os_linux.cpp revision 8225:eb02bcd73927
1/* 2 * Copyright (c) 1999, 2015, 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_linux.inline.hpp" 40#include "os_share_linux.hpp" 41#include "prims/jniFastGetField.hpp" 42#include "prims/jvm.h" 43#include "prims/jvm_misc.hpp" 44#include "runtime/arguments.hpp" 45#include "runtime/atomic.inline.hpp" 46#include "runtime/extendedPC.hpp" 47#include "runtime/globals.hpp" 48#include "runtime/interfaceSupport.hpp" 49#include "runtime/init.hpp" 50#include "runtime/java.hpp" 51#include "runtime/javaCalls.hpp" 52#include "runtime/mutexLocker.hpp" 53#include "runtime/objectMonitor.hpp" 54#include "runtime/orderAccess.inline.hpp" 55#include "runtime/osThread.hpp" 56#include "runtime/perfMemory.hpp" 57#include "runtime/sharedRuntime.hpp" 58#include "runtime/statSampler.hpp" 59#include "runtime/stubRoutines.hpp" 60#include "runtime/thread.inline.hpp" 61#include "runtime/threadCritical.hpp" 62#include "runtime/timer.hpp" 63#include "services/attachListener.hpp" 64#include "services/memTracker.hpp" 65#include "services/runtimeService.hpp" 66#include "utilities/decoder.hpp" 67#include "utilities/defaultStream.hpp" 68#include "utilities/events.hpp" 69#include "utilities/elfFile.hpp" 70#include "utilities/growableArray.hpp" 71#include "utilities/macros.hpp" 72#include "utilities/vmError.hpp" 73 74// put OS-includes here 75# include <sys/types.h> 76# include <sys/mman.h> 77# include <sys/stat.h> 78# include <sys/select.h> 79# include <pthread.h> 80# include <signal.h> 81# include <errno.h> 82# include <dlfcn.h> 83# include <stdio.h> 84# include <unistd.h> 85# include <sys/resource.h> 86# include <pthread.h> 87# include <sys/stat.h> 88# include <sys/time.h> 89# include <sys/times.h> 90# include <sys/utsname.h> 91# include <sys/socket.h> 92# include <sys/wait.h> 93# include <pwd.h> 94# include <poll.h> 95# include <semaphore.h> 96# include <fcntl.h> 97# include <string.h> 98# include <syscall.h> 99# include <sys/sysinfo.h> 100# include <gnu/libc-version.h> 101# include <sys/ipc.h> 102# include <sys/shm.h> 103# include <link.h> 104# include <stdint.h> 105# include <inttypes.h> 106# include <sys/ioctl.h> 107 108PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC 109 110// if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 111// getrusage() is prepared to handle the associated failure. 112#ifndef RUSAGE_THREAD 113 #define RUSAGE_THREAD (1) /* only the calling thread */ 114#endif 115 116#define MAX_PATH (2 * K) 117 118#define MAX_SECS 100000000 119 120// for timer info max values which include all bits 121#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 122 123#define LARGEPAGES_BIT (1 << 6) 124//////////////////////////////////////////////////////////////////////////////// 125// global variables 126julong os::Linux::_physical_memory = 0; 127 128address os::Linux::_initial_thread_stack_bottom = NULL; 129uintptr_t os::Linux::_initial_thread_stack_size = 0; 130 131int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 132int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 133int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL; 134Mutex* os::Linux::_createThread_lock = NULL; 135pthread_t os::Linux::_main_thread; 136int os::Linux::_page_size = -1; 137const int os::Linux::_vm_default_page_size = (8 * K); 138bool os::Linux::_is_floating_stack = false; 139bool os::Linux::_is_NPTL = false; 140bool os::Linux::_supports_fast_thread_cpu_time = false; 141const char * os::Linux::_glibc_version = NULL; 142const char * os::Linux::_libpthread_version = NULL; 143pthread_condattr_t os::Linux::_condattr[1]; 144 145static jlong initial_time_count=0; 146 147static int clock_tics_per_sec = 100; 148 149// For diagnostics to print a message once. see run_periodic_checks 150static sigset_t check_signal_done; 151static bool check_signals = true; 152 153static pid_t _initial_pid = 0; 154 155// Signal number used to suspend/resume a thread 156 157// do not