os_linux.cpp revision 6223:09edc8e9fa4d
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#define MAX_SECS 100000000 113 114// for timer info max values which include all bits 115#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 116 117#define LARGEPAGES_BIT (1 << 6) 118//////////////////////////////////////////////////////////////////////////////// 119// global variables 120julong os::Linux::_physical_memory = 0; 121 122address os::Linux::_initial_thread_stack_bottom = NULL; 123uintptr_t os::Linux::_initial_thread_stack_size = 0; 124 125int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 126int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 127Mutex* os::Linux::_createThread_lock = NULL; 128pthread_t os::Linux::_main_thread; 129int os::Linux::_page_size = -1; 130const int os::Linux::_vm_default_page_size = (8 * K); 131bool os::Linux::_is_floating_stack = false; 132bool os::Linux::_is_NPTL = false; 133bool os::Linux::_supports_fast_thread_cpu_time = false; 134const char * os::Linux::_glibc_version = NULL; 135const char * os::Linux::_libpthread_version = NULL; 136pthread_condattr_t os::Linux::_condattr[1]; 137 138static jlong initial_time_count=0; 139 140static int clock_tics_per_sec = 100; 141 142// For diagnostics to print a message once. see run_periodic_checks 143static sigset_t check_signal_done; 144static bool check_signals = true; 145 146static pid_t _initial_pid = 0; 147 148/* Signal number used to suspend/resume a thread */ 149 150/* do not use any signal number less than SIGSEGV, see 4355769 */ 151static int SR_signum = SIGUSR2; 152sigset_t SR_sigset; 153 154/* Used to protect dlsym() calls */ 155static pthread_mutex_t dl_mutex; 156 157// Declarations 158static void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 159 160#ifdef JAVASE_EMBEDDED 161class MemNotifyThread: public Thread { 162 friend class VMStructs; 163 public: 164 virtual void run(); 165 166 private: 167 static MemNotifyThread* _memnotify_thread; 168 int _fd; 169 170 public: 171 172 // Constructor 173 MemNotifyThread(int fd); 174 175 // Tester 176 bool is_memnotify_thread() const { return true; } 177 178 // Printing 179 char* name() const { return (char*)"Linux MemNotify Thread"; } 180 181 // Returns the single instance of the MemNotifyThread 182 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; } 183 184 // Create and start the single instance of MemNotifyThread 185 static void start(); 186}; 187#endif // JAVASE_EMBEDDED 188 189// utility functions 190 191static int SR_initialize(); 192 193julong os::available_memory() { 194 return Linux::available_memory(); 195} 196 197julong os::Linux::available_memory() { 198 // values in struct sysinfo are "unsigned long" 199 struct sysinfo si; 200 sysinfo(&si); 201 202 return (julong)si.freeram * si.mem_unit; 203} 204 205julong os::physical_memory() { 206 return Linux::physical_memory(); 207} 208 209//////////////////////////////////////////////////////////////////////////////// 210// environment support 211 212bool os::getenv(const char* name, char* buf, int len) { 213 const char* val = ::getenv(name); 214 if (val != NULL && strlen(val) < (size_t)len) { 215 strcpy(buf, val); 216 return true; 217 } 218 if (len > 0) buf[0] = 0; // return a null string 219 return false; 220} 221 222 223// Return true if user is running as root. 224 225bool os::have_special_privileges() { 226 static bool init = false; 227 static bool privileges = false; 228 if (!init) { 229 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 230 init = true; 231 } 232 return privileges; 233} 234 235 236#ifndef SYS_gettid 237// i386: 224, ia64: 1105, amd64: 186, sparc 143 238#ifdef __ia64__ 239#define SYS_gettid 1105 240#elif __i386__ 241#define SYS_gettid 224 242#elif __amd64__ 243#define SYS_gettid 186 244#elif __sparc__ 245#define SYS_gettid 143 246#else 247#error define gettid for the arch 248#endif 249#endif 250 251// Cpu architecture string 252#if defined(ZERO) 253static char cpu_arch[] = ZERO_LIBARCH; 254#elif defined(IA64) 255static char cpu_arch[] = "ia64"; 256#elif defined(IA32) 257static char cpu_arch[] = "i386"; 258#elif defined(AMD64) 259static char cpu_arch[] = "amd64"; 260#elif defined(ARM) 261static char cpu_arch[] = "arm"; 262#elif defined(PPC32) 263static char cpu_arch[] = "ppc"; 264#elif defined(PPC64) 265static char cpu_arch[] = "ppc64"; 266#elif defined(SPARC) 267# ifdef _LP64 268static char cpu_arch[] = "sparcv9"; 269# else 270static char cpu_arch[] = "sparc"; 271# endif 272#else 273#error Add appropriate cpu_arch setting 274#endif 275 276 277// pid_t gettid() 278// 279// Returns the kernel thread id of the currently running thread. Kernel 280// thread id is used to access /proc. 281// 282// (Note that getpid() on LinuxThreads returns kernel thread id too; but 283// on NPTL, it returns the same pid for all threads, as required by POSIX.) 284// 285pid_t os::Linux::gettid() { 286 int rslt = syscall(SYS_gettid); 287 if (rslt == -1) { 288 // old kernel, no NPTL support 289 return getpid(); 290 } else { 291 return (pid_t)rslt; 292 } 293} 294 295// Most versions of linux have a bug where the number of processors are 296// determined by looking at the /proc file system. In a chroot environment, 297// the system call returns 1. This causes the VM to act as if it is 298// a single processor and elide locking (see is_MP() call). 299static bool unsafe_chroot_detected = false; 300static const char *unstable_chroot_error = "/proc file system not found.\n" 301 "Java may be unstable running multithreaded in a chroot " 302 "environment on Linux when /proc filesystem is not mounted."; 303 304void os::Linux::initialize_system_info() { 305 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 306 if (processor_count() == 1) { 307 pid_t pid = os::Linux::gettid(); 308 char fname[32]; 309 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 310 FILE *fp = fopen(fname, "r"); 311 if (fp == NULL) { 312 unsafe_chroot_detected = true; 313 } else { 314 fclose(fp); 315 } 316 } 317 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 318 assert(processor_count() > 0, "linux error"); 319} 320 321void os::init_system_properties_values() { 322// char arch[12]; 323// sysinfo(SI_ARCHITECTURE, arch, sizeof(arch)); 324 325 // The next steps are taken in the product version: 326 // 327 // Obtain the JAVA_HOME value from the location of libjvm.so. 328 // This library should be located at: 329 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 330 // 331 // If "/jre/lib/" appears at the right place in the path, then we 332 // assume libjvm.so is installed in a JDK and we use this path. 333 // 334 // Otherwise exit with message: "Could not create the Java virtual machine." 335 // 336 // The following extra steps are taken in the debugging version: 337 // 338 // If "/jre/lib/" does NOT appear at the right place in the path 339 // instead of exit check for $JAVA_HOME environment variable. 340 // 341 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 342 // then we append a fake suffix "hotspot/libjvm.so" to this path so 343 // it looks like libjvm.so is installed there 344 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 345 // 346 // Otherwise exit. 347 // 348 // Important note: if the location of libjvm.so changes this 349 // code needs to be changed accordingly. 350 351 // The next few definitions allow the code to be verbatim: 352#define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal) 353#define getenv(n) ::getenv(n) 354 355/* 356 * See ld(1): 357 * The linker uses the following search paths to locate required 358 * shared libraries: 359 * 1: ... 360 * ... 361 * 7: The default directories, normally /lib and /usr/lib. 362 */ 363#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 364#define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 365#else 366#define DEFAULT_LIBPATH "/lib:/usr/lib" 367#endif 368 369#define EXTENSIONS_DIR "/lib/ext" 370#define ENDORSED_DIR "/lib/endorsed" 371#define REG_DIR "/usr/java/packages" 372 373 { 374 /* sysclasspath, java_home, dll_dir */ 375 { 376 char *home_path; 377 char *dll_path; 378 char *pslash; 379 char buf[MAXPATHLEN]; 380 os::jvm_path(buf, sizeof(buf)); 381 382 // Found the full path to libjvm.so. 383 // Now cut the path to <java_home>/jre if we can. 384 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */ 385 pslash = strrchr(buf, '/'); 386 if (pslash != NULL) 387 *pslash = '\0'; /* get rid of /{client|server|hotspot} */ 388 dll_path = malloc(strlen(buf) + 1); 389 if (dll_path == NULL) 390 return; 391 strcpy(dll_path, buf); 392 Arguments::set_dll_dir(dll_path); 393 394 if (pslash != NULL) { 395 pslash = strrchr(buf, '/'); 396 if (pslash != NULL) { 397 *pslash = '\0'; /* get rid of /<arch> */ 398 pslash = strrchr(buf, '/'); 399 if (pslash != NULL) 400 *pslash = '\0'; /* get rid of /lib */ 401 } 402 } 403 404 home_path = malloc(strlen(buf) + 1); 405 if (home_path == NULL) 406 return; 407 strcpy(home_path, buf); 408 Arguments::set_java_home(home_path); 409 410 if (!set_boot_path('/', ':')) 411 return; 412 } 413 414 /* 415 * Where to look for native libraries 416 * 417 * Note: Due to a legacy implementation, most of the library path 418 * is set in the launcher. This was to accomodate linking restrictions 419 * on legacy Linux implementations (which are no longer supported). 420 * Eventually, all the library path setting will be done here. 421 * 422 * However, to prevent the proliferation of improperly built native 423 * libraries, the new path component /usr/java/packages is added here. 424 * Eventually, all the library path setting will be done here. 425 */ 426 { 427 char *ld_library_path; 428 429 /* 430 * Construct the invariant part of ld_library_path. Note that the 431 * space for the colon and the trailing null are provided by the 432 * nulls included by the sizeof operator (so actually we allocate 433 * a byte more than necessary). 434 */ 435 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") + 436 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH)); 437 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch); 438 439 /* 440 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It 441 * should always exist (until the legacy problem cited above is 442 * addressed). 443 */ 444 char *v = getenv("LD_LIBRARY_PATH"); 445 if (v != NULL) { 446 char *t = ld_library_path; 447 /* That's +1 for the colon and +1 for the trailing '\0' */ 448 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1); 449 sprintf(ld_library_path, "%s:%s", v, t); 450 } 451 Arguments::set_library_path(ld_library_path); 452 } 453 454 /* 455 * Extensions directories. 456 * 457 * Note that the space for the colon and the trailing null are provided 458 * by the nulls included by the sizeof operator (so actually one byte more 459 * than necessary is allocated). 460 */ 461 { 462 char *buf = malloc(strlen(Arguments::get_java_home()) + 463 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR)); 464 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, 465 Arguments::get_java_home()); 466 Arguments::set_ext_dirs(buf); 467 } 468 469 /* Endorsed standards default directory. */ 470 { 471 char * buf; 472 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR)); 473 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 474 Arguments::set_endorsed_dirs(buf); 475 } 476 } 477 478#undef malloc 479#undef getenv 480#undef EXTENSIONS_DIR 481#undef ENDORSED_DIR 482 483 // Done 484 return; 485} 486 487//////////////////////////////////////////////////////////////////////////////// 488// breakpoint support 489 490void os::breakpoint() { 491 BREAKPOINT; 492} 493 494extern "C" void breakpoint() { 495 // use debugger to set breakpoint here 496} 497 498//////////////////////////////////////////////////////////////////////////////// 499// signal support 500 501debug_only(static bool signal_sets_initialized = false); 502static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 503 504bool os::Linux::is_sig_ignored(int sig) { 505 struct sigaction oact; 506 sigaction(sig, (struct sigaction*)NULL, &oact); 507 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 508 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 509 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 510 return true; 511 else 512 return false; 513} 514 515void os::Linux::signal_sets_init() { 516 // Should also have an assertion stating we are still single-threaded. 517 assert(!signal_sets_initialized, "Already initialized"); 518 // Fill in signals that are necessarily unblocked for all threads in 519 // the VM. Currently, we unblock the following signals: 520 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 521 // by -Xrs (=ReduceSignalUsage)); 522 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 523 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 524 // the dispositions or masks wrt these signals. 525 // Programs embedding the VM that want to use the above signals for their 526 // own purposes must, at this time, use the "-Xrs" option to prevent 527 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 528 // (See bug 4345157, and other related bugs). 529 // In reality, though, unblocking these signals is really a nop, since 530 // these signals are not blocked by default. 531 sigemptyset(&unblocked_sigs); 532 sigemptyset(&allowdebug_blocked_sigs); 533 sigaddset(&unblocked_sigs, SIGILL); 534 sigaddset(&unblocked_sigs, SIGSEGV); 535 sigaddset(&unblocked_sigs, SIGBUS); 536 sigaddset(&unblocked_sigs, SIGFPE); 537#if defined(PPC64) 538 sigaddset(&unblocked_sigs, SIGTRAP); 539#endif 540 sigaddset(&unblocked_sigs, SR_signum); 541 542 if (!ReduceSignalUsage) { 543 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 544 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 545 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 546 } 547 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 548 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 549 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 550 } 551 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 552 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 553 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 554 } 555 } 556 // Fill in signals that are blocked by all but the VM thread. 557 sigemptyset(&vm_sigs); 558 if (!ReduceSignalUsage) 559 sigaddset(&vm_sigs, BREAK_SIGNAL); 560 debug_only(signal_sets_initialized = true); 561 562} 563 564// These are signals that are unblocked while a thread is running Java. 565// (For some reason, they get blocked by default.) 566sigset_t* os::Linux::unblocked_signals() { 567 assert(signal_sets_initialized, "Not initialized"); 568 return &unblocked_sigs; 569} 570 571// These are the signals that are blocked while a (non-VM) thread is 572// running Java. Only the VM thread handles these signals. 573sigset_t* os::Linux::vm_signals() { 574 assert(signal_sets_initialized, "Not initialized"); 575 return &vm_sigs; 576} 577 578// These are signals that are blocked during cond_wait to allow debugger in 579sigset_t* os::Linux::allowdebug_blocked_signals() { 580 assert(signal_sets_initialized, "Not initialized"); 581 return &allowdebug_blocked_sigs; 582} 583 584void os::Linux::hotspot_sigmask(Thread* thread) { 585 586 //Save caller's signal mask before setting VM signal mask 587 sigset_t caller_sigmask; 588 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 589 590 OSThread* osthread = thread->osthread(); 591 osthread->set_caller_sigmask(caller_sigmask); 592 593 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 594 595 if (!ReduceSignalUsage) { 596 if (thread->is_VM_thread()) { 597 // Only the VM thread handles BREAK_SIGNAL ... 598 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 599 } else { 600 // ... all other threads block BREAK_SIGNAL 601 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 602 } 603 } 604} 605 606////////////////////////////////////////////////////////////////////////////// 607// detecting pthread library 608 609void os::Linux::libpthread_init() { 610 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 611 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 612 // generic name for earlier versions. 613 // Define macros here so we can build HotSpot on old systems. 614# ifndef _CS_GNU_LIBC_VERSION 615# define _CS_GNU_LIBC_VERSION 2 616# endif 617# ifndef _CS_GNU_LIBPTHREAD_VERSION 618# define _CS_GNU_LIBPTHREAD_VERSION 3 619# endif 620 621 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 622 if (n > 0) { 623 char *str = (char *)malloc(n, mtInternal); 624 confstr(_CS_GNU_LIBC_VERSION, str, n); 625 os::Linux::set_glibc_version(str); 626 } else { 627 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 628 static char _gnu_libc_version[32]; 629 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 630 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 631 os::Linux::set_glibc_version(_gnu_libc_version); 632 } 633 634 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 635 if (n > 0) { 636 char *str = (char *)malloc(n, mtInternal); 637 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 638 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 639 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 640 // is the case. LinuxThreads has a hard limit on max number of threads. 641 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 642 // On the other hand, NPTL does not have such a limit, sysconf() 643 // will return -1 and errno is not changed. Check if it is really NPTL. 644 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 645 strstr(str, "NPTL") && 646 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 647 free(str); 648 os::Linux::set_libpthread_version("linuxthreads"); 649 } else { 650 os::Linux::set_libpthread_version(str); 651 } 652 } else { 653 // glibc before 2.3.2 only has LinuxThreads. 654 os::Linux::set_libpthread_version("linuxthreads"); 655 } 656 657 if (strstr(libpthread_version(), "NPTL")) { 658 os::Linux::set_is_NPTL(); 659 } else { 660 os::Linux::set_is_LinuxThreads(); 661 } 662 663 // LinuxThreads have two flavors: floating-stack mode, which allows variable 664 // stack size; and fixed-stack mode. NPTL is always floating-stack. 665 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 666 os::Linux::set_is_floating_stack(); 667 } 668} 669 670///////////////////////////////////////////////////////////////////////////// 671// thread stack 672 673// Force Linux kernel to expand current thread stack. If "bottom" is close 674// to the stack guard, caller should block all signals. 675// 676// MAP_GROWSDOWN: 677// A special mmap() flag that is used to implement thread stacks. It tells 678// kernel that the memory region should extend downwards when needed. This 679// allows early versions of LinuxThreads to only mmap the first few pages 680// when creating a new thread. Linux kernel will automatically expand thread 681// stack as needed (on page faults). 682// 683// However, because the memory region of a MAP_GROWSDOWN stack can grow on 684// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 685// region, it's hard to tell if the fault is due to a legitimate stack 686// access or because of reading/writing non-exist memory (e.g. buffer 687// overrun). As a rule, if the fault happens below current stack pointer, 688// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 689// application (see Linux kernel fault.c). 690// 691// This Linux feature can cause SIGSEGV when VM bangs thread stack for 692// stack overflow detection. 693// 694// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 695// not use this flag. However, the stack of initial thread is not created 696// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 697// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 698// and then attach the thread to JVM. 699// 700// To get around the problem and allow stack banging on Linux, we need to 701// manually expand thread stack after receiving the SIGSEGV. 702// 703// There are two ways to expand thread stack to address "bottom", we used 704// both of them in JVM before 1.5: 705// 1. adjust stack pointer first so that it is below "bottom", and then 706// touch "bottom" 707// 2. mmap() the page in question 708// 709// Now alternate signal stack is gone, it's harder to use 2. For instance, 710// if current sp is already near the lower end of page 101, and we need to 711// call mmap() to map page 100, it is possible that part of the mmap() frame 712// will be placed in page 100. When page 100 is mapped, it is zero-filled. 713// That will destroy the mmap() frame and cause VM to crash. 714// 715// The following code works by adjusting sp first, then accessing the "bottom" 716// page to force a page fault. Linux kernel will then automatically expand the 717// stack mapping. 718// 719// _expand_stack_to() assumes its frame size is less than page size, which 720// should always be true if the function is not inlined. 721 722#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 723#define NOINLINE 724#else 725#define NOINLINE __attribute__ ((noinline)) 726#endif 727 728static void _expand_stack_to(address bottom) NOINLINE; 729 730static void _expand_stack_to(address bottom) { 731 address sp; 732 size_t size; 733 volatile char *p; 734 735 // Adjust bottom to point to the largest address within the same page, it 736 // gives us a one-page buffer if alloca() allocates slightly more memory. 737 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 738 bottom += os::Linux::page_size() - 1; 739 740 // sp might be slightly above current stack pointer; if that's the case, we 741 // will alloca() a little more space than necessary, which is OK. Don't use 742 // os::current_stack_pointer(), as its result can be slightly below current 743 // stack pointer, causing us to not alloca enough to reach "bottom". 744 sp = (address)&sp; 745 746 if (sp > bottom) { 747 size = sp - bottom; 748 p = (volatile char *)alloca(size); 749 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 750 p[0] = '\0'; 751 } 752} 753 754bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 755 assert(t!=NULL, "just checking"); 756 assert(t->osthread()->expanding_stack(), "expand should be set"); 757 assert(t->stack_base() != NULL, "stack_base was not initialized"); 758 759 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 760 sigset_t mask_all, old_sigset; 761 sigfillset(&mask_all); 762 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 763 _expand_stack_to(addr); 764 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 765 return true; 766 } 767 return false; 768} 769 770////////////////////////////////////////////////////////////////////////////// 771// create new thread 772 773static address highest_vm_reserved_address(); 774 775// check if it's safe to start a new thread 776static bool _thread_safety_check(Thread* thread) { 777 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 778 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 779 // Heap is mmap'ed at lower end of memory space. Thread stacks are 780 // allocated (MAP_FIXED) from high address space. Every thread stack 781 // occupies a fixed size slot (usually 2Mbytes, but user can change 782 // it to other values if they rebuild LinuxThreads). 783 // 784 // Problem with MAP_FIXED is that mmap() can still succeed even part of 785 // the memory region has already been mmap'ed. That means if we have too 786 // many threads and/or very large heap, eventually thread stack will 787 // collide with heap. 788 // 789 // Here we try to prevent heap/stack collision by comparing current 790 // stack bottom with the highest address that has been mmap'ed by JVM 791 // plus a safety margin for memory maps created by native code. 792 // 793 // This feature can be disabled by setting ThreadSafetyMargin to 0 794 // 795 if (ThreadSafetyMargin > 0) { 796 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 797 798 // not safe if our stack extends below the safety margin 799 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 800 } else { 801 return true; 802 } 803 } else { 804 // Floating stack LinuxThreads or NPTL: 805 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 806 // there's not enough space left, pthread_create() will fail. If we come 807 // here, that means enough space has been reserved for stack. 808 return true; 809 } 810} 811 812// Thread start routine for all newly created threads 813static void *java_start(Thread *thread) { 814 // Try to randomize the cache line index of hot stack frames. 815 // This helps when threads of the same stack traces evict each other's 816 // cache lines. The threads can be either from the same JVM instance, or 817 // from different JVM instances. The benefit is especially true for 818 // processors with hyperthreading technology. 819 static int counter = 0; 820 int pid = os::current_process_id(); 821 alloca(((pid ^ counter++) & 7) * 128); 822 823 ThreadLocalStorage::set_thread(thread); 824 825 OSThread* osthread = thread->osthread(); 826 Monitor* sync = osthread->startThread_lock(); 827 828 // non floating stack LinuxThreads needs extra check, see above 829 if (!