuspended); 2565 } 2566 } 2567 2568 int os::signal_lookup() { 2569 return check_pending_signals(false); 2570 } 2571 2572 int os::signal_wait() { 2573 return check_pending_signals(true); 2574 } 2575 2576 //////////////////////////////////////////////////////////////////////////////// 2577 // Virtual Memory 2578 2579 int os::vm_page_size() { 2580 // Seems redundant as all get out 2581 assert(os::Linux::page_size() != -1, "must call os::init"); 2582 return os::Linux::page_size(); 2583 } 2584 2585 // Solaris allocates memory by pages. 2586 int os::vm_allocation_granularity() { 2587 assert(os::Linux::page_size() != -1, "must call os::init"); 2588 return os::Linux::page_size(); 2589 } 2590 2591 // Rationale behind this function: 2592 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2593 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2594 // samples for JITted code. Here we create private executable mapping over the code cache 2595 // and then we can use standard (well, almost, as mapping can change) way to provide 2596 // info for the reporting script by storing timestamp and location of symbol 2597 void linux_wrap_code(char* base, size_t size) { 2598 static volatile jint cnt = 0; 2599 2600 if (!UseOprofile) { 2601 return; 2602 } 2603 2604 char buf[PATH_MAX+1]; 2605 int num = Atomic::add(1, &cnt); 2606 2607 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2608 os::get_temp_directory(), os::current_process_id(), num); 2609 unlink(buf); 2610 2611 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2612 2613 if (fd != -1) { 2614 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2615 if (rv != (off_t)-1) { 2616 if (::write(fd, "", 1) == 1) { 2617 mmap(base, size, 2618 PROT_READ|PROT_WRITE|PROT_EXEC, 2619 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2620 } 2621 } 2622 ::close(fd); 2623 unlink(buf); 2624 } 2625 } 2626 2627 static bool recoverable_mmap_error(int err) { 2628 // See if the error is one we can let the caller handle. This 2629 // list of errno values comes from JBS-6843484. I can't find a 2630 // Linux man page that documents this specific set of errno 2631 // values so while this list currently matches Solaris, it may 2632 // change as we gain experience with this failure mode. 2633 switch (err) { 2634 case EBADF: 2635 case EINVAL: 2636 case ENOTSUP: 2637 // let the caller deal with these errors 2638 return true; 2639 2640 default: 2641 // Any remaining errors on this OS can cause our reserved mapping 2642 // to be lost. That can cause confusion where different data 2643 // structures think they have the same memory mapped. The worst 2644 // scenario is if both the VM and a library think they have the 2645 // same memory mapped. 2646 return false; 2647 } 2648 } 2649 2650 static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2651 int err) { 2652 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2653 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2654 strerror(err), err); 2655 } 2656 2657 static void warn_fail_commit_memory(char* addr, size_t size, 2658 size_t alignment_hint, bool exec, 2659 int err) { 2660 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2661 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2662 alignment_hint, exec, strerror(err), err); 2663 } 2664 2665 // NOTE: Linux kernel does not really reserve the pages for us. 2666 // All it does is to check if there are enough free pages 2667 // left at the time of mmap(). This could be a potential 2668 // problem. 2669 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2670 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2671 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2672 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2673 if (res != (uintptr_t) MAP_FAILED) { 2674 if (UseNUMAInterleaving) { 2675 numa_make_global(addr, size); 2676 } 2677 return 0; 2678 } 2679 2680 int err = errno; // save errno from mmap() call above 2681 2682 if (!recoverable_mmap_error(err)) { 2683 warn_fail_commit_memory(addr, size, exec, err); 2684 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2685 } 2686 2687 return err; 2688 } 2689 2690 bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2691 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2692 } 2693 2694 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2695 const char* mesg) { 2696 assert(mesg != NULL, "mesg must be specified"); 2697 int err = os::Linux::commit_memory_impl(addr, size, exec); 2698 if (err != 0) { 2699 // the caller wants all commit errors to exit with the specified mesg: 2700 warn_fail_commit_memory(addr, size, exec, err); 2701 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2702 } 2703 } 2704 2705 // Define MAP_HUGETLB here so we can build HotSpot on old systems. 2706 #ifndef MAP_HUGETLB 2707 #define MAP_HUGETLB 0x40000 2708 #endif 2709 2710 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2711 #ifndef MADV_HUGEPAGE 2712 #define MADV_HUGEPAGE 14 2713 #endif 2714 2715 int os::Linux::commit_memory_impl(char* addr, size_t size, 2716 size_t alignment_hint, bool exec) { 2717 int err = os::Linux::commit_memory_impl(addr, size, exec); 2718 if (err == 0) { 2719 realign_memory(addr, size, alignment_hint); 2720 } 2721 return err; 2722 } 2723 2724 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2725 bool exec) { 2726 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2727 } 2728 2729 void os::pd_commit_memory_or_exit(char* addr, size_t size, 2730 size_t alignment_hint, bool exec, 2731 const char* mesg) { 2732 assert(mesg != NULL, "mesg must be specified"); 2733 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2734 if (err != 0) { 2735 // the caller wants all commit errors to exit with the specified mesg: 2736 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2737 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2738 } 2739 } 2740 2741 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2742 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2743 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2744 // be supported or the memory may already be backed by huge pages. 2745 ::madvise(addr, bytes, MADV_HUGEPAGE); 2746 } 2747 } 2748 2749 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2750 // This method works by doing an mmap over an existing mmaping and effectively discarding 2751 // the existing pages. However it won't work for SHM-based large pages that cannot be 2752 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2753 // small pages on top of the SHM segment. This method always works for small pages, so we 2754 // allow that in any case. 2755 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2756 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2757 } 2758 } 2759 2760 void os::numa_make_global(char *addr, size_t bytes) { 2761 Linux::numa_interleave_memory(addr, bytes); 2762 } 2763 2764 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2765 // bind policy to MPOL_PREFERRED for the current thread. 2766 #define USE_MPOL_PREFERRED 0 2767 2768 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2769 // To make NUMA and large pages more robust when both enabled, we need to ease 2770 // the requirements on where the memory should be allocated. MPOL_BIND is the 2771 // default policy and it will force memory to be allocated on the specified 2772 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2773 // the specified node, but will not force it. Using this policy will prevent 2774 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2775 // free large pages. 2776 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2777 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2778 } 2779 2780 bool os::numa_topology_changed() { return false; } 2781 2782 size_t os::numa_get_groups_num() { 2783 int max_node = Linux::numa_max_node(); 2784 return max_node > 0 ? max_node + 1 : 1; 2785 } 2786 2787 int os::numa_get_group_id() { 2788 int cpu_id = Linux::sched_getcpu(); 2789 if (cpu_id != -1) { 2790 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2791 if (lgrp_id != -1) { 2792 return lgrp_id; 2793 } 2794 } 2795 return 0; 2796 } 2797 2798 size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2799 for (size_t i = 0; i < size; i++) { 2800 ids[i] = i; 2801 } 2802 return size; 2803 } 2804 2805 bool os::get_page_info(char *start, page_info* info) { 2806 return false; 2807 } 2808 2809 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2810 return end; 2811 } 2812 2813 2814 int os::Linux::sched_getcpu_syscall(void) { 2815 unsigned int cpu; 2816 int retval = -1; 2817 2818 #if defined(IA32) 2819 # ifndef SYS_getcpu 2820 # define SYS_getcpu 318 2821 # endif 2822 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2823 #elif defined(AMD64) 2824 // Unfortunately we have to bring all these macros here from vsyscall.h 2825 // to be able to compile on old linuxes. 2826 # define __NR_vgetcpu 2 2827 # define VSYSCALL_START (-10UL << 20) 2828 # define VSYSCALL_SIZE 1024 2829 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2830 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2831 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2832 retval = vgetcpu(&cpu, NULL, NULL); 2833 #endif 2834 2835 return (retval == -1) ? retval : cpu; 2836 } 2837 2838 // Something to do with the numa-aware allocator needs these symbols 2839 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2840 extern "C" JNIEXPORT void numa_error(char *where) { } 2841 extern "C" JNIEXPORT int fork1() { return fork(); } 2842 2843 2844 // If we are running with libnuma version > 2, then we should 2845 // be trying to use symbols with versions 1.1 2846 // If we are running with earlier version, which did not have symbol versions, 2847 // we should use the base version. 2848 void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2849 void *f = dlvsym(handle, name, "libnuma_1.1"); 2850 if (f == NULL) { 2851 f = dlsym(handle, name); 2852 } 2853 return f; 2854 } 2855 2856 bool os::Linux::libnuma_init() { 2857 // sched_getcpu() should be in libc. 2858 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2859 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2860 2861 // If it's not, try a direct syscall. 2862 if (sched_getcpu() == -1) 2863 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall)); 2864 2865 if (sched_getcpu() != -1) { // Does it work? 2866 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2867 if (handle != NULL) { 2868 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2869 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2870 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2871 libnuma_dlsym(handle, "numa_max_node"))); 2872 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2873 libnuma_dlsym(handle, "numa_available"))); 2874 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2875 libnuma_dlsym(handle, "numa_tonode_memory"))); 2876 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2877 libnuma_dlsym(handle, "numa_interleave_memory"))); 2878 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2879 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2880 2881 2882 if (numa_available() != -1) { 2883 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2884 // Create a cpu -> node mapping 2885 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2886 rebuild_cpu_to_node_map(); 2887 return true; 2888 } 2889 } 2890 } 2891 return false; 2892 } 2893 2894 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2895 // The table is later used in get_node_by_cpu(). 2896 void os::Linux::rebuild_cpu_to_node_map() { 2897 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2898 // in libnuma (possible values are starting from 16, 2899 // and continuing up with every other power of 2, but less 2900 // than the maximum number of CPUs supported by kernel), and 2901 // is a subject to change (in libnuma version 2 the requirements 2902 // are more reasonable) we'll just hardcode the number they use 2903 // in the library. 2904 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2905 2906 size_t cpu_num = os::active_processor_count(); 2907 size_t cpu_map_size = NCPUS / BitsPerCLong; 2908 size_t cpu_map_valid_size = 2909 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2910 2911 cpu_to_node()->clear(); 2912 cpu_to_node()->at_grow(cpu_num - 1); 2913 size_t node_num = numa_get_groups_num(); 2914 2915 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2916 for (size_t i = 0; i < node_num; i++) { 2917 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2918 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2919 if (cpu_map[j] != 0) { 2920 for (size_t k = 0; k < BitsPerCLong; k++) { 2921 if (cpu_map[j] & (1UL << k)) { 2922 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2923 } 2924 } 2925 } 2926 } 2927 } 2928 } 2929 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2930 } 2931 2932 int os::Linux::get_node_by_cpu(int cpu_id) { 2933 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2934 return cpu_to_node()->at(cpu_id); 2935 } 2936 return -1; 2937 } 2938 2939 GrowableArray<int>* os::Linux::_cpu_to_node; 2940 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2941 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2942 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2943 os::Linux::numa_available_func_t os::Linux::_numa_available; 2944 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2945 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2946 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2947 unsigned long* os::Linux::_numa_all_nodes; 2948 2949 bool os::pd_uncommit_memory(char* addr, size_t size) { 2950 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2951 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2952 return res != (uintptr_t) MAP_FAILED; 2953 } 2954 2955 static 2956 address get_stack_commited_bottom(address bottom, size_t size) { 2957 address nbot = bottom; 2958 address ntop = bottom + size; 2959 2960 size_t page_sz = os::vm_page_size(); 2961 unsigned pages = size / page_sz; 2962 2963 unsigned char vec[1]; 2964 unsigned imin = 1, imax = pages + 1, imid; 2965 int mincore_return_value = 0; 2966 2967 assert(imin <= imax, "Unexpected page size"); 2968 2969 while (imin < imax) { 2970 imid = (imax + imin) / 2; 2971 nbot = ntop - (imid * page_sz); 2972 2973 // Use a trick with mincore to check whether the page is mapped or not. 2974 // mincore sets vec to 1 if page resides in memory and to 0 if page 2975 // is swapped output but if page we are asking for is unmapped 2976 // it returns -1,ENOMEM 2977 mincore_return_value = mincore(nbot, page_sz, vec); 2978 2979 if (mincore_return_value == -1) { 2980 // Page is not mapped go up 2981 // to find first mapped page 2982 if (errno != EAGAIN) { 2983 assert(errno == ENOMEM, "Unexpected mincore errno"); 2984 imax = imid; 2985 } 2986 } else { 2987 // Page is mapped go down 2988 // to find first not mapped page 2989 imin = imid + 1; 2990 } 2991 } 2992 2993 nbot = nbot + page_sz; 2994 2995 // Adjust stack bottom one page up if last checked page is not mapped 2996 if (mincore_return_value == -1) { 2997 nbot = nbot + page_sz; 2998 } 2999 3000 return nbot; 3001 } 3002 3003 3004 // Linux uses a growable mapping for the stack, and if the mapping for 3005 // the stack guard pages is not removed when we detach a thread the 3006 // stack cannot grow beyond the pages where the stack guard was 3007 // mapped. If at some point later in the process the stack expands to 3008 // that point, the Linux kernel cannot expand the stack any further 3009 // because the guard pages are in the way, and a segfault occurs. 