use any signal number less than SIGSEGV, see 4355769 158static int SR_signum = SIGUSR2; 159sigset_t SR_sigset; 160 161// Declarations 162static void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 163 164// utility functions 165 166static int SR_initialize(); 167 168julong os::available_memory() { 169 return Linux::available_memory(); 170} 171 172julong os::Linux::available_memory() { 173 // values in struct sysinfo are "unsigned long" 174 struct sysinfo si; 175 sysinfo(&si); 176 177 return (julong)si.freeram * si.mem_unit; 178} 179 180julong os::physical_memory() { 181 return Linux::physical_memory(); 182} 183 184// Return true if user is running as root. 185 186bool os::have_special_privileges() { 187 static bool init = false; 188 static bool privileges = false; 189 if (!init) { 190 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 191 init = true; 192 } 193 return privileges; 194} 195 196 197#ifndef SYS_gettid 198// i386: 224, ia64: 1105, amd64: 186, sparc 143 199 #ifdef __ia64__ 200 #define SYS_gettid 1105 201 #elif __i386__ 202 #define SYS_gettid 224 203 #elif __amd64__ 204 #define SYS_gettid 186 205 #elif __sparc__ 206 #define SYS_gettid 143 207 #else 208 #error define gettid for the arch 209 #endif 210#endif 211 212// Cpu architecture string 213static char cpu_arch[] = HOTSPOT_LIB_ARCH; 214 215 216// pid_t gettid() 217// 218// Returns the kernel thread id of the currently running thread. Kernel 219// thread id is used to access /proc. 220// 221// (Note that getpid() on LinuxThreads returns kernel thread id too; but 222// on NPTL, it returns the same pid for all threads, as required by POSIX.) 223// 224pid_t os::Linux::gettid() { 225 int rslt = syscall(SYS_gettid); 226 if (rslt == -1) { 227 // old kernel, no NPTL support 228 return getpid(); 229 } else { 230 return (pid_t)rslt; 231 } 232} 233 234// Most versions of linux have a bug where the number of processors are 235// determined by looking at the /proc file system. In a chroot environment, 236// the system call returns 1. This causes the VM to act as if it is 237// a single processor and elide locking (see is_MP() call). 238static bool unsafe_chroot_detected = false; 239static const char *unstable_chroot_error = "/proc file system not found.\n" 240 "Java may be unstable running multithreaded in a chroot " 241 "environment on Linux when /proc filesystem is not mounted."; 242 243void os::Linux::initialize_system_info() { 244 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 245 if (processor_count() == 1) { 246 pid_t pid = os::Linux::gettid(); 247 char fname[32]; 248 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 249 FILE *fp = fopen(fname, "r"); 250 if (fp == NULL) { 251 unsafe_chroot_detected = true; 252 } else { 253 fclose(fp); 254 } 255 } 256 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 257 assert(processor_count() > 0, "linux error"); 258} 259 260void os::init_system_properties_values() { 261 // The next steps are taken in the product version: 262 // 263 // Obtain the JAVA_HOME value from the location of libjvm.so. 264 // This library should be located at: 265 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 266 // 267 // If "/jre/lib/" appears at the right place in the path, then we 268 // assume libjvm.so is installed in a JDK and we use this path. 269 // 270 // Otherwise exit with message: "Could not create the Java virtual machine." 271 // 272 // The following extra steps are taken in the debugging version: 273 // 274 // If "/jre/lib/" does NOT appear at the right place in the path 275 // instead of exit check for $JAVA_HOME environment variable. 276 // 277 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 278 // then we append a fake suffix "hotspot/libjvm.so" to this path so 279 // it looks like libjvm.so is installed there 280 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 281 // 282 // Otherwise exit. 283 // 284 // Important note: if the location of libjvm.so changes