_thread_safety_check(thread)) { 830 // notify parent thread 831 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 832 osthread->set_state(ZOMBIE); 833 sync->notify_all(); 834 return NULL; 835 } 836 837 // thread_id is kernel thread id (similar to Solaris LWP id) 838 osthread->set_thread_id(os::Linux::gettid()); 839 840 if (UseNUMA) { 841 int lgrp_id = os::numa_get_group_id(); 842 if (lgrp_id != -1) { 843 thread->set_lgrp_id(lgrp_id); 844 } 845 } 846 // initialize signal mask for this thread 847 os::Linux::hotspot_sigmask(thread); 848 849 // initialize floating point control register 850 os::Linux::init_thread_fpu_state(); 851 852 // handshaking with parent thread 853 { 854 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 855 856 // notify parent thread 857 osthread->set_state(INITIALIZED); 858 sync->notify_all(); 859 860 // wait until os::start_thread() 861 while (osthread->get_state() == INITIALIZED) { 862 sync->wait(Mutex::_no_safepoint_check_flag); 863 } 864 } 865 866 // call one more level start routine 867 thread->run(); 868 869 return 0; 870} 871 872bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 873 assert(thread->osthread() == NULL, "caller responsible"); 874 875 // Allocate the OSThread object 876 OSThread* osthread = new OSThread(NULL, NULL); 877 if (osthread == NULL) { 878 return false; 879 } 880 881 // set the correct thread state 882 osthread->set_thread_type(thr_type); 883 884 // Initial state is ALLOCATED but not INITIALIZED 885 osthread->set_state(ALLOCATED); 886 887 thread->set_osthread(osthread); 888 889 // init thread attributes 890 pthread_attr_t attr; 891 pthread_attr_init(&attr); 892 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 893 894 // stack size 895 if (os::Linux::supports_variable_stack_size()) { 896 // calculate stack size if it's not specified by caller 897 if (stack_size == 0) { 898 stack_size = os::Linux::default_stack_size(thr_type); 899 900 switch (thr_type) { 901 case os::java_thread: 902 // Java threads use ThreadStackSize which default value can be 903 // changed with the flag -Xss 904 assert (JavaThread::stack_size_at_create() > 0, "this should be set"); 905 stack_size = JavaThread::stack_size_at_create(); 906 break; 907 case os::compiler_thread: 908 if (CompilerThreadStackSize > 0) { 909 stack_size = (size_t)(CompilerThreadStackSize * K); 910 break; 911 } // else fall through: 912 // use VMThreadStackSize if CompilerThreadStackSize is not defined 913 case os::vm_thread: 914 case os::pgc_thread: 915 case os::cgc_thread: 916 case os::watcher_thread: 917 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 918 break; 919 } 920 } 921 922 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 923 pthread_attr_setstacksize(&attr, stack_size); 924 } else { 925 // let pthread_create() pick the default value. 926 } 927 928 // glibc guard page 929 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 930 931 ThreadState state; 932 933 { 934 // Serialize thread creation if we are running with fixed stack LinuxThreads 935 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 936 if (lock) { 937 os::Linux::createThread_lock()->lock_without_safepoint_check(); 938 } 939 940 pthread_t tid; 941 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 942 943 pthread_attr_destroy(&attr); 944 945 if (ret != 0) { 946 if (PrintMiscellaneous && (Verbose || WizardMode)) { 947 perror("pthread_create()"); 948 } 949 // Need to clean up stuff we've allocated so far 950 thread->set_osthread(NULL); 951 delete osthread; 952 if (lock) os::Linux::createThread_lock()->unlock(); 953 return false; 954 } 955 956 // Store pthread info into the OSThread 957 osthread->set_pthread_id(tid); 958 959 // Wait until child thread is either initialized or aborted 960 { 961 Monitor* sync_with_child = osthread->startThread_lock(); 962 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 963 while ((state = osthread->get_state()) == ALLOCATED) { 964 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 965 } 966 } 967 968 if (lock) { 969 os::Linux::createThread_lock()->unlock(); 970 } 971 } 972 973 // Aborted due to thread limit being reached 974 if (state == ZOMBIE) { 975 thread->set_osthread(NULL); 976 delete osthread; 977 return false; 978 } 979 980 // The thread is returned suspended (in state INITIALIZED), 981 // and is started higher up in the call chain 982 assert(state == INITIALIZED, "race condition"); 983 return true; 984} 985 986///////////////////////////////////////////////////////////////////////////// 987// attach existing thread 988 989// bootstrap the main thread 990bool os::create_main_thread(JavaThread* thread) { 991 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 992 return create_attached_thread(thread); 993} 994 995bool os::create_attached_thread(JavaThread* thread) { 996#ifdef ASSERT 997 thread->verify_not_published(); 998#endif 999 1000 // Allocate the OSThread object 1001 OSThread* osthread = new OSThread(NULL, NULL); 1002 1003 if (osthread == NULL) { 1004 return false; 1005 } 1006 1007 // Store pthread info into the OSThread 1008 osthread->set_thread_id(os::Linux::gettid()); 1009 osthread->set_pthread_id(::pthread_self()); 1010 1011 // initialize floating point control register 1012 os::Linux::init_thread_fpu_state(); 1013 1014 // Initial thread state is RUNNABLE 1015 osthread->set_state(RUNNABLE); 1016 1017 thread->set_osthread(osthread); 1018 1019 if (UseNUMA) { 1020 int lgrp_id = os::numa_get_group_id(); 1021 if (lgrp_id != -1) { 1022 thread->set_lgrp_id(lgrp_id); 1023 } 1024 } 1025 1026 if (os::Linux::is_initial_thread()) { 1027 // If current thread is initial thread, its stack is mapped on demand, 1028 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 1029 // the entire stack region to avoid SEGV in stack banging. 1030 // It is also useful to get around the heap-stack-gap problem on SuSE 1031 // kernel (see 4821821 for details). We first expand stack to the top 1032 // of yellow zone, then enable stack yellow zone (order is significant, 1033 // enabling yellow zone first will crash JVM on SuSE Linux), so there 1034 // is no gap between the last two virtual memory regions. 1035 1036 JavaThread *jt = (JavaThread *)thread; 1037 address addr = jt->stack_yellow_zone_base(); 1038 assert(addr != NULL, "initialization problem?"); 1039 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 1040 1041 osthread->set_expanding_stack(); 1042 os::Linux::manually_expand_stack(jt, addr); 1043 osthread->clear_expanding_stack(); 1044 } 1045 1046 // initialize signal mask for this thread 1047 // and save the caller's signal mask 1048 os::Linux::hotspot_sigmask(thread); 1049 1050 return true; 1051} 1052 1053void os::pd_start_thread(Thread* thread) { 1054 OSThread * osthread = thread->osthread(); 1055 assert(osthread->get_state() != INITIALIZED, "just checking"); 1056 Monitor* sync_with_child = osthread->startThread_lock(); 1057 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1058 sync_with_child->notify(); 1059} 1060 1061// Free Linux resources related to the OSThread 1062void os::free_thread(OSThread* osthread) { 1063 assert(osthread != NULL, "osthread not set"); 1064 1065 if (Thread::current()->osthread() == osthread) { 1066 // Restore caller's signal mask 1067 sigset_t sigmask = osthread->caller_sigmask(); 1068 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1069 } 1070 1071 delete osthread; 1072} 1073 1074////////////////////////////////////////////////////////////////////////////// 1075// thread local storage 1076 1077int os::allocate_thread_local_storage() { 1078 pthread_key_t key; 1079 int rslt = pthread_key_create(&key, NULL); 1080 assert(rslt == 0, "cannot allocate thread local storage"); 1081 return (int)key; 1082} 1083 1084// Note: This is currently not used by VM, as we don't destroy TLS key 1085// on VM exit. 1086void os::free_thread_local_storage(int index) { 1087 int rslt = pthread_key_delete((pthread_key_t)index); 1088 assert(rslt == 0, "invalid index"); 1089} 1090 1091void os::thread_local_storage_at_put(int index, void* value) { 1092 int rslt = pthread_setspecific((pthread_key_t)index, value); 1093 assert(rslt == 0, "pthread_setspecific failed"); 1094} 1095 1096extern "C" Thread* get_thread() { 1097 return ThreadLocalStorage::thread(); 1098} 1099 1100////////////////////////////////////////////////////////////////////////////// 1101// initial thread 1102 1103// Check if current thread is the initial thread, similar to Solaris thr_main. 1104bool os::Linux::is_initial_thread(void) { 1105 char dummy; 1106 // If called before init complete, thread stack bottom will be null. 1107 // Can be called if fatal error occurs before initialization. 1108 if (initial_thread_stack_bottom() == NULL) return false; 1109 assert(initial_thread_stack_bottom() != NULL && 1110 initial_thread_stack_size() != 0, 1111 "os::init did not locate initial thread's stack region"); 1112 if ((address)&dummy >= initial_thread_stack_bottom() && 1113 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1114 return true; 1115 else return false; 1116} 1117 1118// Find the virtual memory area that contains addr 1119static bool find_vma(address addr, address* vma_low, address* vma_high) { 1120 FILE *fp = fopen("/proc/self/maps", "r"); 1121 if (fp) { 1122 address low, high; 1123 while (!feof(fp)) { 1124 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1125 if (low <= addr && addr < high) { 1126 if (vma_low) *vma_low = low; 1127 if (vma_high) *vma_high = high; 1128 fclose (fp); 1129 return true; 1130 } 1131 } 1132 for (;;) { 1133 int ch = fgetc(fp); 1134 if (ch == EOF || ch == (int)'\n') break; 1135 } 1136 } 1137 fclose(fp); 1138 } 1139 return false; 1140} 1141 1142// Locate initial thread stack. This special handling of initial thread stack 1143// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1144// bogus value for initial thread. 1145void os::Linux::capture_initial_stack(size_t max_size) { 1146 // stack size is the easy part, get it from RLIMIT_STACK 1147 size_t stack_size; 1148 struct rlimit rlim; 1149 getrlimit(RLIMIT_STACK, &rlim); 1150 stack_size = rlim.rlim_cur; 1151 1152 // 6308388: a bug in ld.so will relocate its own .data section to the 1153 // lower end of primordial stack; reduce ulimit -s value a little bit 1154 // so we won't install guard page on ld.so's data section. 1155 stack_size -= 2 * page_size(); 1156 1157 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1158 // 7.1, in both cases we will get 2G in return value. 1159 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1160 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1161 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1162 // in case other parts in glibc still assumes 2M max stack size. 1163 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1164 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1165 if (stack_size > 2 * K * K IA64_ONLY(*2)) 1166 stack_size = 2 * K * K IA64_ONLY(*2); 1167 // Try to figure out where the stack base (top) is. This is harder. 1168 // 1169 // When an application is started, glibc saves the initial stack pointer in 1170 // a global variable "__libc_stack_end", which is then used by system 1171 // libraries. __libc_stack_end should be pretty close to stack top. The 1172 // variable is available since the very early days. However, because it is 1173 // a private interface, it could disappear in the future. 1174 // 1175 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1176 // to __libc_stack_end, it is very close to stack top, but isn't the real 1177 // stack top. Note that /proc may not exist if VM is running as a chroot 1178 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1179 // /proc/<pid>/stat could change in the future (though unlikely). 1180 // 1181 // We try __libc_stack_end first. If that doesn't work, look for 1182 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1183 // as a hint, which should work well in most cases. 1184 1185 uintptr_t stack_start; 1186 1187 // try __libc_stack_end first 1188 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1189 if (p && *p) { 1190 stack_start = *p; 1191 } else { 1192 // see if we can get the start_stack field from /proc/self/stat 1193 FILE *fp; 1194 int pid; 1195 char state; 1196 int ppid; 1197 int pgrp; 1198 int session; 1199 int nr; 1200 int tpgrp; 1201 unsigned long flags; 1202 unsigned long minflt; 1203 unsigned long cminflt; 1204 unsigned long majflt; 1205 unsigned long cmajflt; 1206 unsigned long utime; 1207 unsigned long stime; 1208 long cutime; 1209 long cstime; 1210 long prio; 1211 long nice; 1212 long junk; 1213 long it_real; 1214 uintptr_t start; 1215 uintptr_t vsize; 1216 intptr_t rss; 1217 uintptr_t rsslim; 1218 uintptr_t scodes; 1219 uintptr_t ecode; 1220 int i; 1221 1222 // Figure what the primordial thread stack base is. Code is inspired 1223 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1224 // followed by command name surrounded by parentheses, state, etc. 1225 char stat[2048]; 1226 int statlen; 1227 1228 fp = fopen("/proc/self/stat", "r"); 1229 if (fp) { 1230 statlen = fread(stat, 1, 2047, fp); 1231 stat[statlen] = '\0'; 1232 fclose(fp); 1233 1234 // Skip pid and the command string. Note that we could be dealing with 1235 // weird command names, e.g. user could decide to rename java launcher 1236 // to "java 1.4.2 :)", then the stat file would look like 1237 // 1234 (java 1.4.2 :)) R ... ... 1238 // We don't really need to know the command string, just find the last 1239 // occurrence of ")" and then start parsing from there. See bug 4726580. 1240 char * s = strrchr(stat, ')'); 1241 1242 i = 0; 1243 if (s) { 1244 // Skip blank chars 1245 do s++; while (isspace(*s)); 1246 1247#define _UFM UINTX_FORMAT 1248#define _DFM INTX_FORMAT 1249 1250 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1251 /* 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 */ 1252 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, 1253 &state, /* 3 %c */ 1254 &ppid, /* 4 %d */ 1255 &pgrp, /* 5 %d */ 1256 &session, /* 6 %d */ 1257 &nr, /* 7 %d */ 1258 &tpgrp, /* 8 %d */ 1259 &flags, /* 9 %lu */ 1260 &minflt, /* 10 %lu */ 1261 &cminflt, /* 11 %lu */ 1262 &majflt, /* 12 %lu */ 1263 &cmajflt, /* 13 %lu */ 1264 &utime, /* 14 %lu */ 1265 &stime, /* 15 %lu */ 1266 &cutime, /* 16 %ld */ 1267 &cstime, /* 17 %ld */ 1268 &prio, /* 18 %ld */ 1269 &nice, /* 19 %ld */ 1270 &junk, /* 20 %ld */ 1271 &it_real, /* 21 %ld */ 1272 &start, /* 22 UINTX_FORMAT */ 1273 &vsize, /* 23 UINTX_FORMAT */ 1274 &rss, /* 24 INTX_FORMAT */ 1275 &rsslim, /* 25 UINTX_FORMAT */ 1276 &scodes, /* 26 UINTX_FORMAT */ 1277 &ecode, /* 27 UINTX_FORMAT */ 1278 &stack_start); /* 28 UINTX_FORMAT */ 1279 } 1280 1281#undef _UFM 1282#undef _DFM 1283 1284 if (i != 28 - 2) { 1285 assert(false, "Bad conversion from /proc/self/stat"); 1286 // product mode - assume we are the initial thread, good luck in the 1287 // embedded case. 1288 warning("Can't detect initial thread stack location - bad conversion"); 1289 stack_start = (uintptr_t) &rlim; 1290 } 1291 } else { 1292 // For some reason we can't open /proc/self/stat (for example, running on 1293 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1294 // most cases, so don't abort: 1295 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1296 stack_start = (uintptr_t) &rlim; 1297 } 1298 } 1299 1300 // Now we have a pointer (stack_start) very close to the stack top, the 1301 // next thing to do is to figure out the exact location of stack top. We 1302 // can find out the virtual memory area that contains stack_start by 1303 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1304 // and its upper limit is the real stack top. (again, this would fail if 1305 // running inside chroot, because /proc may not exist.) 1306 1307 uintptr_t stack_top; 1308 address low, high; 1309 if (find_vma((address)stack_start, &low, &high)) { 1310 // success, "high" is the true stack top. (ignore "low", because initial 1311 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1312 stack_top = (uintptr_t)high; 1313 } else { 1314 // failed, likely because /proc/self/maps does not exist 1315 warning("Can't detect initial thread stack location - find_vma failed"); 1316 // best effort: stack_start is normally within a few pages below the real 1317 // stack top, use it as stack top, and reduce stack size so we won't put 1318 // guard page outside stack. 1319 stack_top = stack_start; 1320 stack_size -= 16 * page_size(); 1321 } 1322 1323 // stack_top could be partially down the page so align it 1324 stack_top = align_size_up(stack_top, page_size()); 1325 1326 if (max_size && stack_size > max_size) { 1327 _initial_thread_stack_size = max_size; 1328 } else { 1329 _initial_thread_stack_size = stack_size; 1330 } 1331 1332 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1333 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1334} 1335 1336//////////////////////////////////////////////////////////////////////////////// 1337// time support 1338 1339// Time since start-up in seconds to a fine granularity. 1340// Used by VMSelfDestructTimer and the MemProfiler. 1341double os::elapsedTime() { 1342 1343 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1344} 1345 1346jlong os::elapsed_counter() { 1347 return javaTimeNanos() - initial_time_count; 1348} 1349 1350jlong os::elapsed_frequency() { 1351 return NANOSECS_PER_SEC; // nanosecond resolution 1352} 1353 1354bool os::supports_vtime() { return true; } 1355bool os::enable_vtime() { return false; } 1356bool os::vtime_enabled() { return false; } 1357 1358double os::elapsedVTime() { 1359 struct rusage usage; 1360 int retval = getrusage(RUSAGE_THREAD, &usage); 1361 if (retval == 0) { 1362 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); 1363 } else { 1364 // better than nothing, but not much 1365 return elapsedTime(); 1366 } 1367} 1368 1369jlong os::javaTimeMillis() { 1370 timeval time; 1371 int status = gettimeofday(&time, NULL); 1372 assert(status != -1, "linux error"); 1373 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1374} 1375 1376#ifndef CLOCK_MONOTONIC 1377#define CLOCK_MONOTONIC (1) 1378#endif 1379 1380void os::Linux::clock_init() { 1381 // we do dlopen's in this particular order due to bug in linux 1382 // dynamical loader (see 6348968) leading to crash on exit 1383 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1384 if (handle == NULL) { 1385 handle = dlopen("librt.so", RTLD_LAZY); 1386 } 1387 1388 if (handle) { 1389 int (*clock_getres_func)(clockid_t, struct timespec*) = 1390 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1391 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1392 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1393 if (clock_getres_func && clock_gettime_func) { 1394 // See if monotonic clock is supported by the kernel. Note that some 1395 // early implementations simply return kernel jiffies (updated every 1396 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1397 // for nano time (though the monotonic property is still nice to have). 1398 // It's fixed in newer kernels, however clock_getres() still returns 1399 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1400 // resolution for now. Hopefully as people move to new kernels, this 1401 // won't be a problem. 1402 struct timespec res; 1403 struct timespec tp; 1404 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1405 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1406 // yes, monotonic clock is supported 1407 _clock_gettime = clock_gettime_func; 1408 return; 1409 } else { 1410 // close librt if there is no monotonic clock 1411 dlclose(handle); 1412 } 1413 } 1414 } 1415 warning("No monotonic clock was available - timed services may " \ 1416 "be adversely affected if the time-of-day clock changes"); 1417} 1418 1419#ifndef SYS_clock_getres 1420 1421#if defined(IA32) || defined(AMD64) 1422#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1423#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1424#else 1425#warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1426#define sys_clock_getres(x,y) -1 1427#endif 1428 1429#else 1430#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1431#endif 1432 1433void os::Linux::fast_thread_clock_init() { 1434 if (!UseLinuxPosixThreadCPUClocks) { 1435 return; 1436 } 1437 clockid_t clockid; 1438 struct timespec tp; 1439 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1440 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1441 1442 // Switch to using fast clocks for thread cpu time if 1443 // the sys_clock_getres() returns 0 error code. 1444 // Note, that some kernels may support the current thread 1445 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1446 // returned by the pthread_getcpuclockid(). 1447 // If the fast Posix clocks are supported then the sys_clock_getres() 1448 // must return at least tp.tv_sec == 0 which means a resolution 1449 // better than 1 sec. This is extra check for reliability. 1450 1451 if(pthread_getcpuclockid_func && 1452 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1453 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1454 1455 _supports_fast_thread_cpu_time = true; 1456 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1457 } 1458} 1459 1460jlong os::javaTimeNanos() { 1461 if (os::supports_monotonic_clock()) { 1462 struct timespec tp; 1463 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1464 assert(status == 0, "gettime error"); 1465 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1466 return result; 1467 } else { 1468 timeval time; 1469 int status = gettimeofday(&time, NULL); 1470 assert(status != -1, "linux error"); 1471 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1472 return 1000 * usecs; 1473 } 1474} 1475 1476void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1477 if (os::supports_monotonic_clock()) { 1478 info_ptr->max_value = ALL_64_BITS; 1479 1480 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1481 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1482 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1483 } else { 1484 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1485 info_ptr->max_value = ALL_64_BITS; 1486 1487 // gettimeofday is a real time clock so it skips 1488 info_ptr->may_skip_backward = true; 1489 info_ptr->may_skip_forward = true; 1490 } 1491 1492 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1493} 1494 1495// Return the real, user, and system times in seconds from an 1496// arbitrary fixed point in the past. 1497bool os::getTimesSecs(double* process_real_time, 1498 double* process_user_time, 1499 double* process_system_time) { 1500 struct tms ticks; 1501 clock_t real_ticks = times(&ticks); 1502 1503 if (real_ticks == (clock_t) (-1)) { 1504 return false; 1505 } else { 1506 double ticks_per_second = (double) clock_tics_per_sec; 1507 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1508 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1509 *process_real_time = ((double) real_ticks) / ticks_per_second; 1510 1511 return true; 1512 } 1513} 1514 1515 1516char * os::local_time_string(char *buf, size_t buflen) { 1517 struct tm t; 1518 time_t long_time; 1519 time(&long_time); 1520 localtime_r(&long_time, &t); 1521 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1522 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1523 t.tm_hour, t.tm_min, t.tm_sec); 1524 return buf; 1525} 1526 1527struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1528 return localtime_r(clock, res); 1529} 1530 1531//////////////////////////////////////////////////////////////////////////////// 1532// runtime exit support 1533 1534// Note: os::shutdown() might be called very early during initialization, or 1535// called from signal handler. Before adding something to os::shutdown(), make 1536// sure it is async-safe and can handle partially initialized VM. 1537void os::shutdown() { 1538 1539 // allow PerfMemory to attempt cleanup of any persistent resources 1540 perfMemory_exit(); 1541 1542 // needs to remove object in file system 1543 AttachListener::abort(); 1544 1545 // flush buffered output, finish log files 1546 ostream_abort(); 1547 1548 // Check for abort hook 1549 abort_hook_t abort_hook = Arguments::abort_hook(); 1550 if (abort_hook != NULL) { 1551 abort_hook(); 1552 } 1553 1554} 1555 1556// Note: os::abort() might be called very early during initialization, or 1557// called from signal handler. Before adding something to os::abort(), make 1558// sure it is async-safe and can handle partially initialized VM. 1559void os::abort(bool dump_core) { 1560 os::shutdown(); 1561 if (dump_core) { 1562#ifndef PRODUCT 1563 fdStream out(defaultStream::output_fd()); 1564 out.print_raw("Current thread is "); 1565 char buf[16]; 1566 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1567 out.print_raw_cr(buf); 1568 out.print_raw_cr("Dumping core ..."); 1569#endif 1570 ::abort(); // dump core 1571 } 1572 1573 ::exit(1); 1574} 1575 1576// Die immediately, no exit hook, no abort hook, no cleanup. 1577void os::die() { 1578 // _exit() on LinuxThreads only kills current thread 1579 ::abort(); 1580} 1581 1582// unused on linux for now. 1583void os::set_error_file(const char *logfile) {} 1584 1585 1586// This method is a copy of JDK's sysGetLastErrorString 1587// from src/solaris/hpi/src/system_md.c 1588 1589size_t os::lasterror(char *buf, size_t len) { 1590 1591 if (errno == 0) return 0; 1592 1593 const char *s = ::strerror(errno); 1594 size_t n = ::strlen(s); 1595 if (n >= len) { 1596 n = len - 1; 1597 } 1598 ::strncpy(buf, s, n); 1599 buf[n] = '\0'; 1600 return n; 1601} 1602 1603intx os::current_thread_id() { return (intx)pthread_self(); } 1604int os::current_process_id() { 1605 1606 // Under the old linux thread library, linux gives each thread 1607 // its own process id. Because of this each thread will return 1608 // a different pid if this method were to return the result 1609 // of getpid(2). Linux provides no api that returns the pid 1610 // of the launcher thread for the vm. This implementation 1611 // returns a unique pid, the pid of the launcher thread 1612 // that starts the vm 'process'. 1613 1614 // Under the NPTL, getpid() returns the same pid as the 1615 // launcher thread rather than a unique pid per thread. 1616 // Use gettid() if you want the old pre NPTL behaviour. 