3010 // 3011 // However, it's essential not to split the stack region by unmapping 3012 // a region (leaving a hole) that's already part of the stack mapping, 3013 // so if the stack mapping has already grown beyond the guard pages at 3014 // the time we create them, we have to truncate the stack mapping. 3015 // So, we need to know the extent of the stack mapping when 3016 // create_stack_guard_pages() is called. 3017 3018 // We only need this for stacks that are growable: at the time of 3019 // writing thread stacks don't use growable mappings (i.e. those 3020 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3021 // only applies to the main thread. 3022 3023 // If the (growable) stack mapping already extends beyond the point 3024 // where we're going to put our guard pages, truncate the mapping at 3025 // that point by munmap()ping it. This ensures that when we later 3026 // munmap() the guard pages we don't leave a hole in the stack 3027 // mapping. This only affects the main/initial thread 3028 3029 bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3030 3031 if (os::Linux::is_initial_thread()) { 3032 // As we manually grow stack up to bottom inside create_attached_thread(), 3033 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3034 // we don't need to do anything special. 3035 // Check it first, before calling heavy function. 3036 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3037 unsigned char vec[1]; 3038 3039 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3040 // Fallback to slow path on all errors, including EAGAIN 3041 stack_extent = (uintptr_t) get_stack_commited_bottom( 3042 os::Linux::initial_thread_stack_bottom(), 3043 (size_t)addr - stack_extent); 3044 } 3045 3046 if (stack_extent < (uintptr_t)addr) { 3047 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3048 } 3049 } 3050 3051 return os::commit_memory(addr, size, !ExecMem); 3052 } 3053 3054 // If this is a growable mapping, remove the guard pages entirely by 3055 // munmap()ping them. If not, just call uncommit_memory(). This only 3056 // affects the main/initial thread, but guard against future OS changes 3057 // It's safe to always unmap guard pages for initial thread because we 3058 // always place it right after end of the mapped region 3059 3060 bool os::remove_stack_guard_pages(char* addr, size_t size) { 3061 uintptr_t stack_extent, stack_base; 3062 3063 if (os::Linux::is_initial_thread()) { 3064 return ::munmap(addr, size) == 0; 3065 } 3066 3067 return os::uncommit_memory(addr, size); 3068 } 3069 3070 static address _highest_vm_reserved_address = NULL; 3071 3072 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3073 // at 'requested_addr'. If there are existing memory mappings at the same 3074 // location, however, they will be overwritten. If 'fixed' is false, 3075 // 'requested_addr' is only treated as a hint, the return value may or 3076 // may not start from the requested address. Unlike Linux mmap(), this 3077 // function returns NULL to indicate failure. 3078 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3079 char * addr; 3080 int flags; 3081 3082 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3083 if (fixed) { 3084 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3085 flags |= MAP_FIXED; 3086 } 3087 3088 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3089 // touch an uncommitted page. Otherwise, the read/write might 3090 // succeed if we have enough swap space to back the physical page. 3091 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3092 flags, -1, 0); 3093 3094 if (addr != MAP_FAILED) { 3095 // anon_mmap() should only get called during VM initialization, 3096 // don't need lock (actually we can skip locking even it can be called 3097 // from multiple threads, because _highest_vm_reserved_address is just a 3098 // hint about the upper limit of non-stack memory regions.) 3099 if ((address)addr + bytes > _highest_vm_reserved_address) { 3100 _highest_vm_reserved_address = (address)addr + bytes; 3101 } 3102 } 3103 3104 return addr == MAP_FAILED ? NULL : addr; 3105 } 3106 3107 // Don't update _highest_vm_reserved_address, because there might be memory 3108 // regions above addr + size. If so, releasing a memory region only creates 3109 // a hole in the address space, it doesn't help prevent heap-stack collision. 3110 // 3111 static int anon_munmap(char * addr, size_t size) { 3112 return ::munmap(addr, size) == 0; 3113 } 3114 3115 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3116 size_t alignment_hint) { 3117 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3118 } 3119 3120 bool os::pd_release_memory(char* addr, size_t size) { 3121 return anon_munmap(addr, size); 3122 } 3123 3124 static address highest_vm_reserved_address() { 3125 return _highest_vm_reserved_address; 3126 } 3127 3128 static bool linux_mprotect(char* addr, size_t size, int prot) { 3129 // Linux wants the mprotect address argument to be page aligned. 3130 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3131 3132 // According to SUSv3, mprotect() should only be used with mappings 3133 // established by mmap(), and mmap() always maps whole pages. Unaligned 3134 // 'addr' likely indicates problem in the VM (e.g. trying to change 3135 // protection of malloc'ed or statically allocated memory). Check the 3136 // caller if you hit this assert. 3137 assert(addr == bottom, "sanity check"); 3138 3139 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3140 return ::mprotect(bottom, size, prot) == 0; 3141 } 3142 3143 // Set protections specified 3144 bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3145 bool is_committed) { 3146 unsigned int p = 0; 3147 switch (prot) { 3148 case MEM_PROT_NONE: p = PROT_NONE; break; 3149 case MEM_PROT_READ: p = PROT_READ; break; 3150 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3151 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3152 default: 3153 ShouldNotReachHere(); 3154 } 3155 // is_committed is unused. 3156 return linux_mprotect(addr, bytes, p); 3157 } 3158 3159 bool os::guard_memory(char* addr, size_t size) { 3160 return linux_mprotect(addr, size, PROT_NONE); 3161 } 3162 3163 bool os::unguard_memory(char* addr, size_t size) { 3164 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3165 } 3166 3167 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) { 3168 bool result = false; 3169 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3170 MAP_ANONYMOUS|MAP_PRIVATE, 3171 -1, 0); 3172 if (p != MAP_FAILED) { 3173 void *aligned_p = align_ptr_up(p, page_size); 3174 3175 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3176 3177 munmap(p, page_size * 2); 3178 } 3179 3180 if (warn && !result) { 3181 warning("TransparentHugePages is not supported by the operating system."); 3182 } 3183 3184 return result; 3185 } 3186 3187 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3188 bool result = false; 3189 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3190 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3191 -1, 0); 3192 3193 if (p != MAP_FAILED) { 3194 // We don't know if this really is a huge page or not. 3195 FILE *fp = fopen("/proc/self/maps", "r"); 3196 if (fp) { 3197 while (!feof(fp)) { 3198 char chars[257]; 3199 long x = 0; 3200 if (fgets(chars, sizeof(chars), fp)) { 3201 if (sscanf(chars, "%lx-%*x", &x) == 1 3202 && x == (long)p) { 3203 if (strstr (chars, "hugepage")) { 3204 result = true; 3205 break; 3206 } 3207 } 3208 } 3209 } 3210 fclose(fp); 3211 } 3212 munmap(p, page_size); 3213 } 3214 3215 if (warn && !result) { 3216 warning("HugeTLBFS is not supported by the operating system."); 3217 } 3218 3219 return result; 3220 } 3221 3222 /* 3223 * Set the coredump_filter bits to include largepages in core dump (bit 6) 3224 * 3225 * From the coredump_filter documentation: 3226 * 3227 * - (bit 0) anonymous private memory 3228 * - (bit 1) anonymous shared memory 3229 * - (bit 2) file-backed private memory 3230 * - (bit 3) file-backed shared memory 3231 * - (bit 4) ELF header pages in file-backed private memory areas (it is 3232 * effective only if the bit 2 is cleared) 3233 * - (bit 5) hugetlb private memory 3234 * - (bit 6) hugetlb shared memory 3235 */ 3236 static void set_coredump_filter(void) { 3237 FILE *f; 3238 long cdm; 3239 3240 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3241 return; 3242 } 3243 3244 if (fscanf(f, "%lx", &cdm) != 1) { 3245 fclose(f); 3246 return; 3247 } 3248 3249 rewind(f); 3250 3251 if ((cdm & LARGEPAGES_BIT) == 0) { 3252 cdm |= LARGEPAGES_BIT; 3253 fprintf(f, "%#lx", cdm); 3254 } 3255 3256 fclose(f); 3257 } 3258 3259 // Large page support 3260 3261 static size_t _large_page_size = 0; 3262 3263 size_t os::Linux::find_large_page_size() { 3264 size_t large_page_size = 0; 3265 3266 // large_page_size on Linux is used to round up heap size. x86 uses either 3267 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3268 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3269 // page as large as 256M. 3270 // 3271 // Here we try to figure out page size by parsing /proc/meminfo and looking 3272 // for a line with the following format: 3273 // Hugepagesize: 2048 kB 3274 // 3275 // If we can't determine the value (e.g. /proc is not mounted, or the text 3276 // format has been changed), we'll use the largest page size supported by 3277 // the processor. 3278 3279 #ifndef ZERO 3280 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3281 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3282 #endif // ZERO 3283 3284 FILE *fp = fopen("/proc/meminfo", "r"); 3285 if (fp) { 3286 while (!feof(fp)) { 3287 int x = 0; 3288 char buf[16]; 3289 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3290 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3291 large_page_size = x * K; 3292 break; 3293 } 3294 } else { 3295 // skip to next line 3296 for (;;) { 3297 int ch = fgetc(fp); 3298 if (ch == EOF || ch == (int)'\n') break; 3299 } 3300 } 3301 } 3302 fclose(fp); 3303 } 3304 3305 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3306 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3307 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3308 proper_unit_for_byte_size(large_page_size)); 3309 } 3310 3311 return large_page_size; 3312 } 3313 3314 size_t os::Linux::setup_large_page_size() { 3315 _large_page_size = Linux::find_large_page_size(); 3316 const size_t default_page_size = (size_t)Linux::page_size(); 3317 if (_large_page_size > default_page_size) { 3318 _page_sizes[0] = _large_page_size; 3319 _page_sizes[1] = default_page_size; 3320 _page_sizes[2] = 0; 3321 } 3322 3323 return _large_page_size; 3324 } 3325 3326 bool os::Linux::setup_large_page_type(size_t page_size) { 3327 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3328 FLAG_IS_DEFAULT(UseSHM) && 3329 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3330 3331 // The type of large pages has not been specified by the user. 3332 3333 // Try UseHugeTLBFS and then UseSHM. 3334 UseHugeTLBFS = UseSHM = true; 3335 3336 // Don't try UseTransparentHugePages since there are known 3337 // performance issues with it turned on. This might change in the future. 3338 UseTransparentHugePages = false; 3339 } 3340 3341 if (UseTransparentHugePages) { 3342 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3343 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3344 UseHugeTLBFS = false; 3345 UseSHM = false; 3346 return true; 3347 } 3348 UseTransparentHugePages = false; 3349 } 3350 3351 if (UseHugeTLBFS) { 3352 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3353 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3354 UseSHM = false; 3355 return true; 3356 } 3357 UseHugeTLBFS = false; 3358 } 3359 3360 return UseSHM; 3361 } 3362 3363 void os::large_page_init() { 3364 if (!UseLargePages && 3365 !UseTransparentHugePages && 3366 !UseHugeTLBFS && 3367 !UseSHM) { 3368 // Not using large pages. 3369 return; 3370 } 3371 3372 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3373 // The user explicitly turned off large pages. 3374 // Ignore the rest of the large pages flags. 3375 UseTransparentHugePages = false; 3376 UseHugeTLBFS = false; 3377 UseSHM = false; 3378 return; 3379 } 3380 3381 size_t large_page_size = Linux::setup_large_page_size(); 3382 UseLargePages = Linux::setup_large_page_type(large_page_size); 3383 3384 set_coredump_filter(); 3385 } 3386 3387 #ifndef SHM_HUGETLB 3388 #define SHM_HUGETLB 04000 3389 #endif 3390 3391 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3392 // "exec" is passed in but not used. Creating the shared image for 3393 // the code cache doesn't have an SHM_X executable permission to check. 3394 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3395 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3396 3397 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3398 return NULL; // Fallback to small pages. 3399 } 3400 3401 key_t key = IPC_PRIVATE; 3402 char *addr; 3403 3404 bool warn_on_failure = UseLargePages && 3405 (!FLAG_IS_DEFAULT(UseLargePages) || 3406 !FLAG_IS_DEFAULT(UseSHM) || 3407 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 3408 ); 3409 char msg[128]; 3410 3411 // Create a large shared memory region to attach to based on size. 3412 // Currently, size is the total size of the heap 3413 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3414 if (shmid == -1) { 3415 // Possible reasons for shmget failure: 3416 // 1. shmmax is too small for Java heap. 3417 // > check shmmax value: cat /proc/sys/kernel/shmmax 3418 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3419 // 2. not enough large page memory. 3420 // > check available large pages: cat /proc/meminfo 3421 // > increase amount of large pages: 3422 // echo new_value > /proc/sys/vm/nr_hugepages 3423 // Note 1: different Linux may use different name for this property, 3424 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3425 // Note 2: it's possible there's enough physical memory available but 3426 // they are so fragmented after a long run that they can't 3427 // coalesce into large pages. Try to reserve large pages when 3428 // the system is still "fresh". 3429 if (warn_on_failure) { 3430 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3431 warning(msg); 3432 } 3433 return NULL; 3434 } 3435 3436 // attach to the region 3437 addr = (char*)shmat(shmid, req_addr, 0); 3438 int err = errno; 3439 3440 // Remove shmid. If shmat() is successful, the actual shared memory segment 3441 // will be deleted when it's detached by shmdt() or when the process 3442 // terminates. If shmat() is not successful this will remove the shared 3443 // segment immediately. 