os_linux.cpp revision 6266:469835cd5494
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 1035// Restore the thread pointer if the destructor is called. This is in case 1036// someone from JNI code sets up a destructor with pthread_key_create to run 1037// detachCurrentThread on thread death. Unless we restore the thread pointer we 1038// will hang or crash. When detachCurrentThread is called the key will be set 1039// to null and we will not be called again. If detachCurrentThread is never 1040// called we could loop forever depending on the pthread implementation. 1041static void restore_thread_pointer(void* p) { 1042 Thread* thread = (Thread*) p; 1043 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread); 1044} 1045 1046int os::allocate_thread_local_storage() { 1047 pthread_key_t key; 1048 int rslt = pthread_key_create(&key, restore_thread_pointer); 1049 assert(rslt == 0, "cannot allocate thread local storage"); 1050 return (int)key; 1051} 1052 1053// Note: This is currently not used by VM, as we don't destroy TLS key 1054// on VM exit. 1055void os::free_thread_local_storage(int index) { 1056 int rslt = pthread_key_delete((pthread_key_t)index); 1057 assert(rslt == 0, "invalid index"); 1058} 1059 1060void os::thread_local_storage_at_put(int index, void* value) { 1061 int rslt = pthread_setspecific((pthread_key_t)index, value); 1062 assert(rslt == 0, "pthread_setspecific failed"); 1063} 1064 1065extern "C" Thread* get_thread() { 1066 return ThreadLocalStorage::thread(); 1067} 1068 1069////////////////////////////////////////////////////////////////////////////// 1070// initial thread 1071 1072// Check if current thread is the initial thread, similar to Solaris thr_main. 1073bool os::Linux::is_initial_thread(void) { 1074 char dummy; 1075 // If called before init complete, thread stack bottom will be null. 1076 // Can be called if fatal error occurs before initialization. 1077 if (initial_thread_stack_bottom() == NULL) return false; 1078 assert(initial_thread_stack_bottom() != NULL && 1079 initial_thread_stack_size() != 0, 1080 "os::init did not locate initial thread's stack region"); 1081 if ((address)&dummy >= initial_thread_stack_bottom() && 1082 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) 1083 return true; 1084 else return false; 1085} 1086 1087// Find the virtual memory area that contains addr 1088static bool find_vma(address addr, address* vma_low, address* vma_high) { 1089 FILE *fp = fopen("/proc/self/maps", "r"); 1090 if (fp) { 1091 address low, high; 1092 while (!feof(fp)) { 1093 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1094 if (low <= addr && addr < high) { 1095 if (vma_low) *vma_low = low; 1096 if (vma_high) *vma_high = high; 1097 fclose (fp); 1098 return true; 1099 } 1100 } 1101 for (;;) { 1102 int ch = fgetc(fp); 1103 if (ch == EOF || ch == (int)'\n') break; 1104 } 1105 } 1106 fclose(fp); 1107 } 1108 return false; 1109} 1110 1111// Locate initial thread stack. This special handling of initial thread stack 1112// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1113// bogus value for initial thread. 1114void os::Linux::capture_initial_stack(size_t max_size) { 1115 // stack size is the easy part, get it from RLIMIT_STACK 1116 size_t stack_size; 1117 struct rlimit rlim; 1118 getrlimit(RLIMIT_STACK, &rlim); 1119 stack_size = rlim.rlim_cur; 1120 1121 // 6308388: a bug in ld.so will relocate its own .data section to the 1122 // lower end of primordial stack; reduce ulimit -s value a little bit 1123 // so we won't install guard page on ld.so's data section. 1124 stack_size -= 2 * page_size(); 1125 1126 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1127 // 7.1, in both cases we will get 2G in return value. 1128 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1129 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1130 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1131 // in case other parts in glibc still assumes 2M max stack size. 1132 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1133 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1134 if (stack_size > 2 * K * K IA64_ONLY(*2)) 1135 stack_size = 2 * K * K IA64_ONLY(*2); 1136 // Try to figure out where the stack base (top) is. This is harder. 1137 // 1138 // When an application is started, glibc saves the initial stack pointer in 1139 // a global variable "__libc_stack_end", which is then used by system 1140 // libraries. __libc_stack_end should be pretty close to stack top. The 1141 // variable is available since the very early days. However, because it is 1142 // a private interface, it could disappear in the future. 1143 // 1144 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1145 // to __libc_stack_end, it is very close to stack top, but isn't the real 1146 // stack top. Note that /proc may not exist if VM is running as a chroot 1147 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1148 // /proc/<pid>/stat could change in the future (though unlikely). 1149 // 1150 // We try __libc_stack_end first. If that doesn't work, look for 1151 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1152 // as a hint, which should work well in most cases. 1153 1154 uintptr_t stack_start; 1155 1156 // try __libc_stack_end first 1157 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1158 if (p && *p) { 1159 stack_start = *p; 1160 } else { 1161 // see if we can get the start_stack field from /proc/self/stat 1162 FILE *fp; 1163 int pid; 1164 char state; 1165 int ppid; 1166 int pgrp; 1167 int session; 1168 int nr; 1169 int tpgrp; 1170 unsigned long flags; 1171 unsigned long minflt; 1172 unsigned long cminflt; 1173 unsigned long majflt; 1174 unsigned long cmajflt; 1175 unsigned long utime; 1176 unsigned long stime; 1177 long cutime; 1178 long cstime; 1179 long prio; 1180 long nice; 1181 long junk; 1182 long it_real; 1183 uintptr_t start; 1184 uintptr_t vsize; 1185 intptr_t rss; 1186 uintptr_t rsslim; 1187 uintptr_t scodes; 1188 uintptr_t ecode; 1189 int i; 1190 1191 // Figure what the primordial thread stack base is. Code is inspired 1192 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1193 // followed by command name surrounded by parentheses, state, etc. 1194 char stat[2048]; 1195 int statlen; 1196 1197 fp = fopen("/proc/self/stat", "r"); 1198 if (fp) { 1199 statlen = fread(stat, 1, 2047, fp); 1200 stat[statlen] = '\0'; 1201 fclose(fp); 1202 1203 // Skip pid and the command string. Note that we could be dealing with 1204 // weird command names, e.g. user could decide to rename java launcher 1205 // to "java 1.4.2 :)", then the stat file would look like 1206 // 1234 (java 1.4.2 :)) R ... ... 1207 // We don't really need to know the command string, just find the last 1208 // occurrence of ")" and then start parsing from there. See bug 4726580. 1209 char * s = strrchr(stat, ')'); 1210 1211 i = 0; 1212 if (s) { 1213 // Skip blank chars 1214 do s++; while (isspace(*s)); 1215 1216#define _UFM UINTX_FORMAT 1217#define _DFM INTX_FORMAT 1218 1219 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */ 1220 /* 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 */ 1221 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, 1222 &state, /* 3 %c */ 1223 &ppid, /* 4 %d */ 1224 &pgrp, /* 5 %d */ 1225 &session, /* 6 %d */ 1226 &nr, /* 7 %d */ 1227 &tpgrp, /* 8 %d */ 1228 &flags, /* 9 %lu */ 1229 &minflt, /* 10 %lu */ 1230 &cminflt, /* 11 %lu */ 1231 &majflt, /* 12 %lu */ 1232 &cmajflt, /* 13 %lu */ 1233 &utime, /* 14 %lu */ 1234 &stime, /* 15 %lu */ 1235 &cutime, /* 16 %ld */ 1236 &cstime, /* 17 %ld */ 1237 &prio, /* 18 %ld */ 1238 &nice, /* 19 %ld */ 1239 &junk, /* 20 %ld */ 1240 &it_real, /* 21 %ld */ 1241 &start, /* 22 UINTX_FORMAT */ 1242 &vsize, /* 23 UINTX_FORMAT */ 1243 &rss, /* 24 INTX_FORMAT */ 1244 &rsslim, /* 25 UINTX_FORMAT */ 1245 &scodes, /* 26 UINTX_FORMAT */ 1246 &ecode, /* 27 UINTX_FORMAT */ 1247 &stack_start); /* 28 UINTX_FORMAT */ 1248 } 1249 1250#undef _UFM 1251#undef _DFM 1252 1253 if (i != 28 - 2) { 1254 assert(false, "Bad conversion from /proc/self/stat"); 1255 // product mode - assume we are the initial thread, good luck in the 1256 // embedded case. 1257 warning("Can't detect initial thread stack location - bad conversion"); 1258 stack_start = (uintptr_t) &rlim; 1259 } 1260 } else { 1261 // For some reason we can't open /proc/self/stat (for example, running on 1262 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1263 // most cases, so don't abort: 1264 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1265 stack_start = (uintptr_t) &rlim; 1266 } 1267 } 1268 1269 // Now we have a pointer (stack_start) very close to the stack top, the 1270 // next thing to do is to figure out the exact location of stack top. We 1271 // can find out the virtual memory area that contains stack_start by 1272 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1273 // and its upper limit is the real stack top. (again, this would fail if 1274 // running inside chroot, because /proc may not exist.) 1275 1276 uintptr_t stack_top; 1277 address low, high; 1278 if (find_vma((address)stack_start, &low, &high)) { 1279 // success, "high" is the true stack top. (ignore "low", because initial 1280 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1281 stack_top = (uintptr_t)high; 1282 } else { 1283 // failed, likely because /proc/self/maps does not exist 1284 warning("Can't detect initial thread stack location - find_vma failed"); 1285 // best effort: stack_start is normally within a few pages below the real 1286 // stack top, use it as stack top, and reduce stack size so we won't put 1287 // guard page outside stack. 1288 stack_top = stack_start; 1289 stack_size -= 16 * page_size(); 1290 } 1291 1292 // stack_top could be partially down the page so align it 1293 stack_top = align_size_up(stack_top, page_size()); 1294 1295 if (max_size && stack_size > max_size) { 1296 _initial_thread_stack_size = max_size; 1297 } else { 1298 _initial_thread_stack_size = stack_size; 1299 } 1300 1301 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1302 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1303} 1304 1305//////////////////////////////////////////////////////////////////////////////// 1306// time support 1307 1308// Time since start-up in seconds to a fine granularity. 1309// Used by VMSelfDestructTimer and the MemProfiler. 1310double os::elapsedTime() { 1311 1312 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1313} 1314 1315jlong os::elapsed_counter() { 1316 return javaTimeNanos() - initial_time_count; 1317} 1318 1319jlong os::elapsed_frequency() { 1320 return NANOSECS_PER_SEC; // nanosecond resolution 1321} 1322 1323bool os::supports_vtime() { return true; } 1324bool os::enable_vtime() { return false; } 1325bool os::vtime_enabled() { return false; } 1326 1327double os::elapsedVTime() { 1328 struct rusage usage; 1329 int retval = getrusage(RUSAGE_THREAD, &usage); 1330 if (retval == 0) { 1331 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); 1332 } else { 1333 // better than nothing, but not much 1334 return elapsedTime(); 1335 } 1336} 1337 1338jlong os::javaTimeMillis() { 1339 timeval time; 1340 int status = gettimeofday(&time, NULL); 1341 assert(status != -1, "linux error"); 1342 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1343} 1344 1345#ifndef CLOCK_MONOTONIC 1346#define CLOCK_MONOTONIC (1) 1347#endif 1348 1349void os::Linux::clock_init() { 1350 // we do dlopen's in this particular order due to bug in linux 1351 // dynamical loader (see 6348968) leading to crash on exit 1352 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1353 if (handle == NULL) { 1354 handle = dlopen("librt.so", RTLD_LAZY); 1355 } 1356 1357 if (handle) { 1358 int (*clock_getres_func)(clockid_t, struct timespec*) = 1359 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1360 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1361 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1362 if (clock_getres_func && clock_gettime_func) { 1363 // See if monotonic clock is supported by the kernel. Note that some 1364 // early implementations simply return kernel jiffies (updated every 1365 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1366 // for nano time (though the monotonic property is still nice to have). 1367 // It's fixed in newer kernels, however clock_getres() still returns 1368 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1369 // resolution for now. Hopefully as people move to new kernels, this 1370 // won't be a problem. 1371 struct timespec res; 1372 struct timespec tp; 1373 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1374 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1375 // yes, monotonic clock is supported 1376 _clock_gettime = clock_gettime_func; 1377 return; 1378 } else { 1379 // close librt if there is no monotonic clock 1380 dlclose(handle); 1381 } 1382 } 1383 } 1384 warning("No monotonic clock was available - timed services may " \ 1385 "be adversely affected if the time-of-day clock changes"); 1386} 1387 1388#ifndef SYS_clock_getres 1389 1390#if defined(IA32) || defined(AMD64) 1391#define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1392#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1393#else 1394#warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1395#define sys_clock_getres(x,y) -1 1396#endif 1397 1398#else 1399#define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1400#endif 1401 1402void os::Linux::fast_thread_clock_init() { 1403 if (!UseLinuxPosixThreadCPUClocks) { 1404 return; 1405 } 1406 clockid_t clockid; 1407 struct timespec tp; 1408 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1409 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1410 1411 // Switch to using fast clocks for thread cpu time if 1412 // the sys_clock_getres() returns 0 error code. 1413 // Note, that some kernels may support the current thread 1414 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1415 // returned by the pthread_getcpuclockid(). 1416 // If the fast Posix clocks are supported then the sys_clock_getres() 1417 // must return at least tp.tv_sec == 0 which means a resolution 1418 // better than 1 sec. This is extra check for reliability. 1419 1420 if(pthread_getcpuclockid_func && 1421 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1422 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1423 1424 _supports_fast_thread_cpu_time = true; 1425 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1426 } 1427} 1428 1429jlong os::javaTimeNanos() { 1430 if (os::supports_monotonic_clock()) { 1431 struct timespec tp; 1432 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1433 assert(status == 0, "gettime error"); 1434 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1435 return result; 1436 } else { 1437 timeval time; 1438 int status = gettimeofday(&time, NULL); 1439 assert(status != -1, "linux error"); 1440 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1441 return 1000 * usecs; 1442 } 1443} 1444 1445void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1446 if (os::supports_monotonic_clock()) { 1447 info_ptr->max_value = ALL_64_BITS; 1448 1449 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1450 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1451 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1452 } else { 1453 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1454 info_ptr->max_value = ALL_64_BITS; 1455 1456 // gettimeofday is a real time clock so it skips 1457 info_ptr->may_skip_backward = true; 1458 info_ptr->may_skip_forward = true; 1459 } 1460 1461 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1462} 1463 1464// Return the real, user, and system times in seconds from an 1465// arbitrary fixed point in the past. 1466bool os::getTimesSecs(double* process_real_time, 1467 double* process_user_time, 1468 double* process_system_time) { 1469 struct tms ticks; 1470 clock_t real_ticks = times(&ticks); 1471 1472 if (real_ticks == (clock_t) (-1)) { 1473 return false; 1474 } else { 1475 double ticks_per_second = (double) clock_tics_per_sec; 1476 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1477 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1478 *process_real_time = ((double) real_ticks) / ticks_per_second; 1479 1480 return true; 1481 } 1482} 1483 1484 1485char * os::local_time_string(char *buf, size_t buflen) { 1486 struct tm t; 1487 time_t long_time; 1488 time(&long_time); 1489 localtime_r(&long_time, &t); 1490 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1491 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1492 t.tm_hour, t.tm_min, t.tm_sec); 1493 return buf; 1494} 1495 1496struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1497 return localtime_r(clock, res); 1498} 1499 1500//////////////////////////////////////////////////////////////////////////////// 1501// runtime exit support 1502 1503// Note: os::shutdown() might be called very early during initialization, or 1504// called from signal handler. Before adding something to os::shutdown(), make 1505// sure it is async-safe and can handle partially initialized VM. 1506void os::shutdown() { 1507 1508 // allow PerfMemory to attempt cleanup of any persistent resources 1509 perfMemory_exit(); 1510 1511 // needs to remove object in file system 1512 AttachListener::abort(); 1513 1514 // flush buffered output, finish log files 1515 ostream_abort(); 1516 1517 // Check for abort hook 1518 abort_hook_t abort_hook = Arguments::abort_hook(); 1519 if (abort_hook != NULL) { 1520 abort_hook(); 1521 } 1522 1523} 1524 1525// Note: os::abort() might be called very early during initialization, or 1526// called from signal handler. Before adding something to os::abort(), make 1527// sure it is async-safe and can handle partially initialized VM. 1528void os::abort(bool dump_core) { 1529 os::shutdown(); 1530 if (dump_core) { 1531#ifndef PRODUCT 1532 fdStream out(defaultStream::output_fd()); 1533 out.print_raw("Current thread is "); 1534 char buf[16]; 1535 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1536 out.print_raw_cr(buf); 1537 out.print_raw_cr("Dumping core ..."); 1538#endif 1539 ::abort(); // dump core 1540 } 1541 1542 ::exit(1); 1543} 1544 1545// Die immediately, no exit hook, no abort hook, no cleanup. 1546void os::die() { 1547 // _exit() on LinuxThreads only kills current thread 1548 ::abort(); 1549} 1550 1551// unused on linux for now. 1552void os::set_error_file(const char *logfile) {} 1553 1554 1555// This method is a copy of JDK's sysGetLastErrorString 1556// from src/solaris/hpi/src/system_md.c 1557 1558size_t os::lasterror(char *buf, size_t len) { 1559 1560 if (errno == 0) return 0; 1561 1562 const char *s = ::strerror(errno); 1563 size_t n = ::strlen(s); 1564 if (n >= len) { 1565 n = len - 1; 1566 } 1567 ::strncpy(buf, s, n); 1568 buf[n] = '\0'; 1569 return n; 1570} 1571 1572intx os::current_thread_id() { return (intx)pthread_self(); } 1573int os::current_process_id() { 1574 1575 // Under the old linux thread library, linux gives each thread 1576 // its own process id. Because of this each thread will return 1577 // a different pid if this method were to return the result 1578 // of getpid(2). Linux provides no api that returns the pid 1579 // of the launcher thread for the vm. This implementation 1580 // returns a unique pid, the pid of the launcher thread 1581 // that starts the vm 'process'. 1582 1583 // Under the NPTL, getpid() returns the same pid as the 1584 // launcher thread rather than a unique pid per thread. 1585 // Use gettid() if you want the old pre NPTL behaviour. 1586 1587 // if you are looking for the result of a call to getpid() that 1588 // returns a unique pid for the calling thread, then look at the 1589 // OSThread::thread_id() method in osThread_linux.hpp file 1590 1591 return (int)(_initial_pid ? _initial_pid : getpid()); 1592} 1593 1594// DLL functions 1595 1596const char* os::dll_file_extension() { return ".so"; } 1597 1598// This must be hard coded because it's the system's temporary 1599// directory not the java application's temp directory, ala java.io.tmpdir. 1600const char* os::get_temp_directory() { return "/tmp"; } 1601 1602static bool file_exists(const char* filename) { 1603 struct stat statbuf; 1604 if (filename == NULL || strlen(filename) == 0) { 1605 return false; 1606 } 1607 return os::stat(filename, &statbuf) == 0; 1608} 1609 1610bool os::dll_build_name(char* buffer, size_t buflen, 1611 const char* pname, const char* fname) { 1612 bool retval = false; 1613 // Copied from libhpi 1614 const size_t pnamelen = pname ? strlen(pname) : 0; 1615 1616 // Return error on buffer overflow. 