1617 1618 // if you are looking for the result of a call to getpid() that 1619 // returns a unique pid for the calling thread, then look at the 1620 // OSThread::thread_id() method in osThread_linux.hpp file 1621 1622 return (int)(_initial_pid ? _initial_pid : getpid()); 1623} 1624 1625// DLL functions 1626 1627const char* os::dll_file_extension() { return ".so"; } 1628 1629// This must be hard coded because it's the system's temporary 1630// directory not the java application's temp directory, ala java.io.tmpdir. 1631const char* os::get_temp_directory() { return "/tmp"; } 1632 1633static bool file_exists(const char* filename) { 1634 struct stat statbuf; 1635 if (filename == NULL || strlen(filename) == 0) { 1636 return false; 1637 } 1638 return os::stat(filename, &statbuf) == 0; 1639} 1640 1641bool os::dll_build_name(char* buffer, size_t buflen, 1642 const char* pname, const char* fname) { 1643 bool retval = false; 1644 // Copied from libhpi 1645 const size_t pnamelen = pname ? strlen(pname) : 0; 1646 1647 // Return error on buffer overflow. 1648 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1649 return retval; 1650 } 1651 1652 if (pnamelen == 0) { 1653 snprintf(buffer, buflen, "lib%s.so", fname); 1654 retval = true; 1655 } else if (strchr(pname, *os::path_separator()) != NULL) { 1656 int n; 1657 char** pelements = split_path(pname, &n); 1658 if (pelements == NULL) { 1659 return false; 1660 } 1661 for (int i = 0 ; i < n ; i++) { 1662 // Really shouldn't be NULL, but check can't hurt 1663 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1664 continue; // skip the empty path values 1665 } 1666 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1667 if (file_exists(buffer)) { 1668 retval = true; 1669 break; 1670 } 1671 } 1672 // release the storage 1673 for (int i = 0 ; i < n ; i++) { 1674 if (pelements[i] != NULL) { 1675 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1676 } 1677 } 1678 if (pelements != NULL) { 1679 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1680 } 1681 } else { 1682 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1683 retval = true; 1684 } 1685 return retval; 1686} 1687 1688// check if addr is inside libjvm.so 1689bool os::address_is_in_vm(address addr) { 1690 static address libjvm_base_addr; 1691 Dl_info dlinfo; 1692 1693 if (libjvm_base_addr == NULL) { 1694 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1695 libjvm_base_addr = (address)dlinfo.dli_fbase; 1696 } 1697 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1698 } 1699 1700 if (dladdr((void *)addr, &dlinfo) != 0) { 1701 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1702 } 1703 1704 return false; 1705} 1706 1707bool os::dll_address_to_function_name(address addr, char *buf, 1708 int buflen, int *offset) { 1709 // buf is not optional, but offset is optional 1710 assert(buf != NULL, "sanity check"); 1711 1712 Dl_info dlinfo; 1713 1714 if (dladdr((void*)addr, &dlinfo) != 0) { 1715 // see if we have a matching symbol 1716 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1717 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1718 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1719 } 1720 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1721 return true; 1722 } 1723 // no matching symbol so try for just file info 1724 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1725 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1726 buf, buflen, offset, dlinfo.dli_fname)) { 1727 return true; 1728 } 1729 } 1730 } 1731 1732 buf[0] = '\0'; 1733 if (offset != NULL) *offset = -1; 1734 return false; 1735} 1736 1737struct _address_to_library_name { 1738 address addr; // input : memory address 1739 size_t buflen; // size of fname 1740 char* fname; // output: library name 1741 address base; // library base addr 1742}; 1743 1744static int address_to_library_name_callback(struct dl_phdr_info *info, 1745 size_t size, void *data) { 1746 int i; 1747 bool found = false; 1748 address libbase = NULL; 1749 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1750 1751 // iterate through all loadable segments 1752 for (i = 0; i < info->dlpi_phnum; i++) { 1753 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1754 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1755 // base address of a library is the lowest address of its loaded 1756 // segments. 1757 if (libbase == NULL || libbase > segbase) { 1758 libbase = segbase; 1759 } 1760 // see if 'addr' is within current segment 1761 if (segbase <= d->addr && 1762 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1763 found = true; 1764 } 1765 } 1766 } 1767 1768 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1769 // so dll_address_to_library_name() can fall through to use dladdr() which 1770 // can figure out executable name from argv[0]. 1771 if (found && info->dlpi_name && info->dlpi_name[0]) { 1772 d->base = libbase; 1773 if (d->fname) { 1774 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1775 } 1776 return 1; 1777 } 1778 return 0; 1779} 1780 1781bool os::dll_address_to_library_name(address addr, char* buf, 1782 int buflen, int* offset) { 1783 // buf is not optional, but offset is optional 1784 assert(buf != NULL, "sanity check"); 1785 1786 Dl_info dlinfo; 1787 struct _address_to_library_name data; 1788 1789 // There is a bug in old glibc dladdr() implementation that it could resolve 1790 // to wrong library name if the .so file has a base address != NULL. Here 1791 // we iterate through the program headers of all loaded libraries to find 1792 // out which library 'addr' really belongs to. This workaround can be 1793 // removed once the minimum requirement for glibc is moved to 2.3.x. 1794 data.addr = addr; 1795 data.fname = buf; 1796 data.buflen = buflen; 1797 data.base = NULL; 1798 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1799 1800 if (rslt) { 1801 // buf already contains library name 1802 if (offset) *offset = addr - data.base; 1803 return true; 1804 } 1805 if (dladdr((void*)addr, &dlinfo) != 0) { 1806 if (dlinfo.dli_fname != NULL) { 1807 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1808 } 1809 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1810 *offset = addr - (address)dlinfo.dli_fbase; 1811 } 1812 return true; 1813 } 1814 1815 buf[0] = '\0'; 1816 if (offset) *offset = -1; 1817 return false; 1818} 1819 1820 // Loads .dll/.so and 1821 // in case of error it checks if .dll/.so was built for the 1822 // same architecture as Hotspot is running on 1823 1824 1825// Remember the stack's state. The Linux dynamic linker will change 1826// the stack to 'executable' at most once, so we must safepoint only once. 1827bool os::Linux::_stack_is_executable = false; 1828 1829// VM operation that loads a library. This is necessary if stack protection 1830// of the Java stacks can be lost during loading the library. If we 1831// do not stop the Java threads, they can stack overflow before the stacks 1832// are protected again. 1833class VM_LinuxDllLoad: public VM_Operation { 1834 private: 1835 const char *_filename; 1836 char *_ebuf; 1837 int _ebuflen; 1838 void *_lib; 1839 public: 1840 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1841 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1842 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1843 void doit() { 1844 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1845 os::Linux::_stack_is_executable = true; 1846 } 1847 void* loaded_library() { return _lib; } 1848}; 1849 1850void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1851{ 1852 void * result = NULL; 1853 bool load_attempted = false; 1854 1855 // Check whether the library to load might change execution rights 1856 // of the stack. If they are changed, the protection of the stack 1857 // guard pages will be lost. We need a safepoint to fix this. 1858 // 1859 // See Linux man page execstack(8) for more info. 1860 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1861 ElfFile ef(filename); 1862 if (!ef.specifies_noexecstack()) { 1863 if (!is_init_completed()) { 1864 os::Linux::_stack_is_executable = true; 1865 // This is OK - No Java threads have been created yet, and hence no 1866 // stack guard pages to fix. 1867 // 1868 // This should happen only when you are building JDK7 using a very 1869 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1870 // 1871 // Dynamic loader will make all stacks executable after 1872 // this function returns, and will not do that again. 1873 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1874 } else { 1875 warning("You have loaded library %s which might have disabled stack guard. " 1876 "The VM will try to fix the stack guard now.\n" 1877 "It's highly recommended that you fix the library with " 1878 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1879 filename); 1880 1881 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1882 JavaThread *jt = JavaThread::current(); 1883 if (jt->thread_state() != _thread_in_native) { 1884 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1885 // that requires ExecStack. Cannot enter safe point. Let's give up. 1886 warning("Unable to fix stack guard. Giving up."); 1887 } else { 1888 if (!LoadExecStackDllInVMThread) { 1889 // This is for the case where the DLL has an static 1890 // constructor function that executes JNI code. We cannot 1891 // load such DLLs in the VMThread. 1892 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1893 } 1894 1895 ThreadInVMfromNative tiv(jt); 1896 debug_only(VMNativeEntryWrapper vew;) 1897 1898 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1899 VMThread::execute(&op); 1900 if (LoadExecStackDllInVMThread) { 1901 result = op.loaded_library(); 1902 } 1903 load_attempted = true; 1904 } 1905 } 1906 } 1907 } 1908 1909 if (!load_attempted) { 1910 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1911 } 1912 1913 if (result != NULL) { 1914 // Successful loading 1915 return result; 1916 } 1917 1918 Elf32_Ehdr elf_head; 1919 int diag_msg_max_length=ebuflen-strlen(ebuf); 1920 char* diag_msg_buf=ebuf+strlen(ebuf); 1921 1922 if (diag_msg_max_length==0) { 1923 // No more space in ebuf for additional diagnostics message 1924 return NULL; 1925 } 1926 1927 1928 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1929 1930 if (file_descriptor < 0) { 1931 // Can't open library, report dlerror() message 1932 return NULL; 1933 } 1934 1935 bool failed_to_read_elf_head= 1936 (sizeof(elf_head)!= 1937 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1938 1939 ::close(file_descriptor); 1940 if (failed_to_read_elf_head) { 1941 // file i/o error - report dlerror() msg 1942 return NULL; 1943 } 1944 1945 typedef struct { 1946 Elf32_Half code; // Actual value as defined in elf.h 1947 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1948 char elf_class; // 32 or 64 bit 1949 char endianess; // MSB or LSB 1950 char* name; // String representation 1951 } arch_t; 1952 1953 #ifndef EM_486 1954 #define EM_486 6 /* Intel 80486 */ 1955 #endif 1956 1957 static const arch_t arch_array[]={ 1958 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1959 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1960 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1961 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1962 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1963 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1964 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1965 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1966#if defined(VM_LITTLE_ENDIAN) 1967 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"}, 1968#else 1969 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1970#endif 1971 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1972 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1973 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1974 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1975 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1976 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1977 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1978 }; 1979 1980 #if (defined IA32) 1981 static Elf32_Half running_arch_code=EM_386; 1982 #elif (defined AMD64) 1983 static Elf32_Half running_arch_code=EM_X86_64; 1984 #elif (defined IA64) 1985 static Elf32_Half running_arch_code=EM_IA_64; 1986 #elif (defined __sparc) && (defined _LP64) 1987 static Elf32_Half running_arch_code=EM_SPARCV9; 1988 #elif (defined __sparc) && (!defined _LP64) 1989 static Elf32_Half running_arch_code=EM_SPARC; 1990 #elif (defined __powerpc64__) 1991 static Elf32_Half running_arch_code=EM_PPC64; 1992 #elif (defined __powerpc__) 1993 static Elf32_Half running_arch_code=EM_PPC; 1994 #elif (defined ARM) 1995 static Elf32_Half running_arch_code=EM_ARM; 1996 #elif (defined S390) 1997 static Elf32_Half running_arch_code=EM_S390; 1998 #elif (defined ALPHA) 1999 static Elf32_Half running_arch_code=EM_ALPHA; 2000 #elif (defined MIPSEL) 2001 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 2002 #elif (defined PARISC) 2003 static Elf32_Half running_arch_code=EM_PARISC; 2004 #elif (defined MIPS) 2005 static Elf32_Half running_arch_code=EM_MIPS; 2006 #elif (defined M68K) 2007 static Elf32_Half running_arch_code=EM_68K; 2008 #else 2009 #error Method os::dll_load requires that one of following is defined:\ 2010 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 2011 #endif 2012 2013 // Identify compatability class for VM's architecture and library's architecture 2014 // Obtain string descriptions for architectures 2015 2016 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 2017 int running_arch_index=-1; 2018 2019 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 2020 if (running_arch_code == arch_array[i].code) { 2021 running_arch_index = i; 2022 } 2023 if (lib_arch.code == arch_array[i].code) { 2024 lib_arch.compat_class = arch_array[i].compat_class; 2025 lib_arch.name = arch_array[i].name; 2026 } 2027 } 2028 2029 assert(running_arch_index != -1, 2030 "Didn't find running architecture code (running_arch_code) in arch_array"); 2031 if (running_arch_index == -1) { 2032 // Even though running architecture detection failed 2033 // we may still continue with reporting dlerror() message 2034 return NULL; 2035 } 2036 2037 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 2038 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 2039 return NULL; 2040 } 2041 2042#ifndef S390 2043 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 2044 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 2045 return NULL; 2046 } 2047#endif // !S390 2048 2049 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2050 if ( lib_arch.name!=NULL ) { 2051 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2052 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2053 lib_arch.name, arch_array[running_arch_index].name); 2054 } else { 2055 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2056 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2057 lib_arch.code, 2058 arch_array[running_arch_index].name); 2059 } 2060 } 2061 2062 return NULL; 2063} 2064 2065void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) { 2066 void * result = ::dlopen(filename, RTLD_LAZY); 2067 if (result == NULL) { 2068 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2069 ebuf[ebuflen-1] = '\0'; 2070 } 2071 return result; 2072} 2073 2074void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) { 2075 void * result = NULL; 2076 if (LoadExecStackDllInVMThread) { 2077 result = dlopen_helper(filename, ebuf, ebuflen); 2078 } 2079 2080 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2081 // library that requires an executable stack, or which does not have this 2082 // stack attribute set, dlopen changes the stack attribute to executable. The 2083 // read protection of the guard pages gets lost. 2084 // 2085 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2086 // may have been queued at the same time. 2087 2088 if (!_stack_is_executable) { 2089 JavaThread *jt = Threads::first(); 2090 2091 while (jt) { 2092 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2093 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2094 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2095 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2096 warning("Attempt to reguard stack yellow zone failed."); 2097 } 2098 } 2099 jt = jt->next(); 2100 } 2101 } 2102 2103 return result; 2104} 2105 2106/* 2107 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2108 * chances are you might want to run the generated bits against glibc-2.0 2109 * libdl.so, so always use locking for any version of glibc. 2110 */ 2111void* os::dll_lookup(void* handle, const char* name) { 2112 pthread_mutex_lock(&dl_mutex); 2113 void* res = dlsym(handle, name); 2114 pthread_mutex_unlock(&dl_mutex); 2115 return res; 2116} 2117 2118void* os::get_default_process_handle() { 2119 return (void*)::dlopen(NULL, RTLD_LAZY); 2120} 2121 2122static bool _print_ascii_file(const char* filename, outputStream* st) { 2123 int fd = ::open(filename, O_RDONLY); 2124 if (fd == -1) { 2125 return false; 2126 } 2127 2128 char buf[32]; 2129 int bytes; 2130 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2131 st->print_raw(buf, bytes); 2132 } 2133 2134 ::close(fd); 2135 2136 return true; 2137} 2138 2139void os::print_dll_info(outputStream *st) { 2140 st->print_cr("Dynamic libraries:"); 2141 2142 char fname[32]; 2143 pid_t pid = os::Linux::gettid(); 2144 2145 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2146 2147 if (!_print_ascii_file(fname, st)) { 2148 st->print("Can not get library information for pid = %d\n", pid); 2149 } 2150} 2151 2152void os::print_os_info_brief(outputStream* st) { 2153 os::Linux::print_distro_info(st); 2154 2155 os::Posix::print_uname_info(st); 2156 2157 os::Linux::print_libversion_info(st); 2158 2159} 2160 2161void os::print_os_info(outputStream* st) { 2162 st->print("OS:"); 2163 2164 os::Linux::print_distro_info(st); 2165 2166 os::Posix::print_uname_info(st); 2167 2168 // Print warning if unsafe chroot environment detected 2169 if (unsafe_chroot_detected) { 2170 st->print("WARNING!! "); 2171 st->print_cr(unstable_chroot_error); 2172 } 2173 2174 os::Linux::print_libversion_info(st); 2175 2176 os::Posix::print_rlimit_info(st); 2177 2178 os::Posix::print_load_average(st); 2179 2180 os::Linux::print_full_memory_info(st); 2181} 2182 2183// Try to identify popular distros. 2184// Most Linux distributions have a /etc/XXX-release file, which contains 2185// the OS version string. Newer Linux distributions have a /etc/lsb-release 2186// file that also contains the OS version string. Some have more than one 2187// /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2188// /etc/redhat-release.), so the order is important. 2189// Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2190// their own specific XXX-release file as well as a redhat-release file. 2191// Because of this the XXX-release file needs to be searched for before the 2192// redhat-release file. 2193// Since Red Hat has a lsb-release file that is not very descriptive the 2194// search for redhat-release needs to be before lsb-release. 2195// Since the lsb-release file is the new standard it needs to be searched 2196// before the older style release files. 2197// Searching system-release (Red Hat) and os-release (other Linuxes) are a 2198// next to last resort. The os-release file is a new standard that contains 2199// distribution information and the system-release file seems to be an old 2200// standard that has been replaced by the lsb-release and os-release files. 2201// Searching for the debian_version file is the last resort. It contains 2202// an informative string like "6.0.6" or "wheezy/sid". Because of this 2203// "Debian " is printed before the contents of the debian_version file. 2204void os::Linux::print_distro_info(outputStream* st) { 2205 if (!_print_ascii_file("/etc/oracle-release", st) && 2206 !_print_ascii_file("/etc/mandriva-release", st) && 2207 !_print_ascii_file("/etc/mandrake-release", st) && 2208 !_print_ascii_file("/etc/sun-release", st) && 2209 !_print_ascii_file("/etc/redhat-release", st) && 2210 !_print_ascii_file("/etc/lsb-release", st) && 2211 !_print_ascii_file("/etc/SuSE-release", st) && 2212 !_print_ascii_file("/etc/turbolinux-release", st) && 2213 !_print_ascii_file("/etc/gentoo-release", st) && 2214 !_print_ascii_file("/etc/ltib-release", st) && 2215 !_print_ascii_file("/etc/angstrom-version", st) && 2216 !_print_ascii_file("/etc/system-release", st) && 2217 !_print_ascii_file("/etc/os-release", st)) { 2218 2219 if (file_exists("/etc/debian_version")) { 2220 st->print("Debian "); 2221 _print_ascii_file("/etc/debian_version", st); 2222 } else { 2223 st->print("Linux"); 2224 } 2225 } 2226 st->cr(); 2227} 2228 2229void os::Linux::print_libversion_info(outputStream* st) { 2230 // libc, pthread 2231 st->print("libc:"); 2232 st->print(os::Linux::glibc_version()); st->print(" "); 2233 st->print(os::Linux::libpthread_version()); st->print(" "); 2234 if (os::Linux::is_LinuxThreads()) { 2235 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2236 } 2237 st->cr(); 2238} 2239 2240void os::Linux::print_full_memory_info(outputStream* st) { 2241 st->print("\n/proc/meminfo:\n"); 2242 _print_ascii_file("/proc/meminfo", st); 2243 st->cr(); 2244} 2245 2246void os::print_memory_info(outputStream* st) { 2247 2248 st->print("Memory:"); 2249 st->print(" %dk page", os::vm_page_size()>>10); 2250 2251 // values in struct sysinfo are "unsigned long" 2252 struct sysinfo si; 2253 sysinfo(&si); 2254 2255 st->print(", physical " UINT64_FORMAT "k", 2256 os::physical_memory() >> 10); 2257 st->print("(" UINT64_FORMAT "k free)", 2258 os::available_memory() >> 10); 2259 st->print(", swap " UINT64_FORMAT "k", 2260 ((jlong)si.totalswap * si.mem_unit) >> 10); 2261 st->print("(" UINT64_FORMAT "k free)", 2262 ((jlong)si.freeswap * si.mem_unit) >> 10); 2263 st->cr(); 2264} 2265 2266void os::pd_print_cpu_info(outputStream* st) { 2267 st->print("\n/proc/cpuinfo:\n"); 2268 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2269 st->print(" <Not Available>"); 2270 } 2271 st->cr(); 2272} 2273 2274void os::print_siginfo(outputStream* st, void* siginfo) { 2275 const siginfo_t* si = (const siginfo_t*)siginfo; 2276 2277 os::Posix::print_siginfo_brief(st, si); 2278 2279 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2280 UseSharedSpaces) { 2281 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2282 if (mapinfo->is_in_shared_space(si->si_addr)) { 2283 st->print("\n\nError accessing class data sharing archive." \ 2284 " Mapped file inaccessible during execution, " \ 2285 " possible disk/network problem."); 2286 } 2287 } 2288 st->cr(); 2289} 2290 2291 2292static void print_signal_handler(outputStream* st, int sig, 2293 char* buf, size_t buflen); 2294 2295void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2296 st->print_cr("Signal Handlers:"); 2297 print_signal_handler(st, SIGSEGV, buf, buflen); 2298 print_signal_handler(st, SIGBUS , buf, buflen); 2299 print_signal_handler(st, SIGFPE , buf, buflen); 2300 print_signal_handler(st, SIGPIPE, buf, buflen); 2301 print_signal_handler(st, SIGXFSZ, buf, buflen); 2302 print_signal_handler(st, SIGILL , buf, buflen); 2303 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2304 print_signal_handler(st, SR_signum, buf, buflen); 2305 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2306 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2307 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2308 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2309#if defined(PPC64) 2310 print_signal_handler(st, SIGTRAP, buf, buflen); 2311#endif 2312} 2313 2314static char saved_jvm_path[MAXPATHLEN] = {0}; 2315 2316// Find the full path to the current module, libjvm.so 2317void os::jvm_path(char *buf, jint buflen) { 2318 // Error checking. 2319 if (buflen < MAXPATHLEN) { 2320 assert(false, "must use a large-enough buffer"); 2321 buf[0] = '\0'; 2322 return; 2323 } 2324 // Lazy resolve the path to current module. 2325 if (saved_jvm_path[0] != 0) { 2326 strcpy(buf, saved_jvm_path); 2327 return; 2328 } 2329 2330 char dli_fname[MAXPATHLEN]; 2331 bool ret = dll_address_to_library_name( 2332 CAST_FROM_FN_PTR(address, os::jvm_path), 2333 dli_fname, sizeof(dli_fname), NULL); 2334 assert(ret, "cannot locate libjvm"); 2335 char *rp = NULL; 2336 if (ret && dli_fname[0] != '\0') { 2337 rp = realpath(dli_fname, buf); 2338 } 2339 if (rp == NULL) 2340 return; 2341 2342 if (Arguments::sun_java_launcher_is_altjvm()) { 2343 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2344 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2345 // If "/jre/lib/" appears at the right place in the string, then 2346 // assume we are installed in a JDK and we're done. Otherwise, check 2347 // for a JAVA_HOME environment variable and fix up the path so it 2348 // looks like libjvm.so is installed there (append a fake suffix 2349 // hotspot/libjvm.so). 2350 const char *p = buf + strlen(buf) - 1; 2351 for (int count = 0; p > buf && count < 5; ++count) { 2352 for (--p; p > buf && *p != '/'; --p) 2353 /* empty */ ; 2354 } 2355 2356 if (strncmp(p, "/jre/lib/", 9) != 0) { 2357 // Look for JAVA_HOME in the environment. 2358 char* java_home_var = ::getenv("JAVA_HOME"); 2359 if (java_home_var != NULL && java_home_var[0] != 0) { 2360 char* jrelib_p; 2361 int len; 2362 2363 // Check the current module name "libjvm.so". 