3444 shmctl(shmid, IPC_RMID, NULL); 3445 3446 if ((intptr_t)addr == -1) { 3447 if (warn_on_failure) { 3448 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3449 warning(msg); 3450 } 3451 return NULL; 3452 } 3453 3454 return addr; 3455 } 3456 3457 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) { 3458 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3459 3460 bool warn_on_failure = UseLargePages && 3461 (!FLAG_IS_DEFAULT(UseLargePages) || 3462 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3463 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3464 3465 if (warn_on_failure) { 3466 char msg[128]; 3467 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3468 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3469 warning(msg); 3470 } 3471 } 3472 3473 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) { 3474 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3475 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3476 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3477 3478 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3479 char* addr = (char*)::mmap(req_addr, bytes, prot, 3480 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3481 -1, 0); 3482 3483 if (addr == MAP_FAILED) { 3484 warn_on_large_pages_failure(req_addr, bytes, errno); 3485 return NULL; 3486 } 3487 3488 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3489 3490 return addr; 3491 } 3492 3493 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3494 size_t large_page_size = os::large_page_size(); 3495 3496 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3497 3498 // Allocate small pages. 3499 3500 char* start; 3501 if (req_addr != NULL) { 3502 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3503 assert(is_size_aligned(bytes, alignment), "Must be"); 3504 start = os::reserve_memory(bytes, req_addr); 3505 assert(start == NULL || start == req_addr, "Must be"); 3506 } else { 3507 start = os::reserve_memory_aligned(bytes, alignment); 3508 } 3509 3510 if (start == NULL) { 3511 return NULL; 3512 } 3513 3514 assert(is_ptr_aligned(start, alignment), "Must be"); 3515 3516 // os::reserve_memory_special will record this memory area. 3517 // Need to release it here to prevent overlapping reservations. 3518 MemTracker::record_virtual_memory_release((address)start, bytes); 3519 3520 char* end = start + bytes; 3521 3522 // Find the regions of the allocated chunk that can be promoted to large pages. 3523 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3524 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3525 3526 size_t lp_bytes = lp_end - lp_start; 3527 3528 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3529 3530 if (lp_bytes == 0) { 3531 // The mapped region doesn't even span the start and the end of a large page. 3532 // Fall back to allocate a non-special area. 3533 ::munmap(start, end - start); 3534 return NULL; 3535 } 3536 3537 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3538 3539 3540 void* result; 3541 3542 if (start != lp_start) { 3543 result = ::mmap(start, lp_start - start, prot, 3544 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3545 -1, 0); 3546 if (result == MAP_FAILED) { 3547 ::munmap(lp_start, end - lp_start); 3548 return NULL; 3549 } 3550 } 3551 3552 result = ::mmap(lp_start, lp_bytes, prot, 3553 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3554 -1, 0); 3555 if (result == MAP_FAILED) { 3556 warn_on_large_pages_failure(req_addr, bytes, errno); 3557 // If the mmap above fails, the large pages region will be unmapped and we 3558 // have regions before and after with small pages. Release these regions. 3559 // 3560 // | mapped | unmapped | mapped | 3561 // ^ ^ ^ ^ 3562 // start lp_start lp_end end 3563 // 3564 ::munmap(start, lp_start - start); 3565 ::munmap(lp_end, end - lp_end); 3566 return NULL; 3567 } 3568 3569 if (lp_end != end) { 3570 result = ::mmap(lp_end, end - lp_end, prot, 3571 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3572 -1, 0); 3573 if (result == MAP_FAILED) { 3574 ::munmap(start, lp_end - start); 3575 return NULL; 3576 } 3577 } 3578 3579 return start; 3580 } 3581 3582 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3583 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3584 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3585 assert(is_power_of_2(alignment), "Must be"); 3586 assert(is_power_of_2(os::large_page_size()), "Must be"); 3587 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3588 3589 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3590 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3591 } else { 3592 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3593 } 3594 } 3595 3596 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3597 assert(UseLargePages, "only for large pages"); 3598 3599 char* addr; 3600 if (UseSHM) { 3601 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3602 } else { 3603 assert(UseHugeTLBFS, "must be"); 3604 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3605 } 3606 3607 if (addr != NULL) { 3608 if (UseNUMAInterleaving) { 3609 numa_make_global(addr, bytes); 3610 } 3611 3612 // The memory is committed 3613 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC); 3614 } 3615 3616 return addr; 3617 } 3618 3619 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3620 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3621 return shmdt(base) == 0; 3622 } 3623 3624 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3625 return pd_release_memory(base, bytes); 3626 } 3627 3628 bool os::release_memory_special(char* base, size_t bytes) { 3629 assert(UseLargePages, "only for large pages"); 3630 3631 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3632 3633 bool res; 3634 if (UseSHM) { 3635 res = os::Linux::release_memory_special_shm(base, bytes); 3636 } else { 3637 assert(UseHugeTLBFS, "must be"); 3638 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3639 } 3640 3641 if (res) { 3642 tkr.record((address)base, bytes); 3643 } else { 3644 tkr.discard(); 3645 } 3646 3647 return res; 3648 } 3649 3650 size_t os::large_page_size() { 3651 return _large_page_size; 3652 } 3653 3654 // With SysV SHM the entire memory region must be allocated as shared 3655 // memory. 3656 // HugeTLBFS allows application to commit large page memory on demand. 3657 // However, when committing memory with HugeTLBFS fails, the region 3658 // that was supposed to be committed will lose the old reservation 3659 // and allow other threads to steal that memory region. Because of this 3660 // behavior we can't commit HugeTLBFS memory. 3661 bool os::can_commit_large_page_memory() { 3662 return UseTransparentHugePages; 3663 } 3664 3665 bool os::can_execute_large_page_memory() { 3666 return UseTransparentHugePages || UseHugeTLBFS; 3667 } 3668 3669 // Reserve memory at an arbitrary address, only if that area is 3670 // available (and not reserved for something else). 3671 3672 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3673 const int max_tries = 10; 3674 char* base[max_tries]; 3675 size_t size[max_tries]; 3676 const size_t gap = 0x000000; 3677 3678 // Assert only that the size is a multiple of the page size, since 3679 // that's all that mmap requires, and since that's all we really know 3680 // about at this low abstraction level. If we need higher alignment, 3681 // we can either pass an alignment to this method or verify alignment 3682 // in one of the methods further up the call chain. See bug 5044738. 3683 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3684 3685 // Repeatedly allocate blocks until the block is allocated at the 3686 // right spot. Give up after max_tries. Note that reserve_memory() will 3687 // automatically update _highest_vm_reserved_address if the call is 3688 // successful. The variable tracks the highest memory address every reserved 3689 // by JVM. It is used to detect heap-stack collision if running with 3690 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3691 // space than needed, it could confuse the collision detecting code. To 3692 // solve the problem, save current _highest_vm_reserved_address and 3693 // calculate the correct value before return. 3694 address old_highest = _highest_vm_reserved_address; 3695 3696 // Linux mmap allows caller to pass an address as hint; give it a try first, 3697 // if kernel honors the hint then we can return immediately. 3698 char * addr = anon_mmap(requested_addr, bytes, false); 3699 if (addr == requested_addr) { 3700 return requested_addr; 3701 } 3702 3703 if (addr != NULL) { 3704 // mmap() is successful but it fails to reserve at the requested address 3705 anon_munmap(addr, bytes); 3706 } 3707 3708 int i; 3709 for (i = 0; i < max_tries; ++i) { 3710 base[i] = reserve_memory(bytes); 3711 3712 if (base[i] != NULL) { 3713 // Is this the block we wanted? 3714 if (base[i] == requested_addr) { 3715 size[i] = bytes; 3716 break; 3717 } 3718 3719 // Does this overlap the block we wanted? Give back the overlapped 3720 // parts and try again. 3721 3722 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3723 if (top_overlap >= 0 && top_overlap < bytes) { 3724 unmap_memory(base[i], top_overlap); 3725 base[i] += top_overlap; 3726 size[i] = bytes - top_overlap; 3727 } else { 3728 size_t bottom_overlap = base[i] + bytes - requested_addr; 3729 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3730 unmap_memory(requested_addr, bottom_overlap); 3731 size[i] = bytes - bottom_overlap; 3732 } else { 3733 size[i] = bytes; 3734 } 3735 } 3736 } 3737 } 3738 3739 // Give back the unused reserved pieces. 3740 3741 for (int j = 0; j < i; ++j) { 3742 if (base[j] != NULL) { 3743 unmap_memory(base[j], size[j]); 3744 } 3745 } 3746 3747 if (i < max_tries) { 3748 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3749 return requested_addr; 3750 } else { 3751 _highest_vm_reserved_address = old_highest; 3752 return NULL; 3753 } 3754 } 3755 3756 size_t os::read(int fd, void *buf, unsigned int nBytes) { 3757 return ::read(fd, buf, nBytes); 3758 } 3759 3760 // 3761 // Short sleep, direct OS call. 3762 // 3763 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3764 // sched_yield(2) will actually give up the CPU: 3765 // 3766 // * Alone on this pariticular CPU, keeps running. 3767 // * Before the introduction of "skip_buddy" with "compat_yield" disabled 3768 // (pre 2.6.39). 3769 // 3770 // So calling this with 0 is an alternative. 3771 // 3772 void os::naked_short_sleep(jlong ms) { 3773 struct timespec req; 3774 3775 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3776 req.tv_sec = 0; 3777 if (ms > 0) { 3778 req.tv_nsec = (ms % 1000) * 1000000; 3779 } 3780 else { 3781 req.tv_nsec = 1; 3782 } 3783 3784 nanosleep(&req, NULL); 3785 3786 return; 3787 } 3788 3789 // Sleep forever; naked call to OS-specific sleep; use with CAUTION 3790 void os::infinite_sleep() { 3791 while (true) { // sleep forever ... 3792 ::sleep(100); // ... 100 seconds at a time 3793 } 3794 } 3795 3796 // Used to convert frequent JVM_Yield() to nops 3797 bool os::dont_yield() { 3798 return DontYieldALot; 3799 } 3800 3801 void os::yield() { 3802 sched_yield(); 3803 } 3804 3805 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3806 3807 void os::yield_all(int attempts) { 3808 // Yields to all threads, including threads with lower priorities 3809 // Threads on Linux are all with same priority. The Solaris style 3810 // os::yield_all() with nanosleep(1ms) is not necessary. 3811 sched_yield(); 3812 } 3813 3814 // Called from the tight loops to possibly influence time-sharing heuristics 3815 void os::loop_breaker(int attempts) { 3816 os::yield_all(attempts); 3817 } 3818 3819 //////////////////////////////////////////////////////////////////////////////// 3820 // thread priority support 3821 3822 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3823 // only supports dynamic priority, static priority must be zero. For real-time 3824 // applications, Linux supports SCHED_RR which allows static priority (1-99). 3825 // However, for large multi-threaded applications, SCHED_RR is not only slower 3826 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3827 // of 5 runs - Sep 2005). 3828 // 3829 // The following code actually changes the niceness of kernel-thread/LWP. It 3830 // has an assumption that setpriority() only modifies one kernel-thread/LWP, 3831 // not the entire user process, and user level threads are 1:1 mapped to kernel 3832 // threads. It has always been the case, but could change in the future. For 3833 // this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3834 // It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3835 3836 int os::java_to_os_priority[CriticalPriority + 1] = { 3837 19, // 0 Entry should never be used 3838 3839 4, // 1 MinPriority 3840 3, // 2 3841 2, // 3 3842 3843 1, // 4 3844 0, // 5 NormPriority 3845 -1, // 6 3846 3847 -2, // 7 3848 -3, // 8 3849 -4, // 9 NearMaxPriority 3850 3851 -5, // 10 MaxPriority 3852 3853 -5 // 11 CriticalPriority 3854 }; 3855 3856 static int prio_init() { 3857 if (ThreadPriorityPolicy == 1) { 3858 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3859 // if effective uid is not root. Perhaps, a more elegant way of doing 3860 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3861 if (geteuid() != 0) { 3862 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3863 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3864 } 3865 ThreadPriorityPolicy = 0; 3866 } 3867 } 3868 if (UseCriticalJavaThreadPriority) { 3869 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3870 } 3871 return 0; 3872 } 3873 3874 OSReturn os::set_native_priority(Thread* thread, int newpri) { 3875 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3876 3877 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3878 return (ret == 0) ? OS_OK : OS_ERR; 3879 } 3880 3881 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3882 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3883 *priority_ptr = java_to_os_priority[NormPriority]; 3884 return OS_OK; 3885 } 3886 3887 errno = 0; 3888 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3889 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3890 } 3891 3892 // Hint to the underlying OS that a task switch would not be good. 3893 // Void return because it's a hint and can fail. 3894 void os::hint_no_preempt() {} 3895 3896 //////////////////////////////////////////////////////////////////////////////// 3897 // suspend/resume support 3898 3899 // the low-level signal-based suspend/resume support is a remnant from the 3900 // old VM-suspension that used to be for java-suspension, safepoints etc, 3901 // within hotspot. Now there is a single use-case for this: 3902 // - calling get_thread_pc() on the VMThread by the flat-profiler task 3903 // that runs in the watcher thread. 3904 // The remaining code is greatly simplified from the more general suspension 3905 // code that used to be used. 3906 // 3907 // The protocol is quite simple: 3908 // - suspend: 3909 // - sends a signal to the target thread 3910 // - polls the suspend state of the osthread using a yield loop 3911 // - target thread signal handler (SR_handler) sets suspend state 3912 // and blocks in sigsuspend until continued 3913 // - resume: 3914 // - sets target osthread state to continue 3915 // - sends signal to end the sigsuspend loop in the SR_handler 3916 // 3917 // Note that the SR_lock plays no role in this suspend/resume protocol. 3918 // 3919 3920 static void resume_clear_context(OSThread *osthread) { 3921 osthread->set_ucontext(NULL); 3922 osthread->set_siginfo(NULL); 3923 } 3924 3925 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3926 osthread->set_ucontext(context); 3927 osthread->set_siginfo(siginfo); 3928 } 3929 3930 // 3931 // Handler function invoked when a thread's execution is suspended or 3932 // resumed. We have to be careful that only async-safe functions are 3933 // called here (Note: most pthread functions are not async safe and 3934 // should be avoided.) 