1617 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1618 return retval; 1619 } 1620 1621 if (pnamelen == 0) { 1622 snprintf(buffer, buflen, "lib%s.so", fname); 1623 retval = true; 1624 } else if (strchr(pname, *os::path_separator()) != NULL) { 1625 int n; 1626 char** pelements = split_path(pname, &n); 1627 if (pelements == NULL) { 1628 return false; 1629 } 1630 for (int i = 0 ; i < n ; i++) { 1631 // Really shouldn't be NULL, but check can't hurt 1632 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1633 continue; // skip the empty path values 1634 } 1635 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1636 if (file_exists(buffer)) { 1637 retval = true; 1638 break; 1639 } 1640 } 1641 // release the storage 1642 for (int i = 0 ; i < n ; i++) { 1643 if (pelements[i] != NULL) { 1644 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal); 1645 } 1646 } 1647 if (pelements != NULL) { 1648 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal); 1649 } 1650 } else { 1651 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1652 retval = true; 1653 } 1654 return retval; 1655} 1656 1657// check if addr is inside libjvm.so 1658bool os::address_is_in_vm(address addr) { 1659 static address libjvm_base_addr; 1660 Dl_info dlinfo; 1661 1662 if (libjvm_base_addr == NULL) { 1663 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1664 libjvm_base_addr = (address)dlinfo.dli_fbase; 1665 } 1666 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1667 } 1668 1669 if (dladdr((void *)addr, &dlinfo) != 0) { 1670 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1671 } 1672 1673 return false; 1674} 1675 1676bool os::dll_address_to_function_name(address addr, char *buf, 1677 int buflen, int *offset) { 1678 // buf is not optional, but offset is optional 1679 assert(buf != NULL, "sanity check"); 1680 1681 Dl_info dlinfo; 1682 1683 if (dladdr((void*)addr, &dlinfo) != 0) { 1684 // see if we have a matching symbol 1685 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1686 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1687 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1688 } 1689 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1690 return true; 1691 } 1692 // no matching symbol so try for just file info 1693 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1694 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1695 buf, buflen, offset, dlinfo.dli_fname)) { 1696 return true; 1697 } 1698 } 1699 } 1700 1701 buf[0] = '\0'; 1702 if (offset != NULL) *offset = -1; 1703 return false; 1704} 1705 1706struct _address_to_library_name { 1707 address addr; // input : memory address 1708 size_t buflen; // size of fname 1709 char* fname; // output: library name 1710 address base; // library base addr 1711}; 1712 1713static int address_to_library_name_callback(struct dl_phdr_info *info, 1714 size_t size, void *data) { 1715 int i; 1716 bool found = false; 1717 address libbase = NULL; 1718 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1719 1720 // iterate through all loadable segments 1721 for (i = 0; i < info->dlpi_phnum; i++) { 1722 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1723 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1724 // base address of a library is the lowest address of its loaded 1725 // segments. 1726 if (libbase == NULL || libbase > segbase) { 1727 libbase = segbase; 1728 } 1729 // see if 'addr' is within current segment 1730 if (segbase <= d->addr && 1731 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1732 found = true; 1733 } 1734 } 1735 } 1736 1737 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1738 // so dll_address_to_library_name() can fall through to use dladdr() which 1739 // can figure out executable name from argv[0]. 1740 if (found && info->dlpi_name && info->dlpi_name[0]) { 1741 d->base = libbase; 1742 if (d->fname) { 1743 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1744 } 1745 return 1; 1746 } 1747 return 0; 1748} 1749 1750bool os::dll_address_to_library_name(address addr, char* buf, 1751 int buflen, int* offset) { 1752 // buf is not optional, but offset is optional 1753 assert(buf != NULL, "sanity check"); 1754 1755 Dl_info dlinfo; 1756 struct _address_to_library_name data; 1757 1758 // There is a bug in old glibc dladdr() implementation that it could resolve 1759 // to wrong library name if the .so file has a base address != NULL. Here 1760 // we iterate through the program headers of all loaded libraries to find 1761 // out which library 'addr' really belongs to. This workaround can be 1762 // removed once the minimum requirement for glibc is moved to 2.3.x. 1763 data.addr = addr; 1764 data.fname = buf; 1765 data.buflen = buflen; 1766 data.base = NULL; 1767 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1768 1769 if (rslt) { 1770 // buf already contains library name 1771 if (offset) *offset = addr - data.base; 1772 return true; 1773 } 1774 if (dladdr((void*)addr, &dlinfo) != 0) { 1775 if (dlinfo.dli_fname != NULL) { 1776 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1777 } 1778 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1779 *offset = addr - (address)dlinfo.dli_fbase; 1780 } 1781 return true; 1782 } 1783 1784 buf[0] = '\0'; 1785 if (offset) *offset = -1; 1786 return false; 1787} 1788 1789 // Loads .dll/.so and 1790 // in case of error it checks if .dll/.so was built for the 1791 // same architecture as Hotspot is running on 1792 1793 1794// Remember the stack's state. The Linux dynamic linker will change 1795// the stack to 'executable' at most once, so we must safepoint only once. 1796bool os::Linux::_stack_is_executable = false; 1797 1798// VM operation that loads a library. This is necessary if stack protection 1799// of the Java stacks can be lost during loading the library. If we 1800// do not stop the Java threads, they can stack overflow before the stacks 1801// are protected again. 1802class VM_LinuxDllLoad: public VM_Operation { 1803 private: 1804 const char *_filename; 1805 char *_ebuf; 1806 int _ebuflen; 1807 void *_lib; 1808 public: 1809 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1810 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1811 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1812 void doit() { 1813 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1814 os::Linux::_stack_is_executable = true; 1815 } 1816 void* loaded_library() { return _lib; } 1817}; 1818 1819void * os::dll_load(const char *filename, char *ebuf, int ebuflen) 1820{ 1821 void * result = NULL; 1822 bool load_attempted = false; 1823 1824 // Check whether the library to load might change execution rights 1825 // of the stack. If they are changed, the protection of the stack 1826 // guard pages will be lost. We need a safepoint to fix this. 1827 // 1828 // See Linux man page execstack(8) for more info. 1829 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1830 ElfFile ef(filename); 1831 if (!ef.specifies_noexecstack()) { 1832 if (!is_init_completed()) { 1833 os::Linux::_stack_is_executable = true; 1834 // This is OK - No Java threads have been created yet, and hence no 1835 // stack guard pages to fix. 1836 // 1837 // This should happen only when you are building JDK7 using a very 1838 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1839 // 1840 // Dynamic loader will make all stacks executable after 1841 // this function returns, and will not do that again. 1842 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1843 } else { 1844 warning("You have loaded library %s which might have disabled stack guard. " 1845 "The VM will try to fix the stack guard now.\n" 1846 "It's highly recommended that you fix the library with " 1847 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1848 filename); 1849 1850 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1851 JavaThread *jt = JavaThread::current(); 1852 if (jt->thread_state() != _thread_in_native) { 1853 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1854 // that requires ExecStack. Cannot enter safe point. Let's give up. 1855 warning("Unable to fix stack guard. Giving up."); 1856 } else { 1857 if (!LoadExecStackDllInVMThread) { 1858 // This is for the case where the DLL has an static 1859 // constructor function that executes JNI code. We cannot 1860 // load such DLLs in the VMThread. 1861 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1862 } 1863 1864 ThreadInVMfromNative tiv(jt); 1865 debug_only(VMNativeEntryWrapper vew;) 1866 1867 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1868 VMThread::execute(&op); 1869 if (LoadExecStackDllInVMThread) { 1870 result = op.loaded_library(); 1871 } 1872 load_attempted = true; 1873 } 1874 } 1875 } 1876 } 1877 1878 if (!load_attempted) { 1879 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1880 } 1881 1882 if (result != NULL) { 1883 // Successful loading 1884 return result; 1885 } 1886 1887 Elf32_Ehdr elf_head; 1888 int diag_msg_max_length=ebuflen-strlen(ebuf); 1889 char* diag_msg_buf=ebuf+strlen(ebuf); 1890 1891 if (diag_msg_max_length==0) { 1892 // No more space in ebuf for additional diagnostics message 1893 return NULL; 1894 } 1895 1896 1897 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1898 1899 if (file_descriptor < 0) { 1900 // Can't open library, report dlerror() message 1901 return NULL; 1902 } 1903 1904 bool failed_to_read_elf_head= 1905 (sizeof(elf_head)!= 1906 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ; 1907 1908 ::close(file_descriptor); 1909 if (failed_to_read_elf_head) { 1910 // file i/o error - report dlerror() msg 1911 return NULL; 1912 } 1913 1914 typedef struct { 1915 Elf32_Half code; // Actual value as defined in elf.h 1916 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1917 char elf_class; // 32 or 64 bit 1918 char endianess; // MSB or LSB 1919 char* name; // String representation 1920 } arch_t; 1921 1922 #ifndef EM_486 1923 #define EM_486 6 /* Intel 80486 */ 1924 #endif 1925 1926 static const arch_t arch_array[]={ 1927 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1928 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1929 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1930 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1931 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1932 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1933 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1934 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1935 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1936 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1937 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1938 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1939 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1940 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1941 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1942 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"} 1943 }; 1944 1945 #if (defined IA32) 1946 static Elf32_Half running_arch_code=EM_386; 1947 #elif (defined AMD64) 1948 static Elf32_Half running_arch_code=EM_X86_64; 1949 #elif (defined IA64) 1950 static Elf32_Half running_arch_code=EM_IA_64; 1951 #elif (defined __sparc) && (defined _LP64) 1952 static Elf32_Half running_arch_code=EM_SPARCV9; 1953 #elif (defined __sparc) && (!defined _LP64) 1954 static Elf32_Half running_arch_code=EM_SPARC; 1955 #elif (defined __powerpc64__) 1956 static Elf32_Half running_arch_code=EM_PPC64; 1957 #elif (defined __powerpc__) 1958 static Elf32_Half running_arch_code=EM_PPC; 1959 #elif (defined ARM) 1960 static Elf32_Half running_arch_code=EM_ARM; 1961 #elif (defined S390) 1962 static Elf32_Half running_arch_code=EM_S390; 1963 #elif (defined ALPHA) 1964 static Elf32_Half running_arch_code=EM_ALPHA; 1965 #elif (defined MIPSEL) 1966 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1967 #elif (defined PARISC) 1968 static Elf32_Half running_arch_code=EM_PARISC; 1969 #elif (defined MIPS) 1970 static Elf32_Half running_arch_code=EM_MIPS; 1971 #elif (defined M68K) 1972 static Elf32_Half running_arch_code=EM_68K; 1973 #else 1974 #error Method os::dll_load requires that one of following is defined:\ 1975 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K 1976 #endif 1977 1978 // Identify compatability class for VM's architecture and library's architecture 1979 // Obtain string descriptions for architectures 1980 1981 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1982 int running_arch_index=-1; 1983 1984 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) { 1985 if (running_arch_code == arch_array[i].code) { 1986 running_arch_index = i; 1987 } 1988 if (lib_arch.code == arch_array[i].code) { 1989 lib_arch.compat_class = arch_array[i].compat_class; 1990 lib_arch.name = arch_array[i].name; 1991 } 1992 } 1993 1994 assert(running_arch_index != -1, 1995 "Didn't find running architecture code (running_arch_code) in arch_array"); 1996 if (running_arch_index == -1) { 1997 // Even though running architecture detection failed 1998 // we may still continue with reporting dlerror() message 1999 return NULL; 2000 } 2001 2002 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 2003 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 2004 return NULL; 2005 } 2006 2007#ifndef S390 2008 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 2009 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 2010 return NULL; 2011 } 2012#endif // !S390 2013 2014 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 2015 if ( lib_arch.name!=NULL ) { 2016 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2017 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 2018 lib_arch.name, arch_array[running_arch_index].name); 2019 } else { 2020 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 2021 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 2022 lib_arch.code, 2023 arch_array[running_arch_index].name); 2024 } 2025 } 2026 2027 return NULL; 2028} 2029 2030void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) { 2031 void * result = ::dlopen(filename, RTLD_LAZY); 2032 if (result == NULL) { 2033 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 2034 ebuf[ebuflen-1] = '\0'; 2035 } 2036 return result; 2037} 2038 2039void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) { 2040 void * result = NULL; 2041 if (LoadExecStackDllInVMThread) { 2042 result = dlopen_helper(filename, ebuf, ebuflen); 2043 } 2044 2045 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2046 // library that requires an executable stack, or which does not have this 2047 // stack attribute set, dlopen changes the stack attribute to executable. The 2048 // read protection of the guard pages gets lost. 2049 // 2050 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2051 // may have been queued at the same time. 2052 2053 if (!_stack_is_executable) { 2054 JavaThread *jt = Threads::first(); 2055 2056 while (jt) { 2057 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2058 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2059 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2060 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2061 warning("Attempt to reguard stack yellow zone failed."); 2062 } 2063 } 2064 jt = jt->next(); 2065 } 2066 } 2067 2068 return result; 2069} 2070 2071/* 2072 * glibc-2.0 libdl is not MT safe. If you are building with any glibc, 2073 * chances are you might want to run the generated bits against glibc-2.0 2074 * libdl.so, so always use locking for any version of glibc. 2075 */ 2076void* os::dll_lookup(void* handle, const char* name) { 2077 pthread_mutex_lock(&dl_mutex); 2078 void* res = dlsym(handle, name); 2079 pthread_mutex_unlock(&dl_mutex); 2080 return res; 2081} 2082 2083void* os::get_default_process_handle() { 2084 return (void*)::dlopen(NULL, RTLD_LAZY); 2085} 2086 2087static bool _print_ascii_file(const char* filename, outputStream* st) { 2088 int fd = ::open(filename, O_RDONLY); 2089 if (fd == -1) { 2090 return false; 2091 } 2092 2093 char buf[32]; 2094 int bytes; 2095 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2096 st->print_raw(buf, bytes); 2097 } 2098 2099 ::close(fd); 2100 2101 return true; 2102} 2103 2104void os::print_dll_info(outputStream *st) { 2105 st->print_cr("Dynamic libraries:"); 2106 2107 char fname[32]; 2108 pid_t pid = os::Linux::gettid(); 2109 2110 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2111 2112 if (!_print_ascii_file(fname, st)) { 2113 st->print("Can not get library information for pid = %d\n", pid); 2114 } 2115} 2116 2117void os::print_os_info_brief(outputStream* st) { 2118 os::Linux::print_distro_info(st); 2119 2120 os::Posix::print_uname_info(st); 2121 2122 os::Linux::print_libversion_info(st); 2123 2124} 2125 2126void os::print_os_info(outputStream* st) { 2127 st->print("OS:"); 2128 2129 os::Linux::print_distro_info(st); 2130 2131 os::Posix::print_uname_info(st); 2132 2133 // Print warning if unsafe chroot environment detected 2134 if (unsafe_chroot_detected) { 2135 st->print("WARNING!! "); 2136 st->print_cr(unstable_chroot_error); 2137 } 2138 2139 os::Linux::print_libversion_info(st); 2140 2141 os::Posix::print_rlimit_info(st); 2142 2143 os::Posix::print_load_average(st); 2144 2145 os::Linux::print_full_memory_info(st); 2146} 2147 2148// Try to identify popular distros. 2149// Most Linux distributions have a /etc/XXX-release file, which contains 2150// the OS version string. Newer Linux distributions have a /etc/lsb-release 2151// file that also contains the OS version string. Some have more than one 2152// /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2153// /etc/redhat-release.), so the order is important. 2154// Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2155// their own specific XXX-release file as well as a redhat-release file. 2156// Because of this the XXX-release file needs to be searched for before the 2157// redhat-release file. 2158// Since Red Hat has a lsb-release file that is not very descriptive the 2159// search for redhat-release needs to be before lsb-release. 2160// Since the lsb-release file is the new standard it needs to be searched 2161// before the older style release files. 2162// Searching system-release (Red Hat) and os-release (other Linuxes) are a 2163// next to last resort. The os-release file is a new standard that contains 2164// distribution information and the system-release file seems to be an old 2165// standard that has been replaced by the lsb-release and os-release files. 2166// Searching for the debian_version file is the last resort. It contains 2167// an informative string like "6.0.6" or "wheezy/sid". Because of this 2168// "Debian " is printed before the contents of the debian_version file. 2169void os::Linux::print_distro_info(outputStream* st) { 2170 if (!_print_ascii_file("/etc/oracle-release", st) && 2171 !_print_ascii_file("/etc/mandriva-release", st) && 2172 !_print_ascii_file("/etc/mandrake-release", st) && 2173 !_print_ascii_file("/etc/sun-release", st) && 2174 !_print_ascii_file("/etc/redhat-release", st) && 2175 !_print_ascii_file("/etc/lsb-release", st) && 2176 !_print_ascii_file("/etc/SuSE-release", st) && 2177 !_print_ascii_file("/etc/turbolinux-release", st) && 2178 !_print_ascii_file("/etc/gentoo-release", st) && 2179 !_print_ascii_file("/etc/ltib-release", st) && 2180 !_print_ascii_file("/etc/angstrom-version", st) && 2181 !_print_ascii_file("/etc/system-release", st) && 2182 !_print_ascii_file("/etc/os-release", st)) { 2183 2184 if (file_exists("/etc/debian_version")) { 2185 st->print("Debian "); 2186 _print_ascii_file("/etc/debian_version", st); 2187 } else { 2188 st->print("Linux"); 2189 } 2190 } 2191 st->cr(); 2192} 2193 2194void os::Linux::print_libversion_info(outputStream* st) { 2195 // libc, pthread 2196 st->print("libc:"); 2197 st->print(os::Linux::glibc_version()); st->print(" "); 2198 st->print(os::Linux::libpthread_version()); st->print(" "); 2199 if (os::Linux::is_LinuxThreads()) { 2200 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2201 } 2202 st->cr(); 2203} 2204 2205void os::Linux::print_full_memory_info(outputStream* st) { 2206 st->print("\n/proc/meminfo:\n"); 2207 _print_ascii_file("/proc/meminfo", st); 2208 st->cr(); 2209} 2210 2211void os::print_memory_info(outputStream* st) { 2212 2213 st->print("Memory:"); 2214 st->print(" %dk page", os::vm_page_size()>>10); 2215 2216 // values in struct sysinfo are "unsigned long" 2217 struct sysinfo si; 2218 sysinfo(&si); 2219 2220 st->print(", physical " UINT64_FORMAT "k", 2221 os::physical_memory() >> 10); 2222 st->print("(" UINT64_FORMAT "k free)", 2223 os::available_memory() >> 10); 2224 st->print(", swap " UINT64_FORMAT "k", 2225 ((jlong)si.totalswap * si.mem_unit) >> 10); 2226 st->print("(" UINT64_FORMAT "k free)", 2227 ((jlong)si.freeswap * si.mem_unit) >> 10); 2228 st->cr(); 2229} 2230 2231void os::pd_print_cpu_info(outputStream* st) { 2232 st->print("\n/proc/cpuinfo:\n"); 2233 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2234 st->print(" <Not Available>"); 2235 } 2236 st->cr(); 2237} 2238 2239void os::print_siginfo(outputStream* st, void* siginfo) { 2240 const siginfo_t* si = (const siginfo_t*)siginfo; 2241 2242 os::Posix::print_siginfo_brief(st, si); 2243 2244 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2245 UseSharedSpaces) { 2246 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2247 if (mapinfo->is_in_shared_space(si->si_addr)) { 2248 st->print("\n\nError accessing class data sharing archive." \ 2249 " Mapped file inaccessible during execution, " \ 2250 " possible disk/network problem."); 2251 } 2252 } 2253 st->cr(); 2254} 2255 2256 2257static void print_signal_handler(outputStream* st, int sig, 2258 char* buf, size_t buflen); 2259 2260void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2261 st->print_cr("Signal Handlers:"); 2262 print_signal_handler(st, SIGSEGV, buf, buflen); 2263 print_signal_handler(st, SIGBUS , buf, buflen); 2264 print_signal_handler(st, SIGFPE , buf, buflen); 2265 print_signal_handler(st, SIGPIPE, buf, buflen); 2266 print_signal_handler(st, SIGXFSZ, buf, buflen); 2267 print_signal_handler(st, SIGILL , buf, buflen); 2268 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2269 print_signal_handler(st, SR_signum, buf, buflen); 2270 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2271 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2272 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2273 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2274#if defined(PPC64) 2275 print_signal_handler(st, SIGTRAP, buf, buflen); 2276#endif 2277} 2278 2279static char saved_jvm_path[MAXPATHLEN] = {0}; 2280 2281// Find the full path to the current module, libjvm.so 2282void os::jvm_path(char *buf, jint buflen) { 2283 // Error checking. 