2364 p = strrchr(buf, '/'); 2365 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2366 2367 rp = realpath(java_home_var, buf); 2368 if (rp == NULL) 2369 return; 2370 2371 // determine if this is a legacy image or modules image 2372 // modules image doesn't have "jre" subdirectory 2373 len = strlen(buf); 2374 jrelib_p = buf + len; 2375 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2376 if (0 != access(buf, F_OK)) { 2377 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2378 } 2379 2380 if (0 == access(buf, F_OK)) { 2381 // Use current module name "libjvm.so" 2382 len = strlen(buf); 2383 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2384 } else { 2385 // Go back to path of .so 2386 rp = realpath(dli_fname, buf); 2387 if (rp == NULL) 2388 return; 2389 } 2390 } 2391 } 2392 } 2393 2394 strcpy(saved_jvm_path, buf); 2395} 2396 2397void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2398 // no prefix required, not even "_" 2399} 2400 2401void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2402 // no suffix required 2403} 2404 2405//////////////////////////////////////////////////////////////////////////////// 2406// sun.misc.Signal support 2407 2408static volatile jint sigint_count = 0; 2409 2410static void 2411UserHandler(int sig, void *siginfo, void *context) { 2412 // 4511530 - sem_post is serialized and handled by the manager thread. When 2413 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2414 // don't want to flood the manager thread with sem_post requests. 2415 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2416 return; 2417 2418 // Ctrl-C is pressed during error reporting, likely because the error 2419 // handler fails to abort. Let VM die immediately. 2420 if (sig == SIGINT && is_error_reported()) { 2421 os::die(); 2422 } 2423 2424 os::signal_notify(sig); 2425} 2426 2427void* os::user_handler() { 2428 return CAST_FROM_FN_PTR(void*, UserHandler); 2429} 2430 2431class Semaphore : public StackObj { 2432 public: 2433 Semaphore(); 2434 ~Semaphore(); 2435 void signal(); 2436 void wait(); 2437 bool trywait(); 2438 bool timedwait(unsigned int sec, int nsec); 2439 private: 2440 sem_t _semaphore; 2441}; 2442 2443Semaphore::Semaphore() { 2444 sem_init(&_semaphore, 0, 0); 2445} 2446 2447Semaphore::~Semaphore() { 2448 sem_destroy(&_semaphore); 2449} 2450 2451void Semaphore::signal() { 2452 sem_post(&_semaphore); 2453} 2454 2455void Semaphore::wait() { 2456 sem_wait(&_semaphore); 2457} 2458 2459bool Semaphore::trywait() { 2460 return sem_trywait(&_semaphore) == 0; 2461} 2462 2463bool Semaphore::timedwait(unsigned int sec, int nsec) { 2464 2465 struct timespec ts; 2466 // Semaphore's are always associated with CLOCK_REALTIME 2467 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2468 // see unpackTime for discussion on overflow checking 2469 if (sec >= MAX_SECS) { 2470 ts.tv_sec += MAX_SECS; 2471 ts.tv_nsec = 0; 2472 } else { 2473 ts.tv_sec += sec; 2474 ts.tv_nsec += nsec; 2475 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2476 ts.tv_nsec -= NANOSECS_PER_SEC; 2477 ++ts.tv_sec; // note: this must be <= max_secs 2478 } 2479 } 2480 2481 while (1) { 2482 int result = sem_timedwait(&_semaphore, &ts); 2483 if (result == 0) { 2484 return true; 2485 } else if (errno == EINTR) { 2486 continue; 2487 } else if (errno == ETIMEDOUT) { 2488 return false; 2489 } else { 2490 return false; 2491 } 2492 } 2493} 2494 2495extern "C" { 2496 typedef void (*sa_handler_t)(int); 2497 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2498} 2499 2500void* os::signal(int signal_number, void* handler) { 2501 struct sigaction sigAct, oldSigAct; 2502 2503 sigfillset(&(sigAct.sa_mask)); 2504 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2505 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2506 2507 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2508 // -1 means registration failed 2509 return (void *)-1; 2510 } 2511 2512 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2513} 2514 2515void os::signal_raise(int signal_number) { 2516 ::raise(signal_number); 2517} 2518 2519/* 2520 * The following code is moved from os.cpp for making this 2521 * code platform specific, which it is by its very nature. 2522 */ 2523 2524// Will be modified when max signal is changed to be dynamic 2525int os::sigexitnum_pd() { 2526 return NSIG; 2527} 2528 2529// a counter for each possible signal value 2530static volatile jint pending_signals[NSIG+1] = { 0 }; 2531 2532// Linux(POSIX) specific hand shaking semaphore. 2533static sem_t sig_sem; 2534static Semaphore sr_semaphore; 2535 2536void os::signal_init_pd() { 2537 // Initialize signal structures 2538 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2539 2540 // Initialize signal semaphore 2541 ::sem_init(&sig_sem, 0, 0); 2542} 2543 2544void os::signal_notify(int sig) { 2545 Atomic::inc(&pending_signals[sig]); 2546 ::sem_post(&sig_sem); 2547} 2548 2549static int check_pending_signals(bool wait) { 2550 Atomic::store(0, &sigint_count); 2551 for (;;) { 2552 for (int i = 0; i < NSIG + 1; i++) { 2553 jint n = pending_signals[i]; 2554 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2555 return i; 2556 } 2557 } 2558 if (!wait) { 2559 return -1; 2560 } 2561 JavaThread *thread = JavaThread::current(); 2562 ThreadBlockInVM tbivm(thread); 2563 2564 bool threadIsSuspended; 2565 do { 2566 thread->set_suspend_equivalent(); 2567 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2568 ::sem_wait(&sig_sem); 2569 2570 // were we externally suspended while we were waiting? 2571 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2572 if (threadIsSuspended) { 2573 // 2574 // The semaphore has been incremented, but while we were waiting 2575 // another thread suspended us. We don't want to continue running 2576 // while suspended because that would surprise the thread that 2577 // suspended us. 2578 // 2579 ::sem_post(&sig_sem); 2580 2581 thread->java_suspend_self(); 2582 } 2583 } while (threadIsSuspended); 2584 } 2585} 2586 2587int os::signal_lookup() { 2588 return check_pending_signals(false); 2589} 2590 2591int os::signal_wait() { 2592 return check_pending_signals(true); 2593} 2594 2595//////////////////////////////////////////////////////////////////////////////// 2596// Virtual Memory 2597 2598int os::vm_page_size() { 2599 // Seems redundant as all get out 2600 assert(os::Linux::page_size() != -1, "must call os::init"); 2601 return os::Linux::page_size(); 2602} 2603 2604// Solaris allocates memory by pages. 2605int os::vm_allocation_granularity() { 2606 assert(os::Linux::page_size() != -1, "must call os::init"); 2607 return os::Linux::page_size(); 2608} 2609 2610// Rationale behind this function: 2611// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2612// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2613// samples for JITted code. Here we create private executable mapping over the code cache 2614// and then we can use standard (well, almost, as mapping can change) way to provide 2615// info for the reporting script by storing timestamp and location of symbol 2616void linux_wrap_code(char* base, size_t size) { 2617 static volatile jint cnt = 0; 2618 2619 if (!UseOprofile) { 2620 return; 2621 } 2622 2623 char buf[PATH_MAX+1]; 2624 int num = Atomic::add(1, &cnt); 2625 2626 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2627 os::get_temp_directory(), os::current_process_id(), num); 2628 unlink(buf); 2629 2630 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2631 2632 if (fd != -1) { 2633 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2634 if (rv != (off_t)-1) { 2635 if (::write(fd, "", 1) == 1) { 2636 mmap(base, size, 2637 PROT_READ|PROT_WRITE|PROT_EXEC, 2638 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2639 } 2640 } 2641 ::close(fd); 2642 unlink(buf); 2643 } 2644} 2645 2646static bool recoverable_mmap_error(int err) { 2647 // See if the error is one we can let the caller handle. This 2648 // list of errno values comes from JBS-6843484. I can't find a 2649 // Linux man page that documents this specific set of errno 2650 // values so while this list currently matches Solaris, it may 2651 // change as we gain experience with this failure mode. 2652 switch (err) { 2653 case EBADF: 2654 case EINVAL: 2655 case ENOTSUP: 2656 // let the caller deal with these errors 2657 return true; 2658 2659 default: 2660 // Any remaining errors on this OS can cause our reserved mapping 2661 // to be lost. That can cause confusion where different data 2662 // structures think they have the same memory mapped. The worst 2663 // scenario is if both the VM and a library think they have the 2664 // same memory mapped. 2665 return false; 2666 } 2667} 2668 2669static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2670 int err) { 2671 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2672 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2673 strerror(err), err); 2674} 2675 2676static void warn_fail_commit_memory(char* addr, size_t size, 2677 size_t alignment_hint, bool exec, 2678 int err) { 2679 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2680 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2681 alignment_hint, exec, strerror(err), err); 2682} 2683 2684// NOTE: Linux kernel does not really reserve the pages for us. 2685// All it does is to check if there are enough free pages 2686// left at the time of mmap(). This could be a potential 2687// problem. 2688int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2689 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2690 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2691 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2692 if (res != (uintptr_t) MAP_FAILED) { 2693 if (UseNUMAInterleaving) { 2694 numa_make_global(addr, size); 2695 } 2696 return 0; 2697 } 2698 2699 int err = errno; // save errno from mmap() call above 2700 2701 if (!recoverable_mmap_error(err)) { 2702 warn_fail_commit_memory(addr, size, exec, err); 2703 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2704 } 2705 2706 return err; 2707} 2708 2709bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2710 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2711} 2712 2713void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2714 const char* mesg) { 2715 assert(mesg != NULL, "mesg must be specified"); 2716 int err = os::Linux::commit_memory_impl(addr, size, exec); 2717 if (err != 0) { 2718 // the caller wants all commit errors to exit with the specified mesg: 2719 warn_fail_commit_memory(addr, size, exec, err); 2720 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2721 } 2722} 2723 2724// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2725#ifndef MAP_HUGETLB 2726#define MAP_HUGETLB 0x40000 2727#endif 2728 2729// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2730#ifndef MADV_HUGEPAGE 2731#define MADV_HUGEPAGE 14 2732#endif 2733 2734int os::Linux::commit_memory_impl(char* addr, size_t size, 2735 size_t alignment_hint, bool exec) { 2736 int err = os::Linux::commit_memory_impl(addr, size, exec); 2737 if (err == 0) { 2738 realign_memory(addr, size, alignment_hint); 2739 } 2740 return err; 2741} 2742 2743bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2744 bool exec) { 2745 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2746} 2747 2748void os::pd_commit_memory_or_exit(char* addr, size_t size, 2749 size_t alignment_hint, bool exec, 2750 const char* mesg) { 2751 assert(mesg != NULL, "mesg must be specified"); 2752 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2753 if (err != 0) { 2754 // the caller wants all commit errors to exit with the specified mesg: 2755 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2756 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2757 } 2758} 2759 2760void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2761 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2762 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2763 // be supported or the memory may already be backed by huge pages. 2764 ::madvise(addr, bytes, MADV_HUGEPAGE); 2765 } 2766} 2767 2768void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2769 // This method works by doing an mmap over an existing mmaping and effectively discarding 2770 // the existing pages. However it won't work for SHM-based large pages that cannot be 2771 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2772 // small pages on top of the SHM segment. This method always works for small pages, so we 2773 // allow that in any case. 2774 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2775 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2776 } 2777} 2778 2779void os::numa_make_global(char *addr, size_t bytes) { 2780 Linux::numa_interleave_memory(addr, bytes); 2781} 2782 2783// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2784// bind policy to MPOL_PREFERRED for the current thread. 2785#define USE_MPOL_PREFERRED 0 2786 2787void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2788 // To make NUMA and large pages more robust when both enabled, we need to ease 2789 // the requirements on where the memory should be allocated. MPOL_BIND is the 2790 // default policy and it will force memory to be allocated on the specified 2791 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2792 // the specified node, but will not force it. Using this policy will prevent 2793 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2794 // free large pages. 2795 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2796 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2797} 2798 2799bool os::numa_topology_changed() { return false; } 2800 2801size_t os::numa_get_groups_num() { 2802 int max_node = Linux::numa_max_node(); 2803 return max_node > 0 ? max_node + 1 : 1; 2804} 2805 2806int os::numa_get_group_id() { 2807 int cpu_id = Linux::sched_getcpu(); 2808 if (cpu_id != -1) { 2809 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2810 if (lgrp_id != -1) { 2811 return lgrp_id; 2812 } 2813 } 2814 return 0; 2815} 2816 2817size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2818 for (size_t i = 0; i < size; i++) { 2819 ids[i] = i; 2820 } 2821 return size; 2822} 2823 2824bool os::get_page_info(char *start, page_info* info) { 2825 return false; 2826} 2827 2828char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2829 return end; 2830} 2831 2832 2833int os::Linux::sched_getcpu_syscall(void) { 2834 unsigned int cpu; 2835 int retval = -1; 2836 2837#if defined(IA32) 2838# ifndef SYS_getcpu 2839# define SYS_getcpu 318 2840# endif 2841 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2842#elif defined(AMD64) 2843// Unfortunately we have to bring all these macros here from vsyscall.h 2844// to be able to compile on old linuxes. 2845# define __NR_vgetcpu 2 2846# define VSYSCALL_START (-10UL << 20) 2847# define VSYSCALL_SIZE 1024 2848# define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2849 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2850 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2851 retval = vgetcpu(&cpu, NULL, NULL); 2852#endif 2853 2854 return (retval == -1) ? retval : cpu; 2855} 2856 2857// Something to do with the numa-aware allocator needs these symbols 2858extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2859extern "C" JNIEXPORT void numa_error(char *where) { } 2860extern "C" JNIEXPORT int fork1() { return fork(); } 2861 2862 2863// If we are running with libnuma version > 2, then we should 2864// be trying to use symbols with versions 1.1 2865// If we are running with earlier version, which did not have symbol versions, 2866// we should use the base version. 2867void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2868 void *f = dlvsym(handle, name, "libnuma_1.1"); 2869 if (f == NULL) { 2870 f = dlsym(handle, name); 2871 } 2872 return f; 2873} 2874 2875bool os::Linux::libnuma_init() { 2876 // sched_getcpu() should be in libc. 2877 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2878 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2879 2880 // If it's not, try a direct syscall. 2881 if (sched_getcpu() == -1) 2882 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall)); 2883 2884 if (sched_getcpu() != -1) { // Does it work? 2885 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2886 if (handle != NULL) { 2887 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2888 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2889 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2890 libnuma_dlsym(handle, "numa_max_node"))); 2891 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2892 libnuma_dlsym(handle, "numa_available"))); 2893 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2894 libnuma_dlsym(handle, "numa_tonode_memory"))); 2895 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2896 libnuma_dlsym(handle, "numa_interleave_memory"))); 2897 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2898 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2899 2900 2901 if (numa_available() != -1) { 2902 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2903 // Create a cpu -> node mapping 2904 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2905 rebuild_cpu_to_node_map(); 2906 return true; 2907 } 2908 } 2909 } 2910 return false; 2911} 2912 2913// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2914// The table is later used in get_node_by_cpu(). 2915void os::Linux::rebuild_cpu_to_node_map() { 2916 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2917 // in libnuma (possible values are starting from 16, 2918 // and continuing up with every other power of 2, but less 2919 // than the maximum number of CPUs supported by kernel), and 2920 // is a subject to change (in libnuma version 2 the requirements 2921 // are more reasonable) we'll just hardcode the number they use 2922 // in the library. 2923 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2924 2925 size_t cpu_num = os::active_processor_count(); 2926 size_t cpu_map_size = NCPUS / BitsPerCLong; 2927 size_t cpu_map_valid_size = 2928 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2929 2930 cpu_to_node()->clear(); 2931 cpu_to_node()->at_grow(cpu_num - 1); 2932 size_t node_num = numa_get_groups_num(); 2933 2934 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2935 for (size_t i = 0; i < node_num; i++) { 2936 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2937 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2938 if (cpu_map[j] != 0) { 2939 for (size_t k = 0; k < BitsPerCLong; k++) { 2940 if (cpu_map[j] & (1UL << k)) { 2941 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2942 } 2943 } 2944 } 2945 } 2946 } 2947 } 2948 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2949} 2950 2951int os::Linux::get_node_by_cpu(int cpu_id) { 2952 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2953 return cpu_to_node()->at(cpu_id); 2954 } 2955 return -1; 2956} 2957 2958GrowableArray<int>* os::Linux::_cpu_to_node; 2959os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2960os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2961os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2962os::Linux::numa_available_func_t os::Linux::_numa_available; 2963os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2964os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2965os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2966unsigned long* os::Linux::_numa_all_nodes; 2967 2968bool os::pd_uncommit_memory(char* addr, size_t size) { 2969 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2970 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2971 return res != (uintptr_t) MAP_FAILED; 2972} 2973 2974static 2975address get_stack_commited_bottom(address bottom, size_t size) { 2976 address nbot = bottom; 2977 address ntop = bottom + size; 2978 2979 size_t page_sz = os::vm_page_size(); 2980 unsigned pages = size / page_sz; 2981 2982 unsigned char vec[1]; 2983 unsigned imin = 1, imax = pages + 1, imid; 2984 int mincore_return_value = 0; 2985 2986 assert(imin <= imax, "Unexpected page size"); 2987 2988 while (imin < imax) { 2989 imid = (imax + imin) / 2; 2990 nbot = ntop - (imid * page_sz); 2991 2992 // Use a trick with mincore to check whether the page is mapped or not. 2993 // mincore sets vec to 1 if page resides in memory and to 0 if page 2994 // is swapped output but if page we are asking for is unmapped 2995 // it returns -1,ENOMEM 2996 mincore_return_value = mincore(nbot, page_sz, vec); 2997 2998 if (mincore_return_value == -1) { 2999 // Page is not mapped go up 3000 // to find first mapped page 3001 if (errno != EAGAIN) { 3002 assert(errno == ENOMEM, "Unexpected mincore errno"); 3003 imax = imid; 3004 } 3005 } else { 3006 // Page is mapped go down 3007 // to find first not mapped page 3008 imin = imid + 1; 3009 } 3010 } 3011 3012 nbot = nbot + page_sz; 3013 3014 // Adjust stack bottom one page up if last checked page is not mapped 3015 if (mincore_return_value == -1) { 3016 nbot = nbot + page_sz; 3017 } 3018 3019 return nbot; 3020} 3021 3022 3023// Linux uses a growable mapping for the stack, and if the mapping for 3024// the stack guard pages is not removed when we detach a thread the 3025// stack cannot grow beyond the pages where the stack guard was 3026// mapped. If at some point later in the process the stack expands to 3027// that point, the Linux kernel cannot expand the stack any further 3028// because the guard pages are in the way, and a segfault occurs. 3029// 3030// However, it's essential not to split the stack region by unmapping 3031// a region (leaving a hole) that's already part of the stack mapping, 3032// so if the stack mapping has already grown beyond the guard pages at 3033// the time we create them, we have to truncate the stack mapping. 3034// So, we need to know the extent of the stack mapping when 3035// create_stack_guard_pages() is called. 3036 3037// We only need this for stacks that are growable: at the time of 3038// writing thread stacks don't use growable mappings (i.e. those 3039// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3040// only applies to the main thread. 3041 3042// If the (growable) stack mapping already extends beyond the point 3043// where we're going to put our guard pages, truncate the mapping at 3044// that point by munmap()ping it. This ensures that when we later 3045// munmap() the guard pages we don't leave a hole in the stack 3046// mapping. This only affects the main/initial thread 3047 3048bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3049 3050 if (os::Linux::is_initial_thread()) { 3051 // As we manually grow stack up to bottom inside create_attached_thread(), 3052 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3053 // we don't need to do anything special. 3054 // Check it first, before calling heavy function. 3055 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3056 unsigned char vec[1]; 3057 3058 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3059 // Fallback to slow path on all errors, including EAGAIN 3060 stack_extent = (uintptr_t) get_stack_commited_bottom( 3061 os::Linux::initial_thread_stack_bottom(), 3062 (size_t)addr - stack_extent); 3063 } 3064 3065 if (stack_extent < (uintptr_t)addr) { 3066 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3067 } 3068 } 3069 3070 return os::commit_memory(addr, size, !ExecMem); 3071} 3072 3073// If this is a growable mapping, remove the guard pages entirely by 3074// munmap()ping them. If not, just call uncommit_memory(). This only 3075// affects the main/initial thread, but guard against future OS changes 3076// It's safe to always unmap guard pages for initial thread because we 3077// always place it right after end of the mapped region 3078 3079bool os::remove_stack_guard_pages(char* addr, size_t size) { 3080 uintptr_t stack_extent, stack_base; 3081 3082 if (os::Linux::is_initial_thread()) { 3083 return ::munmap(addr, size) == 0; 3084 } 3085 3086 return os::uncommit_memory(addr, size); 3087} 3088 3089static address _highest_vm_reserved_address = NULL; 3090 3091// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3092// at 'requested_addr'. If there are existing memory mappings at the same 3093// location, however, they will be overwritten. If 'fixed' is false, 3094// 'requested_addr' is only treated as a hint, the return value may or 3095// may not start from the requested address. Unlike Linux mmap(), this 3096// function returns NULL to indicate failure. 3097static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3098 char * addr; 3099 int flags; 3100 3101 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3102 if (fixed) { 3103 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3104 flags |= MAP_FIXED; 3105 } 3106 3107 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3108 // touch an uncommitted page. Otherwise, the read/write might 3109 // succeed if we have enough swap space to back the physical page. 3110 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3111 flags, -1, 0); 3112 3113 if (addr != MAP_FAILED) { 3114 // anon_mmap() should only get called during VM initialization, 3115 // don't need lock (actually we can skip locking even it can be called 3116 // from multiple threads, because _highest_vm_reserved_address is just a 3117 // hint about the upper limit of non-stack memory regions.) 3118 if ((address)addr + bytes > _highest_vm_reserved_address) { 3119 _highest_vm_reserved_address = (address)addr + bytes; 3120 } 3121 } 3122 3123 return addr == MAP_FAILED ? NULL : addr; 3124} 3125 3126// Don't update _highest_vm_reserved_address, because there might be memory 3127// regions above addr + size. If so, releasing a memory region only creates 3128// a hole in the address space, it doesn't help prevent heap-stack collision. 