3935 // 3936 // Note: sigwait() is a more natural fit than sigsuspend() from an 3937 // interface point of view, but sigwait() prevents the signal hander 3938 // from being run. libpthread would get very confused by not having 3939 // its signal handlers run and prevents sigwait()'s use with the 3940 // mutex granting granting signal. 3941 // 3942 // Currently only ever called on the VMThread and JavaThreads (PC sampling) 3943 // 3944 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3945 // Save and restore errno to avoid confusing native code with EINTR 3946 // after sigsuspend. 3947 int old_errno = errno; 3948 3949 Thread* thread = Thread::current(); 3950 OSThread* osthread = thread->osthread(); 3951 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3952 3953 os::SuspendResume::State current = osthread->sr.state(); 3954 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3955 suspend_save_context(osthread, siginfo, context); 3956 3957 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 3958 os::SuspendResume::State state = osthread->sr.suspended(); 3959 if (state == os::SuspendResume::SR_SUSPENDED) { 3960 sigset_t suspend_set; // signals for sigsuspend() 3961 3962 // get current set of blocked signals and unblock resume signal 3963 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3964 sigdelset(&suspend_set, SR_signum); 3965 3966 sr_semaphore.signal(); 3967 // wait here until we are resumed 3968 while (1) { 3969 sigsuspend(&suspend_set); 3970 3971 os::SuspendResume::State result = osthread->sr.running(); 3972 if (result == os::SuspendResume::SR_RUNNING) { 3973 sr_semaphore.signal(); 3974 break; 3975 } 3976 } 3977 3978 } else if (state == os::SuspendResume::SR_RUNNING) { 3979 // request was cancelled, continue 3980 } else { 3981 ShouldNotReachHere(); 3982 } 3983 3984 resume_clear_context(osthread); 3985 } else if (current == os::SuspendResume::SR_RUNNING) { 3986 // request was cancelled, continue 3987 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 3988 // ignore 3989 } else { 3990 // ignore 3991 } 3992 3993 errno = old_errno; 3994 } 3995 3996 3997 static int SR_initialize() { 3998 struct sigaction act; 3999 char *s; 4000 /* Get signal number to use for suspend/resume */ 4001 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4002 int sig = ::strtol(s, 0, 10); 4003 if (sig > 0 || sig < _NSIG) { 4004 SR_signum = sig; 4005 } 4006 } 4007 4008 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4009 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4010 4011 sigemptyset(&SR_sigset); 4012 sigaddset(&SR_sigset, SR_signum); 4013 4014 /* Set up signal handler for suspend/resume */ 4015 act.sa_flags = SA_RESTART|SA_SIGINFO; 4016 act.sa_handler = (void (*)(int)) SR_handler; 4017 4018 // SR_signum is blocked by default. 4019 // 4528190 - We also need to block pthread restart signal (32 on all 4020 // supported Linux platforms). Note that LinuxThreads need to block 4021 // this signal for all threads to work properly. So we don't have 4022 // to use hard-coded signal number when setting up the mask. 4023 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4024 4025 if (sigaction(SR_signum, &act, 0) == -1) { 4026 return -1; 4027 } 4028 4029 // Save signal flag 4030 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4031 return 0; 4032 } 4033 4034 static int sr_notify(OSThread* osthread) { 4035 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4036 assert_status(status == 0, status, "pthread_kill"); 4037 return status; 4038 } 4039 4040 // "Randomly" selected value for how long we want to spin 4041 // before bailing out on suspending a thread, also how often 4042 // we send a signal to a thread we want to resume 4043 static const int RANDOMLY_LARGE_INTEGER = 1000000; 4044 static const int RANDOMLY_LARGE_INTEGER2 = 100; 4045 4046 // returns true on success and false on error - really an error is fatal 4047 // but this seems the normal response to library errors 4048 static bool do_suspend(OSThread* osthread) { 4049 assert(osthread->sr.is_running(), "thread should be running"); 4050 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4051 4052 // mark as suspended and send signal 4053 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4054 // failed to switch, state wasn't running? 4055 ShouldNotReachHere(); 4056 return false; 4057 } 4058 4059 if (sr_notify(osthread) != 0) { 4060 ShouldNotReachHere(); 4061 } 4062 4063 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4064 while (true) { 4065 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4066 break; 4067 } else { 4068 // timeout 4069 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4070 if (cancelled == os::SuspendResume::SR_RUNNING) { 4071 return false; 4072 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4073 // make sure that we consume the signal on the semaphore as well 4074 sr_semaphore.wait(); 4075 break; 4076 } else { 4077 ShouldNotReachHere(); 4078 return false; 4079 } 4080 } 4081 } 4082 4083 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4084 return true; 4085 } 4086 4087 static void do_resume(OSThread* osthread) { 4088 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4089 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4090 4091 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4092 // failed to switch to WAKEUP_REQUEST 4093 ShouldNotReachHere(); 4094 return; 4095 } 4096 4097 while (true) { 4098 if (sr_notify(osthread) == 0) { 4099 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4100 if (osthread->sr.is_running()) { 4101 return; 4102 } 4103 } 4104 } else { 4105 ShouldNotReachHere(); 4106 } 4107 } 4108 4109 guarantee(osthread->sr.is_running(), "Must be running!"); 4110 } 4111 4112 /////////////////////////////////////////////////////////////////////////////////// 4113 // signal handling (except suspend/resume) 4114 4115 // This routine may be used by user applications as a "hook" to catch signals. 4116 // The user-defined signal handler must pass unrecognized signals to this 4117 // routine, and if it returns true (non-zero), then the signal handler must 4118 // return immediately. If the flag "abort_if_unrecognized" is true, then this 4119 // routine will never retun false (zero), but instead will execute a VM panic 4120 // routine kill the process. 4121 // 4122 // If this routine returns false, it is OK to call it again. This allows 4123 // the user-defined signal handler to perform checks either before or after 4124 // the VM performs its own checks. Naturally, the user code would be making 4125 // a serious error if it tried to handle an exception (such as a null check 4126 // or breakpoint) that the VM was generating for its own correct operation. 4127 // 4128 // This routine may recognize any of the following kinds of signals: 4129 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4130 // It should be consulted by handlers for any of those signals. 4131 // 4132 // The caller of this routine must pass in the three arguments supplied 4133 // to the function referred to in the "sa_sigaction" (not the "sa_handler") 4134 // field of the structure passed to sigaction(). This routine assumes that 4135 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4136 // 4137 // Note that the VM will print warnings if it detects conflicting signal 4138 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4139 // 4140 extern "C" JNIEXPORT int 4141 JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 4142 void* ucontext, int abort_if_unrecognized); 4143 4144 void signalHandler(int sig, siginfo_t* info, void* uc) { 4145 assert(info != NULL && uc != NULL, "it must be old kernel"); 4146 int orig_errno = errno; // Preserve errno value over signal handler. 4147 JVM_handle_linux_signal(sig, info, uc, true); 4148 errno = orig_errno; 4149 } 4150 4151 4152 // This boolean allows users to forward their own non-matching signals 4153 // to JVM_handle_linux_signal, harmlessly. 4154 bool os::Linux::signal_handlers_are_installed = false; 4155 4156 // For signal-chaining 4157 struct sigaction os::Linux::sigact[MAXSIGNUM]; 4158 unsigned int os::Linux::sigs = 0; 4159 bool os::Linux::libjsig_is_loaded = false; 4160 typedef struct sigaction *(*get_signal_t)(int); 4161 get_signal_t os::Linux::get_signal_action = NULL; 4162 4163 struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4164 struct sigaction *actp = NULL; 4165 4166 if (libjsig_is_loaded) { 4167 // Retrieve the old signal handler from libjsig 4168 actp = (*get_signal_action)(sig); 4169 } 4170 if (actp == NULL) { 4171 // Retrieve the preinstalled signal handler from jvm 4172 actp = get_preinstalled_handler(sig); 4173 } 4174 4175 return actp; 4176 } 4177 4178 static bool call_chained_handler(struct sigaction *actp, int sig, 4179 siginfo_t *siginfo, void *context) { 4180 // Call the old signal handler 4181 if (actp->sa_handler == SIG_DFL) { 4182 // It's more reasonable to let jvm treat it as an unexpected exception 4183 // instead of taking the default action. 4184 return false; 4185 } else if (actp->sa_handler != SIG_IGN) { 4186 if ((actp->sa_flags & SA_NODEFER) == 0) { 4187 // automaticlly block the signal 4188 sigaddset(&(actp->sa_mask), sig); 4189 } 4190 4191 sa_handler_t hand; 4192 sa_sigaction_t sa; 4193 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4194 // retrieve the chained handler 4195 if (siginfo_flag_set) { 4196 sa = actp->sa_sigaction; 4197 } else { 4198 hand = actp->sa_handler; 4199 } 4200 4201 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4202 actp->sa_handler = SIG_DFL; 4203 } 4204 4205 // try to honor the signal mask 4206 sigset_t oset; 4207 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4208 4209 // call into the chained handler 4210 if (siginfo_flag_set) { 4211 (*sa)(sig, siginfo, context); 4212 } else { 4213 (*hand)(sig); 4214 } 4215 4216 // restore the signal mask 4217 pthread_sigmask(SIG_SETMASK, &oset, 0); 4218 } 4219 // Tell jvm's signal handler the signal is taken care of. 4220 return true; 4221 } 4222 4223 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4224 bool chained = false; 4225 // signal-chaining 4226 if (UseSignalChaining) { 4227 struct sigaction *actp = get_chained_signal_action(sig); 4228 if (actp != NULL) { 4229 chained = call_chained_handler(actp, sig, siginfo, context); 4230 } 4231 } 4232 return chained; 4233 } 4234 4235 struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4236 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 4237 return &sigact[sig]; 4238 } 4239 return NULL; 4240 } 4241 4242 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4243 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4244 sigact[sig] = oldAct; 4245 sigs |= (unsigned int)1 << sig; 4246 } 4247 4248 // for diagnostic 4249 int os::Linux::sigflags[MAXSIGNUM]; 4250 4251 int os::Linux::get_our_sigflags(int sig) { 4252 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4253 return sigflags[sig]; 4254 } 4255 4256 void os::Linux::set_our_sigflags(int sig, int flags) { 4257 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4258 sigflags[sig] = flags; 4259 } 4260 4261 void os::Linux::set_signal_handler(int sig, bool set_installed) { 4262 // Check for overwrite. 4263 struct sigaction oldAct; 4264 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4265 4266 void* oldhand = oldAct.sa_sigaction 4267 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4268 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4269 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4270 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4271 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4272 if (AllowUserSignalHandlers || !set_installed) { 4273 // Do not overwrite; user takes responsibility to forward to us. 4274 return; 4275 } else if (UseSignalChaining) { 4276 // save the old handler in jvm 4277 save_preinstalled_handler(sig, oldAct); 4278 // libjsig also interposes the sigaction() call below and saves the 4279 // old sigaction on it own. 4280 } else { 4281 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4282 "%#lx for signal %d.", (long)oldhand, sig)); 4283 } 4284 } 4285 4286 struct sigaction sigAct; 4287 sigfillset(&(sigAct.sa_mask)); 4288 sigAct.sa_handler = SIG_DFL; 4289 if (!set_installed) { 4290 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4291 } else { 4292 sigAct.sa_sigaction = signalHandler; 4293 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4294 } 4295 // Save flags, which are set by ours 4296 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4297 sigflags[sig] = sigAct.sa_flags; 4298 4299 int ret = sigaction(sig, &sigAct, &oldAct); 4300 assert(ret == 0, "check"); 4301 4302 void* oldhand2 = oldAct.sa_sigaction 4303 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4304 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4305 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4306 } 4307 4308 // install signal handlers for signals that HotSpot needs to 4309 // handle in order to support Java-level exception handling. 4310 4311 void os::Linux::install_signal_handlers() { 4312 if (!signal_handlers_are_installed) { 4313 signal_handlers_are_installed = true; 4314 4315 // signal-chaining 4316 typedef void (*signal_setting_t)(); 4317 signal_setting_t begin_signal_setting = NULL; 4318 signal_setting_t end_signal_setting = NULL; 4319 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4320 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4321 if (begin_signal_setting != NULL) { 4322 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4323 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4324 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4325 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4326 libjsig_is_loaded = true; 4327 assert(UseSignalChaining, "should enable signal-chaining"); 4328 } 4329 if (libjsig_is_loaded) { 4330 // Tell libjsig jvm is setting signal handlers 4331 (*begin_signal_setting)(); 4332 } 4333 4334 set_signal_handler(SIGSEGV, true); 4335 set_signal_handler(SIGPIPE, true); 4336 set_signal_handler(SIGBUS, true); 4337 set_signal_handler(SIGILL, true); 4338 set_signal_handler(SIGFPE, true); 4339 #if defined(PPC64) 4340 set_signal_handler(SIGTRAP, true); 4341 #endif 4342 set_signal_handler(SIGXFSZ, true); 4343 4344 if (libjsig_is_loaded) { 4345 // Tell libjsig jvm finishes setting signal handlers 4346 (*end_signal_setting)(); 4347 } 4348 4349 // We don't activate signal checker if libjsig is in place, we trust ourselves 4350 // and if UserSignalHandler is installed all bets are off. 4351 // Log that signal checking is off only if -verbose:jni is specified. 4352 if (CheckJNICalls) { 4353 if (libjsig_is_loaded) { 4354 if (PrintJNIResolving) { 4355 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4356 } 4357 check_signals = false; 4358 } 4359 if (AllowUserSignalHandlers) { 4360 if (PrintJNIResolving) { 4361 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4362 } 4363 check_signals = false; 4364 } 4365 } 4366 } 4367 } 4368 4369 // This is the fastest way to get thread cpu time on Linux. 4370 // Returns cpu time (user+sys) for any thread, not only for current. 4371 // POSIX compliant clocks are implemented in the kernels 2.6.16+. 4372 // It might work on 2.6.10+ with a special kernel/glibc patch. 