os_linux.cpp revision 6267:8f18c8dbc6df
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 // The next steps are taken in the product version: 323 // 324 // Obtain the JAVA_HOME value from the location of libjvm.so. 325 // This library should be located at: 326 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 327 // 328 // If "/jre/lib/" appears at the right place in the path, then we 329 // assume libjvm.so is installed in a JDK and we use this path. 330 // 331 // Otherwise exit with message: "Could not create the Java virtual machine." 332 // 333 // The following extra steps are taken in the debugging version: 334 // 335 // If "/jre/lib/" does NOT appear at the right place in the path 336 // instead of exit check for $JAVA_HOME environment variable. 337 // 338 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 339 // then we append a fake suffix "hotspot/libjvm.so" to this path so 340 // it looks like libjvm.so is installed there 341 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 342 // 343 // Otherwise exit. 344 // 345 // Important note: if the location of libjvm.so changes this 346 // code needs to be changed accordingly. 347 348// See ld(1): 349// The linker uses the following search paths to locate required 350// shared libraries: 351// 1: ... 352// ... 353// 7: The default directories, normally /lib and /usr/lib. 354#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 355#define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 356#else 357#define DEFAULT_LIBPATH "/lib:/usr/lib" 358#endif 359 360// Base path of extensions installed on the system. 361#define SYS_EXT_DIR "/usr/java/packages" 362#define EXTENSIONS_DIR "/lib/ext" 363#define ENDORSED_DIR "/lib/endorsed" 364 365 // Buffer that fits several sprintfs. 366 // Note that the space for the colon and the trailing null are provided 367 // by the nulls included by the sizeof operator. 368 const size_t bufsize = 369 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends. 370 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir 371 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir 372 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 373 374 // sysclasspath, java_home, dll_dir 375 { 376 char *pslash; 377 os::jvm_path(buf, bufsize); 378 379 // Found the full path to libjvm.so. 380 // Now cut the path to <java_home>/jre if we can. 381 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so. 382 pslash = strrchr(buf, '/'); 383 if (pslash != NULL) { 384 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 385 } 386 Arguments::set_dll_dir(buf); 387 388 if (pslash != NULL) { 389 pslash = strrchr(buf, '/'); 390 if (pslash != NULL) { 391 *pslash = '\0'; // Get rid of /<arch>. 392 pslash = strrchr(buf, '/'); 393 if (pslash != NULL) { 394 *pslash = '\0'; // Get rid of /lib. 395 } 396 } 397 } 398 Arguments::set_java_home(buf); 399 set_boot_path('/', ':'); 400 } 401 402 // Where to look for native libraries. 403 // 404 // Note: Due to a legacy implementation, most of the library path 405 // is set in the launcher. This was to accomodate linking restrictions 406 // on legacy Linux implementations (which are no longer supported). 407 // Eventually, all the library path setting will be done here. 408 // 409 // However, to prevent the proliferation of improperly built native 410 // libraries, the new path component /usr/java/packages is added here. 411 // Eventually, all the library path setting will be done here. 412 { 413 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 414 // should always exist (until the legacy problem cited above is 415 // addressed). 416 const char *v = ::getenv("LD_LIBRARY_PATH"); 417 const char *v_colon = ":"; 418 if (v == NULL) { v = ""; v_colon = ""; } 419 // That's +1 for the colon and +1 for the trailing '\0'. 420 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 421 strlen(v) + 1 + 422 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1, 423 mtInternal); 424 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch); 425 Arguments::set_library_path(ld_library_path); 426 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal); 427 } 428 429 // Extensions directories. 430 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 431 Arguments::set_ext_dirs(buf); 432 433 // Endorsed standards default directory. 434 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home()); 435 Arguments::set_endorsed_dirs(buf); 436 437 FREE_C_HEAP_ARRAY(char, buf, mtInternal); 438 439#undef DEFAULT_LIBPATH 440#undef SYS_EXT_DIR 441#undef EXTENSIONS_DIR 442#undef ENDORSED_DIR 443} 444 445//////////////////////////////////////////////////////////////////////////////// 446// breakpoint support 447 448void os::breakpoint() { 449 BREAKPOINT; 450} 451 452extern "C" void breakpoint() { 453 // use debugger to set breakpoint here 454} 455 456//////////////////////////////////////////////////////////////////////////////// 457// signal support 458 459debug_only(static bool signal_sets_initialized = false); 460static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 461 462bool os::Linux::is_sig_ignored(int sig) { 463 struct sigaction oact; 464 sigaction(sig, (struct sigaction*)NULL, &oact); 465 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 466 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 467 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) 468 return true; 469 else 470 return false; 471} 472 473void os::Linux::signal_sets_init() { 474 // Should also have an assertion stating we are still single-threaded. 475 assert(!signal_sets_initialized, "Already initialized"); 476 // Fill in signals that are necessarily unblocked for all threads in 477 // the VM. Currently, we unblock the following signals: 478 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 479 // by -Xrs (=ReduceSignalUsage)); 480 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 481 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 482 // the dispositions or masks wrt these signals. 483 // Programs embedding the VM that want to use the above signals for their 484 // own purposes must, at this time, use the "-Xrs" option to prevent 485 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 486 // (See bug 4345157, and other related bugs). 487 // In reality, though, unblocking these signals is really a nop, since 488 // these signals are not blocked by default. 489 sigemptyset(&unblocked_sigs); 490 sigemptyset(&allowdebug_blocked_sigs); 491 sigaddset(&unblocked_sigs, SIGILL); 492 sigaddset(&unblocked_sigs, SIGSEGV); 493 sigaddset(&unblocked_sigs, SIGBUS); 494 sigaddset(&unblocked_sigs, SIGFPE); 495#if defined(PPC64) 496 sigaddset(&unblocked_sigs, SIGTRAP); 497#endif 498 sigaddset(&unblocked_sigs, SR_signum); 499 500 if (!ReduceSignalUsage) { 501 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 502 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 503 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 504 } 505 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 506 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 507 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 508 } 509 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 510 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 511 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 512 } 513 } 514 // Fill in signals that are blocked by all but the VM thread. 515 sigemptyset(&vm_sigs); 516 if (!ReduceSignalUsage) 517 sigaddset(&vm_sigs, BREAK_SIGNAL); 518 debug_only(signal_sets_initialized = true); 519 520} 521 522// These are signals that are unblocked while a thread is running Java. 523// (For some reason, they get blocked by default.) 524sigset_t* os::Linux::unblocked_signals() { 525 assert(signal_sets_initialized, "Not initialized"); 526 return &unblocked_sigs; 527} 528 529// These are the signals that are blocked while a (non-VM) thread is 530// running Java. Only the VM thread handles these signals. 531sigset_t* os::Linux::vm_signals() { 532 assert(signal_sets_initialized, "Not initialized"); 533 return &vm_sigs; 534} 535 536// These are signals that are blocked during cond_wait to allow debugger in 537sigset_t* os::Linux::allowdebug_blocked_signals() { 538 assert(signal_sets_initialized, "Not initialized"); 539 return &allowdebug_blocked_sigs; 540} 541 542void os::Linux::hotspot_sigmask(Thread* thread) { 543 544 //Save caller's signal mask before setting VM signal mask 545 sigset_t caller_sigmask; 546 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 547 548 OSThread* osthread = thread->osthread(); 549 osthread->set_caller_sigmask(caller_sigmask); 550 551 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 552 553 if (!ReduceSignalUsage) { 554 if (thread->is_VM_thread()) { 555 // Only the VM thread handles BREAK_SIGNAL ... 556 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 557 } else { 558 // ... all other threads block BREAK_SIGNAL 559 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 560 } 561 } 562} 563 564////////////////////////////////////////////////////////////////////////////// 565// detecting pthread library 566 567void os::Linux::libpthread_init() { 568 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 569 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 570 // generic name for earlier versions. 571 // Define macros here so we can build HotSpot on old systems. 572# ifndef _CS_GNU_LIBC_VERSION 573# define _CS_GNU_LIBC_VERSION 2 574# endif 575# ifndef _CS_GNU_LIBPTHREAD_VERSION 576# define _CS_GNU_LIBPTHREAD_VERSION 3 577# endif 578 579 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 580 if (n > 0) { 581 char *str = (char *)malloc(n, mtInternal); 582 confstr(_CS_GNU_LIBC_VERSION, str, n); 583 os::Linux::set_glibc_version(str); 584 } else { 585 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 586 static char _gnu_libc_version[32]; 587 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 588 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 589 os::Linux::set_glibc_version(_gnu_libc_version); 590 } 591 592 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 593 if (n > 0) { 594 char *str = (char *)malloc(n, mtInternal); 595 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 596 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 597 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 598 // is the case. LinuxThreads has a hard limit on max number of threads. 599 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 600 // On the other hand, NPTL does not have such a limit, sysconf() 601 // will return -1 and errno is not changed. Check if it is really NPTL. 602 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 603 strstr(str, "NPTL") && 604 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 605 free(str); 606 os::Linux::set_libpthread_version("linuxthreads"); 607 } else { 608 os::Linux::set_libpthread_version(str); 609 } 610 } else { 611 // glibc before 2.3.2 only has LinuxThreads. 612 os::Linux::set_libpthread_version("linuxthreads"); 613 } 614 615 if (strstr(libpthread_version(), "NPTL")) { 616 os::Linux::set_is_NPTL(); 617 } else { 618 os::Linux::set_is_LinuxThreads(); 619 } 620 621 // LinuxThreads have two flavors: floating-stack mode, which allows variable 622 // stack size; and fixed-stack mode. NPTL is always floating-stack. 623 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 624 os::Linux::set_is_floating_stack(); 625 } 626} 627 628///////////////////////////////////////////////////////////////////////////// 629// thread stack 630 631// Force Linux kernel to expand current thread stack. If "bottom" is close 632// to the stack guard, caller should block all signals. 633// 634// MAP_GROWSDOWN: 635// A special mmap() flag that is used to implement thread stacks. It tells 636// kernel that the memory region should extend downwards when needed. This 637// allows early versions of LinuxThreads to only mmap the first few pages 638// when creating a new thread. Linux kernel will automatically expand thread 639// stack as needed (on page faults). 640// 641// However, because the memory region of a MAP_GROWSDOWN stack can grow on 642// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 643// region, it's hard to tell if the fault is due to a legitimate stack 644// access or because of reading/writing non-exist memory (e.g. buffer 645// overrun). As a rule, if the fault happens below current stack pointer, 646// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 647// application (see Linux kernel fault.c). 648// 649// This Linux feature can cause SIGSEGV when VM bangs thread stack for 650// stack overflow detection. 651// 652// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 653// not use this flag. However, the stack of initial thread is not created 654// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 655// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 656// and then attach the thread to JVM. 657// 658// To get around the problem and allow stack banging on Linux, we need to 659// manually expand thread stack after receiving the SIGSEGV. 660// 661// There are two ways to expand thread stack to address "bottom", we used 662// both of them in JVM before 1.5: 663// 1. adjust stack pointer first so that it is below "bottom", and then 664// touch "bottom" 665// 2. mmap() the page in question 666// 667// Now alternate signal stack is gone, it's harder to use 2. For instance, 668// if current sp is already near the lower end of page 101, and we need to 669// call mmap() to map page 100, it is possible that part of the mmap() frame 670// will be placed in page 100. When page 100 is mapped, it is zero-filled. 671// That will destroy the mmap() frame and cause VM to crash. 672// 673// The following code works by adjusting sp first, then accessing the "bottom" 674// page to force a page fault. Linux kernel will then automatically expand the 675// stack mapping. 676// 677// _expand_stack_to() assumes its frame size is less than page size, which 678// should always be true if the function is not inlined. 679 680#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 681#define NOINLINE 682#else 683#define NOINLINE __attribute__ ((noinline)) 684#endif 685 686static void _expand_stack_to(address bottom) NOINLINE; 687 688static void _expand_stack_to(address bottom) { 689 address sp; 690 size_t size; 691 volatile char *p; 692 693 // Adjust bottom to point to the largest address within the same page, it 694 // gives us a one-page buffer if alloca() allocates slightly more memory. 695 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 696 bottom += os::Linux::page_size() - 1; 697 698 // sp might be slightly above current stack pointer; if that's the case, we 699 // will alloca() a little more space than necessary, which is OK. Don't use 700 // os::current_stack_pointer(), as its result can be slightly below current 701 // stack pointer, causing us to not alloca enough to reach "bottom". 702 sp = (address)&sp; 703 704 if (sp > bottom) { 705 size = sp - bottom; 706 p = (volatile char *)alloca(size); 707 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 708 p[0] = '\0'; 709 } 710} 711 712bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 713 assert(t!=NULL, "just checking"); 714 assert(t->osthread()->expanding_stack(), "expand should be set"); 715 assert(t->stack_base() != NULL, "stack_base was not initialized"); 716 717 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 718 sigset_t mask_all, old_sigset; 719 sigfillset(&mask_all); 720 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 721 _expand_stack_to(addr); 722 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 723 return true; 724 } 725 return false; 726} 727 728////////////////////////////////////////////////////////////////////////////// 729// create new thread 730 731static address highest_vm_reserved_address(); 732 733// check if it's safe to start a new thread 734static bool _thread_safety_check(Thread* thread) { 735 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 736 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 737 // Heap is mmap'ed at lower end of memory space. Thread stacks are 738 // allocated (MAP_FIXED) from high address space. Every thread stack 739 // occupies a fixed size slot (usually 2Mbytes, but user can change 740 // it to other values if they rebuild LinuxThreads). 741 // 742 // Problem with MAP_FIXED is that mmap() can still succeed even part of 743 // the memory region has already been mmap'ed. That means if we have too 744 // many threads and/or very large heap, eventually thread stack will 745 // collide with heap. 746 // 747 // Here we try to prevent heap/stack collision by comparing current 748 // stack bottom with the highest address that has been mmap'ed by JVM 749 // plus a safety margin for memory maps created by native code. 750 // 751 // This feature can be disabled by setting ThreadSafetyMargin to 0 752 // 753 if (ThreadSafetyMargin > 0) { 754 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 755 756 // not safe if our stack extends below the safety margin 757 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 758 } else { 759 return true; 760 } 761 } else { 762 // Floating stack LinuxThreads or NPTL: 763 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 764 // there's not enough space left, pthread_create() will fail. If we come 765 // here, that means enough space has been reserved for stack. 766 return true; 767 } 768} 769 770// Thread start routine for all newly created threads 771static void *java_start(Thread *thread) { 772 // Try to randomize the cache line index of hot stack frames. 773 // This helps when threads of the same stack traces evict each other's 774 // cache lines. The threads can be either from the same JVM instance, or 775 // from different JVM instances. The benefit is especially true for 776 // processors with hyperthreading technology. 777 static int counter = 0; 778 int pid = os::current_process_id(); 779 alloca(((pid ^ counter++) & 7) * 128); 780 781 ThreadLocalStorage::set_thread(thread); 782 783 OSThread* osthread = thread->osthread(); 784 Monitor* sync = osthread->startThread_lock(); 785 786 // non floating stack LinuxThreads needs extra check, see above 787 if (!_thread_safety_check(thread)) { 788 // notify parent thread 789 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 790 osthread->set_state(ZOMBIE); 791 sync->notify_all(); 792 return NULL; 793 } 794 795 // thread_id is kernel thread id (similar to Solaris LWP id) 796 osthread->set_thread_id(os::Linux::gettid()); 797 798 if (UseNUMA) { 799 int lgrp_id = os::numa_get_group_id(); 800 if (lgrp_id != -1) { 801 thread->set_lgrp_id(lgrp_id); 802 } 803 } 804 // initialize signal mask for this thread 805 os::Linux::hotspot_sigmask(thread); 806 807 // initialize floating point control register 808 os::Linux::init_thread_fpu_state(); 809 810 // handshaking with parent thread 811 { 812 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 813 814 // notify parent thread 815 osthread->set_state(INITIALIZED); 816 sync->notify_all(); 817 818 // wait until os::start_thread() 819 while (osthread->get_state() == INITIALIZED) { 820 sync->wait(Mutex::_no_safepoint_check_flag); 821 } 822 } 823 824 // call one more level start routine 825 thread->run(); 826 827 return 0; 828} 829 830bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) { 831 assert(thread->osthread() == NULL, "caller responsible"); 832 833 // Allocate the OSThread object 834 OSThread* osthread = new OSThread(NULL, NULL); 835 if (osthread == NULL) { 836 return false; 837 } 838 839 // set the correct thread state 840 osthread->set_thread_type(thr_type); 841 842 // Initial state is ALLOCATED but not INITIALIZED 843 osthread->set_state(ALLOCATED); 844 845 thread->set_osthread(osthread); 846 847 // init thread attributes 848 pthread_attr_t attr; 849 pthread_attr_init(&attr); 850 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 851 852 // stack size 853 if (os::Linux::supports_variable_stack_size()) { 854 // calculate stack size if it's not specified by caller 855 if (stack_size == 0) { 856 stack_size = os::Linux::default_stack_size(thr_type); 857 858 switch (thr_type) { 859 case os::java_thread: 860 // Java threads use ThreadStackSize which default value can be 861 // changed with the flag -Xss 862 assert (JavaThread::stack_size_at_create() > 0, "this should be set"); 863 stack_size = JavaThread::stack_size_at_create(); 864 break; 865 case os::compiler_thread: 866 if (CompilerThreadStackSize > 0) { 867 stack_size = (size_t)(CompilerThreadStackSize * K); 868 break; 869 } // else fall through: 870 // use VMThreadStackSize if CompilerThreadStackSize is not defined 871 case os::vm_thread: 872 case os::pgc_thread: 873 case os::cgc_thread: 874 case os::watcher_thread: 875 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 876 break; 877 } 878 } 879 880 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 881 pthread_attr_setstacksize(&attr, stack_size); 882 } else { 883 // let pthread_create() pick the default value. 884 } 885 886 // glibc guard page 887 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 888 889 ThreadState state; 890 891 { 892 // Serialize thread creation if we are running with fixed stack LinuxThreads 893 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 894 if (lock) { 895 os::Linux::createThread_lock()->lock_without_safepoint_check(); 896 } 897 898 pthread_t tid; 899 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 900 901 pthread_attr_destroy(&attr); 902 903 if (ret != 0) { 904 if (PrintMiscellaneous && (Verbose || WizardMode)) { 905 perror("pthread_create()"); 906 } 907 // Need to clean up stuff we've allocated so far 908 thread->set_osthread(NULL); 909 delete osthread; 910 if (lock) os::Linux::createThread_lock()->unlock(); 911 return false; 912 } 913 914 // Store pthread info into the OSThread 915 osthread->set_pthread_id(tid); 916 917 // Wait until child thread is either initialized or aborted 918 { 919 Monitor* sync_with_child = osthread->startThread_lock(); 920 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 921 while ((state = osthread->get_state()) == ALLOCATED) { 922 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 923 } 924 } 925 926 if (lock) { 927 os::Linux::createThread_lock()->unlock(); 928 } 929 } 930 931 // Aborted due to thread limit being reached 932 if (state == ZOMBIE) { 933 thread->set_osthread(NULL); 934 delete osthread; 935 return false; 936 } 937 938 // The thread is returned suspended (in state INITIALIZED), 939 // and is started higher up in the call chain 940 assert(state == INITIALIZED, "race condition"); 941 return true; 942} 943 944///////////////////////////////////////////////////////////////////////////// 945// attach existing thread 946 947// bootstrap the main thread 948bool os::create_main_thread(JavaThread* thread) { 949 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 950 return create_attached_thread(thread); 951} 952 953bool os::create_attached_thread(JavaThread* thread) { 954#ifdef ASSERT 955 thread->verify_not_published(); 956#endif 957 958 // Allocate the OSThread object 959 OSThread* osthread = new OSThread(NULL, NULL); 960 961 if (osthread == NULL) { 962 return false; 963 } 964 965 // Store pthread info into the OSThread 966 osthread->set_thread_id(os::Linux::gettid()); 967 osthread->set_pthread_id(::pthread_self()); 968 969 // initialize floating point control register 970 os::Linux::init_thread_fpu_state(); 971 972 // Initial thread state is RUNNABLE 973 osthread->set_state(RUNNABLE); 974 975 thread->set_osthread(osthread); 976 977 if (UseNUMA) { 978 int lgrp_id = os::numa_get_group_id(); 979 if (lgrp_id != -1) { 980 thread->set_lgrp_id(lgrp_id); 981 } 982 } 983 984 if (os::Linux::is_initial_thread()) { 985 // If current thread is initial thread, its stack is mapped on demand, 986 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 987 // the entire stack region to avoid SEGV in stack banging. 988 // It is also useful to get around the heap-stack-gap problem on SuSE 989 // kernel (see 4821821 for details). We first expand stack to the top 990 // of yellow zone, then enable stack yellow zone (order is significant, 991 // enabling yellow zone first will crash JVM on SuSE Linux), so there 992 // is no gap between the last two virtual memory regions. 