3129// 3130static int anon_munmap(char * addr, size_t size) { 3131 return ::munmap(addr, size) == 0; 3132} 3133 3134char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3135 size_t alignment_hint) { 3136 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3137} 3138 3139bool os::pd_release_memory(char* addr, size_t size) { 3140 return anon_munmap(addr, size); 3141} 3142 3143static address highest_vm_reserved_address() { 3144 return _highest_vm_reserved_address; 3145} 3146 3147static bool linux_mprotect(char* addr, size_t size, int prot) { 3148 // Linux wants the mprotect address argument to be page aligned. 3149 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3150 3151 // According to SUSv3, mprotect() should only be used with mappings 3152 // established by mmap(), and mmap() always maps whole pages. Unaligned 3153 // 'addr' likely indicates problem in the VM (e.g. trying to change 3154 // protection of malloc'ed or statically allocated memory). Check the 3155 // caller if you hit this assert. 3156 assert(addr == bottom, "sanity check"); 3157 3158 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3159 return ::mprotect(bottom, size, prot) == 0; 3160} 3161 3162// Set protections specified 3163bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3164 bool is_committed) { 3165 unsigned int p = 0; 3166 switch (prot) { 3167 case MEM_PROT_NONE: p = PROT_NONE; break; 3168 case MEM_PROT_READ: p = PROT_READ; break; 3169 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3170 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3171 default: 3172 ShouldNotReachHere(); 3173 } 3174 // is_committed is unused. 3175 return linux_mprotect(addr, bytes, p); 3176} 3177 3178bool os::guard_memory(char* addr, size_t size) { 3179 return linux_mprotect(addr, size, PROT_NONE); 3180} 3181 3182bool os::unguard_memory(char* addr, size_t size) { 3183 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3184} 3185 3186bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) { 3187 bool result = false; 3188 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3189 MAP_ANONYMOUS|MAP_PRIVATE, 3190 -1, 0); 3191 if (p != MAP_FAILED) { 3192 void *aligned_p = align_ptr_up(p, page_size); 3193 3194 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3195 3196 munmap(p, page_size * 2); 3197 } 3198 3199 if (warn && !result) { 3200 warning("TransparentHugePages is not supported by the operating system."); 3201 } 3202 3203 return result; 3204} 3205 3206bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3207 bool result = false; 3208 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3209 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3210 -1, 0); 3211 3212 if (p != MAP_FAILED) { 3213 // We don't know if this really is a huge page or not. 3214 FILE *fp = fopen("/proc/self/maps", "r"); 3215 if (fp) { 3216 while (!feof(fp)) { 3217 char chars[257]; 3218 long x = 0; 3219 if (fgets(chars, sizeof(chars), fp)) { 3220 if (sscanf(chars, "%lx-%*x", &x) == 1 3221 && x == (long)p) { 3222 if (strstr (chars, "hugepage")) { 3223 result = true; 3224 break; 3225 } 3226 } 3227 } 3228 } 3229 fclose(fp); 3230 } 3231 munmap(p, page_size); 3232 } 3233 3234 if (warn && !result) { 3235 warning("HugeTLBFS is not supported by the operating system."); 3236 } 3237 3238 return result; 3239} 3240 3241/* 3242* Set the coredump_filter bits to include largepages in core dump (bit 6) 3243* 3244* From the coredump_filter documentation: 3245* 3246* - (bit 0) anonymous private memory 3247* - (bit 1) anonymous shared memory 3248* - (bit 2) file-backed private memory 3249* - (bit 3) file-backed shared memory 3250* - (bit 4) ELF header pages in file-backed private memory areas (it is 3251* effective only if the bit 2 is cleared) 3252* - (bit 5) hugetlb private memory 3253* - (bit 6) hugetlb shared memory 3254*/ 3255static void set_coredump_filter(void) { 3256 FILE *f; 3257 long cdm; 3258 3259 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3260 return; 3261 } 3262 3263 if (fscanf(f, "%lx", &cdm) != 1) { 3264 fclose(f); 3265 return; 3266 } 3267 3268 rewind(f); 3269 3270 if ((cdm & LARGEPAGES_BIT) == 0) { 3271 cdm |= LARGEPAGES_BIT; 3272 fprintf(f, "%#lx", cdm); 3273 } 3274 3275 fclose(f); 3276} 3277 3278// Large page support 3279 3280static size_t _large_page_size = 0; 3281 3282size_t os::Linux::find_large_page_size() { 3283 size_t large_page_size = 0; 3284 3285 // large_page_size on Linux is used to round up heap size. x86 uses either 3286 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3287 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3288 // page as large as 256M. 3289 // 3290 // Here we try to figure out page size by parsing /proc/meminfo and looking 3291 // for a line with the following format: 3292 // Hugepagesize: 2048 kB 3293 // 3294 // If we can't determine the value (e.g. /proc is not mounted, or the text 3295 // format has been changed), we'll use the largest page size supported by 3296 // the processor. 3297 3298#ifndef ZERO 3299 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3300 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3301#endif // ZERO 3302 3303 FILE *fp = fopen("/proc/meminfo", "r"); 3304 if (fp) { 3305 while (!feof(fp)) { 3306 int x = 0; 3307 char buf[16]; 3308 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3309 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3310 large_page_size = x * K; 3311 break; 3312 } 3313 } else { 3314 // skip to next line 3315 for (;;) { 3316 int ch = fgetc(fp); 3317 if (ch == EOF || ch == (int)'\n') break; 3318 } 3319 } 3320 } 3321 fclose(fp); 3322 } 3323 3324 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3325 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3326 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3327 proper_unit_for_byte_size(large_page_size)); 3328 } 3329 3330 return large_page_size; 3331} 3332 3333size_t os::Linux::setup_large_page_size() { 3334 _large_page_size = Linux::find_large_page_size(); 3335 const size_t default_page_size = (size_t)Linux::page_size(); 3336 if (_large_page_size > default_page_size) { 3337 _page_sizes[0] = _large_page_size; 3338 _page_sizes[1] = default_page_size; 3339 _page_sizes[2] = 0; 3340 } 3341 3342 return _large_page_size; 3343} 3344 3345bool os::Linux::setup_large_page_type(size_t page_size) { 3346 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3347 FLAG_IS_DEFAULT(UseSHM) && 3348 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3349 3350 // The type of large pages has not been specified by the user. 3351 3352 // Try UseHugeTLBFS and then UseSHM. 3353 UseHugeTLBFS = UseSHM = true; 3354 3355 // Don't try UseTransparentHugePages since there are known 3356 // performance issues with it turned on. This might change in the future. 3357 UseTransparentHugePages = false; 3358 } 3359 3360 if (UseTransparentHugePages) { 3361 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3362 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3363 UseHugeTLBFS = false; 3364 UseSHM = false; 3365 return true; 3366 } 3367 UseTransparentHugePages = false; 3368 } 3369 3370 if (UseHugeTLBFS) { 3371 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3372 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3373 UseSHM = false; 3374 return true; 3375 } 3376 UseHugeTLBFS = false; 3377 } 3378 3379 return UseSHM; 3380} 3381 3382void os::large_page_init() { 3383 if (!UseLargePages && 3384 !UseTransparentHugePages && 3385 !UseHugeTLBFS && 3386 !UseSHM) { 3387 // Not using large pages. 3388 return; 3389 } 3390 3391 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3392 // The user explicitly turned off large pages. 3393 // Ignore the rest of the large pages flags. 3394 UseTransparentHugePages = false; 3395 UseHugeTLBFS = false; 3396 UseSHM = false; 3397 return; 3398 } 3399 3400 size_t large_page_size = Linux::setup_large_page_size(); 3401 UseLargePages = Linux::setup_large_page_type(large_page_size); 3402 3403 set_coredump_filter(); 3404} 3405 3406#ifndef SHM_HUGETLB 3407#define SHM_HUGETLB 04000 3408#endif 3409 3410char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3411 // "exec" is passed in but not used. Creating the shared image for 3412 // the code cache doesn't have an SHM_X executable permission to check. 3413 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3414 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3415 3416 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3417 return NULL; // Fallback to small pages. 3418 } 3419 3420 key_t key = IPC_PRIVATE; 3421 char *addr; 3422 3423 bool warn_on_failure = UseLargePages && 3424 (!FLAG_IS_DEFAULT(UseLargePages) || 3425 !FLAG_IS_DEFAULT(UseSHM) || 3426 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 3427 ); 3428 char msg[128]; 3429 3430 // Create a large shared memory region to attach to based on size. 3431 // Currently, size is the total size of the heap 3432 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3433 if (shmid == -1) { 3434 // Possible reasons for shmget failure: 3435 // 1. shmmax is too small for Java heap. 3436 // > check shmmax value: cat /proc/sys/kernel/shmmax 3437 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3438 // 2. not enough large page memory. 3439 // > check available large pages: cat /proc/meminfo 3440 // > increase amount of large pages: 3441 // echo new_value > /proc/sys/vm/nr_hugepages 3442 // Note 1: different Linux may use different name for this property, 3443 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3444 // Note 2: it's possible there's enough physical memory available but 3445 // they are so fragmented after a long run that they can't 3446 // coalesce into large pages. Try to reserve large pages when 3447 // the system is still "fresh". 3448 if (warn_on_failure) { 3449 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3450 warning(msg); 3451 } 3452 return NULL; 3453 } 3454 3455 // attach to the region 3456 addr = (char*)shmat(shmid, req_addr, 0); 3457 int err = errno; 3458 3459 // Remove shmid. If shmat() is successful, the actual shared memory segment 3460 // will be deleted when it's detached by shmdt() or when the process 3461 // terminates. If shmat() is not successful this will remove the shared 3462 // segment immediately. 3463 shmctl(shmid, IPC_RMID, NULL); 3464 3465 if ((intptr_t)addr == -1) { 3466 if (warn_on_failure) { 3467 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3468 warning(msg); 3469 } 3470 return NULL; 3471 } 3472 3473 return addr; 3474} 3475 3476static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) { 3477 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3478 3479 bool warn_on_failure = UseLargePages && 3480 (!FLAG_IS_DEFAULT(UseLargePages) || 3481 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3482 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3483 3484 if (warn_on_failure) { 3485 char msg[128]; 3486 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3487 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3488 warning(msg); 3489 } 3490} 3491 3492char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) { 3493 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3494 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3495 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3496 3497 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3498 char* addr = (char*)::mmap(req_addr, bytes, prot, 3499 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3500 -1, 0); 3501 3502 if (addr == MAP_FAILED) { 3503 warn_on_large_pages_failure(req_addr, bytes, errno); 3504 return NULL; 3505 } 3506 3507 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3508 3509 return addr; 3510} 3511 3512char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3513 size_t large_page_size = os::large_page_size(); 3514 3515 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3516 3517 // Allocate small pages. 3518 3519 char* start; 3520 if (req_addr != NULL) { 3521 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3522 assert(is_size_aligned(bytes, alignment), "Must be"); 3523 start = os::reserve_memory(bytes, req_addr); 3524 assert(start == NULL || start == req_addr, "Must be"); 3525 } else { 3526 start = os::reserve_memory_aligned(bytes, alignment); 3527 } 3528 3529 if (start == NULL) { 3530 return NULL; 3531 } 3532 3533 assert(is_ptr_aligned(start, alignment), "Must be"); 3534 3535 // os::reserve_memory_special will record this memory area. 3536 // Need to release it here to prevent overlapping reservations. 3537 MemTracker::record_virtual_memory_release((address)start, bytes); 3538 3539 char* end = start + bytes; 3540 3541 // Find the regions of the allocated chunk that can be promoted to large pages. 3542 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3543 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3544 3545 size_t lp_bytes = lp_end - lp_start; 3546 3547 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3548 3549 if (lp_bytes == 0) { 3550 // The mapped region doesn't even span the start and the end of a large page. 3551 // Fall back to allocate a non-special area. 3552 ::munmap(start, end - start); 3553 return NULL; 3554 } 3555 3556 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3557 3558 3559 void* result; 3560 3561 if (start != lp_start) { 3562 result = ::mmap(start, lp_start - start, prot, 3563 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3564 -1, 0); 3565 if (result == MAP_FAILED) { 3566 ::munmap(lp_start, end - lp_start); 3567 return NULL; 3568 } 3569 } 3570 3571 result = ::mmap(lp_start, lp_bytes, prot, 3572 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3573 -1, 0); 3574 if (result == MAP_FAILED) { 3575 warn_on_large_pages_failure(req_addr, bytes, errno); 3576 // If the mmap above fails, the large pages region will be unmapped and we 3577 // have regions before and after with small pages. Release these regions. 3578 // 3579 // | mapped | unmapped | mapped | 3580 // ^ ^ ^ ^ 3581 // start lp_start lp_end end 3582 // 3583 ::munmap(start, lp_start - start); 3584 ::munmap(lp_end, end - lp_end); 3585 return NULL; 3586 } 3587 3588 if (lp_end != end) { 3589 result = ::mmap(lp_end, end - lp_end, prot, 3590 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3591 -1, 0); 3592 if (result == MAP_FAILED) { 3593 ::munmap(start, lp_end - start); 3594 return NULL; 3595 } 3596 } 3597 3598 return start; 3599} 3600 3601char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3602 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3603 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3604 assert(is_power_of_2(alignment), "Must be"); 3605 assert(is_power_of_2(os::large_page_size()), "Must be"); 3606 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3607 3608 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3609 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3610 } else { 3611 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3612 } 3613} 3614 3615char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3616 assert(UseLargePages, "only for large pages"); 3617 3618 char* addr; 3619 if (UseSHM) { 3620 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3621 } else { 3622 assert(UseHugeTLBFS, "must be"); 3623 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3624 } 3625 3626 if (addr != NULL) { 3627 if (UseNUMAInterleaving) { 3628 numa_make_global(addr, bytes); 3629 } 3630 3631 // The memory is committed 3632 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC); 3633 } 3634 3635 return addr; 3636} 3637 3638bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3639 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3640 return shmdt(base) == 0; 3641} 3642 3643bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3644 return pd_release_memory(base, bytes); 3645} 3646 3647bool os::release_memory_special(char* base, size_t bytes) { 3648 assert(UseLargePages, "only for large pages"); 3649 3650 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3651 3652 bool res; 3653 if (UseSHM) { 3654 res = os::Linux::release_memory_special_shm(base, bytes); 3655 } else { 3656 assert(UseHugeTLBFS, "must be"); 3657 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3658 } 3659 3660 if (res) { 3661 tkr.record((address)base, bytes); 3662 } else { 3663 tkr.discard(); 3664 } 3665 3666 return res; 3667} 3668 3669size_t os::large_page_size() { 3670 return _large_page_size; 3671} 3672 3673// With SysV SHM the entire memory region must be allocated as shared 3674// memory. 3675// HugeTLBFS allows application to commit large page memory on demand. 3676// However, when committing memory with HugeTLBFS fails, the region 3677// that was supposed to be committed will lose the old reservation 3678// and allow other threads to steal that memory region. Because of this 3679// behavior we can't commit HugeTLBFS memory. 3680bool os::can_commit_large_page_memory() { 3681 return UseTransparentHugePages; 3682} 3683 3684bool os::can_execute_large_page_memory() { 3685 return UseTransparentHugePages || UseHugeTLBFS; 3686} 3687 3688// Reserve memory at an arbitrary address, only if that area is 3689// available (and not reserved for something else). 3690 3691char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3692 const int max_tries = 10; 3693 char* base[max_tries]; 3694 size_t size[max_tries]; 3695 const size_t gap = 0x000000; 3696 3697 // Assert only that the size is a multiple of the page size, since 3698 // that's all that mmap requires, and since that's all we really know 3699 // about at this low abstraction level. If we need higher alignment, 3700 // we can either pass an alignment to this method or verify alignment 3701 // in one of the methods further up the call chain. See bug 5044738. 3702 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3703 3704 // Repeatedly allocate blocks until the block is allocated at the 3705 // right spot. Give up after max_tries. Note that reserve_memory() will 3706 // automatically update _highest_vm_reserved_address if the call is 3707 // successful. The variable tracks the highest memory address every reserved 3708 // by JVM. It is used to detect heap-stack collision if running with 3709 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3710 // space than needed, it could confuse the collision detecting code. To 3711 // solve the problem, save current _highest_vm_reserved_address and 3712 // calculate the correct value before return. 3713 address old_highest = _highest_vm_reserved_address; 3714 3715 // Linux mmap allows caller to pass an address as hint; give it a try first, 3716 // if kernel honors the hint then we can return immediately. 3717 char * addr = anon_mmap(requested_addr, bytes, false); 3718 if (addr == requested_addr) { 3719 return requested_addr; 3720 } 3721 3722 if (addr != NULL) { 3723 // mmap() is successful but it fails to reserve at the requested address 3724 anon_munmap(addr, bytes); 3725 } 3726 3727 int i; 3728 for (i = 0; i < max_tries; ++i) { 3729 base[i] = reserve_memory(bytes); 3730 3731 if (base[i] != NULL) { 3732 // Is this the block we wanted? 3733 if (base[i] == requested_addr) { 3734 size[i] = bytes; 3735 break; 3736 } 3737 3738 // Does this overlap the block we wanted? Give back the overlapped 3739 // parts and try again. 3740 3741 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3742 if (top_overlap >= 0 && top_overlap < bytes) { 3743 unmap_memory(base[i], top_overlap); 3744 base[i] += top_overlap; 3745 size[i] = bytes - top_overlap; 3746 } else { 3747 size_t bottom_overlap = base[i] + bytes - requested_addr; 3748 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3749 unmap_memory(requested_addr, bottom_overlap); 3750 size[i] = bytes - bottom_overlap; 3751 } else { 3752 size[i] = bytes; 3753 } 3754 } 3755 } 3756 } 3757 3758 // Give back the unused reserved pieces. 3759 3760 for (int j = 0; j < i; ++j) { 3761 if (base[j] != NULL) { 3762 unmap_memory(base[j], size[j]); 3763 } 3764 } 3765 3766 if (i < max_tries) { 3767 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3768 return requested_addr; 3769 } else { 3770 _highest_vm_reserved_address = old_highest; 3771 return NULL; 3772 } 3773} 3774 3775size_t os::read(int fd, void *buf, unsigned int nBytes) { 3776 return ::read(fd, buf, nBytes); 3777} 3778 3779// 3780// Short sleep, direct OS call. 3781// 3782// Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3783// sched_yield(2) will actually give up the CPU: 3784// 3785// * Alone on this pariticular CPU, keeps running. 3786// * Before the introduction of "skip_buddy" with "compat_yield" disabled 3787// (pre 2.6.39). 3788// 3789// So calling this with 0 is an alternative. 3790// 3791void os::naked_short_sleep(jlong ms) { 3792 struct timespec req; 3793 3794 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3795 req.tv_sec = 0; 3796 if (ms > 0) { 3797 req.tv_nsec = (ms % 1000) * 1000000; 3798 } 3799 else { 3800 req.tv_nsec = 1; 3801 } 3802 3803 nanosleep(&req, NULL); 3804 3805 return; 3806} 3807 3808// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3809void os::infinite_sleep() { 3810 while (true) { // sleep forever ... 3811 ::sleep(100); // ... 100 seconds at a time 3812 } 3813} 3814 3815// Used to convert frequent JVM_Yield() to nops 3816bool os::dont_yield() { 3817 return DontYieldALot; 3818} 3819 3820void os::yield() { 3821 sched_yield(); 3822} 3823 3824os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3825 3826void os::yield_all(int attempts) { 3827 // Yields to all threads, including threads with lower priorities 3828 // Threads on Linux are all with same priority. The Solaris style 3829 // os::yield_all() with nanosleep(1ms) is not necessary. 3830 sched_yield(); 3831} 3832 3833// Called from the tight loops to possibly influence time-sharing heuristics 3834void os::loop_breaker(int attempts) { 3835 os::yield_all(attempts); 3836} 3837 3838//////////////////////////////////////////////////////////////////////////////// 3839// thread priority support 3840 3841// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3842// only supports dynamic priority, static priority must be zero. For real-time 3843// applications, Linux supports SCHED_RR which allows static priority (1-99). 3844// However, for large multi-threaded applications, SCHED_RR is not only slower 3845// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3846// of 5 runs - Sep 2005). 3847// 3848// The following code actually changes the niceness of kernel-thread/LWP. It 3849// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3850// not the entire user process, and user level threads are 1:1 mapped to kernel 3851// threads. It has always been the case, but could change in the future. For 3852// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3853// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3854 3855int os::java_to_os_priority[CriticalPriority + 1] = { 3856 19, // 0 Entry should never be used 3857 3858 4, // 1 MinPriority 3859 3, // 2 3860 2, // 3 3861 3862 1, // 4 3863 0, // 5 NormPriority 3864 -1, // 6 3865 3866 -2, // 7 3867 -3, // 8 3868 -4, // 9 NearMaxPriority 3869 3870 -5, // 10 MaxPriority 3871 3872 -5 // 11 CriticalPriority 3873}; 3874 3875static int prio_init() { 3876 if (ThreadPriorityPolicy == 1) { 3877 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3878 // if effective uid is not root. Perhaps, a more elegant way of doing 3879 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3880 if (geteuid() != 0) { 3881 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3882 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3883 } 3884 ThreadPriorityPolicy = 0; 3885 } 3886 } 3887 if (UseCriticalJavaThreadPriority) { 3888 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3889 } 3890 return 0; 3891} 3892 3893OSReturn os::set_native_priority(Thread* thread, int newpri) { 3894 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3895 3896 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3897 return (ret == 0) ? OS_OK : OS_ERR; 3898} 3899 3900OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3901 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3902 *priority_ptr = java_to_os_priority[NormPriority]; 3903 return OS_OK; 3904 } 3905 3906 errno = 0; 3907 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3908 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3909} 3910 3911// Hint to the underlying OS that a task switch would not be good. 3912// Void return because it's a hint and can fail. 3913void os::hint_no_preempt() {} 3914 3915//////////////////////////////////////////////////////////////////////////////// 3916// suspend/resume support 3917 3918// the low-level signal-based suspend/resume support is a remnant from the 3919// old VM-suspension that used to be for java-suspension, safepoints etc, 3920// within hotspot. Now there is a single use-case for this: 3921// - calling get_thread_pc() on the VMThread by the flat-profiler task 3922// that runs in the watcher thread. 3923// The remaining code is greatly simplified from the more general suspension 3924// code that used to be used. 3925// 3926// The protocol is quite simple: 3927// - suspend: 3928// - sends a signal to the target thread 3929// - polls the suspend state of the osthread using a yield loop 3930// - target thread signal handler (SR_handler) sets suspend state 3931// and blocks in sigsuspend until continued 3932// - resume: 3933// - sets target osthread state to continue 3934// - sends signal to end the sigsuspend loop in the SR_handler 3935// 3936// Note that the SR_lock plays no role in this suspend/resume protocol. 