4373 // For reference, please, see IEEE Std 1003.1-2004: 4374 // http://www.unix.org/single_unix_specification 4375 4376 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4377 struct timespec tp; 4378 int rc = os::Linux::clock_gettime(clockid, &tp); 4379 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4380 4381 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4382 } 4383 4384 ///// 4385 // glibc on Linux platform uses non-documented flag 4386 // to indicate, that some special sort of signal 4387 // trampoline is used. 4388 // We will never set this flag, and we should 4389 // ignore this flag in our diagnostic 4390 #ifdef SIGNIFICANT_SIGNAL_MASK 4391 #undef SIGNIFICANT_SIGNAL_MASK 4392 #endif 4393 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4394 4395 static const char* get_signal_handler_name(address handler, 4396 char* buf, int buflen) { 4397 int offset; 4398 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4399 if (found) { 4400 // skip directory names 4401 const char *p1, *p2; 4402 p1 = buf; 4403 size_t len = strlen(os::file_separator()); 4404 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4405 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4406 } else { 4407 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4408 } 4409 return buf; 4410 } 4411 4412 static void print_signal_handler(outputStream* st, int sig, 4413 char* buf, size_t buflen) { 4414 struct sigaction sa; 4415 4416 sigaction(sig, NULL, &sa); 4417 4418 // See comment for SIGNIFICANT_SIGNAL_MASK define 4419 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4420 4421 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4422 4423 address handler = (sa.sa_flags & SA_SIGINFO) 4424 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4425 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4426 4427 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4428 st->print("SIG_DFL"); 4429 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4430 st->print("SIG_IGN"); 4431 } else { 4432 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4433 } 4434 4435 st->print(", sa_mask[0]="); 4436 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4437 4438 address rh = VMError::get_resetted_sighandler(sig); 4439 // May be, handler was resetted by VMError? 4440 if(rh != NULL) { 4441 handler = rh; 4442 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4443 } 4444 4445 st->print(", sa_flags="); 4446 os::Posix::print_sa_flags(st, sa.sa_flags); 4447 4448 // Check: is it our handler? 4449 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4450 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4451 // It is our signal handler 4452 // check for flags, reset system-used one! 4453 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4454 st->print( 4455 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4456 os::Linux::get_our_sigflags(sig)); 4457 } 4458 } 4459 st->cr(); 4460 } 4461 4462 4463 #define DO_SIGNAL_CHECK(sig) \ 4464 if (!sigismember(&check_signal_done, sig)) \ 4465 os::Linux::check_signal_handler(sig) 4466 4467 // This method is a periodic task to check for misbehaving JNI applications 4468 // under CheckJNI, we can add any periodic checks here 4469 4470 void os::run_periodic_checks() { 4471 4472 if (check_signals == false) return; 4473 4474 // SEGV and BUS if overridden could potentially prevent 4475 // generation of hs*.log in the event of a crash, debugging 4476 // such a case can be very challenging, so we absolutely 4477 // check the following for a good measure: 4478 DO_SIGNAL_CHECK(SIGSEGV); 4479 DO_SIGNAL_CHECK(SIGILL); 4480 DO_SIGNAL_CHECK(SIGFPE); 4481 DO_SIGNAL_CHECK(SIGBUS); 4482 DO_SIGNAL_CHECK(SIGPIPE); 4483 DO_SIGNAL_CHECK(SIGXFSZ); 4484 #if defined(PPC64) 4485 DO_SIGNAL_CHECK(SIGTRAP); 4486 #endif 4487 4488 // ReduceSignalUsage allows the user to override these handlers 4489 // see comments at the very top and jvm_solaris.h 4490 if (!ReduceSignalUsage) { 4491 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4492 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4493 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4494 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4495 } 4496 4497 DO_SIGNAL_CHECK(SR_signum); 4498 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4499 } 4500 4501 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4502 4503 static os_sigaction_t os_sigaction = NULL; 4504 4505 void os::Linux::check_signal_handler(int sig) { 4506 char buf[O_BUFLEN]; 4507 address jvmHandler = NULL; 4508 4509 4510 struct sigaction act; 4511 if (os_sigaction == NULL) { 4512 // only trust the default sigaction, in case it has been interposed 4513 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4514 if (os_sigaction == NULL) return; 4515 } 4516 4517 os_sigaction(sig, (struct sigaction*)NULL, &act); 4518 4519 4520 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4521 4522 address thisHandler = (act.sa_flags & SA_SIGINFO) 4523 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4524 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 4525 4526 4527 switch(sig) { 4528 case SIGSEGV: 4529 case SIGBUS: 4530 case SIGFPE: 4531 case SIGPIPE: 4532 case SIGILL: 4533 case SIGXFSZ: 4534 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4535 break; 4536 4537 case SHUTDOWN1_SIGNAL: 4538 case SHUTDOWN2_SIGNAL: 4539 case SHUTDOWN3_SIGNAL: 4540 case BREAK_SIGNAL: 4541 jvmHandler = (address)user_handler(); 4542 break; 4543 4544 case INTERRUPT_SIGNAL: 4545 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4546 break; 4547 4548 default: 4549 if (sig == SR_signum) { 4550 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4551 } else { 4552 return; 4553 } 4554 break; 4555 } 4556 4557 if (thisHandler != jvmHandler) { 4558 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4559 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4560 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4561 // No need to check this sig any longer 4562 sigaddset(&check_signal_done, sig); 4563 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4564 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4565 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4566 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4567 // No need to check this sig any longer 4568 sigaddset(&check_signal_done, sig); 4569 } 4570 4571 // Dump all the signal 4572 if (sigismember(&check_signal_done, sig)) { 4573 print_signal_handlers(tty, buf, O_BUFLEN); 4574 } 4575 } 4576 4577 extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 4578 4579 extern bool signal_name(int signo, char* buf, size_t len); 4580 4581 const char* os::exception_name(int exception_code, char* buf, size_t size) { 4582 if (0 < exception_code && exception_code <= SIGRTMAX) { 4583 // signal 4584 if (!signal_name(exception_code, buf, size)) { 4585 jio_snprintf(buf, size, "SIG%d", exception_code); 4586 } 4587 return buf; 4588 } else { 4589 return NULL; 4590 } 4591 } 4592 4593 // this is called _before_ the most of global arguments have been parsed 4594 void os::init(void) { 4595 char dummy; /* used to get a guess on initial stack address */ 4596 // first_hrtime = gethrtime(); 4597 4598 // With LinuxThreads the JavaMain thread pid (primordial thread) 4599 // is different than the pid of the java launcher thread. 4600 // So, on Linux, the launcher thread pid is passed to the VM 4601 // via the sun.java.launcher.pid property. 4602 // Use this property instead of getpid() if it was correctly passed. 4603 // See bug 6351349. 4604 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4605 4606 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4607 4608 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4609 4610 init_random(1234567); 4611 4612 ThreadCritical::initialize(); 4613 4614 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4615 if (Linux::page_size() == -1) { 4616 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4617 strerror(errno))); 4618 } 4619 init_page_sizes((size_t) Linux::page_size()); 4620 4621 Linux::initialize_system_info(); 4622 4623 // main_thread points to the aboriginal thread 4624 Linux::_main_thread = pthread_self(); 4625 4626 Linux::clock_init(); 4627 initial_time_count = javaTimeNanos(); 4628 4629 // pthread_condattr initialization for monotonic clock 4630 int status; 4631 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4632 if ((status = pthread_condattr_init(_condattr)) != 0) { 4633 fatal(err_msg("pthread_condattr_init: %s", strerror(status))); 4634 } 4635 // Only set the clock if CLOCK_MONOTONIC is available 4636 if (os::supports_monotonic_clock()) { 4637 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4638 if (status == EINVAL) { 4639 warning("Unable to use monotonic clock with relative timed-waits" \ 4640 " - changes to the time-of-day clock may have adverse affects"); 4641 } else { 4642 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status))); 4643 } 4644 } 4645 } 4646 // else it defaults to CLOCK_REALTIME 4647 4648 pthread_mutex_init(&dl_mutex, NULL); 4649 4650 // If the pagesize of the VM is greater than 8K determine the appropriate 4651 // number of initial guard pages. The user can change this with the 4652 // command line arguments, if needed. 4653 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4654 StackYellowPages = 1; 4655 StackRedPages = 1; 4656 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4657 } 4658 } 4659 4660 // To install functions for atexit system call 4661 extern "C" { 4662 static void perfMemory_exit_helper() { 4663 perfMemory_exit(); 4664 } 4665 } 4666 4667 // this is called _after_ the global arguments have been parsed 4668 jint os::init_2(void) 4669 { 4670 Linux::fast_thread_clock_init(); 4671 4672 // Allocate a single page and mark it as readable for safepoint polling 4673 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4674 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 4675 4676 os::set_polling_page( polling_page ); 4677 4678 #ifndef PRODUCT 4679 if(Verbose && PrintMiscellaneous) 4680 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 4681 #endif 4682 4683 if (!UseMembar) { 4684 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4685 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4686 os::set_memory_serialize_page( mem_serialize_page ); 4687 4688 #ifndef PRODUCT 4689 if(Verbose && PrintMiscellaneous) 4690 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 4691 #endif 4692 } 4693 4694 // initialize suspend/resume support - must do this before signal_sets_init() 4695 if (SR_initialize() != 0) { 4696 perror("SR_initialize failed"); 4697 return JNI_ERR; 4698 } 4699 4700 Linux::signal_sets_init(); 4701 Linux::install_signal_handlers(); 4702 4703 // Check minimum allowable stack size for thread creation and to initialize 4704 // the java system classes, including StackOverflowError - depends on page 4705 // size. Add a page for compiler2 recursion in main thread. 4706 // Add in 2*BytesPerWord times page size to account for VM stack during 4707 // class initialization depending on 32 or 64 bit VM. 4708 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4709 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4710 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4711 4712 size_t threadStackSizeInBytes = ThreadStackSize * K; 4713 if (threadStackSizeInBytes != 0 && 4714 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4715 tty->print_cr("\nThe stack size specified is too small, " 4716 "Specify at least %dk", 4717 os::Linux::min_stack_allowed/ K); 4718 return JNI_ERR; 4719 } 4720 4721 // Make the stack size a multiple of the page size so that 4722 // the yellow/red zones can be guarded. 4723 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4724 vm_page_size())); 4725 4726 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4727 4728 #if defined(IA32) 4729 workaround_expand_exec_shield_cs_limit(); 4730 #endif 4731 4732 Linux::libpthread_init(); 4733 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4734 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4735 Linux::glibc_version(), Linux::libpthread_version(), 4736 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4737 } 4738 4739 if (UseNUMA) { 4740 if (!Linux::libnuma_init()) { 4741 UseNUMA = false; 4742 } else { 4743 if ((Linux::numa_max_node() < 1)) { 4744 // There's only one node(they start from 0), disable NUMA. 4745 UseNUMA = false; 4746 } 4747 } 4748 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4749 // we can make the adaptive lgrp chunk resizing work. If the user specified 4750 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4751 // disable adaptive resizing. 4752 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4753 if (FLAG_IS_DEFAULT(UseNUMA)) { 4754 UseNUMA = false; 4755 } else { 4756 if (FLAG_IS_DEFAULT(UseLargePages) && 4757 FLAG_IS_DEFAULT(UseSHM) && 4758 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4759 UseLargePages = false; 4760 } else { 4761 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing"); 4762 UseAdaptiveSizePolicy = false; 4763 UseAdaptiveNUMAChunkSizing = false; 4764 } 4765 } 4766 } 4767 if (!UseNUMA && ForceNUMA) { 4768 UseNUMA = true; 4769 } 4770 } 4771 4772 if (MaxFDLimit) { 4773 // set the number of file descriptors to max. print out error 4774 // if getrlimit/setrlimit fails but continue regardless. 4775 struct rlimit nbr_files; 4776 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4777 if (status != 0) { 4778 if (PrintMiscellaneous && (Verbose || WizardMode)) 4779 perror("os::init_2 getrlimit failed"); 4780 } else { 4781 nbr_files.rlim_cur = nbr_files.rlim_max; 4782 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4783 if (status != 0) { 4784 if (PrintMiscellaneous && (Verbose || WizardMode)) 4785 perror("os::init_2 setrlimit failed"); 4786 } 4787 } 4788 } 4789 4790 // Initialize lock used to serialize thread creation (see os::create_thread) 4791 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4792 4793 // at-exit methods are called in the reverse order of their registration. 4794 // atexit functions are called on return from main or as a result of a 4795 // call to exit(3C). There can be only 32 of these functions registered 4796 // and atexit() does not set errno. 4797 4798 if (PerfAllowAtExitRegistration) { 4799 // only register atexit functions if PerfAllowAtExitRegistration is set. 4800 // atexit functions can be delayed until process exit time, which 4801 // can be problematic for embedded VM situations. Embedded VMs should 4802 // call DestroyJavaVM() to assure that VM resources are released. 4803 4804 // note: perfMemory_exit_helper atexit function may be removed in 4805 // the future if the appropriate cleanup code can be added to the 4806 // VM_Exit VMOperation's doit method. 4807 if (atexit(perfMemory_exit_helper) != 0) { 4808 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4809 } 4810 } 4811 4812 // initialize thread priority policy 4813 prio_init(); 4814 4815 return JNI_OK; 4816 } 4817 4818 // this is called at the end of vm_initialization 4819 void os::init_3(void) { 4820 #ifdef JAVASE_EMBEDDED 4821 // Start the MemNotifyThread 4822 if (LowMemoryProtection) { 4823 MemNotifyThread::start(); 4824 } 4825 return; 4826 #endif 4827 } 4828 4829 // Mark the polling page as unreadable 4830 void os::make_polling_page_unreadable(void) { 4831 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4832 fatal("Could not disable polling page"); 4833 }; 4834 4835 // Mark the polling page as readable 4836 void os::make_polling_page_readable(void) { 4837 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4838 fatal("Could not enable polling page"); 4839 } 4840 }; 4841 4842 int os::active_processor_count() { 4843 // Linux doesn't yet have a (official) notion of processor sets, 4844 // so just return the number of online processors. 