993 994 JavaThread *jt = (JavaThread *)thread; 995 address addr = jt->stack_yellow_zone_base(); 996 assert(addr != NULL, "initialization problem?"); 997 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 998 999 osthread->set_expanding_stack(); 1000 os::Linux::manually_expand_stack(jt, addr); 1001 osthread->clear_expanding_stack(); 1002 } 1003 1004 // initialize signal mask for this thread 1005 // and save the caller's signal mask 1006 os::Linux::hotspot_sigmask(thread); 1007 1008 return true; 1009} 1010 1011void os::pd_start_thread(Thread* thread) { 1012 OSThread * osthread = thread->osthread(); 1013 assert(osthread->get_state() != INITIALIZED, "just checking"); 1014 Monitor* sync_with_child = osthread->startThread_lock(); 1015 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 1016 sync_with_child->notify(); 1017} 1018 1019// Free Linux resources related to the OSThread 1020void os::free_thread(OSThread* osthread) { 1021 assert(osthread != NULL, "osthread not set"); 1022 1023 if (Thread::current()->osthread() == osthread) { 1024 // Restore caller's signal mask 1025 sigset_t sigmask = osthread->caller_sigmask(); 1026 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 1027 } 1028 1029 delete osthread; 1030} 1031 1032////////////////////////////////////////////////////////////////////////////// 1033// thread local storage 1034 1035int os::allocate_thread_local_storage() { 1036 pthread_key_t key; 1037 int rslt = pthread_key_create(&key, NULL); 1038 assert(rslt == 0, "cannot allocate thread local storage"); 1039 return (int)key; 1040} 1041 1042// Note: This is currently not used by VM, as we don't destroy TLS key 1043// on VM exit. 1044void os::free_thread_local_storage(int index) { 1045 int rslt = pthread_key_delete((pthread_key_t)index); 1046 assert(rslt == 0, "invalid index"); 1047} 1048 1049void os::thread_local_storage_at_put(int index, void* value) { 1050 int rslt = pthread_setspecific((pthread_key_t)index, value); 1051 assert(rslt == 0, "pthread_setspecific failed"); 1052} 1053 1054extern "C" Thread* get_thread() { 1055 return ThreadLocalStorage::thread(); 1056} 1057 1058////////////////////////////////////////////////////////////////////////////// 1059// initial thread 1060 1061// Check if current thread is the initial thread, similar to Solaris thr_main. 1062bool os::Linux::is_initial_thread(void) { 1063 char dummy; 1064 // If called before init complete, thread stack bottom will be null. 1065 // Can be called if fatal error occurs before initialization. 1066 if (initial_thread_stack_bottom() == NULL) return false; 1067 assert(initial_thread_stack_bottom() != NULL && 1068 initial_thread_stack_size() != 0, 1069 "os::init did not locate initial thread's stack region"); 1070 if ((address)&dummy >= initial_thread_stack_bottom() && 1071 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1072 return true; 1073 else return false; 1074} 1075 1076// Find the virtual memory area that contains addr 1077static bool find_vma(address addr, address* vma_low, address* vma_high) { 1078 FILE *fp = fopen("/proc/self/maps", "r"); 1079 if (fp) { 1080 address low, high; 1081 while (!feof(fp)) { 1082 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1083 if (low <= addr && addr < high) { 1084 if (vma_low) *vma_low = low; 1085 if (vma_high) *vma_high = high; 1086 fclose (fp); 1087 return true; 1088 } 1089 } 1090 for (;;) { 1091 int ch = fgetc(fp); 1092 if (ch == EOF || ch == (int)'\n') break; 1093 } 1094 } 1095 fclose(fp); 1096 } 1097 return false; 1098} 1099 1100// Locate initial thread stack. This special handling of initial thread stack 1101// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1102// bogus value for initial thread. 1103void os::Linux::capture_initial_stack(size_t max_size) { 1104 // stack size is the easy part, get it from RLIMIT_STACK 1105 size_t stack_size; 1106 struct rlimit rlim; 1107 getrlimit(RLIMIT_STACK, &rlim); 1108 stack_size = rlim.rlim_cur; 1109 1110 // 6308388: a bug in ld.so will relocate its own .data section to the 1111 // lower end of primordial stack; reduce ulimit -s value a little bit 1112 // so we won't install guard page on ld.so's data section. 1113 stack_size -= 2 * page_size(); 1114 1115 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1116 // 7.1, in both cases we will get 2G in return value. 1117 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1118 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1119 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1120 // in case other parts in glibc still assumes 2M max stack size. 1121 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1122 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1123 if (stack_size > 2 * K * K IA64_ONLY(*2)) 1124 stack_size = 2 * K * K IA64_ONLY(*2); 1125 // Try to figure out where the stack base (top) is. This is harder. 1126 // 1127 // When an application is started, glibc saves the initial stack pointer in 1128 // a global variable "__libc_stack_end", which is then used by system 1129 // libraries. __libc_stack_end should be pretty close to stack top. The 1130 // variable is available since the very early days. However, because it is 1131 // a private interface, it could disappear in the future. 1132 // 1133 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1134 // to __libc_stack_end, it is very close to stack top, but isn't the real 1135 // stack top. Note that /proc may not exist if VM is running as a chroot 1136 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1137 // /proc/<pid>/stat could change in the future (though unlikely). 1138 // 1139 // We try __libc_stack_end first. If that doesn't work, look for 1140 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1141 // as a hint, which should work well in most cases. 1142 1143 uintptr_t stack_start; 1144 1145 // try __libc_stack_end first 1146 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1147 if (p && *p) { 1148 stack_start = *p; 1149 } else { 1150 // see if we can get the start_stack field from /proc/self/stat 1151 FILE *fp; 1152 int pid; 1153 char state; 1154 int ppid; 1155 int pgrp; 1156 int session; 1157 int nr; 1158 int tpgrp; 1159 unsigned long flags; 1160 unsigned long minflt; 1161 unsigned long cminflt; 1162 unsigned long majflt; 1163 unsigned long cmajflt; 1164 unsigned long utime; 1165 unsigned long stime; 1166 long cutime; 1167 long cstime; 1168 long prio; 1169 long nice; 1170 long junk; 1171 long it_real; 1172 uintptr_t start; 1173 uintptr_t vsize; 1174 intptr_t rss; 1175 uintptr_t rsslim; 1176 uintptr_t scodes; 1177 uintptr_t ecode; 1178 int i; 1179 1180 // Figure what the primordial thread stack base is. Code is inspired 1181 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1182 // followed by command name surrounded by parentheses, state, etc. 1183 char stat[2048]; 1184 int statlen; 1185 1186 fp = fopen("/proc/self/stat", "r"); 1187 if (fp) { 1188 statlen = fread(stat, 1, 2047, fp); 1189 stat[statlen] = '\0'; 1190 fclose(fp); 1191 1192 // Skip pid and the command string. Note that we could be dealing with 1193 // weird command names, e.g. user could decide to rename java launcher 1194 // to "java 1.4.2 :)", then the stat file would look like 1195 // 1234 (java 1.4.2 :)) R ... ... 1196 // We don't really need to know the command string, just find the last 1197 // occurrence of ")" and then start parsing from there. See bug 4726580. 1198 char * s = strrchr(stat, ')'); 1199 1200 i = 0; 1201 if (s) { 1202 // Skip blank chars 1203 do s++; while (isspace(*s)); 1204 1205#define _UFM UINTX_FORMAT 1206#define _DFM INTX_FORMAT 1207 1208 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1209 /* 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 */ 1210 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, 1211 &state, /* 3 %c */ 1212 &ppid, /* 4 %d */ 1213 &pgrp, /* 5 %d */ 1214 &session, /* 6 %d */ 1215 &nr, /* 7 %d */ 1216 &tpgrp, /* 8 %d */ 1217 &flags, /* 9 %lu */ 1218 &minflt, /* 10 %lu */ 1219 &cminflt, /* 11 %lu */ 1220 &majflt, /* 12 %lu */ 1221 &cmajflt, /* 13 %lu */ 1222 &utime, /* 14 %lu */ 1223 &stime, /* 15 %lu */ 1224 &cutime, /* 16 %ld */ 1225 &cstime, /* 17 %ld */ 1226 &prio, /* 18 %ld */ 1227 &nice, /* 19 %ld */ 1228 &junk, /* 20 %ld */ 1229 &it_real, /* 21 %ld */ 1230 &start, /* 22 UINTX_FORMAT */ 1231 &vsize, /* 23 UINTX_FORMAT */ 1232 &rss, /* 24 INTX_FORMAT */ 1233 &rsslim, /* 25 UINTX_FORMAT */ 1234 &scodes, /* 26 UINTX_FORMAT */ 1235 &ecode, /* 27 UINTX_FORMAT */ 1236 &stack_start); /* 28 UINTX_FORMAT */ 1237 } 1238 1239#undef _UFM 1240#undef _DFM 1241 1242 if (i != 28 - 2) { 1243 assert(false, "Bad conversion from /proc/self/stat"); 1244 // product mode - assume we are the initial thread, good luck in the 1245 // embedded case. 1246 warning("Can't detect initial thread stack location - bad conversion"); 1247 stack_start = (uintptr_t) &rlim; 1248 } 1249 } else { 1250 // For some reason we can't open /proc/self/stat (for example, running on 1251 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1252 // most cases, so don't abort: 1253 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1254 stack_start = (uintptr_t) &rlim; 1255 } 1256 } 1257 1258 // Now we have a pointer (stack_start) very close to the stack top, the 1259 // next thing to do is to figure out the exact location of stack top. We 1260 // can find out the virtual memory area that contains stack_start by 1261 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1262 // and its upper limit is the real stack top. (again, this would fail if 1263 // running inside chroot, because /proc may not exist.) 1264 1265 uintptr_t stack_top; 1266 address low, high; 1267 if (find_vma((address)stack_start, &low, &high)) { 1268 // success, "high" is the true stack top. (ignore "low", because initial 1269 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1270 stack_top = (uintptr_t)high; 1271 } else { 1272 // failed, likely because /proc/self/maps does not exist 1273 warning("Can't detect initial thread stack location - find_vma failed"); 1274 // best effort: stack_start is normally within a few pages below the real 1275 // stack top, use it as stack top, and reduce stack size so we won't put 1276 // guard page outside stack. 1277 stack_top = stack_start; 1278 stack_size -= 16 * page_size(); 1279 } 1280 1281 // stack_top could be partially down the page so align it 1282 stack_top = align_size_up(stack_top, page_size()); 1283 1284 if (max_size && stack_size > max_size) { 1285 _initial_thread_stack_size = max_size; 1286 } else { 1287 _initial_thread_stack_size = stack_size; 1288 } 1289 1290 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1291 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1292} 1293 1294//////////////////////////////////////////////////////////////////////////////// 1295// time support 1296 1297// Time since start-up in seconds to a fine granularity. 1298// Used by VMSelfDestructTimer and the MemProfiler. 1299double os::elapsedTime() { 1300 1301 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1302} 1303 1304jlong os::elapsed_counter() { 1305 return javaTimeNanos() - initial_time_count; 1306} 1307 1308jlong os::elapsed_frequency() { 1309 return NANOSECS_PER_SEC; // nanosecond resolution 1310} 1311 1312bool os::supports_vtime() { return true; } 1313bool os::enable_vtime() { return false; } 1314bool os::vtime_enabled() { return false; } 1315 1316double os::elapsedVTime() { 1317 struct rusage usage; 1318 int retval = getrusage(RUSAGE_THREAD, &usage); 1319 if (retval == 0) { 1320 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); 1321 } else { 1322 // better than nothing, but not much 1323 return elapsedTime(); 1324 } 1325} 1326 1327jlong os::javaTimeMillis() { 1328 timeval time; 1329 int status = gettimeofday(&time, NULL); 1330 assert(status != -1, "linux error"); 1331 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1332} 1333 1334#ifndef CLOCK_MONOTONIC 1335#define CLOCK_MONOTONIC (1) 1336#endif 1337 1338void os::Linux::clock_init() { 1339 // we do dlopen's in this particular order due to bug in linux 1340 // dynamical loader (see 6348968) leading to crash on exit 1341 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1342 if (handle == NULL) { 1343 handle = dlopen("librt.so", RTLD_LAZY); 1344 } 1345 1346 if (handle) { 1347 int (*clock_getres_func)(clockid_t, struct timespec*) = 1348 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1349 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1350 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1351 if (clock_getres_func && clock_gettime_func) { 1352 // See if monotonic clock is supported by the kernel. Note that some 1353 // early implementations simply return kernel jiffies (updated every 1354 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1355 // for nano time (though the monotonic property is still nice to have). 1356 // It's fixed in newer kernels, however clock_getres() still returns 1357 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1358 // resolution for now. Hopefully as people move to new kernels, this 1359 // won't be a problem. 1360 struct timespec res; 1361 struct timespec tp; 1362 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1363 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1364 // yes, monotonic clock is supported 1365 _clock_gettime = clock_gettime_func; 1366 return; 1367 } else { 1368 // close librt if there is no monotonic clock 1369 dlclose(handle); 1370 } 1371 } 1372 } 1373 warning("No monotonic clock was available - timed services may " \ 1374 "be adversely affected if the time-of-day clock changes"); 1375} 1376 1377#ifndef SYS_clock_getres 1378 1379#if defined(IA32) || defined(AMD64) 1380#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1381#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1382#else 1383#warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1384#define sys_clock_getres(x,y) -1 1385#endif 1386 1387#else 1388#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1389#endif 1390 1391void os::Linux::fast_thread_clock_init() { 1392 if (!UseLinuxPosixThreadCPUClocks) { 1393 return; 1394 } 1395 clockid_t clockid; 1396 struct timespec tp; 1397 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1398 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1399 1400 // Switch to using fast clocks for thread cpu time if 1401 // the sys_clock_getres() returns 0 error code. 1402 // Note, that some kernels may support the current thread 1403 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1404 // returned by the pthread_getcpuclockid(). 1405 // If the fast Posix clocks are supported then the sys_clock_getres() 1406 // must return at least tp.tv_sec == 0 which means a resolution 1407 // better than 1 sec. This is extra check for reliability. 1408 1409 if(pthread_getcpuclockid_func && 1410 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1411 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1412 1413 _supports_fast_thread_cpu_time = true; 1414 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1415 } 1416} 1417 1418jlong os::javaTimeNanos() { 1419 if (os::supports_monotonic_clock()) { 1420 struct timespec tp; 1421 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1422 assert(status == 0, "gettime error"); 1423 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1424 return result; 1425 } else { 1426 timeval time; 1427 int status = gettimeofday(&time, NULL); 1428 assert(status != -1, "linux error"); 1429 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1430 return 1000 * usecs; 1431 } 1432} 1433 1434void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1435 if (os::supports_monotonic_clock()) { 1436 info_ptr->max_value = ALL_64_BITS; 1437 1438 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1439 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1440 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1441 } else { 1442 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1443 info_ptr->max_value = ALL_64_BITS; 1444 1445 // gettimeofday is a real time clock so it skips 1446 info_ptr->may_skip_backward = true; 1447 info_ptr->may_skip_forward = true; 1448 } 1449 1450 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1451} 1452 1453// Return the real, user, and system times in seconds from an 1454// arbitrary fixed point in the past. 1455bool os::getTimesSecs(double* process_real_time, 1456 double* process_user_time, 1457 double* process_system_time) { 1458 struct tms ticks; 1459 clock_t real_ticks = times(&ticks); 1460 1461 if (real_ticks == (clock_t) (-1)) { 1462 return false; 1463 } else { 1464 double ticks_per_second = (double) clock_tics_per_sec; 1465 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1466 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1467 *process_real_time = ((double) real_ticks) / ticks_per_second; 1468 1469 return true; 1470 } 1471} 1472 1473 1474char * os::local_time_string(char *buf, size_t buflen) { 1475 struct tm t; 1476 time_t long_time; 1477 time(&long_time); 1478 localtime_r(&long_time, &t); 1479 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1480 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1481 t.tm_hour, t.tm_min, t.tm_sec); 1482 return buf; 1483} 1484 1485struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1486 return localtime_r(clock, res); 1487} 1488 1489//////////////////////////////////////////////////////////////////////////////// 1490// runtime exit support 1491 1492// Note: os::shutdown() might be called very early during initialization, or 1493// called from signal handler. Before adding something to os::shutdown(), make 1494// sure it is async-safe and can handle partially initialized VM. 1495void os::shutdown() { 1496 1497 // allow PerfMemory to attempt cleanup of any persistent resources 1498 perfMemory_exit(); 1499 1500 // needs to remove object in file system 1501 AttachListener::abort(); 1502 1503 // flush buffered output, finish log files 1504 ostream_abort(); 1505 1506 // Check for abort hook 1507 abort_hook_t abort_hook = Arguments::abort_hook(); 1508 if (abort_hook != NULL) { 1509 abort_hook(); 1510 } 1511 1512} 1513 1514// Note: os::abort() might be called very early during initialization, or 1515// called from signal handler. Before adding something to os::abort(), make 1516// sure it is async-safe and can handle partially initialized VM. 1517void os::abort(bool dump_core) { 1518 os::shutdown(); 1519 if (dump_core) { 1520#ifndef PRODUCT 1521 fdStream out(defaultStream::output_fd()); 1522 out.print_raw("Current thread is "); 1523 char buf[16]; 1524 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1525 out.print_raw_cr(buf); 1526 out.print_raw_cr("Dumping core ..."); 1527#endif 1528 ::abort(); // dump core 1529 } 1530 1531 ::exit(1); 1532} 1533 1534// Die immediately, no exit hook, no abort hook, no cleanup. 1535void os::die() { 1536 // _exit() on LinuxThreads only kills current thread 1537 ::abort(); 1538} 1539 1540// unused on linux for now. 1541void os::set_error_file(const char *logfile) {} 1542 1543 1544// This method is a copy of JDK's sysGetLastErrorString 1545// from src/solaris/hpi/src/system_md.c 1546 1547size_t os::lasterror(char *buf, size_t len) { 1548 1549 if (errno == 0) return 0; 1550 1551 const char *s = ::strerror(errno); 1552 size_t n = ::strlen(s); 1553 if (n >= len) { 1554 n = len - 1; 1555 } 1556 ::strncpy(buf, s, n); 1557 buf[n] = '\0'; 1558 return n; 1559} 1560 1561intx os::current_thread_id() { return (intx)pthread_self(); } 1562int os::current_process_id() { 1563 1564 // Under the old linux thread library, linux gives each thread 1565 // its own process id. Because of this each thread will return 1566 // a different pid if this method were to return the result 1567 // of getpid(2). Linux provides no api that returns the pid 1568 // of the launcher thread for the vm. This implementation 1569 // returns a unique pid, the pid of the launcher thread 1570 // that starts the vm 'process'. 1571 1572 // Under the NPTL, getpid() returns the same pid as the 1573 // launcher thread rather than a unique pid per thread. 1574 // Use gettid() if you want the old pre NPTL behaviour. 1575 1576 // if you are looking for the result of a call to getpid() that 1577 // returns a unique pid for the calling thread, then look at the 1578 // OSThread::thread_id() method in osThread_linux.hpp file 1579 1580 return (int)(_initial_pid ? _initial_pid : getpid()); 1581} 1582 1583// DLL functions 1584 1585const char* os::dll_file_extension() { return ".so"; } 1586 1587// This must be hard coded because it's the system's temporary 1588// directory not the java application's temp directory, ala java.io.tmpdir. 1589const char* os::get_temp_directory() { return "/tmp"; } 1590 1591static bool file_exists(const char* filename) { 1592 struct stat statbuf; 1593 if (filename == NULL || strlen(filename) == 0) { 1594 return false; 1595 } 1596 return os::stat(filename, &statbuf) == 0; 1597} 1598 1599bool os::dll_build_name(char* buffer, size_t buflen, 1600 const char* pname, const char* fname) { 1601 bool retval = false; 1602 // Copied from libhpi 1603 const size_t pnamelen = pname ? strlen(pname) : 0; 1604 1605 // Return error on buffer overflow. 1606 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1607 return retval; 1608 } 1609 1610 if (pnamelen == 0) { 1611 snprintf(buffer, buflen, "lib%s.so", fname); 1612 retval = true; 1613 } else if (strchr(pname, *os::path_separator()) != NULL) { 1614 int n; 1615 char** pelements = split_path(pname, &n); 1616 if (pelements == NULL) { 1617 return false; 1618 } 1619 for (int i = 0 ; i < n ; i++) { 1620 // Really shouldn't be NULL, but check can't hurt 1621 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1622 continue; // skip the empty path values 1623 } 1624 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1625 if (file_exists(buffer)) { 1626 retval = true; 1627 break; 1628 } 1629 } 1630 // release the storage 1631 for (int i = 0 ; i < n ; i++) { 1632 if (pelements[i] != NULL) { 1633 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1634 } 1635 } 1636 if (pelements != NULL) { 1637 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1638 } 1639 } else { 1640 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1641 retval = true; 1642 } 1643 return retval; 1644} 1645 1646// check if addr is inside libjvm.so 1647bool os::address_is_in_vm(address addr) { 1648 static address libjvm_base_addr; 1649 Dl_info dlinfo; 1650 1651 if (libjvm_base_addr == NULL) { 1652 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1653 libjvm_base_addr = (address)dlinfo.dli_fbase; 1654 } 1655 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1656 } 1657 1658 if (dladdr((void *)addr, &dlinfo) != 0) { 1659 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1660 } 1661 1662 return false; 1663} 1664 1665bool os::dll_address_to_function_name(address addr, char *buf, 1666 int buflen, int *offset) { 1667 // buf is not optional, but offset is optional 1668 assert(buf != NULL, "sanity check"); 1669 1670 Dl_info dlinfo; 1671 1672 if (dladdr((void*)addr, &dlinfo) != 0) { 1673 // see if we have a matching symbol 1674 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1675 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1676 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1677 } 1678 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1679 return true; 1680 } 1681 // no matching symbol so try for just file info 1682 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1683 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1684 buf, buflen, offset, dlinfo.dli_fname)) { 1685 return true; 1686 } 1687 } 1688 } 1689 1690 buf[0] = '\0'; 1691 if (offset != NULL) *offset = -1; 1692 return false; 1693} 1694 1695struct _address_to_library_name { 1696 address addr; // input : memory address 1697 size_t buflen; // size of fname 1698 char* fname; // output: library name 1699 address base; // library base addr 1700}; 1701 1702static int address_to_library_name_callback(struct dl_phdr_info *info, 1703 size_t size, void *data) { 1704 int i; 1705 bool found = false; 1706 address libbase = NULL; 1707 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1708 1709 // iterate through all loadable segments 1710 for (i = 0; i < info->dlpi_phnum; i++) { 1711 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1712 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1713 // base address of a library is the lowest address of its loaded 1714 // segments. 1715 if (libbase == NULL || libbase > segbase) { 1716 libbase = segbase; 1717 } 1718 // see if 'addr' is within current segment 1719 if (segbase <= d->addr && 1720 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1721 found = true; 1722 } 1723 } 1724 } 1725 1726 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1727 // so dll_address_to_library_name() can fall through to use dladdr() which 1728 // can figure out executable name from argv[0]. 