3937// 3938 3939static void resume_clear_context(OSThread *osthread) { 3940 osthread->set_ucontext(NULL); 3941 osthread->set_siginfo(NULL); 3942} 3943 3944static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3945 osthread->set_ucontext(context); 3946 osthread->set_siginfo(siginfo); 3947} 3948 3949// 3950// Handler function invoked when a thread's execution is suspended or 3951// resumed. We have to be careful that only async-safe functions are 3952// called here (Note: most pthread functions are not async safe and 3953// should be avoided.) 3954// 3955// Note: sigwait() is a more natural fit than sigsuspend() from an 3956// interface point of view, but sigwait() prevents the signal hander 3957// from being run. libpthread would get very confused by not having 3958// its signal handlers run and prevents sigwait()'s use with the 3959// mutex granting granting signal. 3960// 3961// Currently only ever called on the VMThread and JavaThreads (PC sampling) 3962// 3963static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3964 // Save and restore errno to avoid confusing native code with EINTR 3965 // after sigsuspend. 3966 int old_errno = errno; 3967 3968 Thread* thread = Thread::current(); 3969 OSThread* osthread = thread->osthread(); 3970 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3971 3972 os::SuspendResume::State current = osthread->sr.state(); 3973 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3974 suspend_save_context(osthread, siginfo, context); 3975 3976 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 3977 os::SuspendResume::State state = osthread->sr.suspended(); 3978 if (state == os::SuspendResume::SR_SUSPENDED) { 3979 sigset_t suspend_set; // signals for sigsuspend() 3980 3981 // get current set of blocked signals and unblock resume signal 3982 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3983 sigdelset(&suspend_set, SR_signum); 3984 3985 sr_semaphore.signal(); 3986 // wait here until we are resumed 3987 while (1) { 3988 sigsuspend(&suspend_set); 3989 3990 os::SuspendResume::State result = osthread->sr.running(); 3991 if (result == os::SuspendResume::SR_RUNNING) { 3992 sr_semaphore.signal(); 3993 break; 3994 } 3995 } 3996 3997 } else if (state == os::SuspendResume::SR_RUNNING) { 3998 // request was cancelled, continue 3999 } else { 4000 ShouldNotReachHere(); 4001 } 4002 4003 resume_clear_context(osthread); 4004 } else if (current == os::SuspendResume::SR_RUNNING) { 4005 // request was cancelled, continue 4006 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4007 // ignore 4008 } else { 4009 // ignore 4010 } 4011 4012 errno = old_errno; 4013} 4014 4015 4016static int SR_initialize() { 4017 struct sigaction act; 4018 char *s; 4019 /* Get signal number to use for suspend/resume */ 4020 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4021 int sig = ::strtol(s, 0, 10); 4022 if (sig > 0 || sig < _NSIG) { 4023 SR_signum = sig; 4024 } 4025 } 4026 4027 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4028 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4029 4030 sigemptyset(&SR_sigset); 4031 sigaddset(&SR_sigset, SR_signum); 4032 4033 /* Set up signal handler for suspend/resume */ 4034 act.sa_flags = SA_RESTART|SA_SIGINFO; 4035 act.sa_handler = (void (*)(int)) SR_handler; 4036 4037 // SR_signum is blocked by default. 4038 // 4528190 - We also need to block pthread restart signal (32 on all 4039 // supported Linux platforms). Note that LinuxThreads need to block 4040 // this signal for all threads to work properly. So we don't have 4041 // to use hard-coded signal number when setting up the mask. 4042 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4043 4044 if (sigaction(SR_signum, &act, 0) == -1) { 4045 return -1; 4046 } 4047 4048 // Save signal flag 4049 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4050 return 0; 4051} 4052 4053static int sr_notify(OSThread* osthread) { 4054 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4055 assert_status(status == 0, status, "pthread_kill"); 4056 return status; 4057} 4058 4059// "Randomly" selected value for how long we want to spin 4060// before bailing out on suspending a thread, also how often 4061// we send a signal to a thread we want to resume 4062static const int RANDOMLY_LARGE_INTEGER = 1000000; 4063static const int RANDOMLY_LARGE_INTEGER2 = 100; 4064 4065// returns true on success and false on error - really an error is fatal 4066// but this seems the normal response to library errors 4067static bool do_suspend(OSThread* osthread) { 4068 assert(osthread->sr.is_running(), "thread should be running"); 4069 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4070 4071 // mark as suspended and send signal 4072 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4073 // failed to switch, state wasn't running? 4074 ShouldNotReachHere(); 4075 return false; 4076 } 4077 4078 if (sr_notify(osthread) != 0) { 4079 ShouldNotReachHere(); 4080 } 4081 4082 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4083 while (true) { 4084 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4085 break; 4086 } else { 4087 // timeout 4088 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4089 if (cancelled == os::SuspendResume::SR_RUNNING) { 4090 return false; 4091 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4092 // make sure that we consume the signal on the semaphore as well 4093 sr_semaphore.wait(); 4094 break; 4095 } else { 4096 ShouldNotReachHere(); 4097 return false; 4098 } 4099 } 4100 } 4101 4102 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4103 return true; 4104} 4105 4106static void do_resume(OSThread* osthread) { 4107 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4108 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4109 4110 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4111 // failed to switch to WAKEUP_REQUEST 4112 ShouldNotReachHere(); 4113 return; 4114 } 4115 4116 while (true) { 4117 if (sr_notify(osthread) == 0) { 4118 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4119 if (osthread->sr.is_running()) { 4120 return; 4121 } 4122 } 4123 } else { 4124 ShouldNotReachHere(); 4125 } 4126 } 4127 4128 guarantee(osthread->sr.is_running(), "Must be running!"); 4129} 4130 4131/////////////////////////////////////////////////////////////////////////////////// 4132// signal handling (except suspend/resume) 4133 4134// This routine may be used by user applications as a "hook" to catch signals. 4135// The user-defined signal handler must pass unrecognized signals to this 4136// routine, and if it returns true (non-zero), then the signal handler must 4137// return immediately. If the flag "abort_if_unrecognized" is true, then this 4138// routine will never retun false (zero), but instead will execute a VM panic 4139// routine kill the process. 4140// 4141// If this routine returns false, it is OK to call it again. This allows 4142// the user-defined signal handler to perform checks either before or after 4143// the VM performs its own checks. Naturally, the user code would be making 4144// a serious error if it tried to handle an exception (such as a null check 4145// or breakpoint) that the VM was generating for its own correct operation. 4146// 4147// This routine may recognize any of the following kinds of signals: 4148// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4149// It should be consulted by handlers for any of those signals. 4150// 4151// The caller of this routine must pass in the three arguments supplied 4152// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4153// field of the structure passed to sigaction(). This routine assumes that 4154// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4155// 4156// Note that the VM will print warnings if it detects conflicting signal 4157// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4158// 4159extern "C" JNIEXPORT int 4160JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 4161 void* ucontext, int abort_if_unrecognized); 4162 4163void signalHandler(int sig, siginfo_t* info, void* uc) { 4164 assert(info != NULL && uc != NULL, "it must be old kernel"); 4165 int orig_errno = errno; // Preserve errno value over signal handler. 4166 JVM_handle_linux_signal(sig, info, uc, true); 4167 errno = orig_errno; 4168} 4169 4170 4171// This boolean allows users to forward their own non-matching signals 4172// to JVM_handle_linux_signal, harmlessly. 4173bool os::Linux::signal_handlers_are_installed = false; 4174 4175// For signal-chaining 4176struct sigaction os::Linux::sigact[MAXSIGNUM]; 4177unsigned int os::Linux::sigs = 0; 4178bool os::Linux::libjsig_is_loaded = false; 4179typedef struct sigaction *(*get_signal_t)(int); 4180get_signal_t os::Linux::get_signal_action = NULL; 4181 4182struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4183 struct sigaction *actp = NULL; 4184 4185 if (libjsig_is_loaded) { 4186 // Retrieve the old signal handler from libjsig 4187 actp = (*get_signal_action)(sig); 4188 } 4189 if (actp == NULL) { 4190 // Retrieve the preinstalled signal handler from jvm 4191 actp = get_preinstalled_handler(sig); 4192 } 4193 4194 return actp; 4195} 4196 4197static bool call_chained_handler(struct sigaction *actp, int sig, 4198 siginfo_t *siginfo, void *context) { 4199 // Call the old signal handler 4200 if (actp->sa_handler == SIG_DFL) { 4201 // It's more reasonable to let jvm treat it as an unexpected exception 4202 // instead of taking the default action. 4203 return false; 4204 } else if (actp->sa_handler != SIG_IGN) { 4205 if ((actp->sa_flags & SA_NODEFER) == 0) { 4206 // automaticlly block the signal 4207 sigaddset(&(actp->sa_mask), sig); 4208 } 4209 4210 sa_handler_t hand; 4211 sa_sigaction_t sa; 4212 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4213 // retrieve the chained handler 4214 if (siginfo_flag_set) { 4215 sa = actp->sa_sigaction; 4216 } else { 4217 hand = actp->sa_handler; 4218 } 4219 4220 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4221 actp->sa_handler = SIG_DFL; 4222 } 4223 4224 // try to honor the signal mask 4225 sigset_t oset; 4226 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4227 4228 // call into the chained handler 4229 if (siginfo_flag_set) { 4230 (*sa)(sig, siginfo, context); 4231 } else { 4232 (*hand)(sig); 4233 } 4234 4235 // restore the signal mask 4236 pthread_sigmask(SIG_SETMASK, &oset, 0); 4237 } 4238 // Tell jvm's signal handler the signal is taken care of. 4239 return true; 4240} 4241 4242bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4243 bool chained = false; 4244 // signal-chaining 4245 if (UseSignalChaining) { 4246 struct sigaction *actp = get_chained_signal_action(sig); 4247 if (actp != NULL) { 4248 chained = call_chained_handler(actp, sig, siginfo, context); 4249 } 4250 } 4251 return chained; 4252} 4253 4254struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4255 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 4256 return &sigact[sig]; 4257 } 4258 return NULL; 4259} 4260 4261void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4262 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4263 sigact[sig] = oldAct; 4264 sigs |= (unsigned int)1 << sig; 4265} 4266 4267// for diagnostic 4268int os::Linux::sigflags[MAXSIGNUM]; 4269 4270int os::Linux::get_our_sigflags(int sig) { 4271 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4272 return sigflags[sig]; 4273} 4274 4275void os::Linux::set_our_sigflags(int sig, int flags) { 4276 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4277 sigflags[sig] = flags; 4278} 4279 4280void os::Linux::set_signal_handler(int sig, bool set_installed) { 4281 // Check for overwrite. 4282 struct sigaction oldAct; 4283 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4284 4285 void* oldhand = oldAct.sa_sigaction 4286 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4287 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4288 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4289 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4290 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4291 if (AllowUserSignalHandlers || !set_installed) { 4292 // Do not overwrite; user takes responsibility to forward to us. 4293 return; 4294 } else if (UseSignalChaining) { 4295 // save the old handler in jvm 4296 save_preinstalled_handler(sig, oldAct); 4297 // libjsig also interposes the sigaction() call below and saves the 4298 // old sigaction on it own. 4299 } else { 4300 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4301 "%#lx for signal %d.", (long)oldhand, sig)); 4302 } 4303 } 4304 4305 struct sigaction sigAct; 4306 sigfillset(&(sigAct.sa_mask)); 4307 sigAct.sa_handler = SIG_DFL; 4308 if (!set_installed) { 4309 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4310 } else { 4311 sigAct.sa_sigaction = signalHandler; 4312 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4313 } 4314 // Save flags, which are set by ours 4315 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4316 sigflags[sig] = sigAct.sa_flags; 4317 4318 int ret = sigaction(sig, &sigAct, &oldAct); 4319 assert(ret == 0, "check"); 4320 4321 void* oldhand2 = oldAct.sa_sigaction 4322 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4323 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4324 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4325} 4326 4327// install signal handlers for signals that HotSpot needs to 4328// handle in order to support Java-level exception handling. 4329 4330void os::Linux::install_signal_handlers() { 4331 if (!signal_handlers_are_installed) { 4332 signal_handlers_are_installed = true; 4333 4334 // signal-chaining 4335 typedef void (*signal_setting_t)(); 4336 signal_setting_t begin_signal_setting = NULL; 4337 signal_setting_t end_signal_setting = NULL; 4338 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4339 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4340 if (begin_signal_setting != NULL) { 4341 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4342 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4343 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4344 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4345 libjsig_is_loaded = true; 4346 assert(UseSignalChaining, "should enable signal-chaining"); 4347 } 4348 if (libjsig_is_loaded) { 4349 // Tell libjsig jvm is setting signal handlers 4350 (*begin_signal_setting)(); 4351 } 4352 4353 set_signal_handler(SIGSEGV, true); 4354 set_signal_handler(SIGPIPE, true); 4355 set_signal_handler(SIGBUS, true); 4356 set_signal_handler(SIGILL, true); 4357 set_signal_handler(SIGFPE, true); 4358#if defined(PPC64) 4359 set_signal_handler(SIGTRAP, true); 4360#endif 4361 set_signal_handler(SIGXFSZ, true); 4362 4363 if (libjsig_is_loaded) { 4364 // Tell libjsig jvm finishes setting signal handlers 4365 (*end_signal_setting)(); 4366 } 4367 4368 // We don't activate signal checker if libjsig is in place, we trust ourselves 4369 // and if UserSignalHandler is installed all bets are off. 4370 // Log that signal checking is off only if -verbose:jni is specified. 4371 if (CheckJNICalls) { 4372 if (libjsig_is_loaded) { 4373 if (PrintJNIResolving) { 4374 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4375 } 4376 check_signals = false; 4377 } 4378 if (AllowUserSignalHandlers) { 4379 if (PrintJNIResolving) { 4380 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4381 } 4382 check_signals = false; 4383 } 4384 } 4385 } 4386} 4387 4388// This is the fastest way to get thread cpu time on Linux. 4389// Returns cpu time (user+sys) for any thread, not only for current. 4390// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4391// It might work on 2.6.10+ with a special kernel/glibc patch. 4392// For reference, please, see IEEE Std 1003.1-2004: 4393// http://www.unix.org/single_unix_specification 4394 4395jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4396 struct timespec tp; 4397 int rc = os::Linux::clock_gettime(clockid, &tp); 4398 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4399 4400 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4401} 4402 4403///// 4404// glibc on Linux platform uses non-documented flag 4405// to indicate, that some special sort of signal 4406// trampoline is used. 4407// We will never set this flag, and we should 4408// ignore this flag in our diagnostic 4409#ifdef SIGNIFICANT_SIGNAL_MASK 4410#undef SIGNIFICANT_SIGNAL_MASK 4411#endif 4412#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4413 4414static const char* get_signal_handler_name(address handler, 4415 char* buf, int buflen) { 4416 int offset; 4417 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4418 if (found) { 4419 // skip directory names 4420 const char *p1, *p2; 4421 p1 = buf; 4422 size_t len = strlen(os::file_separator()); 4423 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4424 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4425 } else { 4426 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4427 } 4428 return buf; 4429} 4430 4431static void print_signal_handler(outputStream* st, int sig, 4432 char* buf, size_t buflen) { 4433 struct sigaction sa; 4434 4435 sigaction(sig, NULL, &sa); 4436 4437 // See comment for SIGNIFICANT_SIGNAL_MASK define 4438 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4439 4440 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4441 4442 address handler = (sa.sa_flags & SA_SIGINFO) 4443 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4444 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4445 4446 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4447 st->print("SIG_DFL"); 4448 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4449 st->print("SIG_IGN"); 4450 } else { 4451 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4452 } 4453 4454 st->print(", sa_mask[0]="); 4455 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4456 4457 address rh = VMError::get_resetted_sighandler(sig); 4458 // May be, handler was resetted by VMError? 4459 if(rh != NULL) { 4460 handler = rh; 4461 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4462 } 4463 4464 st->print(", sa_flags="); 4465 os::Posix::print_sa_flags(st, sa.sa_flags); 4466 4467 // Check: is it our handler? 4468 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4469 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4470 // It is our signal handler 4471 // check for flags, reset system-used one! 4472 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4473 st->print( 4474 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4475 os::Linux::get_our_sigflags(sig)); 4476 } 4477 } 4478 st->cr(); 4479} 4480 4481 4482#define DO_SIGNAL_CHECK(sig) \ 4483 if (!sigismember(&check_signal_done, sig)) \ 4484 os::Linux::check_signal_handler(sig) 4485 4486// This method is a periodic task to check for misbehaving JNI applications 4487// under CheckJNI, we can add any periodic checks here 4488 4489void os::run_periodic_checks() { 4490 4491 if (check_signals == false) return; 4492 4493 // SEGV and BUS if overridden could potentially prevent 4494 // generation of hs*.log in the event of a crash, debugging 4495 // such a case can be very challenging, so we absolutely 4496 // check the following for a good measure: 4497 DO_SIGNAL_CHECK(SIGSEGV); 4498 DO_SIGNAL_CHECK(SIGILL); 4499 DO_SIGNAL_CHECK(SIGFPE); 4500 DO_SIGNAL_CHECK(SIGBUS); 4501 DO_SIGNAL_CHECK(SIGPIPE); 4502 DO_SIGNAL_CHECK(SIGXFSZ); 4503#if defined(PPC64) 4504 DO_SIGNAL_CHECK(SIGTRAP); 4505#endif 4506 4507 // ReduceSignalUsage allows the user to override these handlers 4508 // see comments at the very top and jvm_solaris.h 4509 if (!ReduceSignalUsage) { 4510 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4511 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4512 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4513 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4514 } 4515 4516 DO_SIGNAL_CHECK(SR_signum); 4517 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4518} 4519 4520typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4521 4522static os_sigaction_t os_sigaction = NULL; 4523 4524void os::Linux::check_signal_handler(int sig) { 4525 char buf[O_BUFLEN]; 4526 address jvmHandler = NULL; 4527 4528 4529 struct sigaction act; 4530 if (os_sigaction == NULL) { 4531 // only trust the default sigaction, in case it has been interposed 4532 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4533 if (os_sigaction == NULL) return; 4534 } 4535 4536 os_sigaction(sig, (struct sigaction*)NULL, &act); 4537 4538 4539 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4540 4541 address thisHandler = (act.sa_flags & SA_SIGINFO) 4542 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4543 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 4544 4545 4546 switch(sig) { 4547 case SIGSEGV: 4548 case SIGBUS: 4549 case SIGFPE: 4550 case SIGPIPE: 4551 case SIGILL: 4552 case SIGXFSZ: 4553 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4554 break; 4555 4556 case SHUTDOWN1_SIGNAL: 4557 case SHUTDOWN2_SIGNAL: 4558 case SHUTDOWN3_SIGNAL: 4559 case BREAK_SIGNAL: 4560 jvmHandler = (address)user_handler(); 4561 break; 4562 4563 case INTERRUPT_SIGNAL: 4564 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4565 break; 4566 4567 default: 4568 if (sig == SR_signum) { 4569 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4570 } else { 4571 return; 4572 } 4573 break; 4574 } 4575 4576 if (thisHandler != jvmHandler) { 4577 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4578 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4579 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4580 // No need to check this sig any longer 4581 sigaddset(&check_signal_done, sig); 4582 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4583 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4584 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4585 exception_name(sig, buf, O_BUFLEN)); 4586 } 4587 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4588 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4589 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4590 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4591 // No need to check this sig any longer 4592 sigaddset(&check_signal_done, sig); 4593 } 4594 4595 // Dump all the signal 4596 if (sigismember(&check_signal_done, sig)) { 4597 print_signal_handlers(tty, buf, O_BUFLEN); 4598 } 4599} 4600 4601extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 4602 4603extern bool signal_name(int signo, char* buf, size_t len); 4604 4605const char* os::exception_name(int exception_code, char* buf, size_t size) { 4606 if (0 < exception_code && exception_code <= SIGRTMAX) { 4607 // signal 4608 if (!signal_name(exception_code, buf, size)) { 4609 jio_snprintf(buf, size, "SIG%d", exception_code); 4610 } 4611 return buf; 4612 } else { 4613 return NULL; 4614 } 4615} 4616 4617// this is called _before_ the most of global arguments have been parsed 4618void os::init(void) { 4619 char dummy; /* used to get a guess on initial stack address */ 4620// first_hrtime = gethrtime(); 4621 4622 // With LinuxThreads the JavaMain thread pid (primordial thread) 4623 // is different than the pid of the java launcher thread. 4624 // So, on Linux, the launcher thread pid is passed to the VM 4625 // via the sun.java.launcher.pid property. 4626 // Use this property instead of getpid() if it was correctly passed. 4627 // See bug 6351349. 4628 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4629 4630 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4631 4632 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4633 4634 init_random(1234567); 4635 4636 ThreadCritical::initialize(); 4637 4638 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4639 if (Linux::page_size() == -1) { 4640 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4641 strerror(errno))); 4642 } 4643 init_page_sizes((size_t) Linux::page_size()); 4644 4645 Linux::initialize_system_info(); 4646 4647 // main_thread points to the aboriginal thread 4648 Linux::_main_thread = pthread_self(); 4649 4650 Linux::clock_init(); 4651 initial_time_count = javaTimeNanos(); 4652 4653 // pthread_condattr initialization for monotonic clock 4654 int status; 4655 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4656 if ((status = pthread_condattr_init(_condattr)) != 0) { 4657 fatal(err_msg("pthread_condattr_init: %s", strerror(status))); 4658 } 4659 // Only set the clock if CLOCK_MONOTONIC is available 4660 if (os::supports_monotonic_clock()) { 4661 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4662 if (status == EINVAL) { 4663 warning("Unable to use monotonic clock with relative timed-waits" \ 4664 " - changes to the time-of-day clock may have adverse affects"); 4665 } else { 4666 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status))); 4667 } 4668 } 4669 } 4670 // else it defaults to CLOCK_REALTIME 4671 4672 pthread_mutex_init(&dl_mutex, NULL); 4673 4674 // If the pagesize of the VM is greater than 8K determine the appropriate 4675 // number of initial guard pages. The user can change this with the 4676 // command line arguments, if needed. 