4845 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4846 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4847 return online_cpus; 4848 } 4849 4850 void os::set_native_thread_name(const char *name) { 4851 // Not yet implemented. 4852 return; 4853 } 4854 4855 bool os::distribute_processes(uint length, uint* distribution) { 4856 // Not yet implemented. 4857 return false; 4858 } 4859 4860 bool os::bind_to_processor(uint processor_id) { 4861 // Not yet implemented. 4862 return false; 4863 } 4864 4865 /// 4866 4867 void os::SuspendedThreadTask::internal_do_task() { 4868 if (do_suspend(_thread->osthread())) { 4869 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4870 do_task(context); 4871 do_resume(_thread->osthread()); 4872 } 4873 } 4874 4875 class PcFetcher : public os::SuspendedThreadTask { 4876 public: 4877 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4878 ExtendedPC result(); 4879 protected: 4880 void do_task(const os::SuspendedThreadTaskContext& context); 4881 private: 4882 ExtendedPC _epc; 4883 }; 4884 4885 ExtendedPC PcFetcher::result() { 4886 guarantee(is_done(), "task is not done yet."); 4887 return _epc; 4888 } 4889 4890 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4891 Thread* thread = context.thread(); 4892 OSThread* osthread = thread->osthread(); 4893 if (osthread->ucontext() != NULL) { 4894 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4895 } else { 4896 // NULL context is unexpected, double-check this is the VMThread 4897 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4898 } 4899 } 4900 4901 // Suspends the target using the signal mechanism and then grabs the PC before 4902 // resuming the target. Used by the flat-profiler only 4903 ExtendedPC os::get_thread_pc(Thread* thread) { 4904 // Make sure that it is called by the watcher for the VMThread 4905 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4906 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4907 4908 PcFetcher fetcher(thread); 4909 fetcher.run(); 4910 return fetcher.result(); 4911 } 4912 4913 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4914 { 4915 if (is_NPTL()) { 4916 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4917 } else { 4918 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4919 // word back to default 64bit precision if condvar is signaled. Java 4920 // wants 53bit precision. Save and restore current value. 4921 int fpu = get_fpu_control_word(); 4922 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4923 set_fpu_control_word(fpu); 4924 return status; 4925 } 4926 } 4927 4928 //////////////////////////////////////////////////////////////////////////////// 4929 // debug support 4930 4931 bool os::find(address addr, outputStream* st) { 4932 Dl_info dlinfo; 4933 memset(&dlinfo, 0, sizeof(dlinfo)); 4934 if (dladdr(addr, &dlinfo) != 0) { 4935 st->print(PTR_FORMAT ": ", addr); 4936 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 4937 st->print("%s+%#x", dlinfo.dli_sname, 4938 addr - (intptr_t)dlinfo.dli_saddr); 4939 } else if (dlinfo.dli_fbase != NULL) { 4940 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4941 } else { 4942 st->print("<absolute address>"); 4943 } 4944 if (dlinfo.dli_fname != NULL) { 4945 st->print(" in %s", dlinfo.dli_fname); 4946 } 4947 if (dlinfo.dli_fbase != NULL) { 4948 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4949 } 4950 st->cr(); 4951 4952 if (Verbose) { 4953 // decode some bytes around the PC 4954 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 4955 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 4956 address lowest = (address) dlinfo.dli_sname; 4957 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4958 if (begin < lowest) begin = lowest; 4959 Dl_info dlinfo2; 4960 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 4961 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 4962 end = (address) dlinfo2.dli_saddr; 4963 Disassembler::decode(begin, end, st); 4964 } 4965 return true; 4966 } 4967 return false; 4968 } 4969 4970 //////////////////////////////////////////////////////////////////////////////// 4971 // misc 4972 4973 // This does not do anything on Linux. This is basically a hook for being 4974 // able to use structured exception handling (thread-local exception filters) 4975 // on, e.g., Win32. 4976 void 4977 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4978 JavaCallArguments* args, Thread* thread) { 4979 f(value, method, args, thread); 4980 } 4981 4982 void os::print_statistics() { 4983 } 4984 4985 int os::message_box(const char* title, const char* message) { 4986 int i; 4987 fdStream err(defaultStream::error_fd()); 4988 for (i = 0; i < 78; i++) err.print_raw("="); 4989 err.cr(); 4990 err.print_raw_cr(title); 4991 for (i = 0; i < 78; i++) err.print_raw("-"); 4992 err.cr(); 4993 err.print_raw_cr(message); 4994 for (i = 0; i < 78; i++) err.print_raw("="); 4995 err.cr(); 4996 4997 char buf[16]; 4998 // Prevent process from exiting upon "read error" without consuming all CPU 4999 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5000 5001 return buf[0] == 'y' || buf[0] == 'Y'; 5002 } 5003 5004 int os::stat(const char *path, struct stat *sbuf) { 5005 char pathbuf[MAX_PATH]; 5006 if (strlen(path) > MAX_PATH - 1) { 5007 errno = ENAMETOOLONG; 5008 return -1; 5009 } 5010 os::native_path(strcpy(pathbuf, path)); 5011 return ::stat(pathbuf, sbuf); 5012 } 5013 5014 bool os::check_heap(bool force) { 5015 return true; 5016 } 5017 5018 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 5019 return ::vsnprintf(buf, count, format, args); 5020 } 5021 5022 // Is a (classpath) directory empty? 5023 bool os::dir_is_empty(const char* path) { 5024 DIR *dir = NULL; 5025 struct dirent *ptr; 5026 5027 dir = opendir(path); 5028 if (dir == NULL) return true; 5029 5030 /* Scan the directory */ 5031 bool result = true; 5032 char buf[sizeof(struct dirent) + MAX_PATH]; 5033 while (result && (ptr = ::readdir(dir)) != NULL) { 5034 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5035 result = false; 5036 } 5037 } 5038 closedir(dir); 5039 return result; 5040 } 5041 5042 // This code originates from JDK's sysOpen and open64_w 5043 // from src/solaris/hpi/src/system_md.c 5044 5045 #ifndef O_DELETE 5046 #define O_DELETE 0x10000 5047 #endif 5048 5049 // Open a file. Unlink the file immediately after open returns 5050 // if the specified oflag has the O_DELETE flag set. 5051 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 5052 5053 int os::open(const char *path, int oflag, int mode) { 5054 5055 if (strlen(path) > MAX_PATH - 1) { 5056 errno = ENAMETOOLONG; 5057 return -1; 5058 } 5059 int fd; 5060 int o_delete = (oflag & O_DELETE); 5061 oflag = oflag & ~O_DELETE; 5062 5063 fd = ::open64(path, oflag, mode); 5064 if (fd == -1) return -1; 5065 5066 //If the open succeeded, the file might still be a directory 5067 { 5068 struct stat64 buf64; 5069 int ret = ::fstat64(fd, &buf64); 5070 int st_mode = buf64.st_mode; 5071 5072 if (ret != -1) { 5073 if ((st_mode & S_IFMT) == S_IFDIR) { 5074 errno = EISDIR; 5075 ::close(fd); 5076 return -1; 5077 } 5078 } else { 5079 ::close(fd); 5080 return -1; 5081 } 5082 } 5083 5084 /* 5085 * All file descriptors that are opened in the JVM and not 5086 * specifically destined for a subprocess should have the 5087 * close-on-exec flag set. If we don't set it, then careless 3rd 5088 * party native code might fork and exec without closing all 5089 * appropriate file descriptors (e.g. as we do in closeDescriptors in 5090 * UNIXProcess.c), and this in turn might: 5091 * 5092 * - cause end-of-file to fail to be detected on some file 5093 * descriptors, resulting in mysterious hangs, or 5094 * 5095 * - might cause an fopen in the subprocess to fail on a system 5096 * suffering from bug 1085341. 5097 * 5098 * (Yes, the default setting of the close-on-exec flag is a Unix 5099 * design flaw) 5100 * 5101 * See: 5102 * 1085341: 32-bit stdio routines should support file descriptors >255 5103 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5104 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5105 */ 5106 #ifdef FD_CLOEXEC 5107 { 5108 int flags = ::fcntl(fd, F_GETFD); 5109 if (flags != -1) 5110 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5111 } 5112 #endif 5113 5114 if (o_delete != 0) { 5115 ::unlink(path); 5116 } 5117 return fd; 5118 } 5119 5120 5121 // create binary file, rewriting existing file if required 5122 int os::create_binary_file(const char* path, bool rewrite_existing) { 5123 int oflags = O_WRONLY | O_CREAT; 5124 if (!rewrite_existing) { 5125 oflags |= O_EXCL; 5126 } 5127 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5128 } 5129 5130 // return current position of file pointer 5131 jlong os::current_file_offset(int fd) { 5132 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5133 } 5134 5135 // move file pointer to the specified offset 5136 jlong os::seek_to_file_offset(int fd, jlong offset) { 5137 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5138 } 5139 5140 // This code originates from JDK's sysAvailable 5141 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5142 5143 int os::available(int fd, jlong *bytes) { 5144 jlong cur, end; 5145 int mode; 5146 struct stat64 buf64; 5147 5148 if (::fstat64(fd, &buf64) >= 0) { 5149 mode = buf64.st_mode; 5150 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5151 /* 5152 * XXX: is the following call interruptible? If so, this might 5153 * need to go through the INTERRUPT_IO() wrapper as for other 5154 * blocking, interruptible calls in this file. 5155 */ 5156 int n; 5157 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5158 *bytes = n; 5159 return 1; 5160 } 5161 } 5162 } 5163 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5164 return 0; 5165 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5166 return 0; 5167 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5168 return 0; 5169 } 5170 *bytes = end - cur; 5171 return 1; 5172 } 5173 5174 int os::socket_available(int fd, jint *pbytes) { 5175 // Linux doc says EINTR not returned, unlike Solaris 5176 int ret = ::ioctl(fd, FIONREAD, pbytes); 5177 5178 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 5179 // is expected to return 0 on failure and 1 on success to the jdk. 5180 return (ret < 0) ? 0 : 1; 5181 } 5182 5183 // Map a block of memory. 5184 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5185 char *addr, size_t bytes, bool read_only, 5186 bool allow_exec) { 5187 int prot; 5188 int flags = MAP_PRIVATE; 5189 5190 if (read_only) { 5191 prot = PROT_READ; 5192 } else { 5193 prot = PROT_READ | PROT_WRITE; 5194 } 5195 5196 if (allow_exec) { 5197 prot |= PROT_EXEC; 5198 } 5199 5200 if (addr != NULL) { 5201 flags |= MAP_FIXED; 5202 } 5203 5204 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5205 fd, file_offset); 5206 if (mapped_address == MAP_FAILED) { 5207 return NULL; 5208 } 5209 return mapped_address; 5210 } 5211 5212 5213 // Remap a block of memory. 5214 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5215 char *addr, size_t bytes, bool read_only, 5216 bool allow_exec) { 5217 // same as map_memory() on this OS 5218 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5219 allow_exec); 5220 } 5221 5222 5223 // Unmap a block of memory. 5224 bool os::pd_unmap_memory(char* addr, size_t bytes) { 5225 return munmap(addr, bytes) == 0; 5226 } 5227 5228 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5229 5230 static clockid_t thread_cpu_clockid(Thread* thread) { 5231 pthread_t tid = thread->osthread()->pthread_id(); 5232 clockid_t clockid; 5233 5234 // Get thread clockid 5235 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5236 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5237 return clockid; 5238 } 5239 5240 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5241 // are used by JVM M&M and JVMTI to get user+sys or user CPU time 5242 // of a thread. 5243 // 5244 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5245 // the fast estimate available on the platform. 5246 5247 jlong os::current_thread_cpu_time() { 5248 if (os::Linux::supports_fast_thread_cpu_time()) { 5249 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5250 } else { 5251 // return user + sys since the cost is the same 5252 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5253 } 5254 } 5255 5256 jlong os::thread_cpu_time(Thread* thread) { 5257 // consistent with what current_thread_cpu_time() returns 5258 if (os::Linux::supports_fast_thread_cpu_time()) { 5259 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5260 } else { 5261 return slow_thread_cpu_time(thread, true /* user + sys */); 5262 } 5263 } 5264 5265 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5266 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5267 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5268 } else { 5269 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5270 } 5271 } 5272 5273 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5274 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5275 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5276 } else { 5277 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5278 } 5279 } 5280 5281 // 5282 // -1 on error. 5283 // 5284 5285 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5286 static bool proc_task_unchecked = true; 5287 pid_t tid = thread->osthread()->thread_id(); 5288 char *s; 5289 char stat[2048]; 5290 int statlen; 5291 char proc_name[64]; 5292 int count; 5293 long sys_time, user_time; 5294 char cdummy; 5295 int idummy; 5296 long ldummy; 5297 FILE *fp; 5298 5299 snprintf(proc_name, 64, "/proc/%d/stat", tid); 5300 5301 // The /proc/<tid>/stat aggregates per-process usage on 5302 // new Linux kernels 2.6+ where NPTL is supported. 5303 // The /proc/self/task/<tid>/stat still has the per-thread usage. 5304 // See bug 6328462. 5305 // There possibly can be cases where there is no directory 5306 // /proc/self/task, so we check its availability. 5307 if (proc_task_unchecked && os::Linux::is_NPTL()) { 5308 // This is executed only once 5309 proc_task_unchecked = false; 5310 fp = fopen("/proc/self/task", "r"); 5311 if (fp != NULL) { 5312 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5313 fclose(fp); 5314 } 5315 } 5316 5317 fp = fopen(proc_name, "r"); 5318 if ( fp == NULL ) return -1; 5319 statlen = fread(stat, 1, 2047, fp); 5320 stat[statlen] = '\0'; 5321 fclose(fp); 5322 5323 // Skip pid and the command string. Note that we could be dealing with 5324 // weird command names, e.g. user could decide to rename java launcher 5325 // to "java 1.4.2 :)", then the stat file would look like 5326 // 1234 (java 1.4.2 :)) R ... ... 5327 // We don't really need to know the command string, just find the last 5328 // occurrence of ")" and then start parsing from there. See bug 4726580. 