2284 if (buflen < MAXPATHLEN) { 2285 assert(false, "must use a large-enough buffer"); 2286 buf[0] = '\0'; 2287 return; 2288 } 2289 // Lazy resolve the path to current module. 2290 if (saved_jvm_path[0] != 0) { 2291 strcpy(buf, saved_jvm_path); 2292 return; 2293 } 2294 2295 char dli_fname[MAXPATHLEN]; 2296 bool ret = dll_address_to_library_name( 2297 CAST_FROM_FN_PTR(address, os::jvm_path), 2298 dli_fname, sizeof(dli_fname), NULL); 2299 assert(ret, "cannot locate libjvm"); 2300 char *rp = NULL; 2301 if (ret && dli_fname[0] != '\0') { 2302 rp = realpath(dli_fname, buf); 2303 } 2304 if (rp == NULL) 2305 return; 2306 2307 if (Arguments::sun_java_launcher_is_altjvm()) { 2308 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2309 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2310 // If "/jre/lib/" appears at the right place in the string, then 2311 // assume we are installed in a JDK and we're done. Otherwise, check 2312 // for a JAVA_HOME environment variable and fix up the path so it 2313 // looks like libjvm.so is installed there (append a fake suffix 2314 // hotspot/libjvm.so). 2315 const char *p = buf + strlen(buf) - 1; 2316 for (int count = 0; p > buf && count < 5; ++count) { 2317 for (--p; p > buf && *p != '/'; --p) 2318 /* empty */ ; 2319 } 2320 2321 if (strncmp(p, "/jre/lib/", 9) != 0) { 2322 // Look for JAVA_HOME in the environment. 2323 char* java_home_var = ::getenv("JAVA_HOME"); 2324 if (java_home_var != NULL && java_home_var[0] != 0) { 2325 char* jrelib_p; 2326 int len; 2327 2328 // Check the current module name "libjvm.so". 2329 p = strrchr(buf, '/'); 2330 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2331 2332 rp = realpath(java_home_var, buf); 2333 if (rp == NULL) 2334 return; 2335 2336 // determine if this is a legacy image or modules image 2337 // modules image doesn't have "jre" subdirectory 2338 len = strlen(buf); 2339 jrelib_p = buf + len; 2340 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2341 if (0 != access(buf, F_OK)) { 2342 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2343 } 2344 2345 if (0 == access(buf, F_OK)) { 2346 // Use current module name "libjvm.so" 2347 len = strlen(buf); 2348 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2349 } else { 2350 // Go back to path of .so 2351 rp = realpath(dli_fname, buf); 2352 if (rp == NULL) 2353 return; 2354 } 2355 } 2356 } 2357 } 2358 2359 strcpy(saved_jvm_path, buf); 2360} 2361 2362void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2363 // no prefix required, not even "_" 2364} 2365 2366void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2367 // no suffix required 2368} 2369 2370//////////////////////////////////////////////////////////////////////////////// 2371// sun.misc.Signal support 2372 2373static volatile jint sigint_count = 0; 2374 2375static void 2376UserHandler(int sig, void *siginfo, void *context) { 2377 // 4511530 - sem_post is serialized and handled by the manager thread. When 2378 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2379 // don't want to flood the manager thread with sem_post requests. 2380 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) 2381 return; 2382 2383 // Ctrl-C is pressed during error reporting, likely because the error 2384 // handler fails to abort. Let VM die immediately. 2385 if (sig == SIGINT && is_error_reported()) { 2386 os::die(); 2387 } 2388 2389 os::signal_notify(sig); 2390} 2391 2392void* os::user_handler() { 2393 return CAST_FROM_FN_PTR(void*, UserHandler); 2394} 2395 2396class Semaphore : public StackObj { 2397 public: 2398 Semaphore(); 2399 ~Semaphore(); 2400 void signal(); 2401 void wait(); 2402 bool trywait(); 2403 bool timedwait(unsigned int sec, int nsec); 2404 private: 2405 sem_t _semaphore; 2406}; 2407 2408Semaphore::Semaphore() { 2409 sem_init(&_semaphore, 0, 0); 2410} 2411 2412Semaphore::~Semaphore() { 2413 sem_destroy(&_semaphore); 2414} 2415 2416void Semaphore::signal() { 2417 sem_post(&_semaphore); 2418} 2419 2420void Semaphore::wait() { 2421 sem_wait(&_semaphore); 2422} 2423 2424bool Semaphore::trywait() { 2425 return sem_trywait(&_semaphore) == 0; 2426} 2427 2428bool Semaphore::timedwait(unsigned int sec, int nsec) { 2429 2430 struct timespec ts; 2431 // Semaphore's are always associated with CLOCK_REALTIME 2432 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2433 // see unpackTime for discussion on overflow checking 2434 if (sec >= MAX_SECS) { 2435 ts.tv_sec += MAX_SECS; 2436 ts.tv_nsec = 0; 2437 } else { 2438 ts.tv_sec += sec; 2439 ts.tv_nsec += nsec; 2440 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2441 ts.tv_nsec -= NANOSECS_PER_SEC; 2442 ++ts.tv_sec; // note: this must be <= max_secs 2443 } 2444 } 2445 2446 while (1) { 2447 int result = sem_timedwait(&_semaphore, &ts); 2448 if (result == 0) { 2449 return true; 2450 } else if (errno == EINTR) { 2451 continue; 2452 } else if (errno == ETIMEDOUT) { 2453 return false; 2454 } else { 2455 return false; 2456 } 2457 } 2458} 2459 2460extern "C" { 2461 typedef void (*sa_handler_t)(int); 2462 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2463} 2464 2465void* os::signal(int signal_number, void* handler) { 2466 struct sigaction sigAct, oldSigAct; 2467 2468 sigfillset(&(sigAct.sa_mask)); 2469 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2470 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2471 2472 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2473 // -1 means registration failed 2474 return (void *)-1; 2475 } 2476 2477 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2478} 2479 2480void os::signal_raise(int signal_number) { 2481 ::raise(signal_number); 2482} 2483 2484/* 2485 * The following code is moved from os.cpp for making this 2486 * code platform specific, which it is by its very nature. 2487 */ 2488 2489// Will be modified when max signal is changed to be dynamic 2490int os::sigexitnum_pd() { 2491 return NSIG; 2492} 2493 2494// a counter for each possible signal value 2495static volatile jint pending_signals[NSIG+1] = { 0 }; 2496 2497// Linux(POSIX) specific hand shaking semaphore. 2498static sem_t sig_sem; 2499static Semaphore sr_semaphore; 2500 2501void os::signal_init_pd() { 2502 // Initialize signal structures 2503 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2504 2505 // Initialize signal semaphore 2506 ::sem_init(&sig_sem, 0, 0); 2507} 2508 2509void os::signal_notify(int sig) { 2510 Atomic::inc(&pending_signals[sig]); 2511 ::sem_post(&sig_sem); 2512} 2513 2514static int check_pending_signals(bool wait) { 2515 Atomic::store(0, &sigint_count); 2516 for (;;) { 2517 for (int i = 0; i < NSIG + 1; i++) { 2518 jint n = pending_signals[i]; 2519 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2520 return i; 2521 } 2522 } 2523 if (!wait) { 2524 return -1; 2525 } 2526 JavaThread *thread = JavaThread::current(); 2527 ThreadBlockInVM tbivm(thread); 2528 2529 bool threadIsSuspended; 2530 do { 2531 thread->set_suspend_equivalent(); 2532 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2533 ::sem_wait(&sig_sem); 2534 2535 // were we externally suspended while we were waiting? 2536 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2537 if (threadIsSuspended) { 2538 // 2539 // The semaphore has been incremented, but while we were waiting 2540 // another thread suspended us. We don't want to continue running 2541 // while suspended because that would surprise the thread that 2542 // suspended us. 2543 // 2544 ::sem_post(&sig_sem); 2545 2546 thread->java_suspend_self(); 2547 } 2548 } while (threadIsSuspended); 2549 } 2550} 2551 2552int os::signal_lookup() { 2553 return check_pending_signals(false); 2554} 2555 2556int os::signal_wait() { 2557 return check_pending_signals(true); 2558} 2559 2560//////////////////////////////////////////////////////////////////////////////// 2561// Virtual Memory 2562 2563int os::vm_page_size() { 2564 // Seems redundant as all get out 2565 assert(os::Linux::page_size() != -1, "must call os::init"); 2566 return os::Linux::page_size(); 2567} 2568 2569// Solaris allocates memory by pages. 2570int os::vm_allocation_granularity() { 2571 assert(os::Linux::page_size() != -1, "must call os::init"); 2572 return os::Linux::page_size(); 2573} 2574 2575// Rationale behind this function: 2576// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2577// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2578// samples for JITted code. Here we create private executable mapping over the code cache 2579// and then we can use standard (well, almost, as mapping can change) way to provide 2580// info for the reporting script by storing timestamp and location of symbol 2581void linux_wrap_code(char* base, size_t size) { 2582 static volatile jint cnt = 0; 2583 2584 if (!UseOprofile) { 2585 return; 2586 } 2587 2588 char buf[PATH_MAX+1]; 2589 int num = Atomic::add(1, &cnt); 2590 2591 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2592 os::get_temp_directory(), os::current_process_id(), num); 2593 unlink(buf); 2594 2595 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2596 2597 if (fd != -1) { 2598 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2599 if (rv != (off_t)-1) { 2600 if (::write(fd, "", 1) == 1) { 2601 mmap(base, size, 2602 PROT_READ|PROT_WRITE|PROT_EXEC, 2603 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2604 } 2605 } 2606 ::close(fd); 2607 unlink(buf); 2608 } 2609} 2610 2611static bool recoverable_mmap_error(int err) { 2612 // See if the error is one we can let the caller handle. This 2613 // list of errno values comes from JBS-6843484. I can't find a 2614 // Linux man page that documents this specific set of errno 2615 // values so while this list currently matches Solaris, it may 2616 // change as we gain experience with this failure mode. 2617 switch (err) { 2618 case EBADF: 2619 case EINVAL: 2620 case ENOTSUP: 2621 // let the caller deal with these errors 2622 return true; 2623 2624 default: 2625 // Any remaining errors on this OS can cause our reserved mapping 2626 // to be lost. That can cause confusion where different data 2627 // structures think they have the same memory mapped. The worst 2628 // scenario is if both the VM and a library think they have the 2629 // same memory mapped. 2630 return false; 2631 } 2632} 2633 2634static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2635 int err) { 2636 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2637 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2638 strerror(err), err); 2639} 2640 2641static void warn_fail_commit_memory(char* addr, size_t size, 2642 size_t alignment_hint, bool exec, 2643 int err) { 2644 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2645 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2646 alignment_hint, exec, strerror(err), err); 2647} 2648 2649// NOTE: Linux kernel does not really reserve the pages for us. 2650// All it does is to check if there are enough free pages 2651// left at the time of mmap(). This could be a potential 2652// problem. 2653int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2654 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2655 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2656 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2657 if (res != (uintptr_t) MAP_FAILED) { 2658 if (UseNUMAInterleaving) { 2659 numa_make_global(addr, size); 2660 } 2661 return 0; 2662 } 2663 2664 int err = errno; // save errno from mmap() call above 2665 2666 if (!recoverable_mmap_error(err)) { 2667 warn_fail_commit_memory(addr, size, exec, err); 2668 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2669 } 2670 2671 return err; 2672} 2673 2674bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2675 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2676} 2677 2678void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2679 const char* mesg) { 2680 assert(mesg != NULL, "mesg must be specified"); 2681 int err = os::Linux::commit_memory_impl(addr, size, exec); 2682 if (err != 0) { 2683 // the caller wants all commit errors to exit with the specified mesg: 2684 warn_fail_commit_memory(addr, size, exec, err); 2685 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2686 } 2687} 2688 2689// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2690#ifndef MAP_HUGETLB 2691#define MAP_HUGETLB 0x40000 2692#endif 2693 2694// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2695#ifndef MADV_HUGEPAGE 2696#define MADV_HUGEPAGE 14 2697#endif 2698 2699int os::Linux::commit_memory_impl(char* addr, size_t size, 2700 size_t alignment_hint, bool exec) { 2701 int err = os::Linux::commit_memory_impl(addr, size, exec); 2702 if (err == 0) { 2703 realign_memory(addr, size, alignment_hint); 2704 } 2705 return err; 2706} 2707 2708bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2709 bool exec) { 2710 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2711} 2712 2713void os::pd_commit_memory_or_exit(char* addr, size_t size, 2714 size_t alignment_hint, bool exec, 2715 const char* mesg) { 2716 assert(mesg != NULL, "mesg must be specified"); 2717 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2718 if (err != 0) { 2719 // the caller wants all commit errors to exit with the specified mesg: 2720 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2721 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2722 } 2723} 2724 2725void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2726 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2727 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2728 // be supported or the memory may already be backed by huge pages. 2729 ::madvise(addr, bytes, MADV_HUGEPAGE); 2730 } 2731} 2732 2733void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2734 // This method works by doing an mmap over an existing mmaping and effectively discarding 2735 // the existing pages. However it won't work for SHM-based large pages that cannot be 2736 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2737 // small pages on top of the SHM segment. This method always works for small pages, so we 2738 // allow that in any case. 2739 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2740 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2741 } 2742} 2743 2744void os::numa_make_global(char *addr, size_t bytes) { 2745 Linux::numa_interleave_memory(addr, bytes); 2746} 2747 2748// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2749// bind policy to MPOL_PREFERRED for the current thread. 2750#define USE_MPOL_PREFERRED 0 2751 2752void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2753 // To make NUMA and large pages more robust when both enabled, we need to ease 2754 // the requirements on where the memory should be allocated. MPOL_BIND is the 2755 // default policy and it will force memory to be allocated on the specified 2756 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2757 // the specified node, but will not force it. Using this policy will prevent 2758 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2759 // free large pages. 2760 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2761 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2762} 2763 2764bool os::numa_topology_changed() { return false; } 2765 2766size_t os::numa_get_groups_num() { 2767 int max_node = Linux::numa_max_node(); 2768 return max_node > 0 ? max_node + 1 : 1; 2769} 2770 2771int os::numa_get_group_id() { 2772 int cpu_id = Linux::sched_getcpu(); 2773 if (cpu_id != -1) { 2774 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2775 if (lgrp_id != -1) { 2776 return lgrp_id; 2777 } 2778 } 2779 return 0; 2780} 2781 2782size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2783 for (size_t i = 0; i < size; i++) { 2784 ids[i] = i; 2785 } 2786 return size; 2787} 2788 2789bool os::get_page_info(char *start, page_info* info) { 2790 return false; 2791} 2792 2793char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) { 2794 return end; 2795} 2796 2797 2798int os::Linux::sched_getcpu_syscall(void) { 2799 unsigned int cpu; 2800 int retval = -1; 2801 2802#if defined(IA32) 2803# ifndef SYS_getcpu 2804# define SYS_getcpu 318 2805# endif 2806 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2807#elif defined(AMD64) 2808// Unfortunately we have to bring all these macros here from vsyscall.h 2809// to be able to compile on old linuxes. 2810# define __NR_vgetcpu 2 2811# define VSYSCALL_START (-10UL << 20) 2812# define VSYSCALL_SIZE 1024 2813# define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2814 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2815 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2816 retval = vgetcpu(&cpu, NULL, NULL); 2817#endif 2818 2819 return (retval == -1) ? retval : cpu; 2820} 2821 2822// Something to do with the numa-aware allocator needs these symbols 2823extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2824extern "C" JNIEXPORT void numa_error(char *where) { } 2825extern "C" JNIEXPORT int fork1() { return fork(); } 2826 2827 2828// If we are running with libnuma version > 2, then we should 2829// be trying to use symbols with versions 1.1 2830// If we are running with earlier version, which did not have symbol versions, 2831// we should use the base version. 2832void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2833 void *f = dlvsym(handle, name, "libnuma_1.1"); 2834 if (f == NULL) { 2835 f = dlsym(handle, name); 2836 } 2837 return f; 2838} 2839 2840bool os::Linux::libnuma_init() { 2841 // sched_getcpu() should be in libc. 2842 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2843 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2844 2845 // If it's not, try a direct syscall. 2846 if (sched_getcpu() == -1) 2847 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall)); 2848 2849 if (sched_getcpu() != -1) { // Does it work? 2850 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2851 if (handle != NULL) { 2852 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2853 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2854 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2855 libnuma_dlsym(handle, "numa_max_node"))); 2856 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2857 libnuma_dlsym(handle, "numa_available"))); 2858 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2859 libnuma_dlsym(handle, "numa_tonode_memory"))); 2860 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2861 libnuma_dlsym(handle, "numa_interleave_memory"))); 2862 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2863 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2864 2865 2866 if (numa_available() != -1) { 2867 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2868 // Create a cpu -> node mapping 2869 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2870 rebuild_cpu_to_node_map(); 2871 return true; 2872 } 2873 } 2874 } 2875 return false; 2876} 2877 2878// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2879// The table is later used in get_node_by_cpu(). 2880void os::Linux::rebuild_cpu_to_node_map() { 2881 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2882 // in libnuma (possible values are starting from 16, 2883 // and continuing up with every other power of 2, but less 2884 // than the maximum number of CPUs supported by kernel), and 2885 // is a subject to change (in libnuma version 2 the requirements 2886 // are more reasonable) we'll just hardcode the number they use 2887 // in the library. 2888 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2889 2890 size_t cpu_num = os::active_processor_count(); 2891 size_t cpu_map_size = NCPUS / BitsPerCLong; 2892 size_t cpu_map_valid_size = 2893 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2894 2895 cpu_to_node()->clear(); 2896 cpu_to_node()->at_grow(cpu_num - 1); 2897 size_t node_num = numa_get_groups_num(); 2898 2899 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2900 for (size_t i = 0; i < node_num; i++) { 2901 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2902 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2903 if (cpu_map[j] != 0) { 2904 for (size_t k = 0; k < BitsPerCLong; k++) { 2905 if (cpu_map[j] & (1UL << k)) { 2906 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2907 } 2908 } 2909 } 2910 } 2911 } 2912 } 2913 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal); 2914} 2915 2916int os::Linux::get_node_by_cpu(int cpu_id) { 2917 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2918 return cpu_to_node()->at(cpu_id); 2919 } 2920 return -1; 2921} 2922 2923GrowableArray<int>* os::Linux::_cpu_to_node; 2924os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2925os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2926os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2927os::Linux::numa_available_func_t os::Linux::_numa_available; 2928os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2929os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2930os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2931unsigned long* os::Linux::_numa_all_nodes; 2932 2933bool os::pd_uncommit_memory(char* addr, size_t size) { 2934 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2935 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2936 return res != (uintptr_t) MAP_FAILED; 2937} 2938 2939static 2940address get_stack_commited_bottom(address bottom, size_t size) { 2941 address nbot = bottom; 2942 address ntop = bottom + size; 2943 2944 size_t page_sz = os::vm_page_size(); 2945 unsigned pages = size / page_sz; 2946 2947 unsigned char vec[1]; 2948 unsigned imin = 1, imax = pages + 1, imid; 2949 int mincore_return_value = 0; 2950 2951 assert(imin <= imax, "Unexpected page size"); 2952 2953 while (imin < imax) { 2954 imid = (imax + imin) / 2; 2955 nbot = ntop - (imid * page_sz); 2956 2957 // Use a trick with mincore to check whether the page is mapped or not. 2958 // mincore sets vec to 1 if page resides in memory and to 0 if page 2959 // is swapped output but if page we are asking for is unmapped 2960 // it returns -1,ENOMEM 2961 mincore_return_value = mincore(nbot, page_sz, vec); 2962 2963 if (mincore_return_value == -1) { 2964 // Page is not mapped go up 2965 // to find first mapped page 2966 if (errno != EAGAIN) { 2967 assert(errno == ENOMEM, "Unexpected mincore errno"); 2968 imax = imid; 2969 } 2970 } else { 2971 // Page is mapped go down 2972 // to find first not mapped page 2973 imin = imid + 1; 2974 } 2975 } 2976 2977 nbot = nbot + page_sz; 2978 2979 // Adjust stack bottom one page up if last checked page is not mapped 2980 if (mincore_return_value == -1) { 2981 nbot = nbot + page_sz; 2982 } 2983 2984 return nbot; 2985} 2986 2987 2988// Linux uses a growable mapping for the stack, and if the mapping for 2989// the stack guard pages is not removed when we detach a thread the 2990// stack cannot grow beyond the pages where the stack guard was 2991// mapped. If at some point later in the process the stack expands to 2992// that point, the Linux kernel cannot expand the stack any further 2993// because the guard pages are in the way, and a segfault occurs. 