4677 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4678 StackYellowPages = 1; 4679 StackRedPages = 1; 4680 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4681 } 4682} 4683 4684// To install functions for atexit system call 4685extern "C" { 4686 static void perfMemory_exit_helper() { 4687 perfMemory_exit(); 4688 } 4689} 4690 4691// this is called _after_ the global arguments have been parsed 4692jint os::init_2(void) 4693{ 4694 Linux::fast_thread_clock_init(); 4695 4696 // Allocate a single page and mark it as readable for safepoint polling 4697 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4698 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 4699 4700 os::set_polling_page( polling_page ); 4701 4702#ifndef PRODUCT 4703 if(Verbose && PrintMiscellaneous) 4704 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 4705#endif 4706 4707 if (!UseMembar) { 4708 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4709 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4710 os::set_memory_serialize_page( mem_serialize_page ); 4711 4712#ifndef PRODUCT 4713 if(Verbose && PrintMiscellaneous) 4714 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 4715#endif 4716 } 4717 4718 // initialize suspend/resume support - must do this before signal_sets_init() 4719 if (SR_initialize() != 0) { 4720 perror("SR_initialize failed"); 4721 return JNI_ERR; 4722 } 4723 4724 Linux::signal_sets_init(); 4725 Linux::install_signal_handlers(); 4726 4727 // Check minimum allowable stack size for thread creation and to initialize 4728 // the java system classes, including StackOverflowError - depends on page 4729 // size. Add a page for compiler2 recursion in main thread. 4730 // Add in 2*BytesPerWord times page size to account for VM stack during 4731 // class initialization depending on 32 or 64 bit VM. 4732 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4733 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4734 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4735 4736 size_t threadStackSizeInBytes = ThreadStackSize * K; 4737 if (threadStackSizeInBytes != 0 && 4738 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4739 tty->print_cr("\nThe stack size specified is too small, " 4740 "Specify at least %dk", 4741 os::Linux::min_stack_allowed/ K); 4742 return JNI_ERR; 4743 } 4744 4745 // Make the stack size a multiple of the page size so that 4746 // the yellow/red zones can be guarded. 4747 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4748 vm_page_size())); 4749 4750 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4751 4752#if defined(IA32) 4753 workaround_expand_exec_shield_cs_limit(); 4754#endif 4755 4756 Linux::libpthread_init(); 4757 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4758 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4759 Linux::glibc_version(), Linux::libpthread_version(), 4760 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4761 } 4762 4763 if (UseNUMA) { 4764 if (!Linux::libnuma_init()) { 4765 UseNUMA = false; 4766 } else { 4767 if ((Linux::numa_max_node() < 1)) { 4768 // There's only one node(they start from 0), disable NUMA. 4769 UseNUMA = false; 4770 } 4771 } 4772 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4773 // we can make the adaptive lgrp chunk resizing work. If the user specified 4774 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4775 // disable adaptive resizing. 4776 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4777 if (FLAG_IS_DEFAULT(UseNUMA)) { 4778 UseNUMA = false; 4779 } else { 4780 if (FLAG_IS_DEFAULT(UseLargePages) && 4781 FLAG_IS_DEFAULT(UseSHM) && 4782 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4783 UseLargePages = false; 4784 } else { 4785 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing"); 4786 UseAdaptiveSizePolicy = false; 4787 UseAdaptiveNUMAChunkSizing = false; 4788 } 4789 } 4790 } 4791 if (!UseNUMA && ForceNUMA) { 4792 UseNUMA = true; 4793 } 4794 } 4795 4796 if (MaxFDLimit) { 4797 // set the number of file descriptors to max. print out error 4798 // if getrlimit/setrlimit fails but continue regardless. 4799 struct rlimit nbr_files; 4800 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4801 if (status != 0) { 4802 if (PrintMiscellaneous && (Verbose || WizardMode)) 4803 perror("os::init_2 getrlimit failed"); 4804 } else { 4805 nbr_files.rlim_cur = nbr_files.rlim_max; 4806 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4807 if (status != 0) { 4808 if (PrintMiscellaneous && (Verbose || WizardMode)) 4809 perror("os::init_2 setrlimit failed"); 4810 } 4811 } 4812 } 4813 4814 // Initialize lock used to serialize thread creation (see os::create_thread) 4815 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4816 4817 // at-exit methods are called in the reverse order of their registration. 4818 // atexit functions are called on return from main or as a result of a 4819 // call to exit(3C). There can be only 32 of these functions registered 4820 // and atexit() does not set errno. 4821 4822 if (PerfAllowAtExitRegistration) { 4823 // only register atexit functions if PerfAllowAtExitRegistration is set. 4824 // atexit functions can be delayed until process exit time, which 4825 // can be problematic for embedded VM situations. Embedded VMs should 4826 // call DestroyJavaVM() to assure that VM resources are released. 4827 4828 // note: perfMemory_exit_helper atexit function may be removed in 4829 // the future if the appropriate cleanup code can be added to the 4830 // VM_Exit VMOperation's doit method. 4831 if (atexit(perfMemory_exit_helper) != 0) { 4832 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4833 } 4834 } 4835 4836 // initialize thread priority policy 4837 prio_init(); 4838 4839 return JNI_OK; 4840} 4841 4842// this is called at the end of vm_initialization 4843void os::init_3(void) { 4844#ifdef JAVASE_EMBEDDED 4845 // Start the MemNotifyThread 4846 if (LowMemoryProtection) { 4847 MemNotifyThread::start(); 4848 } 4849 return; 4850#endif 4851} 4852 4853// Mark the polling page as unreadable 4854void os::make_polling_page_unreadable(void) { 4855 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4856 fatal("Could not disable polling page"); 4857}; 4858 4859// Mark the polling page as readable 4860void os::make_polling_page_readable(void) { 4861 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4862 fatal("Could not enable polling page"); 4863 } 4864}; 4865 4866int os::active_processor_count() { 4867 // Linux doesn't yet have a (official) notion of processor sets, 4868 // so just return the number of online processors. 4869 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4870 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4871 return online_cpus; 4872} 4873 4874void os::set_native_thread_name(const char *name) { 4875 // Not yet implemented. 4876 return; 4877} 4878 4879bool os::distribute_processes(uint length, uint* distribution) { 4880 // Not yet implemented. 4881 return false; 4882} 4883 4884bool os::bind_to_processor(uint processor_id) { 4885 // Not yet implemented. 4886 return false; 4887} 4888 4889/// 4890 4891void os::SuspendedThreadTask::internal_do_task() { 4892 if (do_suspend(_thread->osthread())) { 4893 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4894 do_task(context); 4895 do_resume(_thread->osthread()); 4896 } 4897} 4898 4899class PcFetcher : public os::SuspendedThreadTask { 4900public: 4901 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4902 ExtendedPC result(); 4903protected: 4904 void do_task(const os::SuspendedThreadTaskContext& context); 4905private: 4906 ExtendedPC _epc; 4907}; 4908 4909ExtendedPC PcFetcher::result() { 4910 guarantee(is_done(), "task is not done yet."); 4911 return _epc; 4912} 4913 4914void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4915 Thread* thread = context.thread(); 4916 OSThread* osthread = thread->osthread(); 4917 if (osthread->ucontext() != NULL) { 4918 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4919 } else { 4920 // NULL context is unexpected, double-check this is the VMThread 4921 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4922 } 4923} 4924 4925// Suspends the target using the signal mechanism and then grabs the PC before 4926// resuming the target. Used by the flat-profiler only 4927ExtendedPC os::get_thread_pc(Thread* thread) { 4928 // Make sure that it is called by the watcher for the VMThread 4929 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4930 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4931 4932 PcFetcher fetcher(thread); 4933 fetcher.run(); 4934 return fetcher.result(); 4935} 4936 4937int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4938{ 4939 if (is_NPTL()) { 4940 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4941 } else { 4942 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4943 // word back to default 64bit precision if condvar is signaled. Java 4944 // wants 53bit precision. Save and restore current value. 4945 int fpu = get_fpu_control_word(); 4946 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4947 set_fpu_control_word(fpu); 4948 return status; 4949 } 4950} 4951 4952//////////////////////////////////////////////////////////////////////////////// 4953// debug support 4954 4955bool os::find(address addr, outputStream* st) { 4956 Dl_info dlinfo; 4957 memset(&dlinfo, 0, sizeof(dlinfo)); 4958 if (dladdr(addr, &dlinfo) != 0) { 4959 st->print(PTR_FORMAT ": ", addr); 4960 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 4961 st->print("%s+%#x", dlinfo.dli_sname, 4962 addr - (intptr_t)dlinfo.dli_saddr); 4963 } else if (dlinfo.dli_fbase != NULL) { 4964 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4965 } else { 4966 st->print("<absolute address>"); 4967 } 4968 if (dlinfo.dli_fname != NULL) { 4969 st->print(" in %s", dlinfo.dli_fname); 4970 } 4971 if (dlinfo.dli_fbase != NULL) { 4972 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4973 } 4974 st->cr(); 4975 4976 if (Verbose) { 4977 // decode some bytes around the PC 4978 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 4979 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 4980 address lowest = (address) dlinfo.dli_sname; 4981 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4982 if (begin < lowest) begin = lowest; 4983 Dl_info dlinfo2; 4984 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 4985 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 4986 end = (address) dlinfo2.dli_saddr; 4987 Disassembler::decode(begin, end, st); 4988 } 4989 return true; 4990 } 4991 return false; 4992} 4993 4994//////////////////////////////////////////////////////////////////////////////// 4995// misc 4996 4997// This does not do anything on Linux. This is basically a hook for being 4998// able to use structured exception handling (thread-local exception filters) 4999// on, e.g., Win32. 5000void 5001os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 5002 JavaCallArguments* args, Thread* thread) { 5003 f(value, method, args, thread); 5004} 5005 5006void os::print_statistics() { 5007} 5008 5009int os::message_box(const char* title, const char* message) { 5010 int i; 5011 fdStream err(defaultStream::error_fd()); 5012 for (i = 0; i < 78; i++) err.print_raw("="); 5013 err.cr(); 5014 err.print_raw_cr(title); 5015 for (i = 0; i < 78; i++) err.print_raw("-"); 5016 err.cr(); 5017 err.print_raw_cr(message); 5018 for (i = 0; i < 78; i++) err.print_raw("="); 5019 err.cr(); 5020 5021 char buf[16]; 5022 // Prevent process from exiting upon "read error" without consuming all CPU 5023 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5024 5025 return buf[0] == 'y' || buf[0] == 'Y'; 5026} 5027 5028int os::stat(const char *path, struct stat *sbuf) { 5029 char pathbuf[MAX_PATH]; 5030 if (strlen(path) > MAX_PATH - 1) { 5031 errno = ENAMETOOLONG; 5032 return -1; 5033 } 5034 os::native_path(strcpy(pathbuf, path)); 5035 return ::stat(pathbuf, sbuf); 5036} 5037 5038bool os::check_heap(bool force) { 5039 return true; 5040} 5041 5042int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 5043 return ::vsnprintf(buf, count, format, args); 5044} 5045 5046// Is a (classpath) directory empty? 5047bool os::dir_is_empty(const char* path) { 5048 DIR *dir = NULL; 5049 struct dirent *ptr; 5050 5051 dir = opendir(path); 5052 if (dir == NULL) return true; 5053 5054 /* Scan the directory */ 5055 bool result = true; 5056 char buf[sizeof(struct dirent) + MAX_PATH]; 5057 while (result && (ptr = ::readdir(dir)) != NULL) { 5058 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5059 result = false; 5060 } 5061 } 5062 closedir(dir); 5063 return result; 5064} 5065 5066// This code originates from JDK's sysOpen and open64_w 5067// from src/solaris/hpi/src/system_md.c 5068 5069#ifndef O_DELETE 5070#define O_DELETE 0x10000 5071#endif 5072 5073// Open a file. Unlink the file immediately after open returns 5074// if the specified oflag has the O_DELETE flag set. 5075// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 5076 5077int os::open(const char *path, int oflag, int mode) { 5078 5079 if (strlen(path) > MAX_PATH - 1) { 5080 errno = ENAMETOOLONG; 5081 return -1; 5082 } 5083 int fd; 5084 int o_delete = (oflag & O_DELETE); 5085 oflag = oflag & ~O_DELETE; 5086 5087 fd = ::open64(path, oflag, mode); 5088 if (fd == -1) return -1; 5089 5090 //If the open succeeded, the file might still be a directory 5091 { 5092 struct stat64 buf64; 5093 int ret = ::fstat64(fd, &buf64); 5094 int st_mode = buf64.st_mode; 5095 5096 if (ret != -1) { 5097 if ((st_mode & S_IFMT) == S_IFDIR) { 5098 errno = EISDIR; 5099 ::close(fd); 5100 return -1; 5101 } 5102 } else { 5103 ::close(fd); 5104 return -1; 5105 } 5106 } 5107 5108 /* 5109 * All file descriptors that are opened in the JVM and not 5110 * specifically destined for a subprocess should have the 5111 * close-on-exec flag set. If we don't set it, then careless 3rd 5112 * party native code might fork and exec without closing all 5113 * appropriate file descriptors (e.g. as we do in closeDescriptors in 5114 * UNIXProcess.c), and this in turn might: 5115 * 5116 * - cause end-of-file to fail to be detected on some file 5117 * descriptors, resulting in mysterious hangs, or 5118 * 5119 * - might cause an fopen in the subprocess to fail on a system 5120 * suffering from bug 1085341. 5121 * 5122 * (Yes, the default setting of the close-on-exec flag is a Unix 5123 * design flaw) 5124 * 5125 * See: 5126 * 1085341: 32-bit stdio routines should support file descriptors >255 5127 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5128 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5129 */ 5130#ifdef FD_CLOEXEC 5131 { 5132 int flags = ::fcntl(fd, F_GETFD); 5133 if (flags != -1) 5134 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5135 } 5136#endif 5137 5138 if (o_delete != 0) { 5139 ::unlink(path); 5140 } 5141 return fd; 5142} 5143 5144 5145// create binary file, rewriting existing file if required 5146int os::create_binary_file(const char* path, bool rewrite_existing) { 5147 int oflags = O_WRONLY | O_CREAT; 5148 if (!rewrite_existing) { 5149 oflags |= O_EXCL; 5150 } 5151 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5152} 5153 5154// return current position of file pointer 5155jlong os::current_file_offset(int fd) { 5156 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5157} 5158 5159// move file pointer to the specified offset 5160jlong os::seek_to_file_offset(int fd, jlong offset) { 5161 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5162} 5163 5164// This code originates from JDK's sysAvailable 5165// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5166 5167int os::available(int fd, jlong *bytes) { 5168 jlong cur, end; 5169 int mode; 5170 struct stat64 buf64; 5171 5172 if (::fstat64(fd, &buf64) >= 0) { 5173 mode = buf64.st_mode; 5174 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5175 /* 5176 * XXX: is the following call interruptible? If so, this might 5177 * need to go through the INTERRUPT_IO() wrapper as for other 5178 * blocking, interruptible calls in this file. 5179 */ 5180 int n; 5181 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5182 *bytes = n; 5183 return 1; 5184 } 5185 } 5186 } 5187 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5188 return 0; 5189 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5190 return 0; 5191 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5192 return 0; 5193 } 5194 *bytes = end - cur; 5195 return 1; 5196} 5197 5198int os::socket_available(int fd, jint *pbytes) { 5199 // Linux doc says EINTR not returned, unlike Solaris 5200 int ret = ::ioctl(fd, FIONREAD, pbytes); 5201 5202 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 5203 // is expected to return 0 on failure and 1 on success to the jdk. 5204 return (ret < 0) ? 0 : 1; 5205} 5206 5207// Map a block of memory. 5208char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5209 char *addr, size_t bytes, bool read_only, 5210 bool allow_exec) { 5211 int prot; 5212 int flags = MAP_PRIVATE; 5213 5214 if (read_only) { 5215 prot = PROT_READ; 5216 } else { 5217 prot = PROT_READ | PROT_WRITE; 5218 } 5219 5220 if (allow_exec) { 5221 prot |= PROT_EXEC; 5222 } 5223 5224 if (addr != NULL) { 5225 flags |= MAP_FIXED; 5226 } 5227 5228 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5229 fd, file_offset); 5230 if (mapped_address == MAP_FAILED) { 5231 return NULL; 5232 } 5233 return mapped_address; 5234} 5235 5236 5237// Remap a block of memory. 5238char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5239 char *addr, size_t bytes, bool read_only, 5240 bool allow_exec) { 5241 // same as map_memory() on this OS 5242 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5243 allow_exec); 5244} 5245 5246 5247// Unmap a block of memory. 5248bool os::pd_unmap_memory(char* addr, size_t bytes) { 5249 return munmap(addr, bytes) == 0; 5250} 5251 5252static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5253 5254static clockid_t thread_cpu_clockid(Thread* thread) { 5255 pthread_t tid = thread->osthread()->pthread_id(); 5256 clockid_t clockid; 5257 5258 // Get thread clockid 5259 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5260 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5261 return clockid; 5262} 5263 5264// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5265// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5266// of a thread. 5267// 5268// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5269// the fast estimate available on the platform. 5270 5271jlong os::current_thread_cpu_time() { 5272 if (os::Linux::supports_fast_thread_cpu_time()) { 5273 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5274 } else { 5275 // return user + sys since the cost is the same 5276 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5277 } 5278} 5279 5280jlong os::thread_cpu_time(Thread* thread) { 5281 // consistent with what current_thread_cpu_time() returns 5282 if (os::Linux::supports_fast_thread_cpu_time()) { 5283 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5284 } else { 5285 return slow_thread_cpu_time(thread, true /* user + sys */); 5286 } 5287} 5288 5289jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5290 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5291 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5292 } else { 5293 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5294 } 5295} 5296 5297jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5298 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5299 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5300 } else { 5301 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5302 } 5303} 5304 5305// 5306// -1 on error. 5307// 5308 5309static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5310 static bool proc_task_unchecked = true; 5311 pid_t tid = thread->osthread()->thread_id(); 5312 char *s; 5313 char stat[2048]; 5314 int statlen; 5315 char proc_name[64]; 5316 int count; 5317 long sys_time, user_time; 5318 char cdummy; 5319 int idummy; 5320 long ldummy; 5321 FILE *fp; 5322 5323 snprintf(proc_name, 64, "/proc/%d/stat", tid); 5324 5325 // The /proc/<tid>/stat aggregates per-process usage on 5326 // new Linux kernels 2.6+ where NPTL is supported. 5327 // The /proc/self/task/<tid>/stat still has the per-thread usage. 5328 // See bug 6328462. 5329 // There possibly can be cases where there is no directory 5330 // /proc/self/task, so we check its availability. 5331 if (proc_task_unchecked && os::Linux::is_NPTL()) { 5332 // This is executed only once 5333 proc_task_unchecked = false; 5334 fp = fopen("/proc/self/task", "r"); 5335 if (fp != NULL) { 5336 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5337 fclose(fp); 5338 } 5339 } 5340 5341 fp = fopen(proc_name, "r"); 5342 if ( fp == NULL ) return -1; 5343 statlen = fread(stat, 1, 2047, fp); 5344 stat[statlen] = '\0'; 5345 fclose(fp); 5346 5347 // Skip pid and the command string. Note that we could be dealing with 5348 // weird command names, e.g. user could decide to rename java launcher 5349 // to "java 1.4.2 :)", then the stat file would look like 5350 // 1234 (java 1.4.2 :)) R ... ... 5351 // We don't really need to know the command string, just find the last 5352 // occurrence of ")" and then start parsing from there. See bug 4726580. 5353 s = strrchr(stat, ')'); 5354 if (s == NULL ) return -1; 5355 5356 // Skip blank chars 5357 do s++; while (isspace(*s)); 5358 5359 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5360 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5361 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5362 &user_time, &sys_time); 5363 if ( count != 13 ) return -1; 5364 if (user_sys_cpu_time) { 5365 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5366 } else { 5367 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5368 } 5369} 5370 5371void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5372 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5373 info_ptr->may_skip_backward = false; // elapsed time not wall time 5374 info_ptr->may_skip_forward = false; // elapsed time not wall time 5375 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5376} 5377 5378void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5379 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5380 info_ptr->may_skip_backward = false; // elapsed time not wall time 5381 info_ptr->may_skip_forward = false; // elapsed time not wall time 5382 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5383} 5384 5385bool os::is_thread_cpu_time_supported() { 5386 return true; 5387} 5388 5389// System loadavg support. Returns -1 if load average cannot be obtained. 5390// Linux doesn't yet have a (official) notion of processor sets, 5391// so just return the system wide load average. 5392int os::loadavg(double loadavg[], int nelem) { 5393 return ::getloadavg(loadavg, nelem); 5394} 5395 5396void os::pause() { 5397 char filename[MAX_PATH]; 5398 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5399 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5400 } else { 5401 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5402 } 5403 5404 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5405 if (fd != -1) { 5406 struct stat buf; 5407 ::close(fd); 5408 while (::stat(filename, &buf) == 0) { 5409 (void)::poll(NULL, 0, 100); 5410 } 5411 } else { 5412 jio_fprintf(stderr, 5413 "Could not open pause file '%s', continuing immediately.\n", filename); 5414 } 5415} 5416 5417 5418// Refer to the comments in os_solaris.cpp park-unpark. 5419// 5420// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5421// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5422// For specifics regarding the bug see GLIBC BUGID 261237 : 5423// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5424// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5425// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5426// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5427// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5428// and monitorenter when we're using 1-0 locking. All those operations may result in 5429// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5430// of libpthread avoids the problem, but isn't practical. 5431// 5432// Possible remedies: 5433// 5434// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5435// This is palliative and probabilistic, however. If the thread is preempted 5436// between the call to compute_abstime() and pthread_cond_timedwait(), more 5437// than the minimum period may have passed, and the abstime may be stale (in the 5438// past) resultin in a hang. Using this technique reduces the odds of a hang 5439// but the JVM is still vulnerable, particularly on heavily loaded systems. 5440// 5441// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5442// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5443// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5444// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5445// thread. 5446// 5447// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5448// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5449// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5450// This also works well. In fact it avoids kernel-level scalability impediments 5451// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5452// timers in a graceful fashion. 5453// 5454// 4. When the abstime value is in the past it appears that control returns 5455// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5456// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5457// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5458// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5459// It may be possible to avoid reinitialization by checking the return 5460// value from pthread_cond_timedwait(). In addition to reinitializing the 5461// condvar we must establish the invariant that cond_signal() is only called 5462// within critical sections protected by the adjunct mutex. This prevents 5463// cond_signal() from "seeing" a condvar that's in the midst of being 5464// reinitialized or that is corrupt. Sadly, this invariant obviates the 5465// desirable signal-after-unlock optimization that avoids futile context switching. 