5329 s = strrchr(stat, ')'); 5330 if (s == NULL ) return -1; 5331 5332 // Skip blank chars 5333 do s++; while (isspace(*s)); 5334 5335 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5336 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5337 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5338 &user_time, &sys_time); 5339 if ( count != 13 ) return -1; 5340 if (user_sys_cpu_time) { 5341 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5342 } else { 5343 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5344 } 5345 } 5346 5347 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5348 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5349 info_ptr->may_skip_backward = false; // elapsed time not wall time 5350 info_ptr->may_skip_forward = false; // elapsed time not wall time 5351 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5352 } 5353 5354 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5355 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5356 info_ptr->may_skip_backward = false; // elapsed time not wall time 5357 info_ptr->may_skip_forward = false; // elapsed time not wall time 5358 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5359 } 5360 5361 bool os::is_thread_cpu_time_supported() { 5362 return true; 5363 } 5364 5365 // System loadavg support. Returns -1 if load average cannot be obtained. 5366 // Linux doesn't yet have a (official) notion of processor sets, 5367 // so just return the system wide load average. 5368 int os::loadavg(double loadavg[], int nelem) { 5369 return ::getloadavg(loadavg, nelem); 5370 } 5371 5372 void os::pause() { 5373 char filename[MAX_PATH]; 5374 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5375 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5376 } else { 5377 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5378 } 5379 5380 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5381 if (fd != -1) { 5382 struct stat buf; 5383 ::close(fd); 5384 while (::stat(filename, &buf) == 0) { 5385 (void)::poll(NULL, 0, 100); 5386 } 5387 } else { 5388 jio_fprintf(stderr, 5389 "Could not open pause file '%s', continuing immediately.\n", filename); 5390 } 5391 } 5392 5393 5394 // Refer to the comments in os_solaris.cpp park-unpark. 5395 // 5396 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5397 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5398 // For specifics regarding the bug see GLIBC BUGID 261237 : 5399 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5400 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5401 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5402 // is used. (The simple C test-case provided in the GLIBC bug report manifests the 5403 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5404 // and monitorenter when we're using 1-0 locking. All those operations may result in 5405 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5406 // of libpthread avoids the problem, but isn't practical. 5407 // 5408 // Possible remedies: 5409 // 5410 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5411 // This is palliative and probabilistic, however. If the thread is preempted 5412 // between the call to compute_abstime() and pthread_cond_timedwait(), more 5413 // than the minimum period may have passed, and the abstime may be stale (in the 5414 // past) resultin in a hang. Using this technique reduces the odds of a hang 5415 // but the JVM is still vulnerable, particularly on heavily loaded systems. 5416 // 5417 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5418 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5419 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5420 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5421 // thread. 5422 // 5423 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5424 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5425 // a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5426 // This also works well. In fact it avoids kernel-level scalability impediments 5427 // on certain platforms that don't handle lots of active pthread_cond_timedwait() 5428 // timers in a graceful fashion. 5429 // 5430 // 4. When the abstime value is in the past it appears that control returns 5431 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5432 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5433 // can avoid the problem by reinitializing the condvar -- by cond_destroy() 5434 // followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5435 // It may be possible to avoid reinitialization by checking the return 5436 // value from pthread_cond_timedwait(). In addition to reinitializing the 5437 // condvar we must establish the invariant that cond_signal() is only called 5438 // within critical sections protected by the adjunct mutex. This prevents 5439 // cond_signal() from "seeing" a condvar that's in the midst of being 5440 // reinitialized or that is corrupt. Sadly, this invariant obviates the 5441 // desirable signal-after-unlock optimization that avoids futile context switching. 5442 // 5443 // I'm also concerned that some versions of NTPL might allocate an auxilliary 5444 // structure when a condvar is used or initialized. cond_destroy() would 5445 // release the helper structure. Our reinitialize-after-timedwait fix 5446 // put excessive stress on malloc/free and locks protecting the c-heap. 5447 // 5448 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5449 // It may be possible to refine (4) by checking the kernel and NTPL verisons 5450 // and only enabling the work-around for vulnerable environments. 5451 5452 // utility to compute the abstime argument to timedwait: 5453 // millis is the relative timeout time 5454 // abstime will be the absolute timeout time 5455 // TODO: replace compute_abstime() with unpackTime() 5456 5457 static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5458 if (millis < 0) millis = 0; 5459 5460 jlong seconds = millis / 1000; 5461 millis %= 1000; 5462 if (seconds > 50000000) { // see man cond_timedwait(3T) 5463 seconds = 50000000; 5464 } 5465 5466 if (os::supports_monotonic_clock()) { 5467 struct timespec now; 5468 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5469 assert_status(status == 0, status, "clock_gettime"); 5470 abstime->tv_sec = now.tv_sec + seconds; 5471 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5472 if (nanos >= NANOSECS_PER_SEC) { 5473 abstime->tv_sec += 1; 5474 nanos -= NANOSECS_PER_SEC; 5475 } 5476 abstime->tv_nsec = nanos; 5477 } else { 5478 struct timeval now; 5479 int status = gettimeofday(&now, NULL); 5480 assert(status == 0, "gettimeofday"); 5481 abstime->tv_sec = now.tv_sec + seconds; 5482 long usec = now.tv_usec + millis * 1000; 5483 if (usec >= 1000000) { 5484 abstime->tv_sec += 1; 5485 usec -= 1000000; 5486 } 5487 abstime->tv_nsec = usec * 1000; 5488 } 5489 return abstime; 5490 } 5491 5492 5493 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 5494 // Conceptually TryPark() should be equivalent to park(0). 5495 5496 int os::PlatformEvent::TryPark() { 5497 for (;;) { 5498 const int v = _Event ; 5499 guarantee ((v == 0) || (v == 1), "invariant") ; 5500 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 5501 } 5502 } 5503 5504 void os::PlatformEvent::park() { // AKA "down()" 5505 // Invariant: Only the thread associated with the Event/PlatformEvent 5506 // may call park(). 5507 // TODO: assert that _Assoc != NULL or _Assoc == Self 5508 int v ; 5509 for (;;) { 5510 v = _Event ; 5511 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5512 } 5513 guarantee (v >= 0, "invariant") ; 5514 if (v == 0) { 5515 // Do this the hard way by blocking ... 5516 int status = pthread_mutex_lock(_mutex); 5517 assert_status(status == 0, status, "mutex_lock"); 5518 guarantee (_nParked == 0, "invariant") ; 5519 ++ _nParked ; 5520 while (_Event < 0) { 5521 status = pthread_cond_wait(_cond, _mutex); 5522 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5523 // Treat this the same as if the wait was interrupted 5524 if (status == ETIME) { status = EINTR; } 5525 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5526 } 5527 -- _nParked ; 5528 5529 _Event = 0 ; 5530 status = pthread_mutex_unlock(_mutex); 5531 assert_status(status == 0, status, "mutex_unlock"); 5532 // Paranoia to ensure our locked and lock-free paths interact 5533 // correctly with each other. 5534 OrderAccess::fence(); 5535 } 5536 guarantee (_Event >= 0, "invariant") ; 5537 } 5538 5539 int os::PlatformEvent::park(jlong millis) { 5540 guarantee (_nParked == 0, "invariant") ; 5541 5542 int v ; 5543 for (;;) { 5544 v = _Event ; 5545 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5546 } 5547 guarantee (v >= 0, "invariant") ; 5548 if (v != 0) return OS_OK ; 5549 5550 // We do this the hard way, by blocking the thread. 5551 // Consider enforcing a minimum timeout value. 5552 struct timespec abst; 5553 compute_abstime(&abst, millis); 5554 5555 int ret = OS_TIMEOUT; 5556 int status = pthread_mutex_lock(_mutex); 5557 assert_status(status == 0, status, "mutex_lock"); 5558 guarantee (_nParked == 0, "invariant") ; 5559 ++_nParked ; 5560 5561 // Object.wait(timo) will return because of 5562 // (a) notification 5563 // (b) timeout 5564 // (c) thread.interrupt 5565 // 5566 // Thread.interrupt and object.notify{All} both call Event::set. 5567 // That is, we treat thread.interrupt as a special case of notification. 5568 // The underlying Solaris implementation, cond_timedwait, admits 5569 // spurious/premature wakeups, but the JLS/JVM spec prevents the 5570 // JVM from making those visible to Java code. As such, we must 5571 // filter out spurious wakeups. We assume all ETIME returns are valid. 5572 // 5573 // TODO: properly differentiate simultaneous notify+interrupt. 5574 // In that case, we should propagate the notify to another waiter. 5575 5576 while (_Event < 0) { 5577 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5578 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5579 pthread_cond_destroy (_cond); 5580 pthread_cond_init (_cond, os::Linux::condAttr()) ; 5581 } 5582 assert_status(status == 0 || status == EINTR || 5583 status == ETIME || status == ETIMEDOUT, 5584 status, "cond_timedwait"); 5585 if (!FilterSpuriousWakeups) break ; // previous semantics 5586 if (status == ETIME || status == ETIMEDOUT) break ; 5587 // We consume and ignore EINTR and spurious wakeups. 5588 } 5589 --_nParked ; 5590 if (_Event >= 0) { 5591 ret = OS_OK; 5592 } 5593 _Event = 0 ; 5594 status = pthread_mutex_unlock(_mutex); 5595 assert_status(status == 0, status, "mutex_unlock"); 5596 assert (_nParked == 0, "invariant") ; 5597 // Paranoia to ensure our locked and lock-free paths interact 5598 // correctly with each other. 5599 OrderAccess::fence(); 5600 return ret; 5601 } 5602 5603 void os::PlatformEvent::unpark() { 5604 // Transitions for _Event: 5605 // 0 :=> 1 5606 // 1 :=> 1 5607 // -1 :=> either 0 or 1; must signal target thread 5608 // That is, we can safely transition _Event from -1 to either 5609 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back 5610 // unpark() calls. 5611 // See also: "Semaphores in Plan 9" by Mullender & Cox 5612 // 5613 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5614 // that it will take two back-to-back park() calls for the owning 5615 // thread to block. This has the benefit of forcing a spurious return 5616 // from the first park() call after an unpark() call which will help 5617 // shake out uses of park() and unpark() without condition variables. 5618 5619 if (Atomic::xchg(1, &_Event) >= 0) return; 5620 5621 // Wait for the thread associated with the event to vacate 5622 int status = pthread_mutex_lock(_mutex); 5623 assert_status(status == 0, status, "mutex_lock"); 5624 int AnyWaiters = _nParked; 5625 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5626 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5627 AnyWaiters = 0; 5628 pthread_cond_signal(_cond); 5629 } 5630 status = pthread_mutex_unlock(_mutex); 5631 assert_status(status == 0, status, "mutex_unlock"); 5632 if (AnyWaiters != 0) { 5633 status = pthread_cond_signal(_cond); 5634 assert_status(status == 0, status, "cond_signal"); 5635 } 5636 5637 // Note that we signal() _after dropping the lock for "immortal" Events. 5638 // This is safe and avoids a common class of futile wakeups. In rare 5639 // circumstances this can cause a thread to return prematurely from 5640 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5641 // simply re-test the condition and re-park itself. 5642 } 5643 5644 5645 // JSR166 5646 // ------------------------------------------------------- 5647 5648 /* 5649 * The solaris and linux implementations of park/unpark are fairly 5650 * conservative for now, but can be improved. They currently use a 5651 * mutex/condvar pair, plus a a count. 5652 * Park decrements count if > 0, else does a condvar wait. Unpark 5653 * sets count to 1 and signals condvar. Only one thread ever waits 5654 * on the condvar. Contention seen when trying to park implies that someone 5655 * is unparking you, so don't wait. And spurious returns are fine, so there 5656 * is no need to track notifications. 5657 */ 5658 5659 #define MAX_SECS 100000000 5660 /* 5661 * This code is common to linux and solaris and will be moved to a 5662 * common place in dolphin. 5663 * 5664 * The passed in time value is either a relative time in nanoseconds 5665 * or an absolute time in milliseconds. Either way it has to be unpacked 5666 * into suitable seconds and nanoseconds components and stored in the 5667 * given timespec structure. 5668 * Given time is a 64-bit value and the time_t used in the timespec is only 5669 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 5670 * overflow if times way in the future are given. Further on Solaris versions 5671 * prior to 10 there is a restriction (see cond_timedwait) that the specified 5672 * number of seconds, in abstime, is less than current_time + 100,000,000. 5673 * As it will be 28 years before "now + 100000000" will overflow we can 5674 * ignore overflow and just impose a hard-limit on seconds using the value 5675 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 5676 * years from "now". 5677 */ 5678 5679 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5680 assert (time > 0, "convertTime"); 5681 time_t max_secs = 0; 5682 5683 if (!os::supports_monotonic_clock() || isAbsolute) { 5684 struct timeval now; 5685 int status = gettimeofday(&now, NULL); 5686 assert(status == 0, "gettimeofday"); 5687 5688 max_secs = now.tv_sec + MAX_SECS; 5689 5690 if (isAbsolute) { 5691 jlong secs = time / 1000; 5692 if (secs > max_secs) { 5693 absTime->tv_sec = max_secs; 5694 } else { 5695 absTime->tv_sec = secs; 5696 } 5697 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5698 } else { 5699 jlong secs = time / NANOSECS_PER_SEC; 5700 if (secs >= MAX_SECS) { 5701 absTime->tv_sec = max_secs; 5702 absTime->tv_nsec = 0; 5703 } else { 5704 absTime->tv_sec = now.tv_sec + secs; 5705 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5706 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5707 absTime->tv_nsec -= NANOSECS_PER_SEC; 5708 ++absTime->tv_sec; // note: this must be <= max_secs 5709 } 5710 } 5711 } 5712 } else { 5713 // must be relative using monotonic clock 5714 struct timespec now; 5715 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5716 assert_status(status == 0, status, "clock_gettime"); 5717 max_secs = now.tv_sec + MAX_SECS; 5718 jlong secs = time / NANOSECS_PER_SEC; 5719 if (secs >= MAX_SECS) { 5720 absTime->tv_sec = max_secs; 5721 absTime->tv_nsec = 0; 5722 } else { 5723 absTime->tv_sec = now.tv_sec + secs; 5724 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5725 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5726 absTime->tv_nsec -= NANOSECS_PER_SEC; 5727 ++absTime->tv_sec; // note: this must be <= max_secs 5728 } 5729 } 5730 } 5731 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5732 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5733 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5734 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5735 } 5736 5737 void Parker::park(bool isAbsolute, jlong time) { 5738 // Ideally we'd do something useful while spinning, such 5739 // as calling unpackTime(). 