2994// 2995// However, it's essential not to split the stack region by unmapping 2996// a region (leaving a hole) that's already part of the stack mapping, 2997// so if the stack mapping has already grown beyond the guard pages at 2998// the time we create them, we have to truncate the stack mapping. 2999// So, we need to know the extent of the stack mapping when 3000// create_stack_guard_pages() is called. 3001 3002// We only need this for stacks that are growable: at the time of 3003// writing thread stacks don't use growable mappings (i.e. those 3004// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3005// only applies to the main thread. 3006 3007// If the (growable) stack mapping already extends beyond the point 3008// where we're going to put our guard pages, truncate the mapping at 3009// that point by munmap()ping it. This ensures that when we later 3010// munmap() the guard pages we don't leave a hole in the stack 3011// mapping. This only affects the main/initial thread 3012 3013bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3014 3015 if (os::Linux::is_initial_thread()) { 3016 // As we manually grow stack up to bottom inside create_attached_thread(), 3017 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3018 // we don't need to do anything special. 3019 // Check it first, before calling heavy function. 3020 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3021 unsigned char vec[1]; 3022 3023 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3024 // Fallback to slow path on all errors, including EAGAIN 3025 stack_extent = (uintptr_t) get_stack_commited_bottom( 3026 os::Linux::initial_thread_stack_bottom(), 3027 (size_t)addr - stack_extent); 3028 } 3029 3030 if (stack_extent < (uintptr_t)addr) { 3031 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3032 } 3033 } 3034 3035 return os::commit_memory(addr, size, !ExecMem); 3036} 3037 3038// If this is a growable mapping, remove the guard pages entirely by 3039// munmap()ping them. If not, just call uncommit_memory(). This only 3040// affects the main/initial thread, but guard against future OS changes 3041// It's safe to always unmap guard pages for initial thread because we 3042// always place it right after end of the mapped region 3043 3044bool os::remove_stack_guard_pages(char* addr, size_t size) { 3045 uintptr_t stack_extent, stack_base; 3046 3047 if (os::Linux::is_initial_thread()) { 3048 return ::munmap(addr, size) == 0; 3049 } 3050 3051 return os::uncommit_memory(addr, size); 3052} 3053 3054static address _highest_vm_reserved_address = NULL; 3055 3056// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3057// at 'requested_addr'. If there are existing memory mappings at the same 3058// location, however, they will be overwritten. If 'fixed' is false, 3059// 'requested_addr' is only treated as a hint, the return value may or 3060// may not start from the requested address. Unlike Linux mmap(), this 3061// function returns NULL to indicate failure. 3062static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3063 char * addr; 3064 int flags; 3065 3066 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3067 if (fixed) { 3068 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3069 flags |= MAP_FIXED; 3070 } 3071 3072 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3073 // touch an uncommitted page. Otherwise, the read/write might 3074 // succeed if we have enough swap space to back the physical page. 3075 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3076 flags, -1, 0); 3077 3078 if (addr != MAP_FAILED) { 3079 // anon_mmap() should only get called during VM initialization, 3080 // don't need lock (actually we can skip locking even it can be called 3081 // from multiple threads, because _highest_vm_reserved_address is just a 3082 // hint about the upper limit of non-stack memory regions.) 3083 if ((address)addr + bytes > _highest_vm_reserved_address) { 3084 _highest_vm_reserved_address = (address)addr + bytes; 3085 } 3086 } 3087 3088 return addr == MAP_FAILED ? NULL : addr; 3089} 3090 3091// Don't update _highest_vm_reserved_address, because there might be memory 3092// regions above addr + size. If so, releasing a memory region only creates 3093// a hole in the address space, it doesn't help prevent heap-stack collision. 3094// 3095static int anon_munmap(char * addr, size_t size) { 3096 return ::munmap(addr, size) == 0; 3097} 3098 3099char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3100 size_t alignment_hint) { 3101 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3102} 3103 3104bool os::pd_release_memory(char* addr, size_t size) { 3105 return anon_munmap(addr, size); 3106} 3107 3108static address highest_vm_reserved_address() { 3109 return _highest_vm_reserved_address; 3110} 3111 3112static bool linux_mprotect(char* addr, size_t size, int prot) { 3113 // Linux wants the mprotect address argument to be page aligned. 3114 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3115 3116 // According to SUSv3, mprotect() should only be used with mappings 3117 // established by mmap(), and mmap() always maps whole pages. Unaligned 3118 // 'addr' likely indicates problem in the VM (e.g. trying to change 3119 // protection of malloc'ed or statically allocated memory). Check the 3120 // caller if you hit this assert. 3121 assert(addr == bottom, "sanity check"); 3122 3123 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3124 return ::mprotect(bottom, size, prot) == 0; 3125} 3126 3127// Set protections specified 3128bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3129 bool is_committed) { 3130 unsigned int p = 0; 3131 switch (prot) { 3132 case MEM_PROT_NONE: p = PROT_NONE; break; 3133 case MEM_PROT_READ: p = PROT_READ; break; 3134 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3135 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3136 default: 3137 ShouldNotReachHere(); 3138 } 3139 // is_committed is unused. 3140 return linux_mprotect(addr, bytes, p); 3141} 3142 3143bool os::guard_memory(char* addr, size_t size) { 3144 return linux_mprotect(addr, size, PROT_NONE); 3145} 3146 3147bool os::unguard_memory(char* addr, size_t size) { 3148 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3149} 3150 3151bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) { 3152 bool result = false; 3153 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3154 MAP_ANONYMOUS|MAP_PRIVATE, 3155 -1, 0); 3156 if (p != MAP_FAILED) { 3157 void *aligned_p = align_ptr_up(p, page_size); 3158 3159 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3160 3161 munmap(p, page_size * 2); 3162 } 3163 3164 if (warn && !result) { 3165 warning("TransparentHugePages is not supported by the operating system."); 3166 } 3167 3168 return result; 3169} 3170 3171bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3172 bool result = false; 3173 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3174 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3175 -1, 0); 3176 3177 if (p != MAP_FAILED) { 3178 // We don't know if this really is a huge page or not. 3179 FILE *fp = fopen("/proc/self/maps", "r"); 3180 if (fp) { 3181 while (!feof(fp)) { 3182 char chars[257]; 3183 long x = 0; 3184 if (fgets(chars, sizeof(chars), fp)) { 3185 if (sscanf(chars, "%lx-%*x", &x) == 1 3186 && x == (long)p) { 3187 if (strstr (chars, "hugepage")) { 3188 result = true; 3189 break; 3190 } 3191 } 3192 } 3193 } 3194 fclose(fp); 3195 } 3196 munmap(p, page_size); 3197 } 3198 3199 if (warn && !result) { 3200 warning("HugeTLBFS is not supported by the operating system."); 3201 } 3202 3203 return result; 3204} 3205 3206/* 3207* Set the coredump_filter bits to include largepages in core dump (bit 6) 3208* 3209* From the coredump_filter documentation: 3210* 3211* - (bit 0) anonymous private memory 3212* - (bit 1) anonymous shared memory 3213* - (bit 2) file-backed private memory 3214* - (bit 3) file-backed shared memory 3215* - (bit 4) ELF header pages in file-backed private memory areas (it is 3216* effective only if the bit 2 is cleared) 3217* - (bit 5) hugetlb private memory 3218* - (bit 6) hugetlb shared memory 3219*/ 3220static void set_coredump_filter(void) { 3221 FILE *f; 3222 long cdm; 3223 3224 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3225 return; 3226 } 3227 3228 if (fscanf(f, "%lx", &cdm) != 1) { 3229 fclose(f); 3230 return; 3231 } 3232 3233 rewind(f); 3234 3235 if ((cdm & LARGEPAGES_BIT) == 0) { 3236 cdm |= LARGEPAGES_BIT; 3237 fprintf(f, "%#lx", cdm); 3238 } 3239 3240 fclose(f); 3241} 3242 3243// Large page support 3244 3245static size_t _large_page_size = 0; 3246 3247size_t os::Linux::find_large_page_size() { 3248 size_t large_page_size = 0; 3249 3250 // large_page_size on Linux is used to round up heap size. x86 uses either 3251 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3252 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3253 // page as large as 256M. 3254 // 3255 // Here we try to figure out page size by parsing /proc/meminfo and looking 3256 // for a line with the following format: 3257 // Hugepagesize: 2048 kB 3258 // 3259 // If we can't determine the value (e.g. /proc is not mounted, or the text 3260 // format has been changed), we'll use the largest page size supported by 3261 // the processor. 3262 3263#ifndef ZERO 3264 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3265 ARM_ONLY(2 * M) PPC_ONLY(4 * M); 3266#endif // ZERO 3267 3268 FILE *fp = fopen("/proc/meminfo", "r"); 3269 if (fp) { 3270 while (!feof(fp)) { 3271 int x = 0; 3272 char buf[16]; 3273 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3274 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3275 large_page_size = x * K; 3276 break; 3277 } 3278 } else { 3279 // skip to next line 3280 for (;;) { 3281 int ch = fgetc(fp); 3282 if (ch == EOF || ch == (int)'\n') break; 3283 } 3284 } 3285 } 3286 fclose(fp); 3287 } 3288 3289 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3290 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3291 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3292 proper_unit_for_byte_size(large_page_size)); 3293 } 3294 3295 return large_page_size; 3296} 3297 3298size_t os::Linux::setup_large_page_size() { 3299 _large_page_size = Linux::find_large_page_size(); 3300 const size_t default_page_size = (size_t)Linux::page_size(); 3301 if (_large_page_size > default_page_size) { 3302 _page_sizes[0] = _large_page_size; 3303 _page_sizes[1] = default_page_size; 3304 _page_sizes[2] = 0; 3305 } 3306 3307 return _large_page_size; 3308} 3309 3310bool os::Linux::setup_large_page_type(size_t page_size) { 3311 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3312 FLAG_IS_DEFAULT(UseSHM) && 3313 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3314 3315 // The type of large pages has not been specified by the user. 3316 3317 // Try UseHugeTLBFS and then UseSHM. 3318 UseHugeTLBFS = UseSHM = true; 3319 3320 // Don't try UseTransparentHugePages since there are known 3321 // performance issues with it turned on. This might change in the future. 3322 UseTransparentHugePages = false; 3323 } 3324 3325 if (UseTransparentHugePages) { 3326 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3327 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3328 UseHugeTLBFS = false; 3329 UseSHM = false; 3330 return true; 3331 } 3332 UseTransparentHugePages = false; 3333 } 3334 3335 if (UseHugeTLBFS) { 3336 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3337 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3338 UseSHM = false; 3339 return true; 3340 } 3341 UseHugeTLBFS = false; 3342 } 3343 3344 return UseSHM; 3345} 3346 3347void os::large_page_init() { 3348 if (!UseLargePages && 3349 !UseTransparentHugePages && 3350 !UseHugeTLBFS && 3351 !UseSHM) { 3352 // Not using large pages. 3353 return; 3354 } 3355 3356 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3357 // The user explicitly turned off large pages. 3358 // Ignore the rest of the large pages flags. 3359 UseTransparentHugePages = false; 3360 UseHugeTLBFS = false; 3361 UseSHM = false; 3362 return; 3363 } 3364 3365 size_t large_page_size = Linux::setup_large_page_size(); 3366 UseLargePages = Linux::setup_large_page_type(large_page_size); 3367 3368 set_coredump_filter(); 3369} 3370 3371#ifndef SHM_HUGETLB 3372#define SHM_HUGETLB 04000 3373#endif 3374 3375char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3376 // "exec" is passed in but not used. Creating the shared image for 3377 // the code cache doesn't have an SHM_X executable permission to check. 3378 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3379 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3380 3381 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3382 return NULL; // Fallback to small pages. 3383 } 3384 3385 key_t key = IPC_PRIVATE; 3386 char *addr; 3387 3388 bool warn_on_failure = UseLargePages && 3389 (!FLAG_IS_DEFAULT(UseLargePages) || 3390 !FLAG_IS_DEFAULT(UseSHM) || 3391 !FLAG_IS_DEFAULT(LargePageSizeInBytes) 3392 ); 3393 char msg[128]; 3394 3395 // Create a large shared memory region to attach to based on size. 3396 // Currently, size is the total size of the heap 3397 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3398 if (shmid == -1) { 3399 // Possible reasons for shmget failure: 3400 // 1. shmmax is too small for Java heap. 3401 // > check shmmax value: cat /proc/sys/kernel/shmmax 3402 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3403 // 2. not enough large page memory. 3404 // > check available large pages: cat /proc/meminfo 3405 // > increase amount of large pages: 3406 // echo new_value > /proc/sys/vm/nr_hugepages 3407 // Note 1: different Linux may use different name for this property, 3408 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3409 // Note 2: it's possible there's enough physical memory available but 3410 // they are so fragmented after a long run that they can't 3411 // coalesce into large pages. Try to reserve large pages when 3412 // the system is still "fresh". 3413 if (warn_on_failure) { 3414 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3415 warning(msg); 3416 } 3417 return NULL; 3418 } 3419 3420 // attach to the region 3421 addr = (char*)shmat(shmid, req_addr, 0); 3422 int err = errno; 3423 3424 // Remove shmid. If shmat() is successful, the actual shared memory segment 3425 // will be deleted when it's detached by shmdt() or when the process 3426 // terminates. If shmat() is not successful this will remove the shared 3427 // segment immediately. 3428 shmctl(shmid, IPC_RMID, NULL); 3429 3430 if ((intptr_t)addr == -1) { 3431 if (warn_on_failure) { 3432 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3433 warning(msg); 3434 } 3435 return NULL; 3436 } 3437 3438 return addr; 3439} 3440 3441static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) { 3442 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3443 3444 bool warn_on_failure = UseLargePages && 3445 (!FLAG_IS_DEFAULT(UseLargePages) || 3446 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3447 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3448 3449 if (warn_on_failure) { 3450 char msg[128]; 3451 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3452 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3453 warning(msg); 3454 } 3455} 3456 3457char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) { 3458 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3459 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3460 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3461 3462 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3463 char* addr = (char*)::mmap(req_addr, bytes, prot, 3464 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3465 -1, 0); 3466 3467 if (addr == MAP_FAILED) { 3468 warn_on_large_pages_failure(req_addr, bytes, errno); 3469 return NULL; 3470 } 3471 3472 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3473 3474 return addr; 3475} 3476 3477char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3478 size_t large_page_size = os::large_page_size(); 3479 3480 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3481 3482 // Allocate small pages. 3483 3484 char* start; 3485 if (req_addr != NULL) { 3486 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3487 assert(is_size_aligned(bytes, alignment), "Must be"); 3488 start = os::reserve_memory(bytes, req_addr); 3489 assert(start == NULL || start == req_addr, "Must be"); 3490 } else { 3491 start = os::reserve_memory_aligned(bytes, alignment); 3492 } 3493 3494 if (start == NULL) { 3495 return NULL; 3496 } 3497 3498 assert(is_ptr_aligned(start, alignment), "Must be"); 3499 3500 // os::reserve_memory_special will record this memory area. 3501 // Need to release it here to prevent overlapping reservations. 3502 MemTracker::record_virtual_memory_release((address)start, bytes); 3503 3504 char* end = start + bytes; 3505 3506 // Find the regions of the allocated chunk that can be promoted to large pages. 3507 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3508 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3509 3510 size_t lp_bytes = lp_end - lp_start; 3511 3512 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3513 3514 if (lp_bytes == 0) { 3515 // The mapped region doesn't even span the start and the end of a large page. 3516 // Fall back to allocate a non-special area. 3517 ::munmap(start, end - start); 3518 return NULL; 3519 } 3520 3521 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3522 3523 3524 void* result; 3525 3526 if (start != lp_start) { 3527 result = ::mmap(start, lp_start - start, prot, 3528 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3529 -1, 0); 3530 if (result == MAP_FAILED) { 3531 ::munmap(lp_start, end - lp_start); 3532 return NULL; 3533 } 3534 } 3535 3536 result = ::mmap(lp_start, lp_bytes, prot, 3537 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3538 -1, 0); 3539 if (result == MAP_FAILED) { 3540 warn_on_large_pages_failure(req_addr, bytes, errno); 3541 // If the mmap above fails, the large pages region will be unmapped and we 3542 // have regions before and after with small pages. Release these regions. 3543 // 3544 // | mapped | unmapped | mapped | 3545 // ^ ^ ^ ^ 3546 // start lp_start lp_end end 3547 // 3548 ::munmap(start, lp_start - start); 3549 ::munmap(lp_end, end - lp_end); 3550 return NULL; 3551 } 3552 3553 if (lp_end != end) { 3554 result = ::mmap(lp_end, end - lp_end, prot, 3555 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3556 -1, 0); 3557 if (result == MAP_FAILED) { 3558 ::munmap(start, lp_end - start); 3559 return NULL; 3560 } 3561 } 3562 3563 return start; 3564} 3565 3566char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3567 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3568 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3569 assert(is_power_of_2(alignment), "Must be"); 3570 assert(is_power_of_2(os::large_page_size()), "Must be"); 3571 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3572 3573 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3574 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3575 } else { 3576 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3577 } 3578} 3579 3580char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) { 3581 assert(UseLargePages, "only for large pages"); 3582 3583 char* addr; 3584 if (UseSHM) { 3585 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3586 } else { 3587 assert(UseHugeTLBFS, "must be"); 3588 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3589 } 3590 3591 if (addr != NULL) { 3592 if (UseNUMAInterleaving) { 3593 numa_make_global(addr, bytes); 3594 } 3595 3596 // The memory is committed 3597 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC); 3598 } 3599 3600 return addr; 3601} 3602 3603bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3604 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3605 return shmdt(base) == 0; 3606} 3607 3608bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3609 return pd_release_memory(base, bytes); 3610} 3611 3612bool os::release_memory_special(char* base, size_t bytes) { 3613 assert(UseLargePages, "only for large pages"); 3614 3615 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3616 3617 bool res; 3618 if (UseSHM) { 3619 res = os::Linux::release_memory_special_shm(base, bytes); 3620 } else { 3621 assert(UseHugeTLBFS, "must be"); 3622 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3623 } 3624 3625 if (res) { 3626 tkr.record((address)base, bytes); 3627 } else { 3628 tkr.discard(); 3629 } 3630 3631 return res; 3632} 3633 3634size_t os::large_page_size() { 3635 return _large_page_size; 3636} 3637 3638// With SysV SHM the entire memory region must be allocated as shared 3639// memory. 3640// HugeTLBFS allows application to commit large page memory on demand. 3641// However, when committing memory with HugeTLBFS fails, the region 3642// that was supposed to be committed will lose the old reservation 3643// and allow other threads to steal that memory region. Because of this 3644// behavior we can't commit HugeTLBFS memory. 3645bool os::can_commit_large_page_memory() { 3646 return UseTransparentHugePages; 3647} 3648 3649bool os::can_execute_large_page_memory() { 3650 return UseTransparentHugePages || UseHugeTLBFS; 3651} 3652 3653// Reserve memory at an arbitrary address, only if that area is 3654// available (and not reserved for something else). 3655 3656char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3657 const int max_tries = 10; 3658 char* base[max_tries]; 3659 size_t size[max_tries]; 3660 const size_t gap = 0x000000; 3661 3662 // Assert only that the size is a multiple of the page size, since 3663 // that's all that mmap requires, and since that's all we really know 3664 // about at this low abstraction level. If we need higher alignment, 3665 // we can either pass an alignment to this method or verify alignment 3666 // in one of the methods further up the call chain. See bug 5044738. 3667 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3668 3669 // Repeatedly allocate blocks until the block is allocated at the 3670 // right spot. Give up after max_tries. Note that reserve_memory() will 3671 // automatically update _highest_vm_reserved_address if the call is 3672 // successful. The variable tracks the highest memory address every reserved 3673 // by JVM. It is used to detect heap-stack collision if running with 3674 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3675 // space than needed, it could confuse the collision detecting code. To 3676 // solve the problem, save current _highest_vm_reserved_address and 3677 // calculate the correct value before return. 3678 address old_highest = _highest_vm_reserved_address; 3679 3680 // Linux mmap allows caller to pass an address as hint; give it a try first, 3681 // if kernel honors the hint then we can return immediately. 3682 char * addr = anon_mmap(requested_addr, bytes, false); 3683 if (addr == requested_addr) { 3684 return requested_addr; 3685 } 3686 3687 if (addr != NULL) { 3688 // mmap() is successful but it fails to reserve at the requested address 3689 anon_munmap(addr, bytes); 3690 } 3691 3692 int i; 3693 for (i = 0; i < max_tries; ++i) { 3694 base[i] = reserve_memory(bytes); 3695 3696 if (base[i] != NULL) { 3697 // Is this the block we wanted? 3698 if (base[i] == requested_addr) { 3699 size[i] = bytes; 3700 break; 3701 } 3702 3703 // Does this overlap the block we wanted? Give back the overlapped 3704 // parts and try again. 3705 3706 size_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3707 if (top_overlap >= 0 && top_overlap < bytes) { 3708 unmap_memory(base[i], top_overlap); 3709 base[i] += top_overlap; 3710 size[i] = bytes - top_overlap; 3711 } else { 3712 size_t bottom_overlap = base[i] + bytes - requested_addr; 3713 if (bottom_overlap >= 0 && bottom_overlap < bytes) { 3714 unmap_memory(requested_addr, bottom_overlap); 3715 size[i] = bytes - bottom_overlap; 3716 } else { 3717 size[i] = bytes; 3718 } 3719 } 3720 } 3721 } 3722 3723 // Give back the unused reserved pieces. 3724 3725 for (int j = 0; j < i; ++j) { 3726 if (base[j] != NULL) { 3727 unmap_memory(base[j], size[j]); 3728 } 3729 } 3730 3731 if (i < max_tries) { 3732 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3733 return requested_addr; 3734 } else { 3735 _highest_vm_reserved_address = old_highest; 3736 return NULL; 3737 } 3738} 3739 3740size_t os::read(int fd, void *buf, unsigned int nBytes) { 3741 return ::read(fd, buf, nBytes); 3742} 3743 3744// 3745// Short sleep, direct OS call. 3746// 3747// Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3748// sched_yield(2) will actually give up the CPU: 3749// 3750// * Alone on this pariticular CPU, keeps running. 3751// * Before the introduction of "skip_buddy" with "compat_yield" disabled 3752// (pre 2.6.39). 3753// 3754// So calling this with 0 is an alternative. 3755// 3756void os::naked_short_sleep(jlong ms) { 3757 struct timespec req; 3758 3759 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3760 req.tv_sec = 0; 3761 if (ms > 0) { 3762 req.tv_nsec = (ms % 1000) * 1000000; 3763 } 3764 else { 3765 req.tv_nsec = 1; 3766 } 3767 3768 nanosleep(&req, NULL); 3769 3770 return; 3771} 3772 3773// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3774void os::infinite_sleep() { 3775 while (true) { // sleep forever ... 3776 ::sleep(100); // ... 100 seconds at a time 3777 } 3778} 3779 3780// Used to convert frequent JVM_Yield() to nops 3781bool os::dont_yield() { 3782 return DontYieldALot; 3783} 3784 3785void os::yield() { 3786 sched_yield(); 3787} 3788 3789os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;} 3790 3791void os::yield_all(int attempts) { 3792 // Yields to all threads, including threads with lower priorities 3793 // Threads on Linux are all with same priority. The Solaris style 3794 // os::yield_all() with nanosleep(1ms) is not necessary. 3795 sched_yield(); 3796} 3797 3798// Called from the tight loops to possibly influence time-sharing heuristics 3799void os::loop_breaker(int attempts) { 3800 os::yield_all(attempts); 3801} 3802 3803//////////////////////////////////////////////////////////////////////////////// 3804// thread priority support 3805 3806// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3807// only supports dynamic priority, static priority must be zero. For real-time 3808// applications, Linux supports SCHED_RR which allows static priority (1-99). 3809// However, for large multi-threaded applications, SCHED_RR is not only slower 3810// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3811// of 5 runs - Sep 2005). 3812// 3813// The following code actually changes the niceness of kernel-thread/LWP. It 3814// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3815// not the entire user process, and user level threads are 1:1 mapped to kernel 3816// threads. It has always been the case, but could change in the future. For 3817// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3818// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3819 3820int os::java_to_os_priority[CriticalPriority + 1] = { 3821 19, // 0 Entry should never be used 3822 3823 4, // 1 MinPriority 3824 3, // 2 3825 2, // 3 3826 3827 1, // 4 3828 0, // 5 NormPriority 3829 -1, // 6 3830 3831 -2, // 7 3832 -3, // 8 3833 -4, // 9 NearMaxPriority 3834 3835 -5, // 10 MaxPriority 3836 3837 -5 // 11 CriticalPriority 3838}; 3839 3840static int prio_init() { 3841 if (ThreadPriorityPolicy == 1) { 3842 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3843 // if effective uid is not root. Perhaps, a more elegant way of doing 3844 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3845 if (geteuid() != 0) { 3846 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3847 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3848 } 3849 ThreadPriorityPolicy = 0; 3850 } 3851 } 3852 if (UseCriticalJavaThreadPriority) { 3853 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3854 } 3855 return 0; 3856} 3857 3858OSReturn os::set_native_priority(Thread* thread, int newpri) { 3859 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK; 3860 3861 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3862 return (ret == 0) ? OS_OK : OS_ERR; 3863} 3864 3865OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) { 3866 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) { 3867 *priority_ptr = java_to_os_priority[NormPriority]; 3868 return OS_OK; 3869 } 3870 3871 errno = 0; 3872 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3873 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3874} 3875 3876// Hint to the underlying OS that a task switch would not be good. 3877// Void return because it's a hint and can fail. 3878void os::hint_no_preempt() {} 3879 3880//////////////////////////////////////////////////////////////////////////////// 3881// suspend/resume support 3882 3883// the low-level signal-based suspend/resume support is a remnant from the 3884// old VM-suspension that used to be for java-suspension, safepoints etc, 3885// within hotspot. Now there is a single use-case for this: 3886// - calling get_thread_pc() on the VMThread by the flat-profiler task 3887// that runs in the watcher thread. 3888// The remaining code is greatly simplified from the more general suspension 3889// code that used to be used. 3890// 3891// The protocol is quite simple: 3892// - suspend: 3893// - sends a signal to the target thread 3894// - polls the suspend state of the osthread using a yield loop 3895// - target thread signal handler (SR_handler) sets suspend state 3896// and blocks in sigsuspend until continued 3897// - resume: 3898// - sets target osthread state to continue 3899// - sends signal to end the sigsuspend loop in the SR_handler 3900// 3901// Note that the SR_lock plays no role in this suspend/resume protocol. 3902// 3903 3904static void resume_clear_context(OSThread *osthread) { 3905 osthread->set_ucontext(NULL); 3906 osthread->set_siginfo(NULL); 3907} 3908 3909static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) { 3910 osthread->set_ucontext(context); 3911 osthread->set_siginfo(siginfo); 3912} 3913 3914// 3915// Handler function invoked when a thread's execution is suspended or 3916// resumed. We have to be careful that only async-safe functions are 3917// called here (Note: most pthread functions are not async safe and 3918// should be avoided.) 3919// 3920// Note: sigwait() is a more natural fit than sigsuspend() from an 3921// interface point of view, but sigwait() prevents the signal hander 3922// from being run. libpthread would get very confused by not having 3923// its signal handlers run and prevents sigwait()'s use with the 3924// mutex granting granting signal. 3925// 3926// Currently only ever called on the VMThread and JavaThreads (PC sampling) 3927// 3928static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3929 // Save and restore errno to avoid confusing native code with EINTR 3930 // after sigsuspend. 3931 int old_errno = errno; 3932 3933 Thread* thread = Thread::current(); 3934 OSThread* osthread = thread->osthread(); 3935 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3936 3937 os::SuspendResume::State current = osthread->sr.state(); 3938 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3939 suspend_save_context(osthread, siginfo, context); 3940 3941 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 3942 os::SuspendResume::State state = osthread->sr.suspended(); 3943 if (state == os::SuspendResume::SR_SUSPENDED) { 3944 sigset_t suspend_set; // signals for sigsuspend() 3945 3946 // get current set of blocked signals and unblock resume signal 3947 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3948 sigdelset(&suspend_set, SR_signum); 3949 3950 sr_semaphore.signal(); 3951 // wait here until we are resumed 3952 while (1) { 3953 sigsuspend(&suspend_set); 3954 3955 os::SuspendResume::State result = osthread->sr.running(); 3956 if (result == os::SuspendResume::SR_RUNNING) { 3957 sr_semaphore.signal(); 3958 break; 3959 } 3960 } 3961 3962 } else if (state == os::SuspendResume::SR_RUNNING) { 3963 // request was cancelled, continue 3964 } else { 3965 ShouldNotReachHere(); 3966 } 3967 3968 resume_clear_context(osthread); 3969 } else if (current == os::SuspendResume::SR_RUNNING) { 3970 // request was cancelled, continue 3971 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 3972 // ignore 3973 } else { 3974 // ignore 3975 } 3976 3977 errno = old_errno; 3978} 3979 3980 3981static int SR_initialize() { 3982 struct sigaction act; 3983 char *s; 3984 /* Get signal number to use for suspend/resume */ 3985 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3986 int sig = ::strtol(s, 0, 10); 3987 if (sig > 0 || sig < _NSIG) { 3988 SR_signum = sig; 3989 } 3990 } 3991 3992 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3993 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3994 3995 sigemptyset(&SR_sigset); 3996 sigaddset(&SR_sigset, SR_signum); 3997 3998 /* Set up signal handler for suspend/resume */ 3999 act.sa_flags = SA_RESTART|SA_SIGINFO; 4000 act.sa_handler = (void (*)(int)) SR_handler; 4001 4002 // SR_signum is blocked by default. 4003 // 4528190 - We also need to block pthread restart signal (32 on all 4004 // supported Linux platforms). Note that LinuxThreads need to block 4005 // this signal for all threads to work properly. So we don't have 4006 // to use hard-coded signal number when setting up the mask. 4007 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4008 4009 if (sigaction(SR_signum, &act, 0) == -1) { 4010 return -1; 4011 } 4012 4013 // Save signal flag 4014 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4015 return 0; 4016} 4017 4018static int sr_notify(OSThread* osthread) { 4019 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4020 assert_status(status == 0, status, "pthread_kill"); 4021 return status; 4022} 4023 4024// "Randomly" selected value for how long we want to spin 4025// before bailing out on suspending a thread, also how often 4026// we send a signal to a thread we want to resume 4027static const int RANDOMLY_LARGE_INTEGER = 1000000; 4028static const int RANDOMLY_LARGE_INTEGER2 = 100; 4029 4030// returns true on success and false on error - really an error is fatal 4031// but this seems the normal response to library errors 4032static bool do_suspend(OSThread* osthread) { 4033 assert(osthread->sr.is_running(), "thread should be running"); 4034 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4035 4036 // mark as suspended and send signal 4037 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4038 // failed to switch, state wasn't running? 4039 ShouldNotReachHere(); 4040 return false; 4041 } 4042 4043 if (sr_notify(osthread) != 0) { 4044 ShouldNotReachHere(); 4045 } 4046 4047 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4048 while (true) { 4049 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4050 break; 4051 } else { 4052 // timeout 4053 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4054 if (cancelled == os::SuspendResume::SR_RUNNING) { 4055 return false; 4056 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4057 // make sure that we consume the signal on the semaphore as well 4058 sr_semaphore.wait(); 4059 break; 4060 } else { 4061 ShouldNotReachHere(); 4062 return false; 4063 } 4064 } 4065 } 4066 4067 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4068 return true; 4069} 4070 4071static void do_resume(OSThread* osthread) { 4072 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4073 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4074 4075 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4076 // failed to switch to WAKEUP_REQUEST 4077 ShouldNotReachHere(); 4078 return; 4079 } 4080 4081 while (true) { 4082 if (sr_notify(osthread) == 0) { 4083 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4084 if (osthread->sr.is_running()) { 4085 return; 4086 } 4087 } 4088 } else { 4089 ShouldNotReachHere(); 4090 } 4091 } 4092 4093 guarantee(osthread->sr.is_running(), "Must be running!"); 4094} 4095 4096/////////////////////////////////////////////////////////////////////////////////// 4097// signal handling (except suspend/resume) 4098 4099// This routine may be used by user applications as a "hook" to catch signals. 4100// The user-defined signal handler must pass unrecognized signals to this 4101// routine, and if it returns true (non-zero), then the signal handler must 4102// return immediately. If the flag "abort_if_unrecognized" is true, then this 4103// routine will never retun false (zero), but instead will execute a VM panic 4104// routine kill the process. 4105// 4106// If this routine returns false, it is OK to call it again. This allows 4107// the user-defined signal handler to perform checks either before or after 4108// the VM performs its own checks. Naturally, the user code would be making 4109// a serious error if it tried to handle an exception (such as a null check 4110// or breakpoint) that the VM was generating for its own correct operation. 4111// 4112// This routine may recognize any of the following kinds of signals: 4113// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4114// It should be consulted by handlers for any of those signals. 4115// 4116// The caller of this routine must pass in the three arguments supplied 4117// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4118// field of the structure passed to sigaction(). This routine assumes that 4119// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4120// 4121// Note that the VM will print warnings if it detects conflicting signal 4122// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4123// 4124extern "C" JNIEXPORT int 4125JVM_handle_linux_signal(int signo, siginfo_t* siginfo, 4126 void* ucontext, int abort_if_unrecognized); 4127 4128void signalHandler(int sig, siginfo_t* info, void* uc) { 4129 assert(info != NULL && uc != NULL, "it must be old kernel"); 4130 int orig_errno = errno; // Preserve errno value over signal handler. 4131 JVM_handle_linux_signal(sig, info, uc, true); 4132 errno = orig_errno; 4133} 4134 4135 4136// This boolean allows users to forward their own non-matching signals 4137// to JVM_handle_linux_signal, harmlessly. 4138bool os::Linux::signal_handlers_are_installed = false; 4139 4140// For signal-chaining 4141struct sigaction os::Linux::sigact[MAXSIGNUM]; 4142unsigned int os::Linux::sigs = 0; 4143bool os::Linux::libjsig_is_loaded = false; 4144typedef struct sigaction *(*get_signal_t)(int); 4145get_signal_t os::Linux::get_signal_action = NULL; 4146 4147struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4148 struct sigaction *actp = NULL; 4149 4150 if (libjsig_is_loaded) { 4151 // Retrieve the old signal handler from libjsig 4152 actp = (*get_signal_action)(sig); 4153 } 4154 if (actp == NULL) { 4155 // Retrieve the preinstalled signal handler from jvm 4156 actp = get_preinstalled_handler(sig); 4157 } 4158 4159 return actp; 4160} 4161 4162static bool call_chained_handler(struct sigaction *actp, int sig, 4163 siginfo_t *siginfo, void *context) { 4164 // Call the old signal handler 4165 if (actp->sa_handler == SIG_DFL) { 4166 // It's more reasonable to let jvm treat it as an unexpected exception 4167 // instead of taking the default action. 4168 return false; 4169 } else if (actp->sa_handler != SIG_IGN) { 4170 if ((actp->sa_flags & SA_NODEFER) == 0) { 4171 // automaticlly block the signal 4172 sigaddset(&(actp->sa_mask), sig); 4173 } 4174 4175 sa_handler_t hand; 4176 sa_sigaction_t sa; 4177 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4178 // retrieve the chained handler 4179 if (siginfo_flag_set) { 4180 sa = actp->sa_sigaction; 4181 } else { 4182 hand = actp->sa_handler; 4183 } 4184 4185 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4186 actp->sa_handler = SIG_DFL; 4187 } 4188 4189 // try to honor the signal mask 4190 sigset_t oset; 4191 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4192 4193 // call into the chained handler 4194 if (siginfo_flag_set) { 4195 (*sa)(sig, siginfo, context); 4196 } else { 4197 (*hand)(sig); 4198 } 4199 4200 // restore the signal mask 4201 pthread_sigmask(SIG_SETMASK, &oset, 0); 4202 } 4203 // Tell jvm's signal handler the signal is taken care of. 4204 return true; 4205} 4206 4207bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4208 bool chained = false; 4209 // signal-chaining 4210 if (UseSignalChaining) { 4211 struct sigaction *actp = get_chained_signal_action(sig); 4212 if (actp != NULL) { 4213 chained = call_chained_handler(actp, sig, siginfo, context); 4214 } 4215 } 4216 return chained; 4217} 4218 4219struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4220 if ((( (unsigned int)1 << sig ) & sigs) != 0) { 4221 return &sigact[sig]; 4222 } 4223 return NULL; 4224} 4225 4226void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4227 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4228 sigact[sig] = oldAct; 4229 sigs |= (unsigned int)1 << sig; 4230} 4231 4232// for diagnostic 4233int os::Linux::sigflags[MAXSIGNUM]; 4234 4235int os::Linux::get_our_sigflags(int sig) { 4236 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4237 return sigflags[sig]; 4238} 4239 4240void os::Linux::set_our_sigflags(int sig, int flags) { 4241 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4242 sigflags[sig] = flags; 4243} 4244 4245void os::Linux::set_signal_handler(int sig, bool set_installed) { 4246 // Check for overwrite. 4247 struct sigaction oldAct; 4248 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4249 4250 void* oldhand = oldAct.sa_sigaction 4251 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4252 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4253 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4254 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4255 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4256 if (AllowUserSignalHandlers || !set_installed) { 4257 // Do not overwrite; user takes responsibility to forward to us. 4258 return; 4259 } else if (UseSignalChaining) { 4260 // save the old handler in jvm 4261 save_preinstalled_handler(sig, oldAct); 4262 // libjsig also interposes the sigaction() call below and saves the 4263 // old sigaction on it own. 4264 } else { 4265 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4266 "%#lx for signal %d.", (long)oldhand, sig)); 4267 } 4268 } 4269 4270 struct sigaction sigAct; 4271 sigfillset(&(sigAct.sa_mask)); 4272 sigAct.sa_handler = SIG_DFL; 4273 if (!set_installed) { 4274 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4275 } else { 4276 sigAct.sa_sigaction = signalHandler; 4277 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4278 } 4279 // Save flags, which are set by ours 4280 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4281 sigflags[sig] = sigAct.sa_flags; 4282 4283 int ret = sigaction(sig, &sigAct, &oldAct); 4284 assert(ret == 0, "check"); 4285 4286 void* oldhand2 = oldAct.sa_sigaction 4287 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4288 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4289 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4290} 4291 4292// install signal handlers for signals that HotSpot needs to 4293// handle in order to support Java-level exception handling. 4294 4295void os::Linux::install_signal_handlers() { 4296 if (!signal_handlers_are_installed) { 4297 signal_handlers_are_installed = true; 4298 4299 // signal-chaining 4300 typedef void (*signal_setting_t)(); 4301 signal_setting_t begin_signal_setting = NULL; 4302 signal_setting_t end_signal_setting = NULL; 4303 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4304 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4305 if (begin_signal_setting != NULL) { 4306 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4307 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4308 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4309 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4310 libjsig_is_loaded = true; 4311 assert(UseSignalChaining, "should enable signal-chaining"); 4312 } 4313 if (libjsig_is_loaded) { 4314 // Tell libjsig jvm is setting signal handlers 4315 (*begin_signal_setting)(); 4316 } 4317 4318 set_signal_handler(SIGSEGV, true); 4319 set_signal_handler(SIGPIPE, true); 4320 set_signal_handler(SIGBUS, true); 4321 set_signal_handler(SIGILL, true); 4322 set_signal_handler(SIGFPE, true); 4323#if defined(PPC64) 4324 set_signal_handler(SIGTRAP, true); 4325#endif 4326 set_signal_handler(SIGXFSZ, true); 4327 4328 if (libjsig_is_loaded) { 4329 // Tell libjsig jvm finishes setting signal handlers 4330 (*end_signal_setting)(); 4331 } 4332 4333 // We don't activate signal checker if libjsig is in place, we trust ourselves 4334 // and if UserSignalHandler is installed all bets are off. 4335 // Log that signal checking is off only if -verbose:jni is specified. 4336 if (CheckJNICalls) { 4337 if (libjsig_is_loaded) { 4338 if (PrintJNIResolving) { 4339 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4340 } 4341 check_signals = false; 4342 } 4343 if (AllowUserSignalHandlers) { 4344 if (PrintJNIResolving) { 4345 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4346 } 4347 check_signals = false; 4348 } 4349 } 4350 } 4351} 4352 4353// This is the fastest way to get thread cpu time on Linux. 4354// Returns cpu time (user+sys) for any thread, not only for current. 4355// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4356// It might work on 2.6.10+ with a special kernel/glibc patch. 4357// For reference, please, see IEEE Std 1003.1-2004: 4358// http://www.unix.org/single_unix_specification 4359 4360jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4361 struct timespec tp; 4362 int rc = os::Linux::clock_gettime(clockid, &tp); 4363 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4364 4365 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4366} 4367 4368///// 4369// glibc on Linux platform uses non-documented flag 4370// to indicate, that some special sort of signal 4371// trampoline is used. 4372// We will never set this flag, and we should 4373// ignore this flag in our diagnostic 4374#ifdef SIGNIFICANT_SIGNAL_MASK 4375#undef SIGNIFICANT_SIGNAL_MASK 4376#endif 4377#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4378 4379static const char* get_signal_handler_name(address handler, 4380 char* buf, int buflen) { 4381 int offset; 4382 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4383 if (found) { 4384 // skip directory names 4385 const char *p1, *p2; 4386 p1 = buf; 4387 size_t len = strlen(os::file_separator()); 4388 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4389 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4390 } else { 4391 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4392 } 4393 return buf; 4394} 4395 4396static void print_signal_handler(outputStream* st, int sig, 4397 char* buf, size_t buflen) { 4398 struct sigaction sa; 4399 4400 sigaction(sig, NULL, &sa); 4401 4402 // See comment for SIGNIFICANT_SIGNAL_MASK define 4403 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4404 4405 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4406 4407 address handler = (sa.sa_flags & SA_SIGINFO) 4408 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4409 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4410 4411 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4412 st->print("SIG_DFL"); 4413 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4414 st->print("SIG_IGN"); 4415 } else { 4416 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4417 } 4418 4419 st->print(", sa_mask[0]="); 4420 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4421 4422 address rh = VMError::get_resetted_sighandler(sig); 4423 // May be, handler was resetted by VMError? 4424 if(rh != NULL) { 4425 handler = rh; 4426 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4427 } 4428 4429 st->print(", sa_flags="); 4430 os::Posix::print_sa_flags(st, sa.sa_flags); 4431 4432 // Check: is it our handler? 4433 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4434 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4435 // It is our signal handler 4436 // check for flags, reset system-used one! 4437 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4438 st->print( 4439 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4440 os::Linux::get_our_sigflags(sig)); 4441 } 4442 } 4443 st->cr(); 4444} 4445 4446 4447#define DO_SIGNAL_CHECK(sig) \ 4448 if (!sigismember(&check_signal_done, sig)) \ 4449 os::Linux::check_signal_handler(sig) 4450 4451// This method is a periodic task to check for misbehaving JNI applications 4452// under CheckJNI, we can add any periodic checks here 4453 4454void os::run_periodic_checks() { 4455 4456 if (check_signals == false) return; 4457 4458 // SEGV and BUS if overridden could potentially prevent 4459 // generation of hs*.log in the event of a crash, debugging 4460 // such a case can be very challenging, so we absolutely 4461 // check the following for a good measure: 4462 DO_SIGNAL_CHECK(SIGSEGV); 4463 DO_SIGNAL_CHECK(SIGILL); 4464 DO_SIGNAL_CHECK(SIGFPE); 4465 DO_SIGNAL_CHECK(SIGBUS); 4466 DO_SIGNAL_CHECK(SIGPIPE); 4467 DO_SIGNAL_CHECK(SIGXFSZ); 4468#if defined(PPC64) 4469 DO_SIGNAL_CHECK(SIGTRAP); 4470#endif 4471 4472 // ReduceSignalUsage allows the user to override these handlers 4473 // see comments at the very top and jvm_solaris.h 4474 if (!ReduceSignalUsage) { 4475 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4476 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4477 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4478 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4479 } 4480 4481 DO_SIGNAL_CHECK(SR_signum); 4482 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4483} 4484 4485typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4486 4487static os_sigaction_t os_sigaction = NULL; 4488 4489void os::Linux::check_signal_handler(int sig) { 4490 char buf[O_BUFLEN]; 4491 address jvmHandler = NULL; 4492 4493 4494 struct sigaction act; 4495 if (os_sigaction == NULL) { 4496 // only trust the default sigaction, in case it has been interposed 4497 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4498 if (os_sigaction == NULL) return; 4499 } 4500 4501 os_sigaction(sig, (struct sigaction*)NULL, &act); 4502 4503 4504 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4505 4506 address thisHandler = (act.sa_flags & SA_SIGINFO) 4507 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4508 : CAST_FROM_FN_PTR(address, act.sa_handler) ; 4509 4510 4511 switch(sig) { 4512 case SIGSEGV: 4513 case SIGBUS: 4514 case SIGFPE: 4515 case SIGPIPE: 4516 case SIGILL: 4517 case SIGXFSZ: 4518 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4519 break; 4520 4521 case SHUTDOWN1_SIGNAL: 4522 case SHUTDOWN2_SIGNAL: 4523 case SHUTDOWN3_SIGNAL: 4524 case BREAK_SIGNAL: 4525 jvmHandler = (address)user_handler(); 4526 break; 4527 4528 case INTERRUPT_SIGNAL: 4529 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4530 break; 4531 4532 default: 4533 if (sig == SR_signum) { 4534 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4535 } else { 4536 return; 4537 } 4538 break; 4539 } 4540 4541 if (thisHandler != jvmHandler) { 4542 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4543 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4544 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4545 // No need to check this sig any longer 4546 sigaddset(&check_signal_done, sig); 4547 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4548 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4549 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4550 exception_name(sig, buf, O_BUFLEN)); 4551 } 4552 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4553 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4554 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4555 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4556 // No need to check this sig any longer 4557 sigaddset(&check_signal_done, sig); 4558 } 4559 4560 // Dump all the signal 4561 if (sigismember(&check_signal_done, sig)) { 4562 print_signal_handlers(tty, buf, O_BUFLEN); 4563 } 4564} 4565 4566extern void report_error(char* file_name, int line_no, char* title, char* format, ...); 4567 4568extern bool signal_name(int signo, char* buf, size_t len); 4569 4570const char* os::exception_name(int exception_code, char* buf, size_t size) { 4571 if (0 < exception_code && exception_code <= SIGRTMAX) { 4572 // signal 4573 if (!signal_name(exception_code, buf, size)) { 4574 jio_snprintf(buf, size, "SIG%d", exception_code); 4575 } 4576 return buf; 4577 } else { 4578 return NULL; 4579 } 4580} 4581 4582// this is called _before_ the most of global arguments have been parsed 4583void os::init(void) { 4584 char dummy; /* used to get a guess on initial stack address */ 4585// first_hrtime = gethrtime(); 4586 4587 // With LinuxThreads the JavaMain thread pid (primordial thread) 4588 // is different than the pid of the java launcher thread. 4589 // So, on Linux, the launcher thread pid is passed to the VM 4590 // via the sun.java.launcher.pid property. 4591 // Use this property instead of getpid() if it was correctly passed. 4592 // See bug 6351349. 4593 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4594 4595 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4596 4597 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4598 4599 init_random(1234567); 4600 4601 ThreadCritical::initialize(); 4602 4603 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4604 if (Linux::page_size() == -1) { 4605 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4606 strerror(errno))); 4607 } 4608 init_page_sizes((size_t) Linux::page_size()); 4609 4610 Linux::initialize_system_info(); 4611 4612 // main_thread points to the aboriginal thread 4613 Linux::_main_thread = pthread_self(); 4614 4615 Linux::clock_init(); 4616 initial_time_count = javaTimeNanos(); 4617 4618 // pthread_condattr initialization for monotonic clock 4619 int status; 4620 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4621 if ((status = pthread_condattr_init(_condattr)) != 0) { 4622 fatal(err_msg("pthread_condattr_init: %s", strerror(status))); 4623 } 4624 // Only set the clock if CLOCK_MONOTONIC is available 4625 if (os::supports_monotonic_clock()) { 4626 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4627 if (status == EINVAL) { 4628 warning("Unable to use monotonic clock with relative timed-waits" \ 4629 " - changes to the time-of-day clock may have adverse affects"); 4630 } else { 4631 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status))); 4632 } 4633 } 4634 } 4635 // else it defaults to CLOCK_REALTIME 4636 4637 pthread_mutex_init(&dl_mutex, NULL); 4638 4639 // If the pagesize of the VM is greater than 8K determine the appropriate 4640 // number of initial guard pages. The user can change this with the 4641 // command line arguments, if needed. 4642 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4643 StackYellowPages = 1; 4644 StackRedPages = 1; 4645 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4646 } 4647} 4648 4649// To install functions for atexit system call 4650extern "C" { 4651 static void perfMemory_exit_helper() { 4652 perfMemory_exit(); 4653 } 4654} 4655 4656// this is called _after_ the global arguments have been parsed 4657jint os::init_2(void) 4658{ 4659 Linux::fast_thread_clock_init(); 4660 4661 // Allocate a single page and mark it as readable for safepoint polling 4662 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4663 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" ); 4664 4665 os::set_polling_page( polling_page ); 4666 4667#ifndef PRODUCT 4668 if(Verbose && PrintMiscellaneous) 4669 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page); 4670#endif 4671 4672 if (!UseMembar) { 4673 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4674 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4675 os::set_memory_serialize_page( mem_serialize_page ); 4676 4677#ifndef PRODUCT 4678 if(Verbose && PrintMiscellaneous) 4679 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page); 4680#endif 4681 } 4682 4683 // initialize suspend/resume support - must do this before signal_sets_init() 4684 if (SR_initialize() != 0) { 4685 perror("SR_initialize failed"); 4686 return JNI_ERR; 4687 } 4688 4689 Linux::signal_sets_init(); 4690 Linux::install_signal_handlers(); 4691 4692 // Check minimum allowable stack size for thread creation and to initialize 4693 // the java system classes, including StackOverflowError - depends on page 4694 // size. Add a page for compiler2 recursion in main thread. 4695 // Add in 2*BytesPerWord times page size to account for VM stack during 4696 // class initialization depending on 32 or 64 bit VM. 4697 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4698 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4699 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4700 4701 size_t threadStackSizeInBytes = ThreadStackSize * K; 4702 if (threadStackSizeInBytes != 0 && 4703 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4704 tty->print_cr("\nThe stack size specified is too small, " 4705 "Specify at least %dk", 4706 os::Linux::min_stack_allowed/ K); 4707 return JNI_ERR; 4708 } 4709 4710 // Make the stack size a multiple of the page size so that 4711 // the yellow/red zones can be guarded. 4712 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4713 vm_page_size())); 4714 4715 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4716 4717#if defined(IA32) 4718 workaround_expand_exec_shield_cs_limit(); 4719#endif 4720 4721 Linux::libpthread_init(); 4722 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4723 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4724 Linux::glibc_version(), Linux::libpthread_version(), 4725 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4726 } 4727 4728 if (UseNUMA) { 4729 if (!Linux::libnuma_init()) { 4730 UseNUMA = false; 4731 } else { 4732 if ((Linux::numa_max_node() < 1)) { 4733 // There's only one node(they start from 0), disable NUMA. 4734 UseNUMA = false; 4735 } 4736 } 4737 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4738 // we can make the adaptive lgrp chunk resizing work. If the user specified 4739 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4740 // disable adaptive resizing. 4741 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4742 if (FLAG_IS_DEFAULT(UseNUMA)) { 4743 UseNUMA = false; 4744 } else { 4745 if (FLAG_IS_DEFAULT(UseLargePages) && 4746 FLAG_IS_DEFAULT(UseSHM) && 4747 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4748 UseLargePages = false; 4749 } else { 4750 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing"); 4751 UseAdaptiveSizePolicy = false; 4752 UseAdaptiveNUMAChunkSizing = false; 4753 } 4754 } 4755 } 4756 if (!UseNUMA && ForceNUMA) { 4757 UseNUMA = true; 4758 } 4759 } 4760 4761 if (MaxFDLimit) { 4762 // set the number of file descriptors to max. print out error 4763 // if getrlimit/setrlimit fails but continue regardless. 4764 struct rlimit nbr_files; 4765 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4766 if (status != 0) { 4767 if (PrintMiscellaneous && (Verbose || WizardMode)) 4768 perror("os::init_2 getrlimit failed"); 4769 } else { 4770 nbr_files.rlim_cur = nbr_files.rlim_max; 4771 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4772 if (status != 0) { 4773 if (PrintMiscellaneous && (Verbose || WizardMode)) 4774 perror("os::init_2 setrlimit failed"); 4775 } 4776 } 4777 } 4778 4779 // Initialize lock used to serialize thread creation (see os::create_thread) 4780 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4781 4782 // at-exit methods are called in the reverse order of their registration. 4783 // atexit functions are called on return from main or as a result of a 4784 // call to exit(3C). There can be only 32 of these functions registered 4785 // and atexit() does not set errno. 4786 4787 if (PerfAllowAtExitRegistration) { 4788 // only register atexit functions if PerfAllowAtExitRegistration is set. 4789 // atexit functions can be delayed until process exit time, which 4790 // can be problematic for embedded VM situations. Embedded VMs should 4791 // call DestroyJavaVM() to assure that VM resources are released. 4792 4793 // note: perfMemory_exit_helper atexit function may be removed in 4794 // the future if the appropriate cleanup code can be added to the 4795 // VM_Exit VMOperation's doit method. 4796 if (atexit(perfMemory_exit_helper) != 0) { 4797 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4798 } 4799 } 4800 4801 // initialize thread priority policy 4802 prio_init(); 4803 4804 return JNI_OK; 4805} 4806 4807// this is called at the end of vm_initialization 4808void os::init_3(void) { 4809#ifdef JAVASE_EMBEDDED 4810 // Start the MemNotifyThread 4811 if (LowMemoryProtection) { 4812 MemNotifyThread::start(); 4813 } 4814 return; 4815#endif 4816} 4817 4818// Mark the polling page as unreadable 4819void os::make_polling_page_unreadable(void) { 4820 if( !guard_memory((char*)_polling_page, Linux::page_size()) ) 4821 fatal("Could not disable polling page"); 4822}; 4823 4824// Mark the polling page as readable 4825void os::make_polling_page_readable(void) { 4826 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4827 fatal("Could not enable polling page"); 4828 } 4829}; 4830 4831int os::active_processor_count() { 4832 // Linux doesn't yet have a (official) notion of processor sets, 4833 // so just return the number of online processors. 4834 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4835 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4836 return online_cpus; 4837} 4838 4839void os::set_native_thread_name(const char *name) { 4840 // Not yet implemented. 4841 return; 4842} 4843 4844bool os::distribute_processes(uint length, uint* distribution) { 4845 // Not yet implemented. 4846 return false; 4847} 4848 4849bool os::bind_to_processor(uint processor_id) { 4850 // Not yet implemented. 4851 return false; 4852} 4853 4854/// 4855 4856void os::SuspendedThreadTask::internal_do_task() { 4857 if (do_suspend(_thread->osthread())) { 4858 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4859 do_task(context); 4860 do_resume(_thread->osthread()); 4861 } 4862} 4863 4864class PcFetcher : public os::SuspendedThreadTask { 4865public: 4866 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4867 ExtendedPC result(); 4868protected: 4869 void do_task(const os::SuspendedThreadTaskContext& context); 4870private: 4871 ExtendedPC _epc; 4872}; 4873 4874ExtendedPC PcFetcher::result() { 4875 guarantee(is_done(), "task is not done yet."); 4876 return _epc; 4877} 4878 4879void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4880 Thread* thread = context.thread(); 4881 OSThread* osthread = thread->osthread(); 4882 if (osthread->ucontext() != NULL) { 4883 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4884 } else { 4885 // NULL context is unexpected, double-check this is the VMThread 4886 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4887 } 4888} 4889 4890// Suspends the target using the signal mechanism and then grabs the PC before 4891// resuming the target. Used by the flat-profiler only 4892ExtendedPC os::get_thread_pc(Thread* thread) { 4893 // Make sure that it is called by the watcher for the VMThread 4894 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4895 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4896 4897 PcFetcher fetcher(thread); 4898 fetcher.run(); 4899 return fetcher.result(); 4900} 4901 4902int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime) 4903{ 4904 if (is_NPTL()) { 4905 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4906 } else { 4907 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4908 // word back to default 64bit precision if condvar is signaled. Java 4909 // wants 53bit precision. Save and restore current value. 4910 int fpu = get_fpu_control_word(); 4911 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4912 set_fpu_control_word(fpu); 4913 return status; 4914 } 4915} 4916 4917//////////////////////////////////////////////////////////////////////////////// 4918// debug support 4919 4920bool os::find(address addr, outputStream* st) { 4921 Dl_info dlinfo; 4922 memset(&dlinfo, 0, sizeof(dlinfo)); 4923 if (dladdr(addr, &dlinfo) != 0) { 4924 st->print(PTR_FORMAT ": ", addr); 4925 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 4926 st->print("%s+%#x", dlinfo.dli_sname, 4927 addr - (intptr_t)dlinfo.dli_saddr); 4928 } else if (dlinfo.dli_fbase != NULL) { 4929 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4930 } else { 4931 st->print("<absolute address>"); 4932 } 4933 if (dlinfo.dli_fname != NULL) { 4934 st->print(" in %s", dlinfo.dli_fname); 4935 } 4936 if (dlinfo.dli_fbase != NULL) { 4937 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4938 } 4939 st->cr(); 4940 4941 if (Verbose) { 4942 // decode some bytes around the PC 4943 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 4944 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 4945 address lowest = (address) dlinfo.dli_sname; 4946 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4947 if (begin < lowest) begin = lowest; 4948 Dl_info dlinfo2; 4949 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 4950 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) 4951 end = (address) dlinfo2.dli_saddr; 4952 Disassembler::decode(begin, end, st); 4953 } 4954 return true; 4955 } 4956 return false; 4957} 4958 4959//////////////////////////////////////////////////////////////////////////////// 4960// misc 4961 4962// This does not do anything on Linux. This is basically a hook for being 4963// able to use structured exception handling (thread-local exception filters) 4964// on, e.g., Win32. 4965void 4966os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4967 JavaCallArguments* args, Thread* thread) { 4968 f(value, method, args, thread); 4969} 4970 4971void os::print_statistics() { 4972} 4973 4974int os::message_box(const char* title, const char* message) { 4975 int i; 4976 fdStream err(defaultStream::error_fd()); 4977 for (i = 0; i < 78; i++) err.print_raw("="); 4978 err.cr(); 4979 err.print_raw_cr(title); 4980 for (i = 0; i < 78; i++) err.print_raw("-"); 4981 err.cr(); 4982 err.print_raw_cr(message); 4983 for (i = 0; i < 78; i++) err.print_raw("="); 4984 err.cr(); 4985 4986 char buf[16]; 4987 // Prevent process from exiting upon "read error" without consuming all CPU 4988 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 4989 4990 return buf[0] == 'y' || buf[0] == 'Y'; 4991} 4992 4993int os::stat(const char *path, struct stat *sbuf) { 4994 char pathbuf[MAX_PATH]; 4995 if (strlen(path) > MAX_PATH - 1) { 4996 errno = ENAMETOOLONG; 4997 return -1; 4998 } 4999 os::native_path(strcpy(pathbuf, path)); 5000 return ::stat(pathbuf, sbuf); 5001} 5002 5003bool os::check_heap(bool force) { 5004 return true; 5005} 5006 5007int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) { 5008 return ::vsnprintf(buf, count, format, args); 5009} 5010 5011// Is a (classpath) directory empty? 5012bool os::dir_is_empty(const char* path) { 5013 DIR *dir = NULL; 5014 struct dirent *ptr; 5015 5016 dir = opendir(path); 5017 if (dir == NULL) return true; 5018 5019 /* Scan the directory */ 5020 bool result = true; 5021 char buf[sizeof(struct dirent) + MAX_PATH]; 5022 while (result && (ptr = ::readdir(dir)) != NULL) { 5023 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5024 result = false; 5025 } 5026 } 5027 closedir(dir); 5028 return result; 5029} 5030 5031// This code originates from JDK's sysOpen and open64_w 5032// from src/solaris/hpi/src/system_md.c 5033 5034#ifndef O_DELETE 5035#define O_DELETE 0x10000 5036#endif 5037 5038// Open a file. Unlink the file immediately after open returns 5039// if the specified oflag has the O_DELETE flag set. 5040// O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c 5041 5042int os::open(const char *path, int oflag, int mode) { 5043 5044 if (strlen(path) > MAX_PATH - 1) { 5045 errno = ENAMETOOLONG; 5046 return -1; 5047 } 5048 int fd; 5049 int o_delete = (oflag & O_DELETE); 5050 oflag = oflag & ~O_DELETE; 5051 5052 fd = ::open64(path, oflag, mode); 5053 if (fd == -1) return -1; 5054 5055 //If the open succeeded, the file might still be a directory 5056 { 5057 struct stat64 buf64; 5058 int ret = ::fstat64(fd, &buf64); 5059 int st_mode = buf64.st_mode; 5060 5061 if (ret != -1) { 5062 if ((st_mode & S_IFMT) == S_IFDIR) { 5063 errno = EISDIR; 5064 ::close(fd); 5065 return -1; 5066 } 5067 } else { 5068 ::close(fd); 5069 return -1; 5070 } 5071 } 5072 5073 /* 5074 * All file descriptors that are opened in the JVM and not 5075 * specifically destined for a subprocess should have the 5076 * close-on-exec flag set. If we don't set it, then careless 3rd 5077 * party native code might fork and exec without closing all 5078 * appropriate file descriptors (e.g. as we do in closeDescriptors in 5079 * UNIXProcess.c), and this in turn might: 5080 * 5081 * - cause end-of-file to fail to be detected on some file 5082 * descriptors, resulting in mysterious hangs, or 5083 * 5084 * - might cause an fopen in the subprocess to fail on a system 5085 * suffering from bug 1085341. 5086 * 5087 * (Yes, the default setting of the close-on-exec flag is a Unix 5088 * design flaw) 5089 * 5090 * See: 5091 * 1085341: 32-bit stdio routines should support file descriptors >255 5092 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5093 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5094 */ 5095#ifdef FD_CLOEXEC 5096 { 5097 int flags = ::fcntl(fd, F_GETFD); 5098 if (flags != -1) 5099 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5100 } 5101#endif 5102 5103 if (o_delete != 0) { 5104 ::unlink(path); 5105 } 5106 return fd; 5107} 5108 5109 5110// create binary file, rewriting existing file if required 5111int os::create_binary_file(const char* path, bool rewrite_existing) { 5112 int oflags = O_WRONLY | O_CREAT; 5113 if (!rewrite_existing) { 5114 oflags |= O_EXCL; 5115 } 5116 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5117} 5118 5119// return current position of file pointer 5120jlong os::current_file_offset(int fd) { 5121 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5122} 5123 5124// move file pointer to the specified offset 5125jlong os::seek_to_file_offset(int fd, jlong offset) { 5126 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5127} 5128 5129// This code originates from JDK's sysAvailable 5130// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5131 5132int os::available(int fd, jlong *bytes) { 5133 jlong cur, end; 5134 int mode; 5135 struct stat64 buf64; 5136 5137 if (::fstat64(fd, &buf64) >= 0) { 5138 mode = buf64.st_mode; 5139 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5140 /* 5141 * XXX: is the following call interruptible? If so, this might 5142 * need to go through the INTERRUPT_IO() wrapper as for other 5143 * blocking, interruptible calls in this file. 5144 */ 5145 int n; 5146 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5147 *bytes = n; 5148 return 1; 5149 } 5150 } 5151 } 5152 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5153 return 0; 5154 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5155 return 0; 5156 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5157 return 0; 5158 } 5159 *bytes = end - cur; 5160 return 1; 5161} 5162 5163int os::socket_available(int fd, jint *pbytes) { 5164 // Linux doc says EINTR not returned, unlike Solaris 5165 int ret = ::ioctl(fd, FIONREAD, pbytes); 5166 5167 //%% note ioctl can return 0 when successful, JVM_SocketAvailable 5168 // is expected to return 0 on failure and 1 on success to the jdk. 5169 return (ret < 0) ? 0 : 1; 5170} 5171 5172// Map a block of memory. 5173char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5174 char *addr, size_t bytes, bool read_only, 5175 bool allow_exec) { 5176 int prot; 5177 int flags = MAP_PRIVATE; 5178 5179 if (read_only) { 5180 prot = PROT_READ; 5181 } else { 5182 prot = PROT_READ | PROT_WRITE; 5183 } 5184 5185 if (allow_exec) { 5186 prot |= PROT_EXEC; 5187 } 5188 5189 if (addr != NULL) { 5190 flags |= MAP_FIXED; 5191 } 5192 5193 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5194 fd, file_offset); 5195 if (mapped_address == MAP_FAILED) { 5196 return NULL; 5197 } 5198 return mapped_address; 5199} 5200 5201 5202// Remap a block of memory. 5203char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5204 char *addr, size_t bytes, bool read_only, 5205 bool allow_exec) { 5206 // same as map_memory() on this OS 5207 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5208 allow_exec); 5209} 5210 5211 5212// Unmap a block of memory. 5213bool os::pd_unmap_memory(char* addr, size_t bytes) { 5214 return munmap(addr, bytes) == 0; 5215} 5216 5217static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5218 5219static clockid_t thread_cpu_clockid(Thread* thread) { 5220 pthread_t tid = thread->osthread()->pthread_id(); 5221 clockid_t clockid; 5222 5223 // Get thread clockid 5224 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5225 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5226 return clockid; 5227} 5228 5229// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5230// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5231// of a thread. 5232// 5233// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5234// the fast estimate available on the platform. 5235 5236jlong os::current_thread_cpu_time() { 5237 if (os::Linux::supports_fast_thread_cpu_time()) { 5238 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5239 } else { 5240 // return user + sys since the cost is the same 5241 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5242 } 5243} 5244 5245jlong os::thread_cpu_time(Thread* thread) { 5246 // consistent with what current_thread_cpu_time() returns 5247 if (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, true /* user + sys */); 5251 } 5252} 5253 5254jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5255 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5256 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5257 } else { 5258 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5259 } 5260} 5261 5262jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5263 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5264 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5265 } else { 5266 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5267 } 5268} 5269 5270// 5271// -1 on error. 5272// 5273 5274static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5275 static bool proc_task_unchecked = true; 5276 pid_t tid = thread->osthread()->thread_id(); 5277 char *s; 5278 char stat[2048]; 5279 int statlen; 5280 char proc_name[64]; 5281 int count; 5282 long sys_time, user_time; 5283 char cdummy; 5284 int idummy; 5285 long ldummy; 5286 FILE *fp; 5287 5288 snprintf(proc_name, 64, "/proc/%d/stat", tid); 5289 5290 // The /proc/<tid>/stat aggregates per-process usage on 5291 // new Linux kernels 2.6+ where NPTL is supported. 5292 // The /proc/self/task/<tid>/stat still has the per-thread usage. 5293 // See bug 6328462. 5294 // There possibly can be cases where there is no directory 5295 // /proc/self/task, so we check its availability. 5296 if (proc_task_unchecked && os::Linux::is_NPTL()) { 5297 // This is executed only once 5298 proc_task_unchecked = false; 5299 fp = fopen("/proc/self/task", "r"); 5300 if (fp != NULL) { 5301 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5302 fclose(fp); 5303 } 5304 } 5305 5306 fp = fopen(proc_name, "r"); 5307 if ( fp == NULL ) return -1; 5308 statlen = fread(stat, 1, 2047, fp); 5309 stat[statlen] = '\0'; 5310 fclose(fp); 5311 5312 // Skip pid and the command string. Note that we could be dealing with 5313 // weird command names, e.g. user could decide to rename java launcher 5314 // to "java 1.4.2 :)", then the stat file would look like 5315 // 1234 (java 1.4.2 :)) R ... ... 5316 // We don't really need to know the command string, just find the last 5317 // occurrence of ")" and then start parsing from there. See bug 4726580. 5318 s = strrchr(stat, ')'); 5319 if (s == NULL ) return -1; 5320 5321 // Skip blank chars 5322 do s++; while (isspace(*s)); 5323 5324 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5325 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5326 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5327 &user_time, &sys_time); 5328 if ( count != 13 ) return -1; 5329 if (user_sys_cpu_time) { 5330 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5331 } else { 5332 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5333 } 5334} 5335 5336void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5337 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5338 info_ptr->may_skip_backward = false; // elapsed time not wall time 5339 info_ptr->may_skip_forward = false; // elapsed time not wall time 5340 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5341} 5342 5343void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5344 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5345 info_ptr->may_skip_backward = false; // elapsed time not wall time 5346 info_ptr->may_skip_forward = false; // elapsed time not wall time 5347 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5348} 5349 5350bool os::is_thread_cpu_time_supported() { 5351 return true; 5352} 5353 5354// System loadavg support. Returns -1 if load average cannot be obtained. 5355// Linux doesn't yet have a (official) notion of processor sets, 5356// so just return the system wide load average. 5357int os::loadavg(double loadavg[], int nelem) { 5358 return ::getloadavg(loadavg, nelem); 5359} 5360 5361void os::pause() { 5362 char filename[MAX_PATH]; 5363 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5364 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5365 } else { 5366 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5367 } 5368 5369 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5370 if (fd != -1) { 5371 struct stat buf; 5372 ::close(fd); 5373 while (::stat(filename, &buf) == 0) { 5374 (void)::poll(NULL, 0, 100); 5375 } 5376 } else { 5377 jio_fprintf(stderr, 5378 "Could not open pause file '%s', continuing immediately.\n", filename); 5379 } 5380} 5381 5382 5383// Refer to the comments in os_solaris.cpp park-unpark. 5384// 5385// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5386// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5387// For specifics regarding the bug see GLIBC BUGID 261237 : 5388// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5389// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5390// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5391// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5392// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5393// and monitorenter when we're using 1-0 locking. All those operations may result in 5394// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5395// of libpthread avoids the problem, but isn't practical. 5396// 5397// Possible remedies: 5398// 5399// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5400// This is palliative and probabilistic, however. If the thread is preempted 5401// between the call to compute_abstime() and pthread_cond_timedwait(), more 5402// than the minimum period may have passed, and the abstime may be stale (in the 5403// past) resultin in a hang. Using this technique reduces the odds of a hang 5404// but the JVM is still vulnerable, particularly on heavily loaded systems. 5405// 5406// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5407// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5408// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5409// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5410// thread. 5411// 5412// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5413// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5414// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5415// This also works well. In fact it avoids kernel-level scalability impediments 5416// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5417// timers in a graceful fashion. 5418// 5419// 4. When the abstime value is in the past it appears that control returns 5420// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5421// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5422// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5423// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5424// It may be possible to avoid reinitialization by checking the return 5425// value from pthread_cond_timedwait(). In addition to reinitializing the 5426// condvar we must establish the invariant that cond_signal() is only called 5427// within critical sections protected by the adjunct mutex. This prevents 5428// cond_signal() from "seeing" a condvar that's in the midst of being 5429// reinitialized or that is corrupt. Sadly, this invariant obviates the 5430// desirable signal-after-unlock optimization that avoids futile context switching. 5431// 5432// I'm also concerned that some versions of NTPL might allocate an auxilliary 5433// structure when a condvar is used or initialized. cond_destroy() would 5434// release the helper structure. Our reinitialize-after-timedwait fix 5435// put excessive stress on malloc/free and locks protecting the c-heap. 5436// 5437// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5438// It may be possible to refine (4) by checking the kernel and NTPL verisons 5439// and only enabling the work-around for vulnerable environments. 5440 5441// utility to compute the abstime argument to timedwait: 5442// millis is the relative timeout time 5443// abstime will be the absolute timeout time 5444// TODO: replace compute_abstime() with unpackTime() 5445 5446static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5447 if (millis < 0) millis = 0; 5448 5449 jlong seconds = millis / 1000; 5450 millis %= 1000; 5451 if (seconds > 50000000) { // see man cond_timedwait(3T) 5452 seconds = 50000000; 5453 } 5454 5455 if (os::supports_monotonic_clock()) { 5456 struct timespec now; 5457 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5458 assert_status(status == 0, status, "clock_gettime"); 5459 abstime->tv_sec = now.tv_sec + seconds; 5460 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5461 if (nanos >= NANOSECS_PER_SEC) { 5462 abstime->tv_sec += 1; 5463 nanos -= NANOSECS_PER_SEC; 5464 } 5465 abstime->tv_nsec = nanos; 5466 } else { 5467 struct timeval now; 5468 int status = gettimeofday(&now, NULL); 5469 assert(status == 0, "gettimeofday"); 5470 abstime->tv_sec = now.tv_sec + seconds; 5471 long usec = now.tv_usec + millis * 1000; 5472 if (usec >= 1000000) { 5473 abstime->tv_sec += 1; 5474 usec -= 1000000; 5475 } 5476 abstime->tv_nsec = usec * 1000; 5477 } 5478 return abstime; 5479} 5480 5481 5482// Test-and-clear _Event, always leaves _Event set to 0, returns immediately. 5483// Conceptually TryPark() should be equivalent to park(0). 5484 5485int os::PlatformEvent::TryPark() { 5486 for (;;) { 5487 const int v = _Event ; 5488 guarantee ((v == 0) || (v == 1), "invariant") ; 5489 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ; 5490 } 5491} 5492 5493void os::PlatformEvent::park() { // AKA "down()" 5494 // Invariant: Only the thread associated with the Event/PlatformEvent 5495 // may call park(). 5496 // TODO: assert that _Assoc != NULL or _Assoc == Self 5497 int v ; 5498 for (;;) { 5499 v = _Event ; 5500 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5501 } 5502 guarantee (v >= 0, "invariant") ; 5503 if (v == 0) { 5504 // Do this the hard way by blocking ... 5505 int status = pthread_mutex_lock(_mutex); 5506 assert_status(status == 0, status, "mutex_lock"); 5507 guarantee (_nParked == 0, "invariant") ; 5508 ++ _nParked ; 5509 while (_Event < 0) { 5510 status = pthread_cond_wait(_cond, _mutex); 5511 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5512 // Treat this the same as if the wait was interrupted 5513 if (status == ETIME) { status = EINTR; } 5514 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5515 } 5516 -- _nParked ; 5517 5518 _Event = 0 ; 5519 status = pthread_mutex_unlock(_mutex); 5520 assert_status(status == 0, status, "mutex_unlock"); 5521 // Paranoia to ensure our locked and lock-free paths interact 5522 // correctly with each other. 5523 OrderAccess::fence(); 5524 } 5525 guarantee (_Event >= 0, "invariant") ; 5526} 5527 5528int os::PlatformEvent::park(jlong millis) { 5529 guarantee (_nParked == 0, "invariant") ; 5530 5531 int v ; 5532 for (;;) { 5533 v = _Event ; 5534 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ; 5535 } 5536 guarantee (v >= 0, "invariant") ; 5537 if (v != 0) return OS_OK ; 5538 5539 // We do this the hard way, by blocking the thread. 5540 // Consider enforcing a minimum timeout value. 5541 struct timespec abst; 5542 compute_abstime(&abst, millis); 5543 5544 int ret = OS_TIMEOUT; 5545 int status = pthread_mutex_lock(_mutex); 5546 assert_status(status == 0, status, "mutex_lock"); 5547 guarantee (_nParked == 0, "invariant") ; 5548 ++_nParked ; 5549 5550 // Object.wait(timo) will return because of 5551 // (a) notification 5552 // (b) timeout 5553 // (c) thread.interrupt 5554 // 5555 // Thread.interrupt and object.notify{All} both call Event::set. 5556 // That is, we treat thread.interrupt as a special case of notification. 5557 // The underlying Solaris implementation, cond_timedwait, admits 5558 // spurious/premature wakeups, but the JLS/JVM spec prevents the 5559 // JVM from making those visible to Java code. As such, we must 5560 // filter out spurious wakeups. We assume all ETIME returns are valid. 5561 // 5562 // TODO: properly differentiate simultaneous notify+interrupt. 5563 // In that case, we should propagate the notify to another waiter. 5564 5565 while (_Event < 0) { 5566 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5567 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5568 pthread_cond_destroy (_cond); 5569 pthread_cond_init (_cond, os::Linux::condAttr()) ; 5570 } 5571 assert_status(status == 0 || status == EINTR || 5572 status == ETIME || status == ETIMEDOUT, 5573 status, "cond_timedwait"); 5574 if (!FilterSpuriousWakeups) break ; // previous semantics 5575 if (status == ETIME || status == ETIMEDOUT) break ; 5576 // We consume and ignore EINTR and spurious wakeups. 5577 } 5578 --_nParked ; 5579 if (_Event >= 0) { 5580 ret = OS_OK; 5581 } 5582 _Event = 0 ; 5583 status = pthread_mutex_unlock(_mutex); 5584 assert_status(status == 0, status, "mutex_unlock"); 5585 assert (_nParked == 0, "invariant") ; 5586 // Paranoia to ensure our locked and lock-free paths interact 5587 // correctly with each other. 5588 OrderAccess::fence(); 5589 return ret; 5590} 5591 5592void os::PlatformEvent::unpark() { 5593 // Transitions for _Event: 5594 // 0 :=> 1 5595 // 1 :=> 1 5596 // -1 :=> either 0 or 1; must signal target thread 5597 // That is, we can safely transition _Event from -1 to either 5598 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back 5599 // unpark() calls. 5600 // See also: "Semaphores in Plan 9" by Mullender & Cox 5601 // 5602 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5603 // that it will take two back-to-back park() calls for the owning 5604 // thread to block. This has the benefit of forcing a spurious return 5605 // from the first park() call after an unpark() call which will help 5606 // shake out uses of park() and unpark() without condition variables. 5607 5608 if (Atomic::xchg(1, &_Event) >= 0) return; 5609 5610 // Wait for the thread associated with the event to vacate 5611 int status = pthread_mutex_lock(_mutex); 5612 assert_status(status == 0, status, "mutex_lock"); 5613 int AnyWaiters = _nParked; 5614 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5615 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5616 AnyWaiters = 0; 5617 pthread_cond_signal(_cond); 5618 } 5619 status = pthread_mutex_unlock(_mutex); 5620 assert_status(status == 0, status, "mutex_unlock"); 5621 if (AnyWaiters != 0) { 5622 status = pthread_cond_signal(_cond); 5623 assert_status(status == 0, status, "cond_signal"); 5624 } 5625 5626 // Note that we signal() _after dropping the lock for "immortal" Events. 5627 // This is safe and avoids a common class of futile wakeups. In rare 5628 // circumstances this can cause a thread to return prematurely from 5629 // cond_{timed}wait() but the spurious wakeup is benign and the victim will 5630 // simply re-test the condition and re-park itself. 5631} 5632 5633 5634// JSR166 5635// ------------------------------------------------------- 5636 5637/* 5638 * The solaris and linux implementations of park/unpark are fairly 5639 * conservative for now, but can be improved. They currently use a 5640 * mutex/condvar pair, plus a a count. 5641 * Park decrements count if > 0, else does a condvar wait. Unpark 5642 * sets count to 1 and signals condvar. Only one thread ever waits 5643 * on the condvar. Contention seen when trying to park implies that someone 5644 * is unparking you, so don't wait. And spurious returns are fine, so there 5645 * is no need to track notifications. 5646 */ 5647 5648/* 5649 * This code is common to linux and solaris and will be moved to a 5650 * common place in dolphin. 5651 * 5652 * The passed in time value is either a relative time in nanoseconds 5653 * or an absolute time in milliseconds. Either way it has to be unpacked 5654 * into suitable seconds and nanoseconds components and stored in the 5655 * given timespec structure. 5656 * Given time is a 64-bit value and the time_t used in the timespec is only 5657 * a signed-32-bit value (except on 64-bit Linux) we have to watch for 5658 * overflow if times way in the future are given. Further on Solaris versions 5659 * prior to 10 there is a restriction (see cond_timedwait) that the specified 5660 * number of seconds, in abstime, is less than current_time + 100,000,000. 5661 * As it will be 28 years before "now + 100000000" will overflow we can 5662 * ignore overflow and just impose a hard-limit on seconds using the value 5663 * of "now + 100,000,000". This places a limit on the timeout of about 3.17 5664 * years from "now". 5665 */ 5666 5667static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5668 assert (time > 0, "convertTime"); 5669 time_t max_secs = 0; 5670 5671 if (!os::supports_monotonic_clock() || isAbsolute) { 5672 struct timeval now; 5673 int status = gettimeofday(&now, NULL); 5674 assert(status == 0, "gettimeofday"); 5675 5676 max_secs = now.tv_sec + MAX_SECS; 5677 5678 if (isAbsolute) { 5679 jlong secs = time / 1000; 5680 if (secs > max_secs) { 5681 absTime->tv_sec = max_secs; 5682 } else { 5683 absTime->tv_sec = secs; 5684 } 5685 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5686 } else { 5687 jlong secs = time / NANOSECS_PER_SEC; 5688 if (secs >= MAX_SECS) { 5689 absTime->tv_sec = max_secs; 5690 absTime->tv_nsec = 0; 5691 } else { 5692 absTime->tv_sec = now.tv_sec + secs; 5693 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5694 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5695 absTime->tv_nsec -= NANOSECS_PER_SEC; 5696 ++absTime->tv_sec; // note: this must be <= max_secs 5697 } 5698 } 5699 } 5700 } else { 5701 // must be relative using monotonic clock 5702 struct timespec now; 5703 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5704 assert_status(status == 0, status, "clock_gettime"); 5705 max_secs = now.tv_sec + MAX_SECS; 5706 jlong secs = time / NANOSECS_PER_SEC; 5707 if (secs >= MAX_SECS) { 5708 absTime->tv_sec = max_secs; 5709 absTime->tv_nsec = 0; 5710 } else { 5711 absTime->tv_sec = now.tv_sec + secs; 5712 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5713 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5714 absTime->tv_nsec -= NANOSECS_PER_SEC; 5715 ++absTime->tv_sec; // note: this must be <= max_secs 5716 } 5717 } 5718 } 5719 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5720 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5721 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5722 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5723} 5724 5725void Parker::park(bool isAbsolute, jlong time) { 5726 // Ideally we'd do something useful while spinning, such 5727 // as calling unpackTime(). 5728 5729 // Optional fast-path check: 5730 // Return immediately if a permit is available. 5731 // We depend on Atomic::xchg() having full barrier semantics 5732 // since we are doing a lock-free update to _counter. 5733 if (Atomic::xchg(0, &_counter) > 0) return; 5734 5735 Thread* thread = Thread::current(); 5736 assert(thread->is_Java_thread(), "Must be JavaThread"); 5737 JavaThread *jt = (JavaThread *)thread; 5738 5739 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5740 // Check interrupt before trying to wait 5741 if (Thread::is_interrupted(thread, false)) { 5742 return; 5743 } 5744 5745 // Next, demultiplex/decode time arguments 5746 timespec absTime; 5747 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all 5748 return; 5749 } 5750 if (time > 0) { 5751 unpackTime(&absTime, isAbsolute, time); 5752 } 5753 5754 5755 // Enter safepoint region 5756 // Beware of deadlocks such as 6317397. 5757 // The per-thread Parker:: mutex is a classic leaf-lock. 5758 // In particular a thread must never block on the Threads_lock while 5759 // holding the Parker:: mutex. If safepoints are pending both the 5760 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5761 ThreadBlockInVM tbivm(jt); 5762 5763 // Don't wait if cannot get lock since interference arises from 5764 // unblocking. Also. check interrupt before trying wait 5765 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5766 return; 5767 } 5768 5769 int status ; 5770 if (_counter > 0) { // no wait needed 5771 _counter = 0; 5772 status = pthread_mutex_unlock(_mutex); 5773 assert (status == 0, "invariant") ; 5774 // Paranoia to ensure our locked and lock-free paths interact 5775 // correctly with each other and Java-level accesses. 5776 OrderAccess::fence(); 5777 return; 5778 } 5779 5780#ifdef ASSERT 5781 // Don't catch signals while blocked; let the running threads have the signals. 5782 // (This allows a debugger to break into the running thread.) 5783 sigset_t oldsigs; 5784 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5785 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5786#endif 5787 5788 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5789 jt->set_suspend_equivalent(); 5790 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5791 5792 assert(_cur_index == -1, "invariant"); 5793 if (time == 0) { 5794 _cur_index = REL_INDEX; // arbitrary choice when not timed 5795 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ; 5796 } else { 5797 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5798 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ; 5799 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5800 pthread_cond_destroy (&_cond[_cur_index]) ; 5801 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr()); 5802 } 5803 } 5804 _cur_index = -1; 5805 assert_status(status == 0 || status == EINTR || 5806 status == ETIME || status == ETIMEDOUT, 5807 status, "cond_timedwait"); 5808 5809#ifdef ASSERT 5810 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5811#endif 5812 5813 _counter = 0 ; 5814 status = pthread_mutex_unlock(_mutex) ; 5815 assert_status(status == 0, status, "invariant") ; 5816 // Paranoia to ensure our locked and lock-free paths interact 5817 // correctly with each other and Java-level accesses. 5818 OrderAccess::fence(); 5819 5820 // If externally suspended while waiting, re-suspend 5821 if (jt->handle_special_suspend_equivalent_condition()) { 5822 jt->java_suspend_self(); 5823 } 5824} 5825 5826void Parker::unpark() { 5827 int s, status ; 5828 status = pthread_mutex_lock(_mutex); 5829 assert (status == 0, "invariant") ; 5830 s = _counter; 5831 _counter = 1; 5832 if (s < 1) { 5833 // thread might be parked 5834 if (_cur_index != -1) { 5835 // thread is definitely parked 5836 if (WorkAroundNPTLTimedWaitHang) { 5837 status = pthread_cond_signal (&_cond[_cur_index]); 5838 assert (status == 0, "invariant"); 5839 status = pthread_mutex_unlock(_mutex); 5840 assert (status == 0, "invariant"); 5841 } else { 5842 status = pthread_mutex_unlock(_mutex); 5843 assert (status == 0, "invariant"); 5844 status = pthread_cond_signal (&_cond[_cur_index]); 5845 assert (status == 0, "invariant"); 5846 } 5847 } else { 5848 pthread_mutex_unlock(_mutex); 5849 assert (status == 0, "invariant") ; 5850 } 5851 } else { 5852 pthread_mutex_unlock(_mutex); 5853 assert (status == 0, "invariant") ; 5854 } 5855} 5856 5857 5858extern char** environ; 5859 5860#ifndef __NR_fork 5861#define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) 5862#endif 5863 5864#ifndef __NR_execve 5865#define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) 5866#endif 5867 5868// Run the specified command in a separate process. Return its exit value, 5869// or -1 on failure (e.g. can't fork a new process). 5870// Unlike system(), this function can be called from signal handler. It 5871// doesn't block SIGINT et al. 5872int os::fork_and_exec(char* cmd) { 5873 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5874 5875 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5876 // pthread_atfork handlers and reset pthread library. All we need is a 5877 // separate process to execve. Make a direct syscall to fork process. 5878 // On IA64 there's no fork syscall, we have to use fork() and hope for 5879 // the best... 5880 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5881 IA64_ONLY(fork();) 5882 5883 if (pid < 0) { 5884 // fork failed 5885 return -1; 5886 5887 } else if (pid == 0) { 5888 // child process 5889 5890 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5891 // first to kill every thread on the thread list. Because this list is 5892 // not reset by fork() (see notes above), execve() will instead kill 5893 // every thread in the parent process. We know this is the only thread 5894 // in the new process, so make a system call directly. 5895 // IA64 should use normal execve() from glibc to match the glibc fork() 5896 // above. 5897 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5898 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5899 5900 // execve failed 5901 _exit(-1); 5902 5903 } else { 5904 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5905 // care about the actual exit code, for now. 5906 5907 int status; 5908 5909 // Wait for the child process to exit. This returns immediately if 5910 // the child has already exited. */ 5911 while (waitpid(pid, &status, 0) < 0) { 5912 switch (errno) { 5913 case ECHILD: return 0; 5914 case EINTR: break; 5915 default: return -1; 5916 } 5917 } 5918 5919 if (WIFEXITED(status)) { 5920 // The child exited normally; get its exit code. 5921 return WEXITSTATUS(status); 5922 } else if (WIFSIGNALED(status)) { 5923 // The child exited because of a signal 5924 // The best value to return is 0x80 + signal number, 5925 // because that is what all Unix shells do, and because 5926 // it allows callers to distinguish between process exit and 5927 // process death by signal. 5928 return 0x80 + WTERMSIG(status); 5929 } else { 5930 // Unknown exit code; pass it through 5931 return status; 5932 } 5933 } 5934} 5935 5936// is_headless_jre() 5937// 5938// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5939// in order to report if we are running in a headless jre 5940// 5941// Since JDK8 xawt/libmawt.so was moved into the same directory 5942// as libawt.so, and renamed libawt_xawt.so 5943// 5944bool os::is_headless_jre() { 5945 struct stat statbuf; 5946 char buf[MAXPATHLEN]; 5947 char libmawtpath[MAXPATHLEN]; 5948 const char *xawtstr = "/xawt/libmawt.so"; 5949 const char *new_xawtstr = "/libawt_xawt.so"; 5950 char *p; 5951 5952 // Get path to libjvm.so 5953 os::jvm_path(buf, sizeof(buf)); 5954 5955 // Get rid of libjvm.so 5956 p = strrchr(buf, '/'); 5957 if (p == NULL) return false; 5958 else *p = '\0'; 5959 5960 // Get rid of client or server 5961 p = strrchr(buf, '/'); 5962 if (p == NULL) return false; 5963 else *p = '\0'; 5964 5965 // check xawt/libmawt.so 5966 strcpy(libmawtpath, buf); 5967 strcat(libmawtpath, xawtstr); 5968 if (::stat(libmawtpath, &statbuf) == 0) return false; 5969 5970 // check libawt_xawt.so 5971 strcpy(libmawtpath, buf); 5972 strcat(libmawtpath, new_xawtstr); 5973 if (::stat(libmawtpath, &statbuf) == 0) return false; 5974 5975 return true; 5976} 5977 5978// Get the default path to the core file 5979// Returns the length of the string 5980int os::get_core_path(char* buffer, size_t bufferSize) { 5981 const char* p = get_current_directory(buffer, bufferSize); 5982 5983 if (p == NULL) { 5984 assert(p != NULL, "failed to get current directory"); 5985 return 0; 5986 } 5987 5988 return strlen(buffer); 5989} 5990 5991#ifdef JAVASE_EMBEDDED 5992// 5993// A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory. 5994// 5995MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL; 5996 5997// ctor 5998// 5999MemNotifyThread::MemNotifyThread(int fd): Thread() { 6000 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread"); 6001 _fd = fd; 6002 6003 if (os::create_thread(this, os::os_thread)) { 6004 _memnotify_thread = this; 6005 os::set_priority(this, NearMaxPriority); 6006 os::start_thread(this); 6007 } 6008} 6009 6010// Where all the work gets done 6011// 6012void MemNotifyThread::run() { 6013 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread"); 6014 6015 // Set up the select arguments 6016 fd_set rfds; 6017 if (_fd != -1) { 6018 FD_ZERO(&rfds); 6019 FD_SET(_fd, &rfds); 6020 } 6021 6022 // Now wait for the mem_notify device to wake up 6023 while (1) { 6024 // Wait for the mem_notify device to signal us.. 6025 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL); 6026 if (rc == -1) { 6027 perror("select!\n"); 6028 break; 6029 } else if (rc) { 6030 //ssize_t free_before = os::available_memory(); 6031 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024); 6032 6033 // The kernel is telling us there is not much memory left... 6034 // try to do something about that 6035 6036 // If we are not already in a GC, try one. 6037 if (!Universe::heap()->is_gc_active()) { 6038 Universe::heap()->collect(GCCause::_allocation_failure); 6039 6040 //ssize_t free_after = os::available_memory(); 6041 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024); 6042 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024); 6043 } 6044 // We might want to do something like the following if we find the GC's are not helping... 6045 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true); 6046 } 6047 } 6048} 6049 6050// 6051// See if the /dev/mem_notify device exists, and if so, start a thread to monitor it. 6052// 6053void MemNotifyThread::start() { 6054 int fd; 6055 fd = open ("/dev/mem_notify", O_RDONLY, 0); 6056 if (fd < 0) { 6057 return; 6058 } 6059 6060 if (memnotify_thread() == NULL) { 6061 new MemNotifyThread(fd); 6062 } 6063} 6064 6065#endif // JAVASE_EMBEDDED 6066 6067 6068/////////////// Unit tests /////////////// 6069 6070#ifndef PRODUCT 6071 6072#define test_log(...) \ 6073 do {\ 6074 if (VerboseInternalVMTests) { \ 6075 tty->print_cr(__VA_ARGS__); \ 6076 tty->flush(); \ 6077 }\ 6078 } while (false) 6079 6080class TestReserveMemorySpecial : AllStatic { 6081 public: 6082 static void small_page_write(void* addr, size_t size) { 6083 size_t page_size = os::vm_page_size(); 6084 6085 char* end = (char*)addr + size; 6086 for (char* p = (char*)addr; p < end; p += page_size) { 6087 *p = 1; 6088 } 6089 } 6090 6091 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6092 if (!UseHugeTLBFS) { 6093 return; 6094 } 6095 6096 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6097 6098 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6099 6100 if (addr != NULL) { 6101 small_page_write(addr, size); 6102 6103 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6104 } 6105 } 6106 6107 static void test_reserve_memory_special_huge_tlbfs_only() { 6108 if (!UseHugeTLBFS) { 6109 return; 6110 } 6111 6112 size_t lp = os::large_page_size(); 6113 6114 for (size_t size = lp; size <= lp * 10; size += lp) { 6115 test_reserve_memory_special_huge_tlbfs_only(size); 6116 } 6117 } 6118 6119 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) { 6120 if (!UseHugeTLBFS) { 6121 return; 6122 } 6123 6124 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")", 6125 size, alignment); 6126 6127 assert(size >= os::large_page_size(), "Incorrect input to test"); 6128 6129 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6130 6131 if (addr != NULL) { 6132 small_page_write(addr, size); 6133 6134 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6135 } 6136 } 6137 6138 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) { 6139 size_t lp = os::large_page_size(); 6140 size_t ag = os::vm_allocation_granularity(); 6141 6142 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6143 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment); 6144 } 6145 } 6146 6147 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6148 size_t lp = os::large_page_size(); 6149 size_t ag = os::vm_allocation_granularity(); 6150 6151 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp); 6152 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag); 6153 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2); 6154 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2); 6155 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag); 6156 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag); 6157 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2); 6158 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10); 6159 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2); 6160 } 6161 6162 static void test_reserve_memory_special_huge_tlbfs() { 6163 if (!UseHugeTLBFS) { 6164 return; 6165 } 6166 6167 test_reserve_memory_special_huge_tlbfs_only(); 6168 test_reserve_memory_special_huge_tlbfs_mixed(); 6169 } 6170 6171 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6172 if (!UseSHM) { 6173 return; 6174 } 6175 6176 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6177 6178 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6179 6180 if (addr != NULL) { 6181 assert(is_ptr_aligned(addr, alignment), "Check"); 6182 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6183 6184 small_page_write(addr, size); 6185 6186 os::Linux::release_memory_special_shm(addr, size); 6187 } 6188 } 6189 6190 static void test_reserve_memory_special_shm() { 6191 size_t lp = os::large_page_size(); 6192 size_t ag = os::vm_allocation_granularity(); 6193 6194 for (size_t size = ag; size < lp * 3; size += ag) { 6195 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6196 test_reserve_memory_special_shm(size, alignment); 6197 } 6198 } 6199 } 6200 6201 static void test() { 6202 test_reserve_memory_special_huge_tlbfs(); 6203 test_reserve_memory_special_shm(); 6204 } 6205}; 6206 6207void TestReserveMemorySpecial_test() { 6208 TestReserveMemorySpecial::test(); 6209} 6210 6211#endif 6212