5466// 5467// I'm also concerned that some versions of NTPL might allocate an auxilliary 5468// structure when a condvar is used or initialized. cond_destroy() would 5469// release the helper structure. Our reinitialize-after-timedwait fix 5470// put excessive stress on malloc/free and locks protecting the c-heap. 5471// 5472// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5473// It may be possible to refine (4) by checking the kernel and NTPL verisons 5474// and only enabling the work-around for vulnerable environments. 5475 5476// utility to compute the abstime argument to timedwait: 5477// millis is the relative timeout time 5478// abstime will be the absolute timeout time 5479// TODO: replace compute_abstime() with unpackTime() 5480 5481static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5482 if (millis < 0) millis = 0; 5483 5484 jlong seconds = millis / 1000; 5485 millis %= 1000; 5486 if (seconds > 50000000) { // see man cond_timedwait(3T) 5487 seconds = 50000000; 5488 } 5489 5490 if (os::supports_monotonic_clock()) { 5491 struct timespec now; 5492 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5493 assert_status(status == 0, status, "clock_gettime"); 5494 abstime->tv_sec = now.tv_sec + seconds; 5495 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5496 if (nanos >= NANOSECS_PER_SEC) { 5497 abstime->tv_sec += 1; 5498 nanos -= NANOSECS_PER_SEC; 5499 } 5500 abstime->tv_nsec = nanos; 5501 } else { 5502 struct timeval now; 5503 int status = gettimeofday(&now, NULL); 5504 assert(status == 0, "gettimeofday"); 5505 abstime->tv_sec = now.tv_sec + seconds; 5506 long usec = now.tv_usec + millis * 1000; 5507 if (usec >= 1000000) { 5508 abstime->tv_sec += 1; 5509 usec -= 1000000; 5510 } 5511 abstime->tv_nsec = usec * 1000; 5512 } 5513 return abstime; 5514} 5515 5516 5517// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 5518// Conceptually TryPark() should be equivalent to park(0). 5519 5520int os::PlatformEvent::TryPark() { 5521 for (;;) { 5522 const int v = _Event ; 5523 guarantee ((v == 0) || (v == 1), "invariant") ; 5524 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 5525 } 5526} 5527 5528void os::PlatformEvent::park() { // AKA "down()" 5529 // Invariant: Only the thread associated with the Event/PlatformEvent 5530 // may call park(). 5531 // TODO: assert that _Assoc != NULL or _Assoc == Self 5532 int v ; 5533 for (;;) { 5534 v = _Event ; 5535 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5536 } 5537 guarantee (v >= 0, "invariant") ; 5538 if (v == 0) { 5539 // Do this the hard way by blocking ... 5540 int status = pthread_mutex_lock(_mutex); 5541 assert_status(status == 0, status, "mutex_lock"); 5542 guarantee (_nParked == 0, "invariant") ; 5543 ++ _nParked ; 5544 while (_Event < 0) { 5545 status = pthread_cond_wait(_cond, _mutex); 5546 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5547 // Treat this the same as if the wait was interrupted 5548 if (status == ETIME) { status = EINTR; } 5549 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5550 } 5551 -- _nParked ; 5552 5553 _Event = 0 ; 5554 status = pthread_mutex_unlock(_mutex); 5555 assert_status(status == 0, status, "mutex_unlock"); 5556 // Paranoia to ensure our locked and lock-free paths interact 5557 // correctly with each other. 5558 OrderAccess::fence(); 5559 } 5560 guarantee (_Event >= 0, "invariant") ; 5561} 5562 5563int os::PlatformEvent::park(jlong millis) { 5564 guarantee (_nParked == 0, "invariant") ; 5565 5566 int v ; 5567 for (;;) { 5568 v = _Event ; 5569 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5570 } 5571 guarantee (v >= 0, "invariant") ; 5572 if (v != 0) return OS_OK ; 5573 5574 // We do this the hard way, by blocking the thread. 5575 // Consider enforcing a minimum timeout value. 5576 struct timespec abst; 5577 compute_abstime(&abst, millis); 5578 5579 int ret = OS_TIMEOUT; 5580 int status = pthread_mutex_lock(_mutex); 5581 assert_status(status == 0, status, "mutex_lock"); 5582 guarantee (_nParked == 0, "invariant") ; 5583 ++_nParked ; 5584 5585 // Object.wait(timo) will return because of 5586 // (a) notification 5587 // (b) timeout 5588 // (c) thread.interrupt 5589 // 5590 // Thread.interrupt and object.notify{All} both call Event::set. 5591 // That is, we treat thread.interrupt as a special case of notification. 5592 // The underlying Solaris implementation, cond_timedwait, admits 5593 // spurious/premature wakeups, but the JLS/JVM spec prevents the 5594 // JVM from making those visible to Java code. As such, we must 5595 // filter out spurious wakeups. We assume all ETIME returns are valid. 5596 // 5597 // TODO: properly differentiate simultaneous notify+interrupt. 5598 // In that case, we should propagate the notify to another waiter. 5599 5600 while (_Event < 0) { 5601 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5602 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5603 pthread_cond_destroy (_cond); 5604 pthread_cond_init (_cond, os::Linux::condAttr()) ; 5605 } 5606 assert_status(status == 0 || status == EINTR || 5607 status == ETIME || status == ETIMEDOUT, 5608 status, "cond_timedwait"); 5609 if (!FilterSpuriousWakeups) break ; // previous semantics 5610 if (status == ETIME || status == ETIMEDOUT) break ; 5611 // We consume and ignore EINTR and spurious wakeups. 5612 } 5613 --_nParked ; 5614 if (_Event >= 0) { 5615 ret = OS_OK; 5616 } 5617 _Event = 0 ; 5618 status = pthread_mutex_unlock(_mutex); 5619 assert_status(status == 0, status, "mutex_unlock"); 5620 assert (_nParked == 0, "invariant") ; 5621 // Paranoia to ensure our locked and lock-free paths interact 5622 // correctly with each other. 5623 OrderAccess::fence(); 5624 return ret; 5625} 5626 5627void os::PlatformEvent::unpark() { 5628 // Transitions for _Event: 5629 // 0 :=> 1 5630 // 1 :=> 1 5631 // -1 :=> either 0 or 1; must signal target thread 5632 // That is, we can safely transition _Event from -1 to either 5633 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back 5634 // unpark() calls. 5635 // See also: "Semaphores in Plan 9" by Mullender & Cox 5636 // 5637 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5638 // that it will take two back-to-back park() calls for the owning 5639 // thread to block. This has the benefit of forcing a spurious return 5640 // from the first park() call after an unpark() call which will help 5641 // shake out uses of park() and unpark() without condition variables. 5642 5643 if (Atomic::xchg(1, &_Event) >= 0) return; 5644 5645 // Wait for the thread associated with the event to vacate 5646 int status = pthread_mutex_lock(_mutex); 5647 assert_status(status == 0, status, "mutex_lock"); 5648 int AnyWaiters = _nParked; 5649 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5650 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5651 AnyWaiters = 0; 5652 pthread_cond_signal(_cond); 5653 } 5654 status = pthread_mutex_unlock(_mutex); 5655 assert_status(status == 0, status, "mutex_unlock"); 5656 if (AnyWaiters != 0) { 5657 status = pthread_cond_signal(_cond); 5658 assert_status(status == 0, status, "cond_signal"); 5659 } 5660 5661 // Note that we signal() _after dropping the lock for "immortal" Events. 5662 // This is safe and avoids a common class of futile wakeups. In rare 5663 // circumstances this can cause a thread to return prematurely from 5664 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5665 // simply re-test the condition and re-park itself. 5666} 5667 5668 5669// JSR166 5670// ------------------------------------------------------- 5671 5672/* 5673 * The solaris and linux implementations of park/unpark are fairly 5674 * conservative for now, but can be improved. They currently use a 5675 * mutex/condvar pair, plus a a count. 5676 * Park decrements count if > 0, else does a condvar wait. Unpark 5677 * sets count to 1 and signals condvar. Only one thread ever waits 5678 * on the condvar. Contention seen when trying to park implies that someone 5679 * is unparking you, so don't wait. And spurious returns are fine, so there 5680 * is no need to track notifications. 5681 */ 5682 5683/* 5684 * This code is common to linux and solaris and will be moved to a 5685 * common place in dolphin. 5686 * 5687 * The passed in time value is either a relative time in nanoseconds 5688 * or an absolute time in milliseconds. Either way it has to be unpacked 5689 * into suitable seconds and nanoseconds components and stored in the 5690 * given timespec structure. 5691 * Given time is a 64-bit value and the time_t used in the timespec is only 5692 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 5693 * overflow if times way in the future are given. Further on Solaris versions 5694 * prior to 10 there is a restriction (see cond_timedwait) that the specified 5695 * number of seconds, in abstime, is less than current_time + 100,000,000. 5696 * As it will be 28 years before "now + 100000000" will overflow we can 5697 * ignore overflow and just impose a hard-limit on seconds using the value 5698 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 5699 * years from "now". 5700 */ 5701 5702static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5703 assert (time > 0, "convertTime"); 5704 time_t max_secs = 0; 5705 5706 if (!os::supports_monotonic_clock() || isAbsolute) { 5707 struct timeval now; 5708 int status = gettimeofday(&now, NULL); 5709 assert(status == 0, "gettimeofday"); 5710 5711 max_secs = now.tv_sec + MAX_SECS; 5712 5713 if (isAbsolute) { 5714 jlong secs = time / 1000; 5715 if (secs > max_secs) { 5716 absTime->tv_sec = max_secs; 5717 } else { 5718 absTime->tv_sec = secs; 5719 } 5720 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5721 } else { 5722 jlong secs = time / NANOSECS_PER_SEC; 5723 if (secs >= MAX_SECS) { 5724 absTime->tv_sec = max_secs; 5725 absTime->tv_nsec = 0; 5726 } else { 5727 absTime->tv_sec = now.tv_sec + secs; 5728 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5729 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5730 absTime->tv_nsec -= NANOSECS_PER_SEC; 5731 ++absTime->tv_sec; // note: this must be <= max_secs 5732 } 5733 } 5734 } 5735 } else { 5736 // must be relative using monotonic clock 5737 struct timespec now; 5738 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5739 assert_status(status == 0, status, "clock_gettime"); 5740 max_secs = now.tv_sec + MAX_SECS; 5741 jlong secs = time / NANOSECS_PER_SEC; 5742 if (secs >= MAX_SECS) { 5743 absTime->tv_sec = max_secs; 5744 absTime->tv_nsec = 0; 5745 } else { 5746 absTime->tv_sec = now.tv_sec + secs; 5747 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5748 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5749 absTime->tv_nsec -= NANOSECS_PER_SEC; 5750 ++absTime->tv_sec; // note: this must be <= max_secs 5751 } 5752 } 5753 } 5754 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5755 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5756 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5757 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5758} 5759 5760void Parker::park(bool isAbsolute, jlong time) { 5761 // Ideally we'd do something useful while spinning, such 5762 // as calling unpackTime(). 5763 5764 // Optional fast-path check: 5765 // Return immediately if a permit is available. 5766 // We depend on Atomic::xchg() having full barrier semantics 5767 // since we are doing a lock-free update to _counter. 5768 if (Atomic::xchg(0, &_counter) > 0) return; 5769 5770 Thread* thread = Thread::current(); 5771 assert(thread->is_Java_thread(), "Must be JavaThread"); 5772 JavaThread *jt = (JavaThread *)thread; 5773 5774 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5775 // Check interrupt before trying to wait 5776 if (Thread::is_interrupted(thread, false)) { 5777 return; 5778 } 5779 5780 // Next, demultiplex/decode time arguments 5781 timespec absTime; 5782 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5783 return; 5784 } 5785 if (time > 0) { 5786 unpackTime(&absTime, isAbsolute, time); 5787 } 5788 5789 5790 // Enter safepoint region 5791 // Beware of deadlocks such as 6317397. 5792 // The per-thread Parker:: mutex is a classic leaf-lock. 5793 // In particular a thread must never block on the Threads_lock while 5794 // holding the Parker:: mutex. If safepoints are pending both the 5795 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5796 ThreadBlockInVM tbivm(jt); 5797 5798 // Don't wait if cannot get lock since interference arises from 5799 // unblocking. Also. check interrupt before trying wait 5800 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5801 return; 5802 } 5803 5804 int status ; 5805 if (_counter > 0) { // no wait needed 5806 _counter = 0; 5807 status = pthread_mutex_unlock(_mutex); 5808 assert (status == 0, "invariant") ; 5809 // Paranoia to ensure our locked and lock-free paths interact 5810 // correctly with each other and Java-level accesses. 5811 OrderAccess::fence(); 5812 return; 5813 } 5814 5815#ifdef ASSERT 5816 // Don't catch signals while blocked; let the running threads have the signals. 5817 // (This allows a debugger to break into the running thread.) 5818 sigset_t oldsigs; 5819 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5820 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5821#endif 5822 5823 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5824 jt->set_suspend_equivalent(); 5825 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5826 5827 assert(_cur_index == -1, "invariant"); 5828 if (time == 0) { 5829 _cur_index = REL_INDEX; // arbitrary choice when not timed 5830 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ; 5831 } else { 5832 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5833 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ; 5834 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5835 pthread_cond_destroy (&_cond[_cur_index]) ; 5836 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr()); 5837 } 5838 } 5839 _cur_index = -1; 5840 assert_status(status == 0 || status == EINTR || 5841 status == ETIME || status == ETIMEDOUT, 5842 status, "cond_timedwait"); 5843 5844#ifdef ASSERT 5845 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5846#endif 5847 5848 _counter = 0 ; 5849 status = pthread_mutex_unlock(_mutex) ; 5850 assert_status(status == 0, status, "invariant") ; 5851 // Paranoia to ensure our locked and lock-free paths interact 5852 // correctly with each other and Java-level accesses. 5853 OrderAccess::fence(); 5854 5855 // If externally suspended while waiting, re-suspend 5856 if (jt->handle_special_suspend_equivalent_condition()) { 5857 jt->java_suspend_self(); 5858 } 5859} 5860 5861void Parker::unpark() { 5862 int s, status ; 5863 status = pthread_mutex_lock(_mutex); 5864 assert (status == 0, "invariant") ; 5865 s = _counter; 5866 _counter = 1; 5867 if (s < 1) { 5868 // thread might be parked 5869 if (_cur_index != -1) { 5870 // thread is definitely parked 5871 if (WorkAroundNPTLTimedWaitHang) { 5872 status = pthread_cond_signal (&_cond[_cur_index]); 5873 assert (status == 0, "invariant"); 5874 status = pthread_mutex_unlock(_mutex); 5875 assert (status == 0, "invariant"); 5876 } else { 5877 status = pthread_mutex_unlock(_mutex); 5878 assert (status == 0, "invariant"); 5879 status = pthread_cond_signal (&_cond[_cur_index]); 5880 assert (status == 0, "invariant"); 5881 } 5882 } else { 5883 pthread_mutex_unlock(_mutex); 5884 assert (status == 0, "invariant") ; 5885 } 5886 } else { 5887 pthread_mutex_unlock(_mutex); 5888 assert (status == 0, "invariant") ; 5889 } 5890} 5891 5892 5893extern char** environ; 5894 5895#ifndef __NR_fork 5896#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5897#endif 5898 5899#ifndef __NR_execve 5900#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5901#endif 5902 5903// Run the specified command in a separate process. Return its exit value, 5904// or -1 on failure (e.g. can't fork a new process). 5905// Unlike system(), this function can be called from signal handler. It 5906// doesn't block SIGINT et al. 5907int os::fork_and_exec(char* cmd) { 5908 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5909 5910 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5911 // pthread_atfork handlers and reset pthread library. All we need is a 5912 // separate process to execve. Make a direct syscall to fork process. 5913 // On IA64 there's no fork syscall, we have to use fork() and hope for 5914 // the best... 5915 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5916 IA64_ONLY(fork();) 5917 5918 if (pid < 0) { 5919 // fork failed 5920 return -1; 5921 5922 } else if (pid == 0) { 5923 // child process 5924 5925 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5926 // first to kill every thread on the thread list. Because this list is 5927 // not reset by fork() (see notes above), execve() will instead kill 5928 // every thread in the parent process. We know this is the only thread 5929 // in the new process, so make a system call directly. 5930 // IA64 should use normal execve() from glibc to match the glibc fork() 5931 // above. 5932 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5933 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5934 5935 // execve failed 5936 _exit(-1); 5937 5938 } else { 5939 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5940 // care about the actual exit code, for now. 5941 5942 int status; 5943 5944 // Wait for the child process to exit. This returns immediately if 5945 // the child has already exited. */ 5946 while (waitpid(pid, &status, 0) < 0) { 5947 switch (errno) { 5948 case ECHILD: return 0; 5949 case EINTR: break; 5950 default: return -1; 5951 } 5952 } 5953 5954 if (WIFEXITED(status)) { 5955 // The child exited normally; get its exit code. 5956 return WEXITSTATUS(status); 5957 } else if (WIFSIGNALED(status)) { 5958 // The child exited because of a signal 5959 // The best value to return is 0x80 + signal number, 5960 // because that is what all Unix shells do, and because 5961 // it allows callers to distinguish between process exit and 5962 // process death by signal. 5963 return 0x80 + WTERMSIG(status); 5964 } else { 5965 // Unknown exit code; pass it through 5966 return status; 5967 } 5968 } 5969} 5970 5971// is_headless_jre() 5972// 5973// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5974// in order to report if we are running in a headless jre 5975// 5976// Since JDK8 xawt/libmawt.so was moved into the same directory 5977// as libawt.so, and renamed libawt_xawt.so 5978// 5979bool os::is_headless_jre() { 5980 struct stat statbuf; 5981 char buf[MAXPATHLEN]; 5982 char libmawtpath[MAXPATHLEN]; 5983 const char *xawtstr = "/xawt/libmawt.so"; 5984 const char *new_xawtstr = "/libawt_xawt.so"; 5985 char *p; 5986 5987 // Get path to libjvm.so 5988 os::jvm_path(buf, sizeof(buf)); 5989 5990 // Get rid of libjvm.so 5991 p = strrchr(buf, '/'); 5992 if (p == NULL) return false; 5993 else *p = '\0'; 5994 5995 // Get rid of client or server 5996 p = strrchr(buf, '/'); 5997 if (p == NULL) return false; 5998 else *p = '\0'; 5999 6000 // check xawt/libmawt.so 6001 strcpy(libmawtpath, buf); 6002 strcat(libmawtpath, xawtstr); 6003 if (::stat(libmawtpath, &statbuf) == 0) return false; 6004 6005 // check libawt_xawt.so 6006 strcpy(libmawtpath, buf); 6007 strcat(libmawtpath, new_xawtstr); 6008 if (::stat(libmawtpath, &statbuf) == 0) return false; 6009 6010 return true; 6011} 6012 6013// Get the default path to the core file 6014// Returns the length of the string 6015int os::get_core_path(char* buffer, size_t bufferSize) { 6016 const char* p = get_current_directory(buffer, bufferSize); 6017 6018 if (p == NULL) { 6019 assert(p != NULL, "failed to get current directory"); 6020 return 0; 6021 } 6022 6023 return strlen(buffer); 6024} 6025 6026#ifdef JAVASE_EMBEDDED 6027// 6028// A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 6029// 6030MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 6031 6032// ctor 6033// 6034MemNotifyThread::MemNotifyThread(int fd): Thread() { 6035 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 6036 _fd = fd; 6037 6038 if (os::create_thread(this, os::os_thread)) { 6039 _memnotify_thread = this; 6040 os::set_priority(this, NearMaxPriority); 6041 os::start_thread(this); 6042 } 6043} 6044 6045// Where all the work gets done 6046// 6047void MemNotifyThread::run() { 6048 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 6049 6050 // Set up the select arguments 6051 fd_set rfds; 6052 if (_fd != -1) { 6053 FD_ZERO(&rfds); 6054 FD_SET(_fd, &rfds); 6055 } 6056 6057 // Now wait for the mem_notify device to wake up 6058 while (1) { 6059 // Wait for the mem_notify device to signal us.. 6060 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 6061 if (rc == -1) { 6062 perror("select!\n"); 6063 break; 6064 } else if (rc) { 6065 //ssize_t free_before = os::available_memory(); 6066 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 6067 6068 // The kernel is telling us there is not much memory left... 6069 // try to do something about that 6070 6071 // If we are not already in a GC, try one. 6072 if (!Universe::heap()->is_gc_active()) { 6073 Universe::heap()->collect(GCCause::_allocation_failure); 6074 6075 //ssize_t free_after = os::available_memory(); 6076 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 6077 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 6078 } 6079 // We might want to do something like the following if we find the GC's are not helping... 6080 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 6081 } 6082 } 6083} 6084 6085// 6086// See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 6087// 6088void MemNotifyThread::start() { 6089 int fd; 6090 fd = open ("/dev/mem_notify", O_RDONLY, 0); 6091 if (fd < 0) { 6092 return; 6093 } 6094 6095 if (memnotify_thread() == NULL) { 6096 new MemNotifyThread(fd); 6097 } 6098} 6099 6100#endif // JAVASE_EMBEDDED 6101 6102 6103/////////////// Unit tests /////////////// 6104 6105#ifndef PRODUCT 6106 6107#define test_log(...) \ 6108 do {\ 6109 if (VerboseInternalVMTests) { \ 6110 tty->print_cr(__VA_ARGS__); \ 6111 tty->flush(); \ 6112 }\ 6113 } while (false) 6114 6115class TestReserveMemorySpecial : AllStatic { 6116 public: 6117 static void small_page_write(void* addr, size_t size) { 6118 size_t page_size = os::vm_page_size(); 6119 6120 char* end = (char*)addr + size; 6121 for (char* p = (char*)addr; p < end; p += page_size) { 6122 *p = 1; 6123 } 6124 } 6125 6126 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6127 if (!UseHugeTLBFS) { 6128 return; 6129 } 6130 6131 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6132 6133 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6134 6135 if (addr != NULL) { 6136 small_page_write(addr, size); 6137 6138 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6139 } 6140 } 6141 6142 static void test_reserve_memory_special_huge_tlbfs_only() { 6143 if (!UseHugeTLBFS) { 6144 return; 6145 } 6146 6147 size_t lp = os::large_page_size(); 6148 6149 for (size_t size = lp; size <= lp * 10; size += lp) { 6150 test_reserve_memory_special_huge_tlbfs_only(size); 6151 } 6152 } 6153 6154 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) { 6155 if (!UseHugeTLBFS) { 6156 return; 6157 } 6158 6159 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")", 6160 size, alignment); 6161 6162 assert(size >= os::large_page_size(), "Incorrect input to test"); 6163 6164 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6165 6166 if (addr != NULL) { 6167 small_page_write(addr, size); 6168 6169 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6170 } 6171 } 6172 6173 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) { 6174 size_t lp = os::large_page_size(); 6175 size_t ag = os::vm_allocation_granularity(); 6176 6177 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6178 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment); 6179 } 6180 } 6181 6182 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6183 size_t lp = os::large_page_size(); 6184 size_t ag = os::vm_allocation_granularity(); 6185 6186 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp); 6187 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag); 6188 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2); 6189 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2); 6190 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag); 6191 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag); 6192 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2); 6193 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10); 6194 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2); 6195 } 6196 6197 static void test_reserve_memory_special_huge_tlbfs() { 6198 if (!UseHugeTLBFS) { 6199 return; 6200 } 6201 6202 test_reserve_memory_special_huge_tlbfs_only(); 6203 test_reserve_memory_special_huge_tlbfs_mixed(); 6204 } 6205 6206 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6207 if (!UseSHM) { 6208 return; 6209 } 6210 6211 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6212 6213 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6214 6215 if (addr != NULL) { 6216 assert(is_ptr_aligned(addr, alignment), "Check"); 6217 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6218 6219 small_page_write(addr, size); 6220 6221 os::Linux::release_memory_special_shm(addr, size); 6222 } 6223 } 6224 6225 static void test_reserve_memory_special_shm() { 6226 size_t lp = os::large_page_size(); 6227 size_t ag = os::vm_allocation_granularity(); 6228 6229 for (size_t size = ag; size < lp * 3; size += ag) { 6230 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6231 test_reserve_memory_special_shm(size, alignment); 6232 } 6233 } 6234 } 6235 6236 static void test() { 6237 test_reserve_memory_special_huge_tlbfs(); 6238 test_reserve_memory_special_shm(); 6239 } 6240}; 6241 6242void TestReserveMemorySpecial_test() { 6243 TestReserveMemorySpecial::test(); 6244} 6245 6246#endif 6247