5740 5741 // Optional fast-path check: 5742 // Return immediately if a permit is available. 5743 // We depend on Atomic::xchg() having full barrier semantics 5744 // since we are doing a lock-free update to _counter. 5745 if (Atomic::xchg(0, &_counter) > 0) return; 5746 5747 Thread* thread = Thread::current(); 5748 assert(thread->is_Java_thread(), "Must be JavaThread"); 5749 JavaThread *jt = (JavaThread *)thread; 5750 5751 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5752 // Check interrupt before trying to wait 5753 if (Thread::is_interrupted(thread, false)) { 5754 return; 5755 } 5756 5757 // Next, demultiplex/decode time arguments 5758 timespec absTime; 5759 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5760 return; 5761 } 5762 if (time > 0) { 5763 unpackTime(&absTime, isAbsolute, time); 5764 } 5765 5766 5767 // Enter safepoint region 5768 // Beware of deadlocks such as 6317397. 5769 // The per-thread Parker:: mutex is a classic leaf-lock. 5770 // In particular a thread must never block on the Threads_lock while 5771 // holding the Parker:: mutex. If safepoints are pending both the 5772 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5773 ThreadBlockInVM tbivm(jt); 5774 5775 // Don't wait if cannot get lock since interference arises from 5776 // unblocking. Also. check interrupt before trying wait 5777 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5778 return; 5779 } 5780 5781 int status ; 5782 if (_counter > 0) { // no wait needed 5783 _counter = 0; 5784 status = pthread_mutex_unlock(_mutex); 5785 assert (status == 0, "invariant") ; 5786 // Paranoia to ensure our locked and lock-free paths interact 5787 // correctly with each other and Java-level accesses. 5788 OrderAccess::fence(); 5789 return; 5790 } 5791 5792 #ifdef ASSERT 5793 // Don't catch signals while blocked; let the running threads have the signals. 5794 // (This allows a debugger to break into the running thread.) 5795 sigset_t oldsigs; 5796 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5797 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5798 #endif 5799 5800 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5801 jt->set_suspend_equivalent(); 5802 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5803 5804 assert(_cur_index == -1, "invariant"); 5805 if (time == 0) { 5806 _cur_index = REL_INDEX; // arbitrary choice when not timed 5807 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ; 5808 } else { 5809 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5810 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ; 5811 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5812 pthread_cond_destroy (&_cond[_cur_index]) ; 5813 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr()); 5814 } 5815 } 5816 _cur_index = -1; 5817 assert_status(status == 0 || status == EINTR || 5818 status == ETIME || status == ETIMEDOUT, 5819 status, "cond_timedwait"); 5820 5821 #ifdef ASSERT 5822 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5823 #endif 5824 5825 _counter = 0 ; 5826 status = pthread_mutex_unlock(_mutex) ; 5827 assert_status(status == 0, status, "invariant") ; 5828 // Paranoia to ensure our locked and lock-free paths interact 5829 // correctly with each other and Java-level accesses. 5830 OrderAccess::fence(); 5831 5832 // If externally suspended while waiting, re-suspend 5833 if (jt->handle_special_suspend_equivalent_condition()) { 5834 jt->java_suspend_self(); 5835 } 5836 } 5837 5838 void Parker::unpark() { 5839 int s, status ; 5840 status = pthread_mutex_lock(_mutex); 5841 assert (status == 0, "invariant") ; 5842 s = _counter; 5843 _counter = 1; 5844 if (s < 1) { 5845 // thread might be parked 5846 if (_cur_index != -1) { 5847 // thread is definitely parked 5848 if (WorkAroundNPTLTimedWaitHang) { 5849 status = pthread_cond_signal (&_cond[_cur_index]); 5850 assert (status == 0, "invariant"); 5851 status = pthread_mutex_unlock(_mutex); 5852 assert (status == 0, "invariant"); 5853 } else { 5854 status = pthread_mutex_unlock(_mutex); 5855 assert (status == 0, "invariant"); 5856 status = pthread_cond_signal (&_cond[_cur_index]); 5857 assert (status == 0, "invariant"); 5858 } 5859 } else { 5860 pthread_mutex_unlock(_mutex); 5861 assert (status == 0, "invariant") ; 5862 } 5863 } else { 5864 pthread_mutex_unlock(_mutex); 5865 assert (status == 0, "invariant") ; 5866 } 5867 } 5868 5869 5870 extern char** environ; 5871 5872 #ifndef __NR_fork 5873 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5874 #endif 5875 5876 #ifndef __NR_execve 5877 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5878 #endif 5879 5880 // Run the specified command in a separate process. Return its exit value, 5881 // or -1 on failure (e.g. can't fork a new process). 5882 // Unlike system(), this function can be called from signal handler. It 5883 // doesn't block SIGINT et al. 5884 int os::fork_and_exec(char* cmd) { 5885 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5886 5887 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5888 // pthread_atfork handlers and reset pthread library. All we need is a 5889 // separate process to execve. Make a direct syscall to fork process. 5890 // On IA64 there's no fork syscall, we have to use fork() and hope for 5891 // the best... 5892 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5893 IA64_ONLY(fork();) 5894 5895 if (pid < 0) { 5896 // fork failed 5897 return -1; 5898 5899 } else if (pid == 0) { 5900 // child process 5901 5902 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5903 // first to kill every thread on the thread list. Because this list is 5904 // not reset by fork() (see notes above), execve() will instead kill 5905 // every thread in the parent process. We know this is the only thread 5906 // in the new process, so make a system call directly. 5907 // IA64 should use normal execve() from glibc to match the glibc fork() 5908 // above. 5909 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5910 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5911 5912 // execve failed 5913 _exit(-1); 5914 5915 } else { 5916 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5917 // care about the actual exit code, for now. 5918 5919 int status; 5920 5921 // Wait for the child process to exit. This returns immediately if 5922 // the child has already exited. */ 5923 while (waitpid(pid, &status, 0) < 0) { 5924 switch (errno) { 5925 case ECHILD: return 0; 5926 case EINTR: break; 5927 default: return -1; 5928 } 5929 } 5930 5931 if (WIFEXITED(status)) { 5932 // The child exited normally; get its exit code. 5933 return WEXITSTATUS(status); 5934 } else if (WIFSIGNALED(status)) { 5935 // The child exited because of a signal 5936 // The best value to return is 0x80 + signal number, 5937 // because that is what all Unix shells do, and because 5938 // it allows callers to distinguish between process exit and 5939 // process death by signal. 5940 return 0x80 + WTERMSIG(status); 5941 } else { 5942 // Unknown exit code; pass it through 5943 return status; 5944 } 5945 } 5946 } 5947 5948 // is_headless_jre() 5949 // 5950 // Test for the existence of xawt/libmawt.so or libawt_xawt.so 5951 // in order to report if we are running in a headless jre 5952 // 5953 // Since JDK8 xawt/libmawt.so was moved into the same directory 5954 // as libawt.so, and renamed libawt_xawt.so 5955 // 5956 bool os::is_headless_jre() { 5957 struct stat statbuf; 5958 char buf[MAXPATHLEN]; 5959 char libmawtpath[MAXPATHLEN]; 5960 const char *xawtstr = "/xawt/libmawt.so"; 5961 const char *new_xawtstr = "/libawt_xawt.so"; 5962 char *p; 5963 5964 // Get path to libjvm.so 5965 os::jvm_path(buf, sizeof(buf)); 5966 5967 // Get rid of libjvm.so 5968 p = strrchr(buf, '/'); 5969 if (p == NULL) return false; 5970 else *p = '\0'; 5971 5972 // Get rid of client or server 5973 p = strrchr(buf, '/'); 5974 if (p == NULL) return false; 5975 else *p = '\0'; 5976 5977 // check xawt/libmawt.so 5978 strcpy(libmawtpath, buf); 5979 strcat(libmawtpath, xawtstr); 5980 if (::stat(libmawtpath, &statbuf) == 0) return false; 5981 5982 // check libawt_xawt.so 5983 strcpy(libmawtpath, buf); 5984 strcat(libmawtpath, new_xawtstr); 5985 if (::stat(libmawtpath, &statbuf) == 0) return false; 5986 5987 return true; 5988 } 5989 5990 // Get the default path to the core file 5991 // Returns the length of the string 5992 int os::get_core_path(char* buffer, size_t bufferSize) { 5993 const char* p = get_current_directory(buffer, bufferSize); 5994 5995 if (p == NULL) { 5996 assert(p != NULL, "failed to get current directory"); 5997 return 0; 5998 } 5999 6000 return strlen(buffer); 6001 } 6002 6003 #ifdef JAVASE_EMBEDDED 6004 // 6005 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 6006 // 6007 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 6008 6009 // ctor 6010 // 6011 MemNotifyThread::MemNotifyThread(int fd): Thread() { 6012 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 6013 _fd = fd; 6014 6015 if (os::create_thread(this, os::os_thread)) { 6016 _memnotify_thread = this; 6017 os::set_priority(this, NearMaxPriority); 6018 os::start_thread(this); 6019 } 6020 } 6021 6022 // Where all the work gets done 6023 // 6024 void MemNotifyThread::run() { 6025 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 6026 6027 // Set up the select arguments 6028 fd_set rfds; 6029 if (_fd != -1) { 6030 FD_ZERO(&rfds); 6031 FD_SET(_fd, &rfds); 6032 } 6033 6034 // Now wait for the mem_notify device to wake up 6035 while (1) { 6036 // Wait for the mem_notify device to signal us.. 6037 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 6038 if (rc == -1) { 6039 perror("select!\n"); 6040 break; 6041 } else if (rc) { 6042 //ssize_t free_before = os::available_memory(); 6043 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 6044 6045 // The kernel is telling us there is not much memory left... 6046 // try to do something about that 6047 6048 // If we are not already in a GC, try one. 6049 if (!Universe::heap()->is_gc_active()) { 6050 Universe::heap()->collect(GCCause::_allocation_failure); 6051 6052 //ssize_t free_after = os::available_memory(); 6053 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 6054 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 6055 } 6056 // We might want to do something like the following if we find the GC's are not helping... 6057 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 6058 } 6059 } 6060 } 6061 6062 // 6063 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 6064 // 6065 void MemNotifyThread::start() { 6066 int fd; 6067 fd = open ("/dev/mem_notify", O_RDONLY, 0); 6068 if (fd < 0) { 6069 return; 6070 } 6071 6072 if (memnotify_thread() == NULL) { 6073 new MemNotifyThread(fd); 6074 } 6075 } 6076 6077 #endif // JAVASE_EMBEDDED 6078 6079 6080 /////////////// Unit tests /////////////// 6081 6082 #ifndef PRODUCT 6083 6084 #define test_log(...) \ 6085 do {\ 6086 if (VerboseInternalVMTests) { \ 6087 tty->print_cr(__VA_ARGS__); \ 6088 tty->flush(); \ 6089 }\ 6090 } while (false) 6091 6092 class TestReserveMemorySpecial : AllStatic { 6093 public: 6094 static void small_page_write(void* addr, size_t size) { 6095 size_t page_size = os::vm_page_size(); 6096 6097 char* end = (char*)addr + size; 6098 for (char* p = (char*)addr; p < end; p += page_size) { 6099 *p = 1; 6100 } 6101 } 6102 6103 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6104 if (!UseHugeTLBFS) { 6105 return; 6106 } 6107 6108 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6109 6110 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6111 6112 if (addr != NULL) { 6113 small_page_write(addr, size); 6114 6115 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6116 } 6117 } 6118 6119 static void test_reserve_memory_special_huge_tlbfs_only() { 6120 if (!UseHugeTLBFS) { 6121 return; 6122 } 6123 6124 size_t lp = os::large_page_size(); 6125 6126 for (size_t size = lp; size <= lp * 10; size += lp) { 6127 test_reserve_memory_special_huge_tlbfs_only(size); 6128 } 6129 } 6130 6131 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) { 6132 if (!UseHugeTLBFS) { 6133 return; 6134 } 6135 6136 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")", 6137 size, alignment); 6138 6139 assert(size >= os::large_page_size(), "Incorrect input to test"); 6140 6141 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6142 6143 if (addr != NULL) { 6144 small_page_write(addr, size); 6145 6146 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6147 } 6148 } 6149 6150 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) { 6151 size_t lp = os::large_page_size(); 6152 size_t ag = os::vm_allocation_granularity(); 6153 6154 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6155 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment); 6156 } 6157 } 6158 6159 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6160 size_t lp = os::large_page_size(); 6161 size_t ag = os::vm_allocation_granularity(); 6162 6163 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp); 6164 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag); 6165 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2); 6166 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2); 6167 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag); 6168 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag); 6169 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2); 6170 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10); 6171 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2); 6172 } 6173 6174 static void test_reserve_memory_special_huge_tlbfs() { 6175 if (!UseHugeTLBFS) { 6176 return; 6177 } 6178 6179 test_reserve_memory_special_huge_tlbfs_only(); 6180 test_reserve_memory_special_huge_tlbfs_mixed(); 6181 } 6182 6183 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6184 if (!UseSHM) { 6185 return; 6186 } 6187 6188 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6189 6190 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6191 6192 if (addr != NULL) { 6193 assert(is_ptr_aligned(addr, alignment), "Check"); 6194 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6195 6196 small_page_write(addr, size); 6197 6198 os::Linux::release_memory_special_shm(addr, size); 6199 } 6200 } 6201 6202 static void test_reserve_memory_special_shm() { 6203 size_t lp = os::large_page_size(); 6204 size_t ag = os::vm_allocation_granularity(); 6205 6206 for (size_t size = ag; size < lp * 3; size += ag) { 6207 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6208 test_reserve_memory_special_shm(size, alignment); 6209 } 6210 } 6211 } 6212 6213 static void test() { 6214 test_reserve_memory_special_huge_tlbfs(); 6215 test_reserve_memory_special_shm(); 6216 } 6217 }; 6218 6219 void TestReserveMemorySpecial_test() { 6220 TestReserveMemorySpecial::test(); 6221 } 6222 6223 #endif