1729 if (found && info->dlpi_name && info->dlpi_name[0]) { 1730 d->base = libbase; 1731 if (d->fname) { 1732 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1733 } 1734 return 1; 1735 } 1736 return 0; 1737} 1738 1739bool os::dll_address_to_library_name(address addr, char* buf, 1740 int buflen, int* offset) { 1741 // buf is not optional, but offset is optional 1742 assert(buf != NULL, "sanity check"); 1743 1744 Dl_info dlinfo; 1745 struct _address_to_library_name data; 1746 1747 // There is a bug in old glibc dladdr() implementation that it could resolve 1748 // to wrong library name if the .so file has a base address != NULL. Here 1749 // we iterate through the program headers of all loaded libraries to find 1750 // out which library 'addr' really belongs to. This workaround can be 1751 // removed once the minimum requirement for glibc is moved to 2.3.x. 1752 data.addr = addr; 1753 data.fname = buf; 1754 data.buflen = buflen; 1755 data.base = NULL; 1756 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1757 1758 if (rslt) { 1759 // buf already contains library name 1760 if (offset) *offset = addr - data.base; 1761 return true; 1762 } 1763 if (dladdr((void*)addr, &dlinfo) != 0) { 1764 if (dlinfo.dli_fname != NULL) { 1765 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1766 } 1767 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1768 *offset = addr - (address)dlinfo.dli_fbase; 1769 } 1770 return true; 1771 } 1772 1773 buf[0] = '\0'; 1774 if (offset) *offset = -1; 1775 return false; 1776} 1777 1778 // Loads .dll/.so and 1779 // in case of error it checks if .dll/.so was built for the 1780 // same architecture as Hotspot is running on 1781 1782 1783// Remember the stack's state. The Linux dynamic linker will change 1784// the stack to 'executable' at most once, so we must safepoint only once. 1785bool os::Linux::_stack_is_executable = false; 1786 1787// VM operation that loads a library. This is necessary if stack protection 1788// of the Java stacks can be lost during loading the library. If we 1789// do not stop the Java threads, they can stack overflow before the stacks 1790// are protected again. 1791class VM_LinuxDllLoad: public VM_Operation { 1792 private: 1793 const char *_filename; 1794 char *_ebuf; 1795 int _ebuflen; 1796 void *_lib; 1797 public: 1798 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1799 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1800 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1801 void doit() { 1802 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1803 os::Linux::_stack_is_executable = true; 1804 } 1805 void* loaded_library() { return _lib; } 1806}; 1807 1808void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1809{ 1810 void * result = NULL; 1811 bool load_attempted = false; 1812 1813 // Check whether the library to load might change execution rights 1814 // of the stack. If they are changed, the protection of the stack 1815 // guard pages will be lost. We need a safepoint to fix this. 1816 // 1817 // See Linux man page execstack(8) for more info. 1818 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1819 ElfFile ef(filename); 1820 if (!ef.specifies_noexecstack()) { 1821 if (!is_init_completed()) { 1822 os::Linux::_stack_is_executable = true; 1823 // This is OK - No Java threads have been created yet, and hence no 1824 // stack guard pages to fix. 1825 // 1826 // This should happen only when you are building JDK7 using a very 1827 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1828 // 1829 // Dynamic loader will make all stacks executable after 1830 // this function returns, and will not do that again. 1831 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1832 } else { 1833 warning("You have loaded library %s which might have disabled stack guard. " 1834 "The VM will try to fix the stack guard now.\n" 1835 "It's highly recommended that you fix the library with " 1836 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1837 filename); 1838 1839 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1840 JavaThread *jt = JavaThread::current(); 1841 if (jt->thread_state() != _thread_in_native) { 1842 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1843 // that requires ExecStack. Cannot enter safe point. Let's give up. 1844 warning("Unable to fix stack guard. Giving up."); 1845 } else { 1846 if (!LoadExecStackDllInVMThread) { 1847 // This is for the case where the DLL has an static 1848 // constructor function that executes JNI code. We cannot 1849 // load such DLLs in the VMThread. 1850 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1851 } 1852 1853 ThreadInVMfromNative tiv(jt); 1854 debug_only(VMNativeEntryWrapper vew;) 1855 1856 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1857 VMThread::execute(&op); 1858 if (LoadExecStackDllInVMThread) { 1859 result = op.loaded_library(); 1860 } 1861 load_attempted = true; 1862 } 1863 } 1864 } 1865 } 1866 1867 if (!load_attempted) { 1868 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1869 } 1870 1871 if (result != NULL) { 1872 // Successful loading 1873 return result; 1874 } 1875 1876 Elf32_Ehdr elf_head; 1877 int diag_msg_max_length=ebuflen-strlen(ebuf); 1878 char* diag_msg_buf=ebuf+strlen(ebuf); 1879 1880 if (diag_msg_max_length==0) { 1881 // No more space in ebuf for additional diagnostics message 1882 return NULL; 1883 } 1884 1885 1886 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1887 1888 if (file_descriptor < 0) { 1889 // Can't open library, report dlerror() message 1890 return NULL; 1891 } 1892 1893 bool failed_to_read_elf_head= 1894 (sizeof(elf_head)!= 1895 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1896 1897 ::close(file_descriptor); 1898 if (failed_to_read_elf_head) { 1899 // file i/o error - report dlerror() msg 1900 return NULL; 1901 } 1902 1903 typedef struct { 1904 Elf32_Half code; // Actual value as defined in elf.h 1905 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1906 char elf_class; // 32 or 64 bit 1907 char endianess; // MSB or LSB 1908 char* name; // String representation 1909 } arch_t; 1910 1911 #ifndef EM_486 1912 #define EM_486 6 /* Intel 80486 */ 1913 #endif 1914 1915 static const arch_t arch_array[]={ 1916 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1917 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1918 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1919 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1920 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1921 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1922 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1923 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1924 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1925 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1926 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1927 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1928 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1929 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1930 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1931 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1932 }; 1933 1934 #if (defined IA32) 1935 static Elf32_Half running_arch_code=EM_386; 1936 #elif (defined AMD64) 1937 static Elf32_Half running_arch_code=EM_X86_64; 1938 #elif (defined IA64) 1939 static Elf32_Half running_arch_code=EM_IA_64; 1940 #elif (defined __sparc) && (defined _LP64) 1941 static Elf32_Half running_arch_code=EM_SPARCV9; 1942 #elif (defined __sparc) && (!defined _LP64) 1943 static Elf32_Half running_arch_code=EM_SPARC; 1944 #elif (defined __powerpc64__) 1945 static Elf32_Half running_arch_code=EM_PPC64; 1946 #elif (defined __powerpc__) 1947 static Elf32_Half running_arch_code=EM_PPC; 1948 #elif (defined ARM) 1949 static Elf32_Half running_arch_code=EM_ARM; 1950 #elif (defined S390) 1951 static Elf32_Half running_arch_code=EM_S390; 1952 #elif (defined ALPHA) 1953 static Elf32_Half running_arch_code=EM_ALPHA; 1954 #elif (defined MIPSEL) 1955 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1956 #elif (defined PARISC) 1957 static Elf32_Half running_arch_code=EM_PARISC; 1958 #elif (defined MIPS) 1959 static Elf32_Half running_arch_code=EM_MIPS; 1960 #elif (defined M68K) 1961 static Elf32_Half running_arch_code=EM_68K; 1962 #else 1963 #error Method os::dll_load requires that one of following is defined:\ 1964 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 1965 #endif 1966 1967 // Identify compatability class for VM's architecture and library's architecture 1968 // Obtain string descriptions for architectures 1969 1970 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1971 int running_arch_index=-1; 1972 1973 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1974 if (running_arch_code == arch_array[i].code) { 1975 running_arch_index = i; 1976 } 1977 if (lib_arch.code == arch_array[i].code) { 1978 lib_arch.compat_class = arch_array[i].compat_class; 1979 lib_arch.name = arch_array[i].name; 1980 } 1981 } 1982 1983 assert(running_arch_index != -1, 1984 "Didn't find running architecture code (running_arch_code) in arch_array"); 1985 if (running_arch_index == -1) { 1986 // Even though running architecture detection failed 1987 // we may still continue with reporting dlerror() message 1988 return NULL; 1989 } 1990 1991 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1992 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1993 return NULL; 1994 } 1995 1996#ifndef S390 1997 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1998 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1999 return NULL; 2000 } 2001#endif // !S390 2002 2003 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2004 if ( lib_arch.name!=NULL ) { 2005 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2006 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2007 lib_arch.name, arch_array[running_arch_index].name); 2008 } else { 2009 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2010 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2011 lib_arch.code, 2012 arch_array[running_arch_index].name); 2013 } 2014 } 2015 2016 return NULL; 2017} 2018 2019void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) { 2020 void * result = ::dlopen(filename, RTLD_LAZY); 2021 if (result == NULL) { 2022 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2023 ebuf[ebuflen-1] = '\0'; 2024 } 2025 return result; 2026} 2027 2028void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) { 2029 void * result = NULL; 2030 if (LoadExecStackDllInVMThread) { 2031 result = dlopen_helper(filename, ebuf, ebuflen); 2032 } 2033 2034 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2035 // library that requires an executable stack, or which does not have this 2036 // stack attribute set, dlopen changes the stack attribute to executable. The 2037 // read protection of the guard pages gets lost. 2038 // 2039 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2040 // may have been queued at the same time. 2041 2042 if (!_stack_is_executable) { 2043 JavaThread *jt = Threads::first(); 2044 2045 while (jt) { 2046 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2047 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2048 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2049 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2050 warning("Attempt to reguard stack yellow zone failed."); 2051 } 2052 } 2053 jt = jt->next(); 2054 } 2055 } 2056 2057 return result; 2058} 2059 2060/* 2061 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2062 * chances are you might want to run the generated bits against glibc-2.0 2063 * libdl.so, so always use locking for any version of glibc. 2064 */ 2065void* os::dll_lookup(void* handle, const char* name) { 2066 pthread_mutex_lock(&dl_mutex); 2067 void* res = dlsym(handle, name); 2068 pthread_mutex_unlock(&dl_mutex); 2069 return res; 2070} 2071 2072void* os::get_default_process_handle() { 2073 return (void*)::dlopen(NULL, RTLD_LAZY); 2074} 2075 2076static bool _print_ascii_file(const char* filename, outputStream* st) { 2077 int fd = ::open(filename, O_RDONLY); 2078 if (fd == -1) { 2079 return false; 2080 } 2081 2082 char buf[32]; 2083 int bytes; 2084 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2085 st->print_raw(buf, bytes); 2086 } 2087 2088 ::close(fd); 2089 2090 return true; 2091} 2092 2093void os::print_dll_info(outputStream *st) { 2094 st->print_cr("Dynamic libraries:"); 2095 2096 char fname[32]; 2097 pid_t pid = os::Linux::gettid(); 2098 2099 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2100 2101 if (!_print_ascii_file(fname, st)) { 2102 st->print("Can not get library information for pid = %d\n", pid); 2103 } 2104} 2105 2106void os::print_os_info_brief(outputStream* st) { 2107 os::Linux::print_distro_info(st); 2108 2109 os::Posix::print_uname_info(st); 2110 2111 os::Linux::print_libversion_info(st); 2112 2113} 2114 2115void os::print_os_info(outputStream* st) { 2116 st->print("OS:"); 2117 2118 os::Linux::print_distro_info(st); 2119 2120 os::Posix::print_uname_info(st); 2121 2122 // Print warning if unsafe chroot environment detected 2123 if (unsafe_chroot_detected) { 2124 st->print("WARNING!! "); 2125 st->print_cr(unstable_chroot_error); 2126 } 2127 2128 os::Linux::print_libversion_info(st); 2129 2130 os::Posix::print_rlimit_info(st); 2131 2132 os::Posix::print_load_average(st); 2133 2134 os::Linux::print_full_memory_info(st); 2135} 2136 2137// Try to identify popular distros. 2138// Most Linux distributions have a /etc/XXX-release file, which contains 2139// the OS version string. Newer Linux distributions have a /etc/lsb-release 2140// file that also contains the OS version string. Some have more than one 2141// /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2142// /etc/redhat-release.), so the order is important. 2143// Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2144// their own specific XXX-release file as well as a redhat-release file. 2145// Because of this the XXX-release file needs to be searched for before the 2146// redhat-release file. 2147// Since Red Hat has a lsb-release file that is not very descriptive the 2148// search for redhat-release needs to be before lsb-release. 2149// Since the lsb-release file is the new standard it needs to be searched 2150// before the older style release files. 2151// Searching system-release (Red Hat) and os-release (other Linuxes) are a 2152// next to last resort. The os-release file is a new standard that contains 2153// distribution information and the system-release file seems to be an old 2154// standard that has been replaced by the lsb-release and os-release files. 2155// Searching for the debian_version file is the last resort. It contains 2156// an informative string like "6.0.6" or "wheezy/sid". Because of this 2157// "Debian " is printed before the contents of the debian_version file. 2158void os::Linux::print_distro_info(outputStream* st) { 2159 if (!_print_ascii_file("/etc/oracle-release", st) && 2160 !_print_ascii_file("/etc/mandriva-release", st) && 2161 !_print_ascii_file("/etc/mandrake-release", st) && 2162 !_print_ascii_file("/etc/sun-release", st) && 2163 !_print_ascii_file("/etc/redhat-release", st) && 2164 !_print_ascii_file("/etc/lsb-release", st) && 2165 !_print_ascii_file("/etc/SuSE-release", st) && 2166 !_print_ascii_file("/etc/turbolinux-release", st) && 2167 !_print_ascii_file("/etc/gentoo-release", st) && 2168 !_print_ascii_file("/etc/ltib-release", st) && 2169 !_print_ascii_file("/etc/angstrom-version", st) && 2170 !_print_ascii_file("/etc/system-release", st) && 2171 !_print_ascii_file("/etc/os-release", st)) { 2172 2173 if (file_exists("/etc/debian_version")) { 2174 st->print("Debian "); 2175 _print_ascii_file("/etc/debian_version", st); 2176 } else { 2177 st->print("Linux"); 2178 } 2179 } 2180 st->cr(); 2181} 2182 2183void os::Linux::print_libversion_info(outputStream* st) { 2184 // libc, pthread 2185 st->print("libc:"); 2186 st->print(os::Linux::glibc_version()); st->print(" "); 2187 st->print(os::Linux::libpthread_version()); st->print(" "); 2188 if (os::Linux::is_LinuxThreads()) { 2189 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2190 } 2191 st->cr(); 2192} 2193 2194void os::Linux::print_full_memory_info(outputStream* st) { 2195 st->print("\n/proc/meminfo:\n"); 2196 _print_ascii_file("/proc/meminfo", st); 2197 st->cr(); 2198} 2199 2200void os::print_memory_info(outputStream* st) { 2201 2202 st->print("Memory:"); 2203 st->print(" %dk page", os::vm_page_size()>>10); 2204 2205 // values in struct sysinfo are "unsigned long" 2206 struct sysinfo si; 2207 sysinfo(&si); 2208 2209 st->print(", physical " UINT64_FORMAT "k", 2210 os::physical_memory() >> 10); 2211 st->print("(" UINT64_FORMAT "k free)", 2212 os::available_memory() >> 10); 2213 st->print(", swap " UINT64_FORMAT "k", 2214 ((jlong)si.totalswap * si.mem_unit) >> 10); 2215 st->print("(" UINT64_FORMAT "k free)", 2216 ((jlong)si.freeswap * si.mem_unit) >> 10); 2217 st->cr(); 2218} 2219 2220void os::pd_print_cpu_info(outputStream* st) { 2221 st->print("\n/proc/cpuinfo:\n"); 2222 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2223 st->print(" <Not Available>"); 2224 } 2225 st->cr(); 2226} 2227 2228void os::print_siginfo(outputStream* st, void* siginfo) { 2229 const siginfo_t* si = (const siginfo_t*)siginfo; 2230 2231 os::Posix::print_siginfo_brief(st, si); 2232 2233 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2234 UseSharedSpaces) { 2235 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2236 if (mapinfo->is_in_shared_space(si->si_addr)) { 2237 st->print("\n\nError accessing class data sharing archive." \ 2238 " Mapped file inaccessible during execution, " \ 2239 " possible disk/network problem."); 2240 } 2241 } 2242 st->cr(); 2243} 2244 2245 2246static void print_signal_handler(outputStream* st, int sig, 2247 char* buf, size_t buflen); 2248 2249void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2250 st->print_cr("Signal Handlers:"); 2251 print_signal_handler(st, SIGSEGV, buf, buflen); 2252 print_signal_handler(st, SIGBUS , buf, buflen); 2253 print_signal_handler(st, SIGFPE , buf, buflen); 2254 print_signal_handler(st, SIGPIPE, buf, buflen); 2255 print_signal_handler(st, SIGXFSZ, buf, buflen); 2256 print_signal_handler(st, SIGILL , buf, buflen); 2257 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2258 print_signal_handler(st, SR_signum, buf, buflen); 2259 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2260 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2261 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2262 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2263#if defined(PPC64) 2264 print_signal_handler(st, SIGTRAP, buf, buflen); 2265#endif 2266} 2267 2268static char saved_jvm_path[MAXPATHLEN] = {0}; 2269 2270// Find the full path to the current module, libjvm.so 2271void os::jvm_path(char *buf, jint buflen) { 2272 // Error checking. 2273 if (buflen < MAXPATHLEN) { 2274 assert(false, "must use a large-enough buffer"); 2275 buf[0] = '\0'; 2276 return; 2277 } 2278 // Lazy resolve the path to current module. 2279 if (saved_jvm_path[0] != 0) { 2280 strcpy(buf, saved_jvm_path); 2281 return; 2282 } 2283 2284 char dli_fname[MAXPATHLEN]; 2285 bool ret = dll_address_to_library_name( 2286 CAST_FROM_FN_PTR(address, os::jvm_path), 2287 dli_fname, sizeof(dli_fname), NULL); 2288 assert(ret, "cannot locate libjvm"); 2289 char *rp = NULL; 2290 if (ret && dli_fname[0] != '\0') { 2291 rp = realpath(dli_fname, buf); 2292 } 2293 if (rp == NULL) 2294 return; 2295 2296 if (Arguments::sun_java_launcher_is_altjvm()) { 2297 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2298 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2299 // If "/jre/lib/" appears at the right place in the string, then 2300 // assume we are installed in a JDK and we're done. Otherwise, check 2301 // for a JAVA_HOME environment variable and fix up the path so it 2302 // looks like libjvm.so is installed there (append a fake suffix 2303 // hotspot/libjvm.so). 2304 const char *p = buf + strlen(buf) - 1; 2305 for (int count = 0; p > buf && count < 5; ++count) { 2306 for (--p; p > buf && *p != '/'; --p) 2307 /* empty */ ; 2308 } 2309 2310 if (strncmp(p, "/jre/lib/", 9) != 0) { 2311 // Look for JAVA_HOME in the environment. 2312 char* java_home_var = ::getenv("JAVA_HOME"); 2313 if (java_home_var != NULL && java_home_var[0] != 0) { 2314 char* jrelib_p; 2315 int len; 2316 2317 // Check the current module name "libjvm.so". 2318 p = strrchr(buf, '/'); 2319 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2320 2321 rp = realpath(java_home_var, buf); 2322 if (rp == NULL) 2323 return; 2324 2325 // determine if this is a legacy image or modules image 2326 // modules image doesn't have "jre" subdirectory 2327 len = strlen(buf); 2328 jrelib_p = buf + len; 2329 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2330 if (0 != access(buf, F_OK)) { 2331 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2332 } 2333 2334 if (0 == access(buf, F_OK)) { 2335 // Use current module name "libjvm.so" 2336 len = strlen(buf); 2337 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2338 } else { 2339 // Go back to path of .so 2340 rp = realpath(dli_fname, buf); 2341 if (rp == NULL) 2342 return; 2343 } 2344 } 2345 } 2346 } 2347 2348 strcpy(saved_jvm_path, buf); 2349} 2350 2351void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2352 // no prefix required, not even "_" 2353} 2354 2355void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2356 // no suffix required 2357} 2358 2359//////////////////////////////////////////////////////////////////////////////// 2360// sun.misc.Signal support 2361 2362static volatile jint sigint_count = 0; 2363 2364static void 2365UserHandler(int sig, void *siginfo, void *context) { 2366 // 4511530 - sem_post is serialized and handled by the manager thread. When 2367 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2368 // don't want to flood the manager thread with sem_post requests. 2369 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2370 return; 2371 2372 // Ctrl-C is pressed during error reporting, likely because the error 2373 // handler fails to abort. Let VM die immediately. 2374 if (sig == SIGINT && is_error_reported()) { 2375 os::die(); 2376 } 2377 2378 os::signal_notify(sig); 2379} 2380 2381void* os::user_handler() { 2382 return CAST_FROM_FN_PTR(void*, UserHandler); 2383} 2384 2385class Semaphore : public StackObj { 2386 public: 2387 Semaphore(); 2388 ~Semaphore(); 2389 void signal(); 2390 void wait(); 2391 bool trywait(); 2392 bool timedwait(unsigned int sec, int nsec); 2393 private: 2394 sem_t _semaphore; 2395}; 2396 2397Semaphore::Semaphore() { 2398 sem_init(&_semaphore, 0, 0); 2399} 2400 2401Semaphore::~Semaphore() { 2402 sem_destroy(&_semaphore); 2403} 2404 2405void Semaphore::signal() { 2406 sem_post(&_semaphore); 2407} 2408 2409void Semaphore::wait() { 2410 sem_wait(&_semaphore); 2411} 2412 2413bool Semaphore::trywait() { 2414 return sem_trywait(&_semaphore) == 0; 2415} 2416 2417bool Semaphore::timedwait(unsigned int sec, int nsec) { 2418 2419 struct timespec ts; 2420 // Semaphore's are always associated with CLOCK_REALTIME 2421 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2422 // see unpackTime for discussion on overflow checking 2423 if (sec >= MAX_SECS) { 2424 ts.tv_sec += MAX_SECS; 2425 ts.tv_nsec = 0; 2426 } else { 2427 ts.tv_sec += sec; 2428 ts.tv_nsec += nsec; 2429 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2430 ts.tv_nsec -= NANOSECS_PER_SEC; 2431 ++ts.tv_sec; // note: this must be <= max_secs 2432 } 2433 } 2434 2435 while (1) { 2436 int result = sem_timedwait(&_semaphore, &ts); 2437 if (result == 0) { 2438 return true; 2439 } else if (errno == EINTR) { 2440 continue; 2441 } else if (errno == ETIMEDOUT) { 2442 return false; 2443 } else { 2444 return false; 2445 } 2446 } 2447} 2448 2449extern "C" { 2450 typedef void (*sa_handler_t)(int); 2451 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2452} 2453 2454void* os::signal(int signal_number, void* handler) { 2455 struct sigaction sigAct, oldSigAct; 2456 2457 sigfillset(&(sigAct.sa_mask)); 2458 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2459 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2460 2461 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2462 // -1 means registration failed 2463 return (void *)-1; 2464 } 2465 2466 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2467} 2468 2469void os::signal_raise(int signal_number) { 2470 ::raise(signal_number); 2471} 2472 2473/* 2474 * The following code is moved from os.cpp for making this 2475 * code platform specific, which it is by its very nature. 2476 */ 2477 2478// Will be modified when max signal is changed to be dynamic 2479int os::sigexitnum_pd() { 2480 return NSIG; 2481} 2482 2483// a counter for each possible signal value 2484static volatile jint pending_signals[NSIG+1] = { 0 }; 2485 2486// Linux(POSIX) specific hand shaking semaphore. 2487static sem_t sig_sem; 2488static Semaphore sr_semaphore; 2489 2490void os::signal_init_pd() { 2491 // Initialize signal structures 2492 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2493 2494 // Initialize signal semaphore 2495 ::sem_init(&sig_sem, 0, 0); 2496} 2497 2498void os::signal_notify(int sig) { 2499 Atomic::inc(&pending_signals[sig]); 2500 ::sem_post(&sig_sem); 2501} 2502 2503static int check_pending_signals(bool wait) { 2504 Atomic::store(0, &sigint_count); 2505 for (;;) { 2506 for (int i = 0; i < NSIG + 1; i++) { 2507 jint n = pending_signals[i]; 2508 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2509 return i; 2510 } 2511 } 2512 if (!wait) { 2513 return -1; 2514 } 2515 JavaThread *thread = JavaThread::current(); 2516 ThreadBlockInVM tbivm(thread); 2517 2518 bool threadIsSuspended; 2519 do { 2520 thread->set_suspend_equivalent(); 2521 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2522 ::sem_wait(&sig_sem); 2523 2524 // were we externally suspended while we were waiting? 2525 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2526 if (threadIsSuspended) { 2527 // 2528 // The semaphore has been incremented, but while we were waiting 2529 // another thread suspended us. We don't want to continue running 2530 // while suspended because that would surprise the thread that 2531 // suspended us. 2532 // 2533 ::sem_post(&sig_sem); 2534 2535 thread->java_suspend_self(); 2536 } 2537 } while (threadIsSuspended); 2538 } 2539} 2540 2541int os::signal_lookup() { 2542 return check_pending_signals(false); 2543} 2544 2545int os::signal_wait() { 2546 return check_pending_signals(true); 2547} 2548 2549//////////////////////////////////////////////////////////////////////////////// 2550// Virtual Memory 2551 2552int os::vm_page_size() { 2553 // Seems redundant as all get out 2554 assert(os::Linux::page_size() != -1, "must call os::init"); 2555 return os::Linux::page_size(); 2556} 2557 2558// Solaris allocates memory by pages. 2559int os::vm_allocation_granularity() { 2560 assert(os::Linux::page_size() != -1, "must call os::init"); 2561 return os::Linux::page_size(); 2562} 2563 2564// Rationale behind this function: 2565// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2566// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2567// samples for JITted code. Here we create private executable mapping over the code cache 2568// and then we can use standard (well, almost, as mapping can change) way to provide 2569// info for the reporting script by storing timestamp and location of symbol 2570void linux_wrap_code(char* base, size_t size) { 2571 static volatile jint cnt = 0; 2572 2573 if (!UseOprofile) { 2574 return; 2575 } 2576 2577 char buf[PATH_MAX+1]; 2578 int num = Atomic::add(1, &cnt); 2579 2580 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2581 os::get_temp_directory(), os::current_process_id(), num); 2582 unlink(buf); 2583 2584 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2585 2586 if (fd != -1) { 2587 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2588 if (rv != (off_t)-1) { 2589 if (::write(fd, "", 1) == 1) { 2590 mmap(base, size, 2591 PROT_READ|PROT_WRITE|PROT_EXEC, 2592 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2593 } 2594 } 2595 ::close(fd); 2596 unlink(buf); 2597 } 2598} 2599 2600static bool recoverable_mmap_error(int err) { 2601 // See if the error is one we can let the caller handle. This 2602 // list of errno values comes from JBS-6843484. I can't find a 2603 // Linux man page that documents this specific set of errno 2604 // values so while this list currently matches Solaris, it may 2605 // change as we gain experience with this failure mode. 2606 switch (err) { 2607 case EBADF: 2608 case EINVAL: 2609 case ENOTSUP: 2610 // let the caller deal with these errors 2611 return true; 2612 2613 default: 2614 // Any remaining errors on this OS can cause our reserved mapping 2615 // to be lost. That can cause confusion where different data 2616 // structures think they have the same memory mapped. The worst 2617 // scenario is if both the VM and a library think they have the 2618 // same memory mapped. 2619 return false; 2620 } 2621} 2622 2623static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2624 int err) { 2625 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2626 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2627 strerror(err), err); 2628} 2629 2630static void warn_fail_commit_memory(char* addr, size_t size, 2631 size_t alignment_hint, bool exec, 2632 int err) { 2633 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2634 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2635 alignment_hint, exec, strerror(err), err); 2636} 2637 2638// NOTE: Linux kernel does not really reserve the pages for us. 2639// All it does is to check if there are enough free pages 2640// left at the time of mmap(). This could be a potential 2641// problem. 2642int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2643 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2644 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2645 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2646 if (res != (uintptr_t) MAP_FAILED) { 2647 if (UseNUMAInterleaving) { 2648 numa_make_global(addr, size); 2649 } 2650 return 0; 2651 } 2652 2653 int err = errno; // save errno from mmap() call above 2654 2655 if (!recoverable_mmap_error(err)) { 2656 warn_fail_commit_memory(addr, size, exec, err); 2657 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2658 } 2659 2660 return err; 2661} 2662 2663bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2664 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2665} 2666 2667void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2668 const char* mesg) { 2669 assert(mesg != NULL, "mesg must be specified"); 2670 int err = os::Linux::commit_memory_impl(addr, size, exec); 2671 if (err != 0) { 2672 // the caller wants all commit errors to exit with the specified mesg: 2673 warn_fail_commit_memory(addr, size, exec, err); 2674 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2675 } 2676} 2677 2678// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2679#ifndef MAP_HUGETLB 2680#define MAP_HUGETLB 0x40000 2681#endif 2682 2683// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2684#ifndef MADV_HUGEPAGE 2685#define MADV_HUGEPAGE 14 2686#endif 2687 2688int os::Linux::commit_memory_impl(char* addr, size_t size, 2689 size_t alignment_hint, bool exec) { 2690 int err = os::Linux::commit_memory_impl(addr, size, exec); 2691 if (err == 0) { 2692 realign_memory(addr, size, alignment_hint); 2693 } 2694 return err; 2695} 2696 2697bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2698 bool exec) { 2699 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2700} 2701 2702void os::pd_commit_memory_or_exit(char* addr, size_t size, 2703 size_t alignment_hint, bool exec, 2704 const char* mesg) { 2705 assert(mesg != NULL, "mesg must be specified"); 2706 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2707 if (err != 0) { 2708 // the caller wants all commit errors to exit with the specified mesg: 2709 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2710 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2711 } 2712} 2713 2714void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2715 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2716 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2717 // be supported or the memory may already be backed by huge pages. 2718 ::madvise(addr, bytes, MADV_HUGEPAGE); 2719 } 2720} 2721 2722void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2723 // This method works by doing an mmap over an existing mmaping and effectively discarding 2724 // the existing pages. However it won't work for SHM-based large pages that cannot be 2725 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2726 // small pages on top of the SHM segment. This method always works for small pages, so we 2727 // allow that in any case. 2728 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2729 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2730 } 2731} 2732 2733void os::numa_make_global(char *addr, size_t bytes) { 2734 Linux::numa_interleave_memory(addr, bytes); 2735} 2736 2737// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2738// bind policy to MPOL_PREFERRED for the current thread. 2739#define USE_MPOL_PREFERRED 0 2740 2741void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2742 // To make NUMA and large pages more robust when both enabled, we need to ease 2743 // the requirements on where the memory should be allocated. MPOL_BIND is the 2744 // default policy and it will force memory to be allocated on the specified 2745 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2746 // the specified node, but will not force it. Using this policy will prevent 2747 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2748 // free large pages. 2749 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2750 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2751} 2752 2753bool os::numa_topology_changed() { return false; } 2754 2755size_t os::numa_get_groups_num() { 2756 int max_node = Linux::numa_max_node(); 2757 return max_node > 0 ? max_node + 1 : 1; 2758} 2759 2760int os::numa_get_group_id() { 2761 int cpu_id = Linux::sched_getcpu(); 2762 if (cpu_id != -1) { 2763 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2764 if (lgrp_id != -1) { 2765 return lgrp_id; 2766 } 2767 } 2768 return 0; 2769} 2770 2771size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2772 for (size_t i = 0; i < size; i++) { 2773 ids[i] = i; 2774 } 2775 return size; 2776} 2777 2778bool os::get_page_info(char *start, page_info* info) { 2779 return false; 2780} 2781 2782char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2783 return end; 2784} 2785 2786 2787int os::Linux::sched_getcpu_syscall(void) { 2788 unsigned int cpu; 2789 int retval = -1; 2790 2791#if defined(IA32) 2792# ifndef SYS_getcpu 2793# define SYS_getcpu 318 2794# endif 2795 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2796#elif defined(AMD64) 2797// Unfortunately we have to bring all these macros here from vsyscall.h 2798// to be able to compile on old linuxes. 2799# define __NR_vgetcpu 2 2800# define VSYSCALL_START (-10UL << 20) 2801# define VSYSCALL_SIZE 1024 2802# define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2803 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2804 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2805 retval = vgetcpu(&cpu, NULL, NULL); 2806#endif 2807 2808 return (retval == -1) ? retval : cpu; 2809} 2810 2811// Something to do with the numa-aware allocator needs these symbols 2812extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2813extern "C" JNIEXPORT void numa_error(char *where) { } 2814extern "C" JNIEXPORT int fork1() { return fork(); } 2815 2816 2817// If we are running with libnuma version > 2, then we should 2818// be trying to use symbols with versions 1.1 2819// If we are running with earlier version, which did not have symbol versions, 2820// we should use the base version. 2821void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2822 void *f = dlvsym(handle, name, "libnuma_1.1"); 2823 if (f == NULL) { 2824 f = dlsym(handle, name); 2825 } 2826 return f; 2827} 2828 2829bool os::Linux::libnuma_init() { 2830 // sched_getcpu() should be in libc. 2831 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2832 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2833 2834 // If it's not, try a direct syscall. 2835 if (sched_getcpu() == -1) 2836 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall)); 2837 2838 if (sched_getcpu() != -1) { // Does it work? 2839 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2840 if (handle != NULL) { 2841 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2842 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2843 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2844 libnuma_dlsym(handle, "numa_max_node"))); 2845 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2846 libnuma_dlsym(handle, "numa_available"))); 2847 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2848 libnuma_dlsym(handle, "numa_tonode_memory"))); 2849 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2850 libnuma_dlsym(handle, "numa_interleave_memory"))); 2851 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2852 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2853 2854 2855 if (numa_available() != -1) { 2856 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2857 // Create a cpu -> node mapping 2858 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2859 rebuild_cpu_to_node_map(); 2860 return true; 2861 } 2862 } 2863 } 2864 return false; 2865} 2866 2867// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2868// The table is later used in get_node_by_cpu(). 2869void os::Linux::rebuild_cpu_to_node_map() { 2870 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2871 // in libnuma (possible values are starting from 16, 2872 // and continuing up with every other power of 2, but less 2873 // than the maximum number of CPUs supported by kernel), and 2874 // is a subject to change (in libnuma version 2 the requirements 2875 // are more reasonable) we'll just hardcode the number they use 2876 // in the library. 2877 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2878 2879 size_t cpu_num = os::active_processor_count(); 2880 size_t cpu_map_size = NCPUS / BitsPerCLong; 2881 size_t cpu_map_valid_size = 2882 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2883 2884 cpu_to_node()->clear(); 2885 cpu_to_node()->at_grow(cpu_num - 1); 2886 size_t node_num = numa_get_groups_num(); 2887 2888 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2889 for (size_t i = 0; i < node_num; i++) { 2890 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2891 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2892 if (cpu_map[j] != 0) { 2893 for (size_t k = 0; k < BitsPerCLong; k++) { 2894 if (cpu_map[j] & (1UL << k)) { 2895 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2896 } 2897 } 2898 } 2899 } 2900 } 2901 } 2902 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2903} 2904 2905int os::Linux::get_node_by_cpu(int cpu_id) { 2906 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2907 return cpu_to_node()->at(cpu_id); 2908 } 2909 return -1; 2910} 2911 2912GrowableArray<int>* os::Linux::_cpu_to_node; 2913os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2914os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2915os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2916os::Linux::numa_available_func_t os::Linux::_numa_available; 2917os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2918os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2919os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2920unsigned long* os::Linux::_numa_all_nodes; 2921 2922bool os::pd_uncommit_memory(char* addr, size_t size) { 2923 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2924 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2925 return res != (uintptr_t) MAP_FAILED; 2926} 2927 2928static 2929address get_stack_commited_bottom(address bottom, size_t size) { 2930 address nbot = bottom; 2931 address ntop = bottom + size; 2932 2933 size_t page_sz = os::vm_page_size(); 2934 unsigned pages = size / page_sz; 2935 2936 unsigned char vec[1]; 2937 unsigned imin = 1, imax = pages + 1, imid; 2938 int mincore_return_value = 0; 2939 2940 assert(imin <= imax, "Unexpected page size"); 2941 2942 while (imin < imax) { 2943 imid = (imax + imin) / 2; 2944 nbot = ntop - (imid * page_sz); 2945 2946 // Use a trick with mincore to check whether the page is mapped or not. 2947 // mincore sets vec to 1 if page resides in memory and to 0 if page 2948 // is swapped output but if page we are asking for is unmapped 2949 // it returns -1,ENOMEM 2950 mincore_return_value = mincore(nbot, page_sz, vec); 2951 2952 if (mincore_return_value == -1) { 2953 // Page is not mapped go up 2954 // to find first mapped page 2955 if (errno != EAGAIN) { 2956 assert(errno == ENOMEM, "Unexpected mincore errno"); 2957 imax = imid; 2958 } 2959 } else { 2960 // Page is mapped go down 2961 // to find first not mapped page 2962 imin = imid + 1; 2963 } 2964 } 2965 2966 nbot = nbot + page_sz; 2967 2968 // Adjust stack bottom one page up if last checked page is not mapped 2969 if (mincore_return_value == -1) { 2970 nbot = nbot + page_sz; 2971 } 2972 2973 return nbot; 2974} 2975 2976 2977// Linux uses a growable mapping for the stack, and if the mapping for 2978// the stack guard pages is not removed when we detach a thread the 2979// stack cannot grow beyond the pages where the stack guard was 2980// mapped. If at some point later in the process the stack expands to 2981// that point, the Linux kernel cannot expand the stack any further 2982// because the guard pages are in the way, and a segfault occurs. 2983// 2984// However, it's essential not to split the stack region by unmapping 2985// a region (leaving a hole) that's already part of the stack mapping, 2986// so if the stack mapping has already grown beyond the guard pages at 2987// the time we create them, we have to truncate the stack mapping. 2988// So, we need to know the extent of the stack mapping when 2989// create_stack_guard_pages() is called. 2990 2991// We only need this for stacks that are growable: at the time of 2992// writing thread stacks don't use growable mappings (i.e. those 2993// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 2994// only applies to the main thread. 2995 2996// If the (growable) stack mapping already extends beyond the point 2997// where we're going to put our guard pages, truncate the mapping at 2998// that point by munmap()ping it. This ensures that when we later 2999// munmap() the guard pages we don't leave a hole in the stack 3000// mapping. This only affects the main/initial thread 3001 3002bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3003 3004 if (os::Linux::is_initial_thread()) { 3005 // As we manually grow stack up to bottom inside create_attached_thread(), 3006 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3007 // we don't need to do anything special. 3008 // Check it first, before calling heavy function. 3009 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3010 unsigned char vec[1]; 3011 3012 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3013 // Fallback to slow path on all errors, including EAGAIN 3014 stack_extent = (uintptr_t) get_stack_commited_bottom( 3015 os::Linux::initial_thread_stack_bottom(), 3016 (size_t)addr - stack_extent); 3017 } 3018 3019 if (stack_extent < (uintptr_t)addr) { 3020 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3021 } 3022 } 3023 3024 return os::commit_memory(addr, size, !ExecMem); 3025} 3026 3027// If this is a growable mapping, remove the guard pages entirely by 3028// munmap()ping them. If not, just call uncommit_memory(). This only 3029// affects the main/initial thread, but guard against future OS changes 3030// It's safe to always unmap guard pages for initial thread because we 3031// always place it right after end of the mapped region 3032 3033bool os::remove_stack_guard_pages(char* addr, size_t size) { 3034 uintptr_t stack_extent, stack_base; 3035 3036 if (os::Linux::is_initial_thread()) { 3037 return ::munmap(addr, size) == 0; 3038 } 3039 3040 return os::uncommit_memory(addr, size); 3041} 3042 3043static address _highest_vm_reserved_address = NULL; 3044 3045// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3046// at 'requested_addr'. If there are existing memory mappings at the same 3047// location, however, they will be overwritten. If 'fixed' is false, 3048// 'requested_addr' is only treated as a hint, the return value may or 3049// may not start from the requested address. Unlike Linux mmap(), this 3050// function returns NULL to indicate failure. 3051static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3052 char * addr; 3053 int flags; 3054 3055 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3056 if (fixed) { 3057 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3058 flags |= MAP_FIXED; 3059 } 3060 3061 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3062 // touch an uncommitted page. Otherwise, the read/write might 3063 // succeed if we have enough swap space to back the physical page. 3064 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3065 flags, -1, 0); 3066 3067 if (addr != MAP_FAILED) { 3068 // anon_mmap() should only get called during VM initialization, 3069 // don't need lock (actually we can skip locking even it can be called 3070 // from multiple threads, because _highest_vm_reserved_address is just a 3071 // hint about the upper limit of non-stack memory regions.) 3072 if ((address)addr + bytes > _highest_vm_reserved_address) { 3073 _highest_vm_reserved_address = (address)addr + bytes; 3074 } 3075 } 3076 3077 return addr == MAP_FAILED ? NULL : addr; 3078} 3079 3080// Don't update _highest_vm_reserved_address, because there might be memory 3081// regions above addr + size. If so, releasing a memory region only creates 3082// a hole in the address space, it doesn't help prevent heap-stack collision. 3083// 3084static int anon_munmap(char * addr, size_t size) { 3085 return ::munmap(addr, size) == 0; 3086} 3087 3088char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3089 size_t alignment_hint) { 3090 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3091} 3092 3093bool os::pd_release_memory(char* addr, size_t size) { 3094 return anon_munmap(addr, size); 3095} 3096 3097static address highest_vm_reserved_address() { 3098 return _highest_vm_reserved_address; 3099} 3100 3101static bool linux_mprotect(char* addr, size_t size, int prot) { 3102 // Linux wants the mprotect address argument to be page aligned. 3103 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3104 3105 // According to SUSv3, mprotect() should only be used with mappings 3106 // established by mmap(), and mmap() always maps whole pages. Unaligned 3107 // 'addr' likely indicates problem in the VM (e.g. trying to change 3108 // protection of malloc'ed or statically allocated memory). Check the 3109 // caller if you hit this assert. 3110 assert(addr == bottom, "sanity check"); 3111 3112 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3113 return ::mprotect(bottom, size, prot) == 0; 3114} 3115 3116// Set protections specified 3117bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3118 bool is_committed) { 3119 unsigned int p = 0; 3120 switch (prot) { 3121 case MEM_PROT_NONE: p = PROT_NONE; break; 3122 case MEM_PROT_READ: p = PROT_READ; break; 3123 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3124 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3125 default: 3126 ShouldNotReachHere(); 3127 } 3128 // is_committed is unused. 3129 return linux_mprotect(addr, bytes, p); 3130} 3131 3132bool os::guard_memory(char* addr, size_t size) { 3133 return linux_mprotect(addr, size, PROT_NONE); 3134} 3135 3136bool os::unguard_memory(char* addr, size_t size) { 3137 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3138} 3139 3140bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) { 3141 bool result = false; 3142 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3143 MAP_ANONYMOUS|MAP_PRIVATE, 3144 -1, 0); 3145 if (p != MAP_FAILED) { 3146 void *aligned_p = align_ptr_up(p, page_size); 3147 3148 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3149 3150 munmap(p, page_size * 2); 3151 } 3152 3153 if (warn && !result) { 3154 warning("TransparentHugePages is not supported by the operating system."); 3155 } 3156 3157 return result; 3158} 3159 3160bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3161 bool result = false; 3162 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3163 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3164 -1, 0); 3165 3166 if (p != MAP_FAILED) { 3167 // We don't know if this really is a huge page or not. 3168 FILE *fp = fopen("/proc/self/maps", "r"); 3169 if (fp) { 3170 while (!feof(fp)) { 3171 char chars[257]; 3172 long x = 0; 3173 if (fgets(chars, sizeof(chars), fp)) { 3174 if (sscanf(chars, "%lx-%*x", &x) == 1 3175 && x == (long)p) { 3176 if (strstr (chars, "hugepage")) { 3177 result = true; 3178 break; 3179 } 3180 } 3181 } 3182 } 3183 fclose(fp); 3184 } 3185 munmap(p, page_size); 3186 } 3187 3188 if (warn && !result) { 3189 warning("HugeTLBFS is not supported by the operating system."); 3190 } 3191 3192 return result; 3193} 3194 3195/* 3196* Set the coredump_filter bits to include largepages in core dump (bit 6) 3197* 3198* From the coredump_filter documentation: 3199* 3200* - (bit 0) anonymous private memory 3201* - (bit 1) anonymous shared memory 3202* - (bit 2) file-backed private memory 3203* - (bit 3) file-backed shared memory 3204* - (bit 4) ELF header pages in file-backed private memory areas (it is 3205* effective only if the bit 2 is cleared) 3206* - (bit 5) hugetlb private memory 3207* - (bit 6) hugetlb shared memory 3208*/ 3209static void set_coredump_filter(void) { 3210 FILE *f; 3211 long cdm; 3212 3213 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3214 return; 3215 } 3216 3217 if (fscanf(f, "%lx", &cdm) != 1) { 3218 fclose(f); 3219 return; 3220 } 3221 3222 rewind(f); 3223 3224 if ((cdm & LARGEPAGES_BIT) == 0) { 3225 cdm |= LARGEPAGES_BIT; 3226 fprintf(f, "%#lx", cdm); 3227 } 3228 3229 fclose(f); 3230} 3231 3232// Large page support 3233 3234static size_t _large_page_size = 0; 3235 3236size_t os::Linux::find_large_page_size() { 3237 size_t large_page_size = 0; 3238 3239 // large_page_size on Linux is used to round up heap size. x86 uses either 3240 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3241 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3242 // page as large as 256M. 3243 // 3244 // Here we try to figure out page size by parsing /proc/meminfo and looking 3245 // for a line with the following format: 3246 // Hugepagesize: 2048 kB 3247 // 3248 // If we can't determine the value (e.g. /proc is not mounted, or the text 3249 // format has been changed), we'll use the largest page size supported by 3250 // the processor. 3251 3252#ifndef ZERO 3253 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3254 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3255#endif // ZERO 3256 3257 FILE *fp = fopen("/proc/meminfo", "r"); 3258 if (fp) { 3259 while (!feof(fp)) { 3260 int x = 0; 3261 char buf[16]; 3262 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3263 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3264 large_page_size = x * K; 3265 break; 3266 } 3267 } else { 3268 // skip to next line 3269 for (;;) { 3270 int ch = fgetc(fp); 3271 if (ch == EOF || ch == (int)'\n') break; 3272 } 3273 } 3274 } 3275 fclose(fp); 3276 } 3277 3278 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3279 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3280 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3281 proper_unit_for_byte_size(large_page_size)); 3282 } 3283 3284 return large_page_size; 3285} 3286 3287size_t os::Linux::setup_large_page_size() { 3288 _large_page_size = Linux::find_large_page_size(); 3289 const size_t default_page_size = (size_t)Linux::page_size(); 3290 if (_large_page_size > default_page_size) { 3291 _page_sizes[0] = _large_page_size; 3292 _page_sizes[1] = default_page_size; 3293 _page_sizes[2] = 0; 3294 } 3295 3296 return _large_page_size; 3297} 3298 3299bool os::Linux::setup_large_page_type(size_t page_size) { 3300 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3301 FLAG_IS_DEFAULT(UseSHM) && 3302 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3303 3304 // The type of large pages has not been specified by the user. 3305 3306 // Try UseHugeTLBFS and then UseSHM. 3307 UseHugeTLBFS = UseSHM = true; 3308 3309 // Don't try UseTransparentHugePages since there are known 3310 // performance issues with it turned on. This might change in the future. 3311 UseTransparentHugePages = false; 3312 } 3313 3314 if (UseTransparentHugePages) { 3315 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3316 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3317 UseHugeTLBFS = false; 3318 UseSHM = false; 3319 return true; 3320 } 3321 UseTransparentHugePages = false; 3322 } 3323 3324 if (UseHugeTLBFS) { 3325 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3326 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3327 UseSHM = false; 3328 return true; 3329 } 3330 UseHugeTLBFS = false; 3331 } 3332 3333 return UseSHM; 3334} 3335 3336void os::large_page_init() { 3337 if (!UseLargePages && 3338 !UseTransparentHugePages && 3339 !UseHugeTLBFS && 3340 !UseSHM) { 3341 // Not using large pages. 3342 return; 3343 } 3344 3345 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3346 // The user explicitly turned off large pages. 3347 // Ignore the rest of the large pages flags. 3348 UseTransparentHugePages = false; 3349 UseHugeTLBFS = false; 3350 UseSHM = false; 3351 return; 3352 } 3353 3354 size_t large_page_size = Linux::setup_large_page_size(); 3355 UseLargePages = Linux::setup_large_page_type(large_page_size); 3356 3357 set_coredump_filter(); 3358} 3359 3360#ifndef SHM_HUGETLB 3361#define SHM_HUGETLB 04000 3362#endif 3363 3364char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3365 // "exec" is passed in but not used. Creating the shared image for 3366 // the code cache doesn't have an SHM_X executable permission to check. 3367 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3368 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3369 3370 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3371 return NULL; // Fallback to small pages. 3372 } 3373 3374 key_t key = IPC_PRIVATE; 3375 char *addr; 3376 3377 bool warn_on_failure = UseLargePages && 3378 (!FLAG_IS_DEFAULT(UseLargePages) || 3379 !FLAG_IS_DEFAULT(UseSHM) || 3380 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 3381 ); 3382 char msg[128]; 3383 3384 // Create a large shared memory region to attach to based on size. 3385 // Currently, size is the total size of the heap 3386 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3387 if (shmid == -1) { 3388 // Possible reasons for shmget failure: 3389 // 1. shmmax is too small for Java heap. 3390 // > check shmmax value: cat /proc/sys/kernel/shmmax 3391 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3392 // 2. not enough large page memory. 3393 // > check available large pages: cat /proc/meminfo 3394 // > increase amount of large pages: 3395 // echo new_value > /proc/sys/vm/nr_hugepages 3396 // Note 1: different Linux may use different name for this property, 3397 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3398 // Note 2: it's possible there's enough physical memory available but 3399 // they are so fragmented after a long run that they can't 3400 // coalesce into large pages. Try to reserve large pages when 3401 // the system is still "fresh". 3402 if (warn_on_failure) { 3403 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3404 warning(msg); 3405 } 3406 return NULL; 3407 } 3408 3409 // attach to the region 3410 addr = (char*)shmat(shmid, req_addr, 0); 3411 int err = errno; 3412 3413 // Remove shmid. If shmat() is successful, the actual shared memory segment 3414 // will be deleted when it's detached by shmdt() or when the process 3415 // terminates. If shmat() is not successful this will remove the shared 3416 // segment immediately. 3417 shmctl(shmid, IPC_RMID, NULL); 3418 3419 if ((intptr_t)addr == -1) { 3420 if (warn_on_failure) { 3421 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3422 warning(msg); 3423 } 3424 return NULL; 3425 } 3426 3427 return addr; 3428} 3429 3430static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) { 3431 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3432 3433 bool warn_on_failure = UseLargePages && 3434 (!FLAG_IS_DEFAULT(UseLargePages) || 3435 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3436 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3437 3438 if (warn_on_failure) { 3439 char msg[128]; 3440 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3441 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3442 warning(msg); 3443 } 3444} 3445 3446char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) { 3447 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3448 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3449 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3450 3451 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3452 char* addr = (char*)::mmap(req_addr, bytes, prot, 3453 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3454 -1, 0); 3455 3456 if (addr == MAP_FAILED) { 3457 warn_on_large_pages_failure(req_addr, bytes, errno); 3458 return NULL; 3459 } 3460 3461 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3462 3463 return addr; 3464} 3465 3466char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3467 size_t large_page_size = os::large_page_size(); 3468 3469 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3470 3471 // Allocate small pages. 3472 3473 char* start; 3474 if (req_addr != NULL) { 3475 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3476 assert(is_size_aligned(bytes, alignment), "Must be"); 3477 start = os::reserve_memory(bytes, req_addr); 3478 assert(start == NULL || start == req_addr, "Must be"); 3479 } else { 3480 start = os::reserve_memory_aligned(bytes, alignment); 3481 } 3482 3483 if (start == NULL) { 3484 return NULL; 3485 } 3486 3487 assert(is_ptr_aligned(start, alignment), "Must be"); 3488 3489 // os::reserve_memory_special will record this memory area. 3490 // Need to release it here to prevent overlapping reservations. 3491 MemTracker::record_virtual_memory_release((address)start, bytes); 3492 3493 char* end = start + bytes; 3494 3495 // Find the regions of the allocated chunk that can be promoted to large pages. 3496 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3497 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3498 3499 size_t lp_bytes = lp_end - lp_start; 3500 3501 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3502 3503 if (lp_bytes == 0) { 3504 // The mapped region doesn't even span the start and the end of a large page. 3505 // Fall back to allocate a non-special area. 3506 ::munmap(start, end - start); 3507 return NULL; 3508 } 3509 3510 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3511 3512 3513 void* result; 3514 3515 if (start != lp_start) { 3516 result = ::mmap(start, lp_start - start, prot, 3517 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3518 -1, 0); 3519 if (result == MAP_FAILED) { 3520 ::munmap(lp_start, end - lp_start); 3521 return NULL; 3522 } 3523 } 3524 3525 result = ::mmap(lp_start, lp_bytes, prot, 3526 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3527 -1, 0); 3528 if (result == MAP_FAILED) { 3529 warn_on_large_pages_failure(req_addr, bytes, errno); 3530 // If the mmap above fails, the large pages region will be unmapped and we 3531 // have regions before and after with small pages. Release these regions. 3532 // 3533 // | mapped | unmapped | mapped | 3534 // ^ ^ ^ ^ 3535 // start lp_start lp_end end 3536 // 3537 ::munmap(start, lp_start - start); 3538 ::munmap(lp_end, end - lp_end); 3539 return NULL; 3540 } 3541 3542 if (lp_end != end) { 3543 result = ::mmap(lp_end, end - lp_end, prot, 3544 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3545 -1, 0); 3546 if (result == MAP_FAILED) { 3547 ::munmap(start, lp_end - start); 3548 return NULL; 3549 } 3550 } 3551 3552 return start; 3553} 3554 3555char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3556 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3557 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3558 assert(is_power_of_2(alignment), "Must be"); 3559 assert(is_power_of_2(os::large_page_size()), "Must be"); 3560 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3561 3562 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3563 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3564 } else { 3565 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3566 } 3567} 3568 3569char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3570 assert(UseLargePages, "only for large pages"); 3571 3572 char* addr; 3573 if (UseSHM) { 3574 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3575 } else { 3576 assert(UseHugeTLBFS, "must be"); 3577 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3578 } 3579 3580 if (addr != NULL) { 3581 if (UseNUMAInterleaving) { 3582 numa_make_global(addr, bytes); 3583 } 3584 3585 // The memory is committed 3586 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC); 3587 } 3588 3589 return addr; 3590} 3591 3592bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3593 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3594 return shmdt(base) == 0; 3595} 3596 3597bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3598 return pd_release_memory(base, bytes); 3599} 3600 3601bool os::release_memory_special(char* base, size_t bytes) { 3602 assert(UseLargePages, "only for large pages"); 3603 3604 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3605 3606 bool res; 3607 if (UseSHM) { 3608 res = os::Linux::release_memory_special_shm(base, bytes); 3609 } else { 3610 assert(UseHugeTLBFS, "must be"); 3611 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3612 } 3613 3614 if (res) { 3615 tkr.record((address)base, bytes); 3616 } else { 3617 tkr.discard(); 3618 } 3619 3620 return res; 3621} 3622 3623size_t os::large_page_size() { 3624 return _large_page_size; 3625} 3626 3627// With SysV SHM the entire memory region must be allocated as shared 3628// memory. 3629// HugeTLBFS allows application to commit large page memory on demand. 3630// However, when committing memory with HugeTLBFS fails, the region 3631// that was supposed to be committed will lose the old reservation 3632// and allow other threads to steal that memory region. Because of this 3633// behavior we can't commit HugeTLBFS memory. 3634bool os::can_commit_large_page_memory() { 3635 return UseTransparentHugePages; 3636} 3637 3638bool os::can_execute_large_page_memory() { 3639 return UseTransparentHugePages || UseHugeTLBFS; 3640} 3641 3642// Reserve memory at an arbitrary address, only if that area is 3643// available (and not reserved for something else). 3644 3645char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3646 const int max_tries = 10; 3647 char* base[max_tries]; 3648 size_t size[max_tries]; 3649 const size_t gap = 0x000000; 3650 3651 // Assert only that the size is a multiple of the page size, since 3652 // that's all that mmap requires, and since that's all we really know 3653 // about at this low abstraction level. If we need higher alignment, 3654 // we can either pass an alignment to this method or verify alignment 3655 // in one of the methods further up the call chain. See bug 5044738. 3656 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3657 3658 // Repeatedly allocate blocks until the block is allocated at the 3659 // right spot. Give up after max_tries. Note that reserve_memory() will 3660 // automatically update _highest_vm_reserved_address if the call is 3661 // successful. The variable tracks the highest memory address every reserved 3662 // by JVM. It is used to detect heap-stack collision if running with 3663 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3664 // space than needed, it could confuse the collision detecting code. To 3665 // solve the problem, save current _highest_vm_reserved_address and 3666 // calculate the correct value before return. 3667 address old_highest = _highest_vm_reserved_address; 3668 3669 // Linux mmap allows caller to pass an address as hint; give it a try first, 3670 // if kernel honors the hint then we can return immediately. 3671 char * addr = anon_mmap(requested_addr, bytes, false); 3672 if (addr == requested_addr) { 3673 return requested_addr; 3674 } 3675 3676 if (addr != NULL) { 3677 // mmap() is successful but it fails to reserve at the requested address 3678 anon_munmap(addr, bytes); 3679 } 3680 3681 int i; 3682 for (i = 0; i < max_tries; ++i) { 3683 base[i] = reserve_memory(bytes); 3684 3685 if (base[i] != NULL) { 3686 // Is this the block we wanted? 3687 if (base[i] == requested_addr) { 3688 size[i] = bytes; 3689 break; 3690 } 3691 3692 // Does this overlap the block we wanted? Give back the overlapped 3693 // parts and try again. 3694 3695 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3696 if (top_overlap >= 0 && top_overlap < bytes) { 3697 unmap_memory(base[i], top_overlap); 3698 base[i] += top_overlap; 3699 size[i] = bytes - top_overlap; 3700 } else { 3701 size_t bottom_overlap = base[i] + bytes - requested_addr; 3702 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3703 unmap_memory(requested_addr, bottom_overlap); 3704 size[i] = bytes - bottom_overlap; 3705 } else { 3706 size[i] = bytes; 3707 } 3708 } 3709 } 3710 } 3711 3712 // Give back the unused reserved pieces. 3713 3714 for (int j = 0; j < i; ++j) { 3715 if (base[j] != NULL) { 3716 unmap_memory(base[j], size[j]); 3717 } 3718 } 3719 3720 if (i < max_tries) { 3721 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3722 return requested_addr; 3723 } else { 3724 _highest_vm_reserved_address = old_highest; 3725 return NULL; 3726 } 3727} 3728 3729size_t os::read(int fd, void *buf, unsigned int nBytes) { 3730 return ::read(fd, buf, nBytes); 3731} 3732 3733// 3734// Short sleep, direct OS call. 3735// 3736// Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3737// sched_yield(2) will actually give up the CPU: 3738// 3739// * Alone on this pariticular CPU, keeps running. 3740// * Before the introduction of "skip_buddy" with "compat_yield" disabled 3741// (pre 2.6.39). 3742// 3743// So calling this with 0 is an alternative. 3744// 3745void os::naked_short_sleep(jlong ms) { 3746 struct timespec req; 3747 3748 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3749 req.tv_sec = 0; 3750 if (ms > 0) { 3751 req.tv_nsec = (ms % 1000) * 1000000; 3752 } 3753 else { 3754 req.tv_nsec = 1; 3755 } 3756 3757 nanosleep(&req, NULL); 3758 3759 return; 3760} 3761 3762// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3763void os::infinite_sleep() { 3764 while (true) { // sleep forever ... 3765 ::sleep(100); // ... 100 seconds at a time 3766 } 3767} 3768 3769// Used to convert frequent JVM_Yield() to nops 3770bool os::dont_yield() { 3771 return DontYieldALot; 3772} 3773 3774void os::yield() { 3775 sched_yield(); 3776} 3777 3778os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3779 3780void os::yield_all() { 3781 // Yields to all threads, including threads with lower priorities 3782 // Threads on Linux are all with same priority. The Solaris style 3783 // os::yield_all() with nanosleep(1ms) is not necessary. 3784 sched_yield(); 3785} 3786 3787//////////////////////////////////////////////////////////////////////////////// 3788// thread priority support 3789 3790// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3791// only supports dynamic priority, static priority must be zero. For real-time 3792// applications, Linux supports SCHED_RR which allows static priority (1-99). 3793// However, for large multi-threaded applications, SCHED_RR is not only slower 3794// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3795// of 5 runs - Sep 2005). 3796// 3797// The following code actually changes the niceness of kernel-thread/LWP. It 3798// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3799// not the entire user process, and user level threads are 1:1 mapped to kernel 3800// threads. It has always been the case, but could change in the future. For 3801// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3802// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3803 3804int os::java_to_os_priority[CriticalPriority + 1] = { 3805 19, // 0 Entry should never be used 3806 3807 4, // 1 MinPriority 3808 3, // 2 3809 2, // 3 3810 3811 1, // 4 3812 0, // 5 NormPriority 3813 -1, // 6 3814 3815 -2, // 7 3816 -3, // 8 3817 -4, // 9 NearMaxPriority 3818 3819 -5, // 10 MaxPriority 3820 3821 -5 // 11 CriticalPriority 3822}; 3823 3824static int prio_init() { 3825 if (ThreadPriorityPolicy == 1) { 3826 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3827 // if effective uid is not root. Perhaps, a more elegant way of doing 3828 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3829 if (geteuid() != 0) { 3830 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3831 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3832 } 3833 ThreadPriorityPolicy = 0; 3834 } 3835 } 3836 if (UseCriticalJavaThreadPriority) { 3837 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3838 } 3839 return 0; 3840} 3841 3842OSReturn os::set_native_priority(Thread* thread, int newpri) { 3843 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3844 3845 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3846 return (ret == 0) ? OS_OK : OS_ERR; 3847} 3848 3849OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3850 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3851 *priority_ptr = java_to_os_priority[NormPriority]; 3852 return OS_OK; 3853 } 3854 3855 errno = 0; 3856 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3857 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3858} 3859 3860// Hint to the underlying OS that a task switch would not be good. 3861// Void return because it's a hint and can fail. 3862void os::hint_no_preempt() {} 3863 3864//////////////////////////////////////////////////////////////////////////////// 3865// suspend/resume support 3866 3867// the low-level signal-based suspend/resume support is a remnant from the 3868// old VM-suspension that used to be for java-suspension, safepoints etc, 3869// within hotspot. Now there is a single use-case for this: 3870// - calling get_thread_pc() on the VMThread by the flat-profiler task 3871// that runs in the watcher thread. 3872// The remaining code is greatly simplified from the more general suspension 3873// code that used to be used. 3874// 3875// The protocol is quite simple: 3876// - suspend: 3877// - sends a signal to the target thread 3878// - polls the suspend state of the osthread using a yield loop 3879// - target thread signal handler (SR_handler) sets suspend state 3880// and blocks in sigsuspend until continued 3881// - resume: 3882// - sets target osthread state to continue 3883// - sends signal to end the sigsuspend loop in the SR_handler 3884// 3885// Note that the SR_lock plays no role in this suspend/resume protocol. 3886// 3887 3888static void resume_clear_context(OSThread *osthread) { 3889 osthread->set_ucontext(NULL); 3890 osthread->set_siginfo(NULL); 3891} 3892 3893static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3894 osthread->set_ucontext(context); 3895 osthread->set_siginfo(siginfo); 3896} 3897 3898// 3899// Handler function invoked when a thread's execution is suspended or 3900// resumed. We have to be careful that only async-safe functions are 3901// called here (Note: most pthread functions are not async safe and 3902// should be avoided.) 3903// 3904// Note: sigwait() is a more natural fit than sigsuspend() from an 3905// interface point of view, but sigwait() prevents the signal hander 3906// from being run. libpthread would get very confused by not having 3907// its signal handlers run and prevents sigwait()'s use with the 3908// mutex granting granting signal. 3909// 3910// Currently only ever called on the VMThread and JavaThreads (PC sampling) 3911// 3912static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3913 // Save and restore errno to avoid confusing native code with EINTR 3914 // after sigsuspend. 3915 int old_errno = errno; 3916 3917 Thread* thread = Thread::current(); 3918 OSThread* osthread = thread->osthread(); 3919 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3920 3921 os::SuspendResume::State current = osthread->sr.state(); 3922 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3923 suspend_save_context(osthread, siginfo, context); 3924 3925 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 3926 os::SuspendResume::State state = osthread->sr.suspended(); 3927 if (state == os::SuspendResume::SR_SUSPENDED) { 3928 sigset_t suspend_set; // signals for sigsuspend() 3929 3930 // get current set of blocked signals and unblock resume signal 3931 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3932 sigdelset(&suspend_set, SR_signum); 3933 3934 sr_semaphore.signal(); 3935 // wait here until we are resumed 3936 while (1) { 3937 sigsuspend(&suspend_set); 3938 3939 os::SuspendResume::State result = osthread->sr.running(); 3940 if (result == os::SuspendResume::SR_RUNNING) { 3941 sr_semaphore.signal(); 3942 break; 3943 } 3944 } 3945 3946 } else if (state == os::SuspendResume::SR_RUNNING) { 3947 // request was cancelled, continue 3948 } else { 3949 ShouldNotReachHere(); 3950 } 3951 3952 resume_clear_context(osthread); 3953 } else if (current == os::SuspendResume::SR_RUNNING) { 3954 // request was cancelled, continue 3955 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 3956 // ignore 3957 } else { 3958 // ignore 3959 } 3960 3961 errno = old_errno; 3962} 3963 3964 3965static int SR_initialize() { 3966 struct sigaction act; 3967 char *s; 3968 /* Get signal number to use for suspend/resume */ 3969 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3970 int sig = ::strtol(s, 0, 10); 3971 if (sig > 0 || sig < _NSIG) { 3972 SR_signum = sig; 3973 } 3974 } 3975 3976 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3977 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3978 3979 sigemptyset(&SR_sigset); 3980 sigaddset(&SR_sigset, SR_signum); 3981 3982 /* Set up signal handler for suspend/resume */ 3983 act.sa_flags = SA_RESTART|SA_SIGINFO; 3984 act.sa_handler = (void (*)(int)) SR_handler; 3985 3986 // SR_signum is blocked by default. 3987 // 4528190 - We also need to block pthread restart signal (32 on all 3988 // supported Linux platforms). Note that LinuxThreads need to block 3989 // this signal for all threads to work properly. So we don't have 3990 // to use hard-coded signal number when setting up the mask. 3991 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 3992 3993 if (sigaction(SR_signum, &act, 0) == -1) { 3994 return -1; 3995 } 3996 3997 // Save signal flag 3998 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 3999 return 0; 4000} 4001 4002static int sr_notify(OSThread* osthread) { 4003 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4004 assert_status(status == 0, status, "pthread_kill"); 4005 return status; 4006} 4007 4008// "Randomly" selected value for how long we want to spin 4009// before bailing out on suspending a thread, also how often 4010// we send a signal to a thread we want to resume 4011static const int RANDOMLY_LARGE_INTEGER = 1000000; 4012static const int RANDOMLY_LARGE_INTEGER2 = 100; 4013 4014// returns true on success and false on error - really an error is fatal 4015// but this seems the normal response to library errors 4016static bool do_suspend(OSThread* osthread) { 4017 assert(osthread->sr.is_running(), "thread should be running"); 4018 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4019 4020 // mark as suspended and send signal 4021 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4022 // failed to switch, state wasn't running? 4023 ShouldNotReachHere(); 4024 return false; 4025 } 4026 4027 if (sr_notify(osthread) != 0) { 4028 ShouldNotReachHere(); 4029 } 4030 4031 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4032 while (true) { 4033 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4034 break; 4035 } else { 4036 // timeout 4037 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4038 if (cancelled == os::SuspendResume::SR_RUNNING) { 4039 return false; 4040 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4041 // make sure that we consume the signal on the semaphore as well 4042 sr_semaphore.wait(); 4043 break; 4044 } else { 4045 ShouldNotReachHere(); 4046 return false; 4047 } 4048 } 4049 } 4050 4051 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4052 return true; 4053} 4054 4055static void do_resume(OSThread* osthread) { 4056 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4057 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4058 4059 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4060 // failed to switch to WAKEUP_REQUEST 4061 ShouldNotReachHere(); 4062 return; 4063 } 4064 4065 while (true) { 4066 if (sr_notify(osthread) == 0) { 4067 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4068 if (osthread->sr.is_running()) { 4069 return; 4070 } 4071 } 4072 } else { 4073 ShouldNotReachHere(); 4074 } 4075 } 4076 4077 guarantee(osthread->sr.is_running(), "Must be running!"); 4078} 4079 4080/////////////////////////////////////////////////////////////////////////////////// 4081// signal handling (except suspend/resume) 4082 4083// This routine may be used by user applications as a "hook" to catch signals. 4084// The user-defined signal handler must pass unrecognized signals to this 4085// routine, and if it returns true (non-zero), then the signal handler must 4086// return immediately. If the flag "abort_if_unrecognized" is true, then this 4087// routine will never retun false (zero), but instead will execute a VM panic 4088// routine kill the process. 4089// 4090// If this routine returns false, it is OK to call it again. This allows 4091// the user-defined signal handler to perform checks either before or after 4092// the VM performs its own checks. Naturally, the user code would be making 4093// a serious error if it tried to handle an exception (such as a null check 4094// or breakpoint) that the VM was generating for its own correct operation. 4095// 4096// This routine may recognize any of the following kinds of signals: 4097// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4098// It should be consulted by handlers for any of those signals. 4099// 4100// The caller of this routine must pass in the three arguments supplied 4101// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4102// field of the structure passed to sigaction(). This routine assumes that 4103// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4104// 4105// Note that the VM will print warnings if it detects conflicting signal 4106// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4107// 4108extern "C" JNIEXPORT int 4109JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 4110 void* ucontext, int abort_if_unrecognized); 4111 4112void signalHandler(int sig, siginfo_t* info, void* uc) { 4113 assert(info != NULL && uc != NULL, "it must be old kernel"); 4114 int orig_errno = errno; // Preserve errno value over signal handler. 4115 JVM_handle_linux_signal(sig, info, uc, true); 4116 errno = orig_errno; 4117} 4118 4119 4120// This boolean allows users to forward their own non-matching signals 4121// to JVM_handle_linux_signal, harmlessly. 4122bool os::Linux::signal_handlers_are_installed = false; 4123 4124// For signal-chaining 4125struct sigaction os::Linux::sigact[MAXSIGNUM]; 4126unsigned int os::Linux::sigs = 0; 4127bool os::Linux::libjsig_is_loaded = false; 4128typedef struct sigaction *(*get_signal_t)(int); 4129get_signal_t os::Linux::get_signal_action = NULL; 4130 4131struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4132 struct sigaction *actp = NULL; 4133 4134 if (libjsig_is_loaded) { 4135 // Retrieve the old signal handler from libjsig 4136 actp = (*get_signal_action)(sig); 4137 } 4138 if (actp == NULL) { 4139 // Retrieve the preinstalled signal handler from jvm 4140 actp = get_preinstalled_handler(sig); 4141 } 4142 4143 return actp; 4144} 4145 4146static bool call_chained_handler(struct sigaction *actp, int sig, 4147 siginfo_t *siginfo, void *context) { 4148 // Call the old signal handler 4149 if (actp->sa_handler == SIG_DFL) { 4150 // It's more reasonable to let jvm treat it as an unexpected exception 4151 // instead of taking the default action. 4152 return false; 4153 } else if (actp->sa_handler != SIG_IGN) { 4154 if ((actp->sa_flags & SA_NODEFER) == 0) { 4155 // automaticlly block the signal 4156 sigaddset(&(actp->sa_mask), sig); 4157 } 4158 4159 sa_handler_t hand; 4160 sa_sigaction_t sa; 4161 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4162 // retrieve the chained handler 4163 if (siginfo_flag_set) { 4164 sa = actp->sa_sigaction; 4165 } else { 4166 hand = actp->sa_handler; 4167 } 4168 4169 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4170 actp->sa_handler = SIG_DFL; 4171 } 4172 4173 // try to honor the signal mask 4174 sigset_t oset; 4175 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4176 4177 // call into the chained handler 4178 if (siginfo_flag_set) { 4179 (*sa)(sig, siginfo, context); 4180 } else { 4181 (*hand)(sig); 4182 } 4183 4184 // restore the signal mask 4185 pthread_sigmask(SIG_SETMASK, &oset, 0); 4186 } 4187 // Tell jvm's signal handler the signal is taken care of. 4188 return true; 4189} 4190 4191bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4192 bool chained = false; 4193 // signal-chaining 4194 if (UseSignalChaining) { 4195 struct sigaction *actp = get_chained_signal_action(sig); 4196 if (actp != NULL) { 4197 chained = call_chained_handler(actp, sig, siginfo, context); 4198 } 4199 } 4200 return chained; 4201} 4202 4203struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4204 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 4205 return &sigact[sig]; 4206 } 4207 return NULL; 4208} 4209 4210void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4211 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4212 sigact[sig] = oldAct; 4213 sigs |= (unsigned int)1 << sig; 4214} 4215 4216// for diagnostic 4217int os::Linux::sigflags[MAXSIGNUM]; 4218 4219int os::Linux::get_our_sigflags(int sig) { 4220 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4221 return sigflags[sig]; 4222} 4223 4224void os::Linux::set_our_sigflags(int sig, int flags) { 4225 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4226 sigflags[sig] = flags; 4227} 4228 4229void os::Linux::set_signal_handler(int sig, bool set_installed) { 4230 // Check for overwrite. 4231 struct sigaction oldAct; 4232 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4233 4234 void* oldhand = oldAct.sa_sigaction 4235 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4236 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4237 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4238 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4239 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4240 if (AllowUserSignalHandlers || !set_installed) { 4241 // Do not overwrite; user takes responsibility to forward to us. 4242 return; 4243 } else if (UseSignalChaining) { 4244 // save the old handler in jvm 4245 save_preinstalled_handler(sig, oldAct); 4246 // libjsig also interposes the sigaction() call below and saves the 4247 // old sigaction on it own. 4248 } else { 4249 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4250 "%#lx for signal %d.", (long)oldhand, sig)); 4251 } 4252 } 4253 4254 struct sigaction sigAct; 4255 sigfillset(&(sigAct.sa_mask)); 4256 sigAct.sa_handler = SIG_DFL; 4257 if (!set_installed) { 4258 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4259 } else { 4260 sigAct.sa_sigaction = signalHandler; 4261 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4262 } 4263 // Save flags, which are set by ours 4264 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4265 sigflags[sig] = sigAct.sa_flags; 4266 4267 int ret = sigaction(sig, &sigAct, &oldAct); 4268 assert(ret == 0, "check"); 4269 4270 void* oldhand2 = oldAct.sa_sigaction 4271 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4272 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4273 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4274} 4275 4276// install signal handlers for signals that HotSpot needs to 4277// handle in order to support Java-level exception handling. 4278 4279void os::Linux::install_signal_handlers() { 4280 if (!signal_handlers_are_installed) { 4281 signal_handlers_are_installed = true; 4282 4283 // signal-chaining 4284 typedef void (*signal_setting_t)(); 4285 signal_setting_t begin_signal_setting = NULL; 4286 signal_setting_t end_signal_setting = NULL; 4287 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4288 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4289 if (begin_signal_setting != NULL) { 4290 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4291 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4292 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4293 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4294 libjsig_is_loaded = true; 4295 assert(UseSignalChaining, "should enable signal-chaining"); 4296 } 4297 if (libjsig_is_loaded) { 4298 // Tell libjsig jvm is setting signal handlers 4299 (*begin_signal_setting)(); 4300 } 4301 4302 set_signal_handler(SIGSEGV, true); 4303 set_signal_handler(SIGPIPE, true); 4304 set_signal_handler(SIGBUS, true); 4305 set_signal_handler(SIGILL, true); 4306 set_signal_handler(SIGFPE, true); 4307#if defined(PPC64) 4308 set_signal_handler(SIGTRAP, true); 4309#endif 4310 set_signal_handler(SIGXFSZ, true); 4311 4312 if (libjsig_is_loaded) { 4313 // Tell libjsig jvm finishes setting signal handlers 4314 (*end_signal_setting)(); 4315 } 4316 4317 // We don't activate signal checker if libjsig is in place, we trust ourselves 4318 // and if UserSignalHandler is installed all bets are off. 4319 // Log that signal checking is off only if -verbose:jni is specified. 4320 if (CheckJNICalls) { 4321 if (libjsig_is_loaded) { 4322 if (PrintJNIResolving) { 4323 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4324 } 4325 check_signals = false; 4326 } 4327 if (AllowUserSignalHandlers) { 4328 if (PrintJNIResolving) { 4329 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4330 } 4331 check_signals = false; 4332 } 4333 } 4334 } 4335} 4336 4337// This is the fastest way to get thread cpu time on Linux. 4338// Returns cpu time (user+sys) for any thread, not only for current. 4339// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4340// It might work on 2.6.10+ with a special kernel/glibc patch. 4341// For reference, please, see IEEE Std 1003.1-2004: 4342// http://www.unix.org/single_unix_specification 4343 4344jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4345 struct timespec tp; 4346 int rc = os::Linux::clock_gettime(clockid, &tp); 4347 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4348 4349 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4350} 4351 4352///// 4353// glibc on Linux platform uses non-documented flag 4354// to indicate, that some special sort of signal 4355// trampoline is used. 4356// We will never set this flag, and we should 4357// ignore this flag in our diagnostic 4358#ifdef SIGNIFICANT_SIGNAL_MASK 4359#undef SIGNIFICANT_SIGNAL_MASK 4360#endif 4361#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4362 4363static const char* get_signal_handler_name(address handler, 4364 char* buf, int buflen) { 4365 int offset; 4366 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4367 if (found) { 4368 // skip directory names 4369 const char *p1, *p2; 4370 p1 = buf; 4371 size_t len = strlen(os::file_separator()); 4372 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4373 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4374 } else { 4375 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4376 } 4377 return buf; 4378} 4379 4380static void print_signal_handler(outputStream* st, int sig, 4381 char* buf, size_t buflen) { 4382 struct sigaction sa; 4383 4384 sigaction(sig, NULL, &sa); 4385 4386 // See comment for SIGNIFICANT_SIGNAL_MASK define 4387 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4388 4389 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4390 4391 address handler = (sa.sa_flags & SA_SIGINFO) 4392 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4393 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4394 4395 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4396 st->print("SIG_DFL"); 4397 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4398 st->print("SIG_IGN"); 4399 } else { 4400 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4401 } 4402 4403 st->print(", sa_mask[0]="); 4404 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4405 4406 address rh = VMError::get_resetted_sighandler(sig); 4407 // May be, handler was resetted by VMError? 4408 if(rh != NULL) { 4409 handler = rh; 4410 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4411 } 4412 4413 st->print(", sa_flags="); 4414 os::Posix::print_sa_flags(st, sa.sa_flags); 4415 4416 // Check: is it our handler? 4417 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4418 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4419 // It is our signal handler 4420 // check for flags, reset system-used one! 4421 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4422 st->print( 4423 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4424 os::Linux::get_our_sigflags(sig)); 4425 } 4426 } 4427 st->cr(); 4428} 4429 4430 4431#define DO_SIGNAL_CHECK(sig) \ 4432 if (!sigismember(&check_signal_done, sig)) \ 4433 os::Linux::check_signal_handler(sig) 4434 4435// This method is a periodic task to check for misbehaving JNI applications 4436// under CheckJNI, we can add any periodic checks here 4437 4438void os::run_periodic_checks() { 4439 4440 if (check_signals == false) return; 4441 4442 // SEGV and BUS if overridden could potentially prevent 4443 // generation of hs*.log in the event of a crash, debugging 4444 // such a case can be very challenging, so we absolutely 4445 // check the following for a good measure: 4446 DO_SIGNAL_CHECK(SIGSEGV); 4447 DO_SIGNAL_CHECK(SIGILL); 4448 DO_SIGNAL_CHECK(SIGFPE); 4449 DO_SIGNAL_CHECK(SIGBUS); 4450 DO_SIGNAL_CHECK(SIGPIPE); 4451 DO_SIGNAL_CHECK(SIGXFSZ); 4452#if defined(PPC64) 4453 DO_SIGNAL_CHECK(SIGTRAP); 4454#endif 4455 4456 // ReduceSignalUsage allows the user to override these handlers 4457 // see comments at the very top and jvm_solaris.h 4458 if (!ReduceSignalUsage) { 4459 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4460 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4461 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4462 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4463 } 4464 4465 DO_SIGNAL_CHECK(SR_signum); 4466 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4467} 4468 4469typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4470 4471static os_sigaction_t os_sigaction = NULL; 4472 4473void os::Linux::check_signal_handler(int sig) { 4474 char buf[O_BUFLEN]; 4475 address jvmHandler = NULL; 4476 4477 4478 struct sigaction act; 4479 if (os_sigaction == NULL) { 4480 // only trust the default sigaction, in case it has been interposed 4481 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4482 if (os_sigaction == NULL) return; 4483 } 4484 4485 os_sigaction(sig, (struct sigaction*)NULL, &act); 4486 4487 4488 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4489 4490 address thisHandler = (act.sa_flags & SA_SIGINFO) 4491 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4492 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 4493 4494 4495 switch(sig) { 4496 case SIGSEGV: 4497 case SIGBUS: 4498 case SIGFPE: 4499 case SIGPIPE: 4500 case SIGILL: 4501 case SIGXFSZ: 4502 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4503 break; 4504 4505 case SHUTDOWN1_SIGNAL: 4506 case SHUTDOWN2_SIGNAL: 4507 case SHUTDOWN3_SIGNAL: 4508 case BREAK_SIGNAL: 4509 jvmHandler = (address)user_handler(); 4510 break; 4511 4512 case INTERRUPT_SIGNAL: 4513 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4514 break; 4515 4516 default: 4517 if (sig == SR_signum) { 4518 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4519 } else { 4520 return; 4521 } 4522 break; 4523 } 4524 4525 if (thisHandler != jvmHandler) { 4526 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4527 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4528 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4529 // No need to check this sig any longer 4530 sigaddset(&check_signal_done, sig); 4531 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4532 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4533 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4534 exception_name(sig, buf, O_BUFLEN)); 4535 } 4536 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4537 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4538 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4539 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4540 // No need to check this sig any longer 4541 sigaddset(&check_signal_done, sig); 4542 } 4543 4544 // Dump all the signal 4545 if (sigismember(&check_signal_done, sig)) { 4546 print_signal_handlers(tty, buf, O_BUFLEN); 4547 } 4548} 4549 4550extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 4551 4552extern bool signal_name(int signo, char* buf, size_t len); 4553 4554const char* os::exception_name(int exception_code, char* buf, size_t size) { 4555 if (0 < exception_code && exception_code <= SIGRTMAX) { 4556 // signal 4557 if (!signal_name(exception_code, buf, size)) { 4558 jio_snprintf(buf, size, "SIG%d", exception_code); 4559 } 4560 return buf; 4561 } else { 4562 return NULL; 4563 } 4564} 4565 4566// this is called _before_ the most of global arguments have been parsed 4567void os::init(void) { 4568 char dummy; /* used to get a guess on initial stack address */ 4569// first_hrtime = gethrtime(); 4570 4571 // With LinuxThreads the JavaMain thread pid (primordial thread) 4572 // is different than the pid of the java launcher thread. 4573 // So, on Linux, the launcher thread pid is passed to the VM 4574 // via the sun.java.launcher.pid property. 4575 // Use this property instead of getpid() if it was correctly passed. 4576 // See bug 6351349. 4577 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4578 4579 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4580 4581 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4582 4583 init_random(1234567); 4584 4585 ThreadCritical::initialize(); 4586 4587 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4588 if (Linux::page_size() == -1) { 4589 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4590 strerror(errno))); 4591 } 4592 init_page_sizes((size_t) Linux::page_size()); 4593 4594 Linux::initialize_system_info(); 4595 4596 // main_thread points to the aboriginal thread 4597 Linux::_main_thread = pthread_self(); 4598 4599 Linux::clock_init(); 4600 initial_time_count = javaTimeNanos(); 4601 4602 // pthread_condattr initialization for monotonic clock 4603 int status; 4604 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4605 if ((status = pthread_condattr_init(_condattr)) != 0) { 4606 fatal(err_msg("pthread_condattr_init: %s", strerror(status))); 4607 } 4608 // Only set the clock if CLOCK_MONOTONIC is available 4609 if (os::supports_monotonic_clock()) { 4610 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4611 if (status == EINVAL) { 4612 warning("Unable to use monotonic clock with relative timed-waits" \ 4613 " - changes to the time-of-day clock may have adverse affects"); 4614 } else { 4615 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status))); 4616 } 4617 } 4618 } 4619 // else it defaults to CLOCK_REALTIME 4620 4621 pthread_mutex_init(&dl_mutex, NULL); 4622 4623 // If the pagesize of the VM is greater than 8K determine the appropriate 4624 // number of initial guard pages. The user can change this with the 4625 // command line arguments, if needed. 4626 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4627 StackYellowPages = 1; 4628 StackRedPages = 1; 4629 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4630 } 4631} 4632 4633// To install functions for atexit system call 4634extern "C" { 4635 static void perfMemory_exit_helper() { 4636 perfMemory_exit(); 4637 } 4638} 4639 4640// this is called _after_ the global arguments have been parsed 4641jint os::init_2(void) 4642{ 4643 Linux::fast_thread_clock_init(); 4644 4645 // Allocate a single page and mark it as readable for safepoint polling 4646 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4647 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 4648 4649 os::set_polling_page( polling_page ); 4650 4651#ifndef PRODUCT 4652 if(Verbose && PrintMiscellaneous) 4653 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 4654#endif 4655 4656 if (!UseMembar) { 4657 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4658 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4659 os::set_memory_serialize_page( mem_serialize_page ); 4660 4661#ifndef PRODUCT 4662 if(Verbose && PrintMiscellaneous) 4663 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 4664#endif 4665 } 4666 4667 // initialize suspend/resume support - must do this before signal_sets_init() 4668 if (SR_initialize() != 0) { 4669 perror("SR_initialize failed"); 4670 return JNI_ERR; 4671 } 4672 4673 Linux::signal_sets_init(); 4674 Linux::install_signal_handlers(); 4675 4676 // Check minimum allowable stack size for thread creation and to initialize 4677 // the java system classes, including StackOverflowError - depends on page 4678 // size. Add a page for compiler2 recursion in main thread. 4679 // Add in 2*BytesPerWord times page size to account for VM stack during 4680 // class initialization depending on 32 or 64 bit VM. 4681 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4682 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4683 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4684 4685 size_t threadStackSizeInBytes = ThreadStackSize * K; 4686 if (threadStackSizeInBytes != 0 && 4687 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4688 tty->print_cr("\nThe stack size specified is too small, " 4689 "Specify at least %dk", 4690 os::Linux::min_stack_allowed/ K); 4691 return JNI_ERR; 4692 } 4693 4694 // Make the stack size a multiple of the page size so that 4695 // the yellow/red zones can be guarded. 4696 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4697 vm_page_size())); 4698 4699 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4700 4701#if defined(IA32) 4702 workaround_expand_exec_shield_cs_limit(); 4703#endif 4704 4705 Linux::libpthread_init(); 4706 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4707 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4708 Linux::glibc_version(), Linux::libpthread_version(), 4709 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4710 } 4711 4712 if (UseNUMA) { 4713 if (!Linux::libnuma_init()) { 4714 UseNUMA = false; 4715 } else { 4716 if ((Linux::numa_max_node() < 1)) { 4717 // There's only one node(they start from 0), disable NUMA. 4718 UseNUMA = false; 4719 } 4720 } 4721 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4722 // we can make the adaptive lgrp chunk resizing work. If the user specified 4723 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4724 // disable adaptive resizing. 4725 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4726 if (FLAG_IS_DEFAULT(UseNUMA)) { 4727 UseNUMA = false; 4728 } else { 4729 if (FLAG_IS_DEFAULT(UseLargePages) && 4730 FLAG_IS_DEFAULT(UseSHM) && 4731 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4732 UseLargePages = false; 4733 } else { 4734 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing"); 4735 UseAdaptiveSizePolicy = false; 4736 UseAdaptiveNUMAChunkSizing = false; 4737 } 4738 } 4739 } 4740 if (!UseNUMA && ForceNUMA) { 4741 UseNUMA = true; 4742 } 4743 } 4744 4745 if (MaxFDLimit) { 4746 // set the number of file descriptors to max. print out error 4747 // if getrlimit/setrlimit fails but continue regardless. 4748 struct rlimit nbr_files; 4749 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4750 if (status != 0) { 4751 if (PrintMiscellaneous && (Verbose || WizardMode)) 4752 perror("os::init_2 getrlimit failed"); 4753 } else { 4754 nbr_files.rlim_cur = nbr_files.rlim_max; 4755 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4756 if (status != 0) { 4757 if (PrintMiscellaneous && (Verbose || WizardMode)) 4758 perror("os::init_2 setrlimit failed"); 4759 } 4760 } 4761 } 4762 4763 // Initialize lock used to serialize thread creation (see os::create_thread) 4764 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4765 4766 // at-exit methods are called in the reverse order of their registration. 4767 // atexit functions are called on return from main or as a result of a 4768 // call to exit(3C). There can be only 32 of these functions registered 4769 // and atexit() does not set errno. 4770 4771 if (PerfAllowAtExitRegistration) { 4772 // only register atexit functions if PerfAllowAtExitRegistration is set. 4773 // atexit functions can be delayed until process exit time, which 4774 // can be problematic for embedded VM situations. Embedded VMs should 4775 // call DestroyJavaVM() to assure that VM resources are released. 4776 4777 // note: perfMemory_exit_helper atexit function may be removed in 4778 // the future if the appropriate cleanup code can be added to the 4779 // VM_Exit VMOperation's doit method. 4780 if (atexit(perfMemory_exit_helper) != 0) { 4781 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4782 } 4783 } 4784 4785 // initialize thread priority policy 4786 prio_init(); 4787 4788 return JNI_OK; 4789} 4790 4791// this is called at the end of vm_initialization 4792void os::init_3(void) { 4793#ifdef JAVASE_EMBEDDED 4794 // Start the MemNotifyThread 4795 if (LowMemoryProtection) { 4796 MemNotifyThread::start(); 4797 } 4798 return; 4799#endif 4800} 4801 4802// Mark the polling page as unreadable 4803void os::make_polling_page_unreadable(void) { 4804 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4805 fatal("Could not disable polling page"); 4806}; 4807 4808// Mark the polling page as readable 4809void os::make_polling_page_readable(void) { 4810 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4811 fatal("Could not enable polling page"); 4812 } 4813}; 4814 4815int os::active_processor_count() { 4816 // Linux doesn't yet have a (official) notion of processor sets, 4817 // so just return the number of online processors. 4818 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4819 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4820 return online_cpus; 4821} 4822 4823void os::set_native_thread_name(const char *name) { 4824 // Not yet implemented. 4825 return; 4826} 4827 4828bool os::distribute_processes(uint length, uint* distribution) { 4829 // Not yet implemented. 4830 return false; 4831} 4832 4833bool os::bind_to_processor(uint processor_id) { 4834 // Not yet implemented. 4835 return false; 4836} 4837 4838/// 4839 4840void os::SuspendedThreadTask::internal_do_task() { 4841 if (do_suspend(_thread->osthread())) { 4842 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4843 do_task(context); 4844 do_resume(_thread->osthread()); 4845 } 4846} 4847 4848class PcFetcher : public os::SuspendedThreadTask { 4849public: 4850 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4851 ExtendedPC result(); 4852protected: 4853 void do_task(const os::SuspendedThreadTaskContext& context); 4854private: 4855 ExtendedPC _epc; 4856}; 4857 4858ExtendedPC PcFetcher::result() { 4859 guarantee(is_done(), "task is not done yet."); 4860 return _epc; 4861} 4862 4863void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4864 Thread* thread = context.thread(); 4865 OSThread* osthread = thread->osthread(); 4866 if (osthread->ucontext() != NULL) { 4867 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4868 } else { 4869 // NULL context is unexpected, double-check this is the VMThread 4870 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4871 } 4872} 4873 4874// Suspends the target using the signal mechanism and then grabs the PC before 4875// resuming the target. Used by the flat-profiler only 4876ExtendedPC os::get_thread_pc(Thread* thread) { 4877 // Make sure that it is called by the watcher for the VMThread 4878 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4879 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4880 4881 PcFetcher fetcher(thread); 4882 fetcher.run(); 4883 return fetcher.result(); 4884} 4885 4886int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4887{ 4888 if (is_NPTL()) { 4889 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4890 } else { 4891 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4892 // word back to default 64bit precision if condvar is signaled. Java 4893 // wants 53bit precision. Save and restore current value. 4894 int fpu = get_fpu_control_word(); 4895 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4896 set_fpu_control_word(fpu); 4897 return status; 4898 } 4899} 4900 4901//////////////////////////////////////////////////////////////////////////////// 4902// debug support 4903 4904bool os::find(address addr, outputStream* st) { 4905 Dl_info dlinfo; 4906 memset(&dlinfo, 0, sizeof(dlinfo)); 4907 if (dladdr(addr, &dlinfo) != 0) { 4908 st->print(PTR_FORMAT ": ", addr); 4909 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 4910 st->print("%s+%#x", dlinfo.dli_sname, 4911 addr - (intptr_t)dlinfo.dli_saddr); 4912 } else if (dlinfo.dli_fbase != NULL) { 4913 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4914 } else { 4915 st->print("<absolute address>"); 4916 } 4917 if (dlinfo.dli_fname != NULL) { 4918 st->print(" in %s", dlinfo.dli_fname); 4919 } 4920 if (dlinfo.dli_fbase != NULL) { 4921 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4922 } 4923 st->cr(); 4924 4925 if (Verbose) { 4926 // decode some bytes around the PC 4927 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 4928 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 4929 address lowest = (address) dlinfo.dli_sname; 4930 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4931 if (begin < lowest) begin = lowest; 4932 Dl_info dlinfo2; 4933 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 4934 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 4935 end = (address) dlinfo2.dli_saddr; 4936 Disassembler::decode(begin, end, st); 4937 } 4938 return true; 4939 } 4940 return false; 4941} 4942 4943//////////////////////////////////////////////////////////////////////////////// 4944// misc 4945 4946// This does not do anything on Linux. This is basically a hook for being 4947// able to use structured exception handling (thread-local exception filters) 4948// on, e.g., Win32. 4949void 4950os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4951 JavaCallArguments* args, Thread* thread) { 4952 f(value, method, args, thread); 4953} 4954 4955void os::print_statistics() { 4956} 4957 4958int os::message_box(const char* title, const char* message) { 4959 int i; 4960 fdStream err(defaultStream::error_fd()); 4961 for (i = 0; i < 78; i++) err.print_raw("="); 4962 err.cr(); 4963 err.print_raw_cr(title); 4964 for (i = 0; i < 78; i++) err.print_raw("-"); 4965 err.cr(); 4966 err.print_raw_cr(message); 4967 for (i = 0; i < 78; i++) err.print_raw("="); 4968 err.cr(); 4969 4970 char buf[16]; 4971 // Prevent process from exiting upon "read error" without consuming all CPU 4972 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 4973 4974 return buf[0] == 'y' || buf[0] == 'Y'; 4975} 4976 4977int os::stat(const char *path, struct stat *sbuf) { 4978 char pathbuf[MAX_PATH]; 4979 if (strlen(path) > MAX_PATH - 1) { 4980 errno = ENAMETOOLONG; 4981 return -1; 4982 } 4983 os::native_path(strcpy(pathbuf, path)); 4984 return ::stat(pathbuf, sbuf); 4985} 4986 4987bool os::check_heap(bool force) { 4988 return true; 4989} 4990 4991int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 4992 return ::vsnprintf(buf, count, format, args); 4993} 4994 4995// Is a (classpath) directory empty? 4996bool os::dir_is_empty(const char* path) { 4997 DIR *dir = NULL; 4998 struct dirent *ptr; 4999 5000 dir = opendir(path); 5001 if (dir == NULL) return true; 5002 5003 /* Scan the directory */ 5004 bool result = true; 5005 char buf[sizeof(struct dirent) + MAX_PATH]; 5006 while (result && (ptr = ::readdir(dir)) != NULL) { 5007 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5008 result = false; 5009 } 5010 } 5011 closedir(dir); 5012 return result; 5013} 5014 5015// This code originates from JDK's sysOpen and open64_w 5016// from src/solaris/hpi/src/system_md.c 5017 5018#ifndef O_DELETE 5019#define O_DELETE 0x10000 5020#endif 5021 5022// Open a file. Unlink the file immediately after open returns 5023// if the specified oflag has the O_DELETE flag set. 5024// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 5025 5026int os::open(const char *path, int oflag, int mode) { 5027 5028 if (strlen(path) > MAX_PATH - 1) { 5029 errno = ENAMETOOLONG; 5030 return -1; 5031 } 5032 int fd; 5033 int o_delete = (oflag & O_DELETE); 5034 oflag = oflag & ~O_DELETE; 5035 5036 fd = ::open64(path, oflag, mode); 5037 if (fd == -1) return -1; 5038 5039 //If the open succeeded, the file might still be a directory 5040 { 5041 struct stat64 buf64; 5042 int ret = ::fstat64(fd, &buf64); 5043 int st_mode = buf64.st_mode; 5044 5045 if (ret != -1) { 5046 if ((st_mode & S_IFMT) == S_IFDIR) { 5047 errno = EISDIR; 5048 ::close(fd); 5049 return -1; 5050 } 5051 } else { 5052 ::close(fd); 5053 return -1; 5054 } 5055 } 5056 5057 /* 5058 * All file descriptors that are opened in the JVM and not 5059 * specifically destined for a subprocess should have the 5060 * close-on-exec flag set. If we don't set it, then careless 3rd 5061 * party native code might fork and exec without closing all 5062 * appropriate file descriptors (e.g. as we do in closeDescriptors in 5063 * UNIXProcess.c), and this in turn might: 5064 * 5065 * - cause end-of-file to fail to be detected on some file 5066 * descriptors, resulting in mysterious hangs, or 5067 * 5068 * - might cause an fopen in the subprocess to fail on a system 5069 * suffering from bug 1085341. 5070 * 5071 * (Yes, the default setting of the close-on-exec flag is a Unix 5072 * design flaw) 5073 * 5074 * See: 5075 * 1085341: 32-bit stdio routines should support file descriptors >255 5076 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5077 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5078 */ 5079#ifdef FD_CLOEXEC 5080 { 5081 int flags = ::fcntl(fd, F_GETFD); 5082 if (flags != -1) 5083 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5084 } 5085#endif 5086 5087 if (o_delete != 0) { 5088 ::unlink(path); 5089 } 5090 return fd; 5091} 5092 5093 5094// create binary file, rewriting existing file if required 5095int os::create_binary_file(const char* path, bool rewrite_existing) { 5096 int oflags = O_WRONLY | O_CREAT; 5097 if (!rewrite_existing) { 5098 oflags |= O_EXCL; 5099 } 5100 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5101} 5102 5103// return current position of file pointer 5104jlong os::current_file_offset(int fd) { 5105 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5106} 5107 5108// move file pointer to the specified offset 5109jlong os::seek_to_file_offset(int fd, jlong offset) { 5110 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5111} 5112 5113// This code originates from JDK's sysAvailable 5114// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5115 5116int os::available(int fd, jlong *bytes) { 5117 jlong cur, end; 5118 int mode; 5119 struct stat64 buf64; 5120 5121 if (::fstat64(fd, &buf64) >= 0) { 5122 mode = buf64.st_mode; 5123 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5124 /* 5125 * XXX: is the following call interruptible? If so, this might 5126 * need to go through the INTERRUPT_IO() wrapper as for other 5127 * blocking, interruptible calls in this file. 5128 */ 5129 int n; 5130 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5131 *bytes = n; 5132 return 1; 5133 } 5134 } 5135 } 5136 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5137 return 0; 5138 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5139 return 0; 5140 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5141 return 0; 5142 } 5143 *bytes = end - cur; 5144 return 1; 5145} 5146 5147int os::socket_available(int fd, jint *pbytes) { 5148 // Linux doc says EINTR not returned, unlike Solaris 5149 int ret = ::ioctl(fd, FIONREAD, pbytes); 5150 5151 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 5152 // is expected to return 0 on failure and 1 on success to the jdk. 5153 return (ret < 0) ? 0 : 1; 5154} 5155 5156// Map a block of memory. 5157char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5158 char *addr, size_t bytes, bool read_only, 5159 bool allow_exec) { 5160 int prot; 5161 int flags = MAP_PRIVATE; 5162 5163 if (read_only) { 5164 prot = PROT_READ; 5165 } else { 5166 prot = PROT_READ | PROT_WRITE; 5167 } 5168 5169 if (allow_exec) { 5170 prot |= PROT_EXEC; 5171 } 5172 5173 if (addr != NULL) { 5174 flags |= MAP_FIXED; 5175 } 5176 5177 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5178 fd, file_offset); 5179 if (mapped_address == MAP_FAILED) { 5180 return NULL; 5181 } 5182 return mapped_address; 5183} 5184 5185 5186// Remap a block of memory. 5187char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5188 char *addr, size_t bytes, bool read_only, 5189 bool allow_exec) { 5190 // same as map_memory() on this OS 5191 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5192 allow_exec); 5193} 5194 5195 5196// Unmap a block of memory. 5197bool os::pd_unmap_memory(char* addr, size_t bytes) { 5198 return munmap(addr, bytes) == 0; 5199} 5200 5201static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5202 5203static clockid_t thread_cpu_clockid(Thread* thread) { 5204 pthread_t tid = thread->osthread()->pthread_id(); 5205 clockid_t clockid; 5206 5207 // Get thread clockid 5208 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5209 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5210 return clockid; 5211} 5212 5213// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5214// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5215// of a thread. 5216// 5217// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5218// the fast estimate available on the platform. 5219 5220jlong os::current_thread_cpu_time() { 5221 if (os::Linux::supports_fast_thread_cpu_time()) { 5222 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5223 } else { 5224 // return user + sys since the cost is the same 5225 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5226 } 5227} 5228 5229jlong os::thread_cpu_time(Thread* thread) { 5230 // consistent with what current_thread_cpu_time() returns 5231 if (os::Linux::supports_fast_thread_cpu_time()) { 5232 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5233 } else { 5234 return slow_thread_cpu_time(thread, true /* user + sys */); 5235 } 5236} 5237 5238jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5239 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5240 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5241 } else { 5242 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5243 } 5244} 5245 5246jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5247 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5248 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5249 } else { 5250 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5251 } 5252} 5253 5254// 5255// -1 on error. 5256// 5257 5258static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5259 static bool proc_task_unchecked = true; 5260 pid_t tid = thread->osthread()->thread_id(); 5261 char *s; 5262 char stat[2048]; 5263 int statlen; 5264 char proc_name[64]; 5265 int count; 5266 long sys_time, user_time; 5267 char cdummy; 5268 int idummy; 5269 long ldummy; 5270 FILE *fp; 5271 5272 snprintf(proc_name, 64, "/proc/%d/stat", tid); 5273 5274 // The /proc/<tid>/stat aggregates per-process usage on 5275 // new Linux kernels 2.6+ where NPTL is supported. 5276 // The /proc/self/task/<tid>/stat still has the per-thread usage. 5277 // See bug 6328462. 5278 // There possibly can be cases where there is no directory 5279 // /proc/self/task, so we check its availability. 5280 if (proc_task_unchecked && os::Linux::is_NPTL()) { 5281 // This is executed only once 5282 proc_task_unchecked = false; 5283 fp = fopen("/proc/self/task", "r"); 5284 if (fp != NULL) { 5285 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5286 fclose(fp); 5287 } 5288 } 5289 5290 fp = fopen(proc_name, "r"); 5291 if ( fp == NULL ) return -1; 5292 statlen = fread(stat, 1, 2047, fp); 5293 stat[statlen] = '\0'; 5294 fclose(fp); 5295 5296 // Skip pid and the command string. Note that we could be dealing with 5297 // weird command names, e.g. user could decide to rename java launcher 5298 // to "java 1.4.2 :)", then the stat file would look like 5299 // 1234 (java 1.4.2 :)) R ... ... 5300 // We don't really need to know the command string, just find the last 5301 // occurrence of ")" and then start parsing from there. See bug 4726580. 5302 s = strrchr(stat, ')'); 5303 if (s == NULL ) return -1; 5304 5305 // Skip blank chars 5306 do s++; while (isspace(*s)); 5307 5308 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5309 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5310 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5311 &user_time, &sys_time); 5312 if ( count != 13 ) return -1; 5313 if (user_sys_cpu_time) { 5314 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5315 } else { 5316 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5317 } 5318} 5319 5320void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5321 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5322 info_ptr->may_skip_backward = false; // elapsed time not wall time 5323 info_ptr->may_skip_forward = false; // elapsed time not wall time 5324 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5325} 5326 5327void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5328 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5329 info_ptr->may_skip_backward = false; // elapsed time not wall time 5330 info_ptr->may_skip_forward = false; // elapsed time not wall time 5331 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5332} 5333 5334bool os::is_thread_cpu_time_supported() { 5335 return true; 5336} 5337 5338// System loadavg support. Returns -1 if load average cannot be obtained. 5339// Linux doesn't yet have a (official) notion of processor sets, 5340// so just return the system wide load average. 5341int os::loadavg(double loadavg[], int nelem) { 5342 return ::getloadavg(loadavg, nelem); 5343} 5344 5345void os::pause() { 5346 char filename[MAX_PATH]; 5347 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5348 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5349 } else { 5350 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5351 } 5352 5353 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5354 if (fd != -1) { 5355 struct stat buf; 5356 ::close(fd); 5357 while (::stat(filename, &buf) == 0) { 5358 (void)::poll(NULL, 0, 100); 5359 } 5360 } else { 5361 jio_fprintf(stderr, 5362 "Could not open pause file '%s', continuing immediately.\n", filename); 5363 } 5364} 5365 5366 5367// Refer to the comments in os_solaris.cpp park-unpark. 5368// 5369// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5370// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5371// For specifics regarding the bug see GLIBC BUGID 261237 : 5372// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5373// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5374// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5375// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5376// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5377// and monitorenter when we're using 1-0 locking. All those operations may result in 5378// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5379// of libpthread avoids the problem, but isn't practical. 5380// 5381// Possible remedies: 5382// 5383// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5384// This is palliative and probabilistic, however. If the thread is preempted 5385// between the call to compute_abstime() and pthread_cond_timedwait(), more 5386// than the minimum period may have passed, and the abstime may be stale (in the 5387// past) resultin in a hang. Using this technique reduces the odds of a hang 5388// but the JVM is still vulnerable, particularly on heavily loaded systems. 5389// 5390// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5391// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5392// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5393// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5394// thread. 5395// 5396// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5397// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5398// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5399// This also works well. In fact it avoids kernel-level scalability impediments 5400// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5401// timers in a graceful fashion. 5402// 5403// 4. When the abstime value is in the past it appears that control returns 5404// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5405// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5406// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5407// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5408// It may be possible to avoid reinitialization by checking the return 5409// value from pthread_cond_timedwait(). In addition to reinitializing the 5410// condvar we must establish the invariant that cond_signal() is only called 5411// within critical sections protected by the adjunct mutex. This prevents 5412// cond_signal() from "seeing" a condvar that's in the midst of being 5413// reinitialized or that is corrupt. Sadly, this invariant obviates the 5414// desirable signal-after-unlock optimization that avoids futile context switching. 5415// 5416// I'm also concerned that some versions of NTPL might allocate an auxilliary 5417// structure when a condvar is used or initialized. cond_destroy() would 5418// release the helper structure. Our reinitialize-after-timedwait fix 5419// put excessive stress on malloc/free and locks protecting the c-heap. 5420// 5421// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5422// It may be possible to refine (4) by checking the kernel and NTPL verisons 5423// and only enabling the work-around for vulnerable environments. 5424 5425// utility to compute the abstime argument to timedwait: 5426// millis is the relative timeout time 5427// abstime will be the absolute timeout time 5428// TODO: replace compute_abstime() with unpackTime() 5429 5430static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5431 if (millis < 0) millis = 0; 5432 5433 jlong seconds = millis / 1000; 5434 millis %= 1000; 5435 if (seconds > 50000000) { // see man cond_timedwait(3T) 5436 seconds = 50000000; 5437 } 5438 5439 if (os::supports_monotonic_clock()) { 5440 struct timespec now; 5441 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5442 assert_status(status == 0, status, "clock_gettime"); 5443 abstime->tv_sec = now.tv_sec + seconds; 5444 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5445 if (nanos >= NANOSECS_PER_SEC) { 5446 abstime->tv_sec += 1; 5447 nanos -= NANOSECS_PER_SEC; 5448 } 5449 abstime->tv_nsec = nanos; 5450 } else { 5451 struct timeval now; 5452 int status = gettimeofday(&now, NULL); 5453 assert(status == 0, "gettimeofday"); 5454 abstime->tv_sec = now.tv_sec + seconds; 5455 long usec = now.tv_usec + millis * 1000; 5456 if (usec >= 1000000) { 5457 abstime->tv_sec += 1; 5458 usec -= 1000000; 5459 } 5460 abstime->tv_nsec = usec * 1000; 5461 } 5462 return abstime; 5463} 5464 5465 5466// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 5467// Conceptually TryPark() should be equivalent to park(0). 5468 5469int os::PlatformEvent::TryPark() { 5470 for (;;) { 5471 const int v = _Event ; 5472 guarantee ((v == 0) || (v == 1), "invariant") ; 5473 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 5474 } 5475} 5476 5477void os::PlatformEvent::park() { // AKA "down()" 5478 // Invariant: Only the thread associated with the Event/PlatformEvent 5479 // may call park(). 5480 // TODO: assert that _Assoc != NULL or _Assoc == Self 5481 int v ; 5482 for (;;) { 5483 v = _Event ; 5484 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5485 } 5486 guarantee (v >= 0, "invariant") ; 5487 if (v == 0) { 5488 // Do this the hard way by blocking ... 5489 int status = pthread_mutex_lock(_mutex); 5490 assert_status(status == 0, status, "mutex_lock"); 5491 guarantee (_nParked == 0, "invariant") ; 5492 ++ _nParked ; 5493 while (_Event < 0) { 5494 status = pthread_cond_wait(_cond, _mutex); 5495 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5496 // Treat this the same as if the wait was interrupted 5497 if (status == ETIME) { status = EINTR; } 5498 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5499 } 5500 -- _nParked ; 5501 5502 _Event = 0 ; 5503 status = pthread_mutex_unlock(_mutex); 5504 assert_status(status == 0, status, "mutex_unlock"); 5505 // Paranoia to ensure our locked and lock-free paths interact 5506 // correctly with each other. 5507 OrderAccess::fence(); 5508 } 5509 guarantee (_Event >= 0, "invariant") ; 5510} 5511 5512int os::PlatformEvent::park(jlong millis) { 5513 guarantee (_nParked == 0, "invariant") ; 5514 5515 int v ; 5516 for (;;) { 5517 v = _Event ; 5518 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5519 } 5520 guarantee (v >= 0, "invariant") ; 5521 if (v != 0) return OS_OK ; 5522 5523 // We do this the hard way, by blocking the thread. 5524 // Consider enforcing a minimum timeout value. 5525 struct timespec abst; 5526 compute_abstime(&abst, millis); 5527 5528 int ret = OS_TIMEOUT; 5529 int status = pthread_mutex_lock(_mutex); 5530 assert_status(status == 0, status, "mutex_lock"); 5531 guarantee (_nParked == 0, "invariant") ; 5532 ++_nParked ; 5533 5534 // Object.wait(timo) will return because of 5535 // (a) notification 5536 // (b) timeout 5537 // (c) thread.interrupt 5538 // 5539 // Thread.interrupt and object.notify{All} both call Event::set. 5540 // That is, we treat thread.interrupt as a special case of notification. 5541 // The underlying Solaris implementation, cond_timedwait, admits 5542 // spurious/premature wakeups, but the JLS/JVM spec prevents the 5543 // JVM from making those visible to Java code. As such, we must 5544 // filter out spurious wakeups. We assume all ETIME returns are valid. 5545 // 5546 // TODO: properly differentiate simultaneous notify+interrupt. 5547 // In that case, we should propagate the notify to another waiter. 5548 5549 while (_Event < 0) { 5550 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5551 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5552 pthread_cond_destroy (_cond); 5553 pthread_cond_init (_cond, os::Linux::condAttr()) ; 5554 } 5555 assert_status(status == 0 || status == EINTR || 5556 status == ETIME || status == ETIMEDOUT, 5557 status, "cond_timedwait"); 5558 if (!FilterSpuriousWakeups) break ; // previous semantics 5559 if (status == ETIME || status == ETIMEDOUT) break ; 5560 // We consume and ignore EINTR and spurious wakeups. 5561 } 5562 --_nParked ; 5563 if (_Event >= 0) { 5564 ret = OS_OK; 5565 } 5566 _Event = 0 ; 5567 status = pthread_mutex_unlock(_mutex); 5568 assert_status(status == 0, status, "mutex_unlock"); 5569 assert (_nParked == 0, "invariant") ; 5570 // Paranoia to ensure our locked and lock-free paths interact 5571 // correctly with each other. 5572 OrderAccess::fence(); 5573 return ret; 5574} 5575 5576void os::PlatformEvent::unpark() { 5577 // Transitions for _Event: 5578 // 0 :=> 1 5579 // 1 :=> 1 5580 // -1 :=> either 0 or 1; must signal target thread 5581 // That is, we can safely transition _Event from -1 to either 5582 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back 5583 // unpark() calls. 5584 // See also: "Semaphores in Plan 9" by Mullender & Cox 5585 // 5586 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5587 // that it will take two back-to-back park() calls for the owning 5588 // thread to block. This has the benefit of forcing a spurious return 5589 // from the first park() call after an unpark() call which will help 5590 // shake out uses of park() and unpark() without condition variables. 5591 5592 if (Atomic::xchg(1, &_Event) >= 0) return; 5593 5594 // Wait for the thread associated with the event to vacate 5595 int status = pthread_mutex_lock(_mutex); 5596 assert_status(status == 0, status, "mutex_lock"); 5597 int AnyWaiters = _nParked; 5598 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5599 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5600 AnyWaiters = 0; 5601 pthread_cond_signal(_cond); 5602 } 5603 status = pthread_mutex_unlock(_mutex); 5604 assert_status(status == 0, status, "mutex_unlock"); 5605 if (AnyWaiters != 0) { 5606 status = pthread_cond_signal(_cond); 5607 assert_status(status == 0, status, "cond_signal"); 5608 } 5609 5610 // Note that we signal() _after dropping the lock for "immortal" Events. 5611 // This is safe and avoids a common class of futile wakeups. In rare 5612 // circumstances this can cause a thread to return prematurely from 5613 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5614 // simply re-test the condition and re-park itself. 5615} 5616 5617 5618// JSR166 5619// ------------------------------------------------------- 5620 5621/* 5622 * The solaris and linux implementations of park/unpark are fairly 5623 * conservative for now, but can be improved. They currently use a 5624 * mutex/condvar pair, plus a a count. 5625 * Park decrements count if > 0, else does a condvar wait. Unpark 5626 * sets count to 1 and signals condvar. Only one thread ever waits 5627 * on the condvar. Contention seen when trying to park implies that someone 5628 * is unparking you, so don't wait. And spurious returns are fine, so there 5629 * is no need to track notifications. 5630 */ 5631 5632/* 5633 * This code is common to linux and solaris and will be moved to a 5634 * common place in dolphin. 5635 * 5636 * The passed in time value is either a relative time in nanoseconds 5637 * or an absolute time in milliseconds. Either way it has to be unpacked 5638 * into suitable seconds and nanoseconds components and stored in the 5639 * given timespec structure. 5640 * Given time is a 64-bit value and the time_t used in the timespec is only 5641 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 5642 * overflow if times way in the future are given. Further on Solaris versions 5643 * prior to 10 there is a restriction (see cond_timedwait) that the specified 5644 * number of seconds, in abstime, is less than current_time + 100,000,000. 5645 * As it will be 28 years before "now + 100000000" will overflow we can 5646 * ignore overflow and just impose a hard-limit on seconds using the value 5647 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 5648 * years from "now". 5649 */ 5650 5651static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5652 assert (time > 0, "convertTime"); 5653 time_t max_secs = 0; 5654 5655 if (!os::supports_monotonic_clock() || isAbsolute) { 5656 struct timeval now; 5657 int status = gettimeofday(&now, NULL); 5658 assert(status == 0, "gettimeofday"); 5659 5660 max_secs = now.tv_sec + MAX_SECS; 5661 5662 if (isAbsolute) { 5663 jlong secs = time / 1000; 5664 if (secs > max_secs) { 5665 absTime->tv_sec = max_secs; 5666 } else { 5667 absTime->tv_sec = secs; 5668 } 5669 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5670 } else { 5671 jlong secs = time / NANOSECS_PER_SEC; 5672 if (secs >= MAX_SECS) { 5673 absTime->tv_sec = max_secs; 5674 absTime->tv_nsec = 0; 5675 } else { 5676 absTime->tv_sec = now.tv_sec + secs; 5677 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5678 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5679 absTime->tv_nsec -= NANOSECS_PER_SEC; 5680 ++absTime->tv_sec; // note: this must be <= max_secs 5681 } 5682 } 5683 } 5684 } else { 5685 // must be relative using monotonic clock 5686 struct timespec now; 5687 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5688 assert_status(status == 0, status, "clock_gettime"); 5689 max_secs = now.tv_sec + MAX_SECS; 5690 jlong secs = time / NANOSECS_PER_SEC; 5691 if (secs >= MAX_SECS) { 5692 absTime->tv_sec = max_secs; 5693 absTime->tv_nsec = 0; 5694 } else { 5695 absTime->tv_sec = now.tv_sec + secs; 5696 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5697 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5698 absTime->tv_nsec -= NANOSECS_PER_SEC; 5699 ++absTime->tv_sec; // note: this must be <= max_secs 5700 } 5701 } 5702 } 5703 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5704 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5705 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5706 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5707} 5708 5709void Parker::park(bool isAbsolute, jlong time) { 5710 // Ideally we'd do something useful while spinning, such 5711 // as calling unpackTime(). 5712 5713 // Optional fast-path check: 5714 // Return immediately if a permit is available. 5715 // We depend on Atomic::xchg() having full barrier semantics 5716 // since we are doing a lock-free update to _counter. 5717 if (Atomic::xchg(0, &_counter) > 0) return; 5718 5719 Thread* thread = Thread::current(); 5720 assert(thread->is_Java_thread(), "Must be JavaThread"); 5721 JavaThread *jt = (JavaThread *)thread; 5722 5723 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5724 // Check interrupt before trying to wait 5725 if (Thread::is_interrupted(thread, false)) { 5726 return; 5727 } 5728 5729 // Next, demultiplex/decode time arguments 5730 timespec absTime; 5731 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5732 return; 5733 } 5734 if (time > 0) { 5735 unpackTime(&absTime, isAbsolute, time); 5736 } 5737 5738 5739 // Enter safepoint region 5740 // Beware of deadlocks such as 6317397. 5741 // The per-thread Parker:: mutex is a classic leaf-lock. 5742 // In particular a thread must never block on the Threads_lock while 5743 // holding the Parker:: mutex. If safepoints are pending both the 5744 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5745 ThreadBlockInVM tbivm(jt); 5746 5747 // Don't wait if cannot get lock since interference arises from 5748 // unblocking. Also. check interrupt before trying wait 5749 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5750 return; 5751 } 5752 5753 int status ; 5754 if (_counter > 0) { // no wait needed 5755 _counter = 0; 5756 status = pthread_mutex_unlock(_mutex); 5757 assert (status == 0, "invariant") ; 5758 // Paranoia to ensure our locked and lock-free paths interact 5759 // correctly with each other and Java-level accesses. 5760 OrderAccess::fence(); 5761 return; 5762 } 5763 5764#ifdef ASSERT 5765 // Don't catch signals while blocked; let the running threads have the signals. 5766 // (This allows a debugger to break into the running thread.) 5767 sigset_t oldsigs; 5768 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5769 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5770#endif 5771 5772 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5773 jt->set_suspend_equivalent(); 5774 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5775 5776 assert(_cur_index == -1, "invariant"); 5777 if (time == 0) { 5778 _cur_index = REL_INDEX; // arbitrary choice when not timed 5779 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ; 5780 } else { 5781 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5782 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ; 5783 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5784 pthread_cond_destroy (&_cond[_cur_index]) ; 5785 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr()); 5786 } 5787 } 5788 _cur_index = -1; 5789 assert_status(status == 0 || status == EINTR || 5790 status == ETIME || status == ETIMEDOUT, 5791 status, "cond_timedwait"); 5792 5793#ifdef ASSERT 5794 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5795#endif 5796 5797 _counter = 0 ; 5798 status = pthread_mutex_unlock(_mutex) ; 5799 assert_status(status == 0, status, "invariant") ; 5800 // Paranoia to ensure our locked and lock-free paths interact 5801 // correctly with each other and Java-level accesses. 5802 OrderAccess::fence(); 5803 5804 // If externally suspended while waiting, re-suspend 5805 if (jt->handle_special_suspend_equivalent_condition()) { 5806 jt->java_suspend_self(); 5807 } 5808} 5809 5810void Parker::unpark() { 5811 int s, status ; 5812 status = pthread_mutex_lock(_mutex); 5813 assert (status == 0, "invariant") ; 5814 s = _counter; 5815 _counter = 1; 5816 if (s < 1) { 5817 // thread might be parked 5818 if (_cur_index != -1) { 5819 // thread is definitely parked 5820 if (WorkAroundNPTLTimedWaitHang) { 5821 status = pthread_cond_signal (&_cond[_cur_index]); 5822 assert (status == 0, "invariant"); 5823 status = pthread_mutex_unlock(_mutex); 5824 assert (status == 0, "invariant"); 5825 } else { 5826 status = pthread_mutex_unlock(_mutex); 5827 assert (status == 0, "invariant"); 5828 status = pthread_cond_signal (&_cond[_cur_index]); 5829 assert (status == 0, "invariant"); 5830 } 5831 } else { 5832 pthread_mutex_unlock(_mutex); 5833 assert (status == 0, "invariant") ; 5834 } 5835 } else { 5836 pthread_mutex_unlock(_mutex); 5837 assert (status == 0, "invariant") ; 5838 } 5839} 5840 5841 5842extern char** environ; 5843 5844#ifndef __NR_fork 5845#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5846#endif 5847 5848#ifndef __NR_execve 5849#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5850#endif 5851 5852// Run the specified command in a separate process. Return its exit value, 5853// or -1 on failure (e.g. can't fork a new process). 5854// Unlike system(), this function can be called from signal handler. It 5855// doesn't block SIGINT et al. 5856int os::fork_and_exec(char* cmd) { 5857 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5858 5859 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5860 // pthread_atfork handlers and reset pthread library. All we need is a 5861 // separate process to execve. Make a direct syscall to fork process. 5862 // On IA64 there's no fork syscall, we have to use fork() and hope for 5863 // the best... 5864 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5865 IA64_ONLY(fork();) 5866 5867 if (pid < 0) { 5868 // fork failed 5869 return -1; 5870 5871 } else if (pid == 0) { 5872 // child process 5873 5874 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5875 // first to kill every thread on the thread list. Because this list is 5876 // not reset by fork() (see notes above), execve() will instead kill 5877 // every thread in the parent process. We know this is the only thread 5878 // in the new process, so make a system call directly. 5879 // IA64 should use normal execve() from glibc to match the glibc fork() 5880 // above. 5881 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5882 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5883 5884 // execve failed 5885 _exit(-1); 5886 5887 } else { 5888 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5889 // care about the actual exit code, for now. 5890 5891 int status; 5892 5893 // Wait for the child process to exit. This returns immediately if 5894 // the child has already exited. */ 5895 while (waitpid(pid, &status, 0) < 0) { 5896 switch (errno) { 5897 case ECHILD: return 0; 5898 case EINTR: break; 5899 default: return -1; 5900 } 5901 } 5902 5903 if (WIFEXITED(status)) { 5904 // The child exited normally; get its exit code. 5905 return WEXITSTATUS(status); 5906 } else if (WIFSIGNALED(status)) { 5907 // The child exited because of a signal 5908 // The best value to return is 0x80 + signal number, 5909 // because that is what all Unix shells do, and because 5910 // it allows callers to distinguish between process exit and 5911 // process death by signal. 5912 return 0x80 + WTERMSIG(status); 5913 } else { 5914 // Unknown exit code; pass it through 5915 return status; 5916 } 5917 } 5918} 5919 5920// is_headless_jre() 5921// 5922// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5923// in order to report if we are running in a headless jre 5924// 5925// Since JDK8 xawt/libmawt.so was moved into the same directory 5926// as libawt.so, and renamed libawt_xawt.so 5927// 5928bool os::is_headless_jre() { 5929 struct stat statbuf; 5930 char buf[MAXPATHLEN]; 5931 char libmawtpath[MAXPATHLEN]; 5932 const char *xawtstr = "/xawt/libmawt.so"; 5933 const char *new_xawtstr = "/libawt_xawt.so"; 5934 char *p; 5935 5936 // Get path to libjvm.so 5937 os::jvm_path(buf, sizeof(buf)); 5938 5939 // Get rid of libjvm.so 5940 p = strrchr(buf, '/'); 5941 if (p == NULL) return false; 5942 else *p = '\0'; 5943 5944 // Get rid of client or server 5945 p = strrchr(buf, '/'); 5946 if (p == NULL) return false; 5947 else *p = '\0'; 5948 5949 // check xawt/libmawt.so 5950 strcpy(libmawtpath, buf); 5951 strcat(libmawtpath, xawtstr); 5952 if (::stat(libmawtpath, &statbuf) == 0) return false; 5953 5954 // check libawt_xawt.so 5955 strcpy(libmawtpath, buf); 5956 strcat(libmawtpath, new_xawtstr); 5957 if (::stat(libmawtpath, &statbuf) == 0) return false; 5958 5959 return true; 5960} 5961 5962// Get the default path to the core file 5963// Returns the length of the string 5964int os::get_core_path(char* buffer, size_t bufferSize) { 5965 const char* p = get_current_directory(buffer, bufferSize); 5966 5967 if (p == NULL) { 5968 assert(p != NULL, "failed to get current directory"); 5969 return 0; 5970 } 5971 5972 return strlen(buffer); 5973} 5974 5975#ifdef JAVASE_EMBEDDED 5976// 5977// A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 5978// 5979MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 5980 5981// ctor 5982// 5983MemNotifyThread::MemNotifyThread(int fd): Thread() { 5984 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 5985 _fd = fd; 5986 5987 if (os::create_thread(this, os::os_thread)) { 5988 _memnotify_thread = this; 5989 os::set_priority(this, NearMaxPriority); 5990 os::start_thread(this); 5991 } 5992} 5993 5994// Where all the work gets done 5995// 5996void MemNotifyThread::run() { 5997 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 5998 5999 // Set up the select arguments 6000 fd_set rfds; 6001 if (_fd != -1) { 6002 FD_ZERO(&rfds); 6003 FD_SET(_fd, &rfds); 6004 } 6005 6006 // Now wait for the mem_notify device to wake up 6007 while (1) { 6008 // Wait for the mem_notify device to signal us.. 6009 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 6010 if (rc == -1) { 6011 perror("select!\n"); 6012 break; 6013 } else if (rc) { 6014 //ssize_t free_before = os::available_memory(); 6015 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 6016 6017 // The kernel is telling us there is not much memory left... 6018 // try to do something about that 6019 6020 // If we are not already in a GC, try one. 6021 if (!Universe::heap()->is_gc_active()) { 6022 Universe::heap()->collect(GCCause::_allocation_failure); 6023 6024 //ssize_t free_after = os::available_memory(); 6025 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 6026 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 6027 } 6028 // We might want to do something like the following if we find the GC's are not helping... 6029 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 6030 } 6031 } 6032} 6033 6034// 6035// See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 6036// 6037void MemNotifyThread::start() { 6038 int fd; 6039 fd = open ("/dev/mem_notify", O_RDONLY, 0); 6040 if (fd < 0) { 6041 return; 6042 } 6043 6044 if (memnotify_thread() == NULL) { 6045 new MemNotifyThread(fd); 6046 } 6047} 6048 6049#endif // JAVASE_EMBEDDED 6050 6051 6052/////////////// Unit tests /////////////// 6053 6054#ifndef PRODUCT 6055 6056#define test_log(...) \ 6057 do {\ 6058 if (VerboseInternalVMTests) { \ 6059 tty->print_cr(__VA_ARGS__); \ 6060 tty->flush(); \ 6061 }\ 6062 } while (false) 6063 6064class TestReserveMemorySpecial : AllStatic { 6065 public: 6066 static void small_page_write(void* addr, size_t size) { 6067 size_t page_size = os::vm_page_size(); 6068 6069 char* end = (char*)addr + size; 6070 for (char* p = (char*)addr; p < end; p += page_size) { 6071 *p = 1; 6072 } 6073 } 6074 6075 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6076 if (!UseHugeTLBFS) { 6077 return; 6078 } 6079 6080 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6081 6082 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6083 6084 if (addr != NULL) { 6085 small_page_write(addr, size); 6086 6087 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6088 } 6089 } 6090 6091 static void test_reserve_memory_special_huge_tlbfs_only() { 6092 if (!UseHugeTLBFS) { 6093 return; 6094 } 6095 6096 size_t lp = os::large_page_size(); 6097 6098 for (size_t size = lp; size <= lp * 10; size += lp) { 6099 test_reserve_memory_special_huge_tlbfs_only(size); 6100 } 6101 } 6102 6103 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) { 6104 if (!UseHugeTLBFS) { 6105 return; 6106 } 6107 6108 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")", 6109 size, alignment); 6110 6111 assert(size >= os::large_page_size(), "Incorrect input to test"); 6112 6113 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6114 6115 if (addr != NULL) { 6116 small_page_write(addr, size); 6117 6118 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6119 } 6120 } 6121 6122 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) { 6123 size_t lp = os::large_page_size(); 6124 size_t ag = os::vm_allocation_granularity(); 6125 6126 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6127 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment); 6128 } 6129 } 6130 6131 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6132 size_t lp = os::large_page_size(); 6133 size_t ag = os::vm_allocation_granularity(); 6134 6135 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp); 6136 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag); 6137 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2); 6138 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2); 6139 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag); 6140 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag); 6141 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2); 6142 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10); 6143 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2); 6144 } 6145 6146 static void test_reserve_memory_special_huge_tlbfs() { 6147 if (!UseHugeTLBFS) { 6148 return; 6149 } 6150 6151 test_reserve_memory_special_huge_tlbfs_only(); 6152 test_reserve_memory_special_huge_tlbfs_mixed(); 6153 } 6154 6155 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6156 if (!UseSHM) { 6157 return; 6158 } 6159 6160 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6161 6162 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6163 6164 if (addr != NULL) { 6165 assert(is_ptr_aligned(addr, alignment), "Check"); 6166 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6167 6168 small_page_write(addr, size); 6169 6170 os::Linux::release_memory_special_shm(addr, size); 6171 } 6172 } 6173 6174 static void test_reserve_memory_special_shm() { 6175 size_t lp = os::large_page_size(); 6176 size_t ag = os::vm_allocation_granularity(); 6177 6178 for (size_t size = ag; size < lp * 3; size += ag) { 6179 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6180 test_reserve_memory_special_shm(size, alignment); 6181 } 6182 } 6183 } 6184 6185 static void test() { 6186 test_reserve_memory_special_huge_tlbfs(); 6187 test_reserve_memory_special_shm(); 6188 } 6189}; 6190 6191void TestReserveMemorySpecial_test() { 6192 TestReserveMemorySpecial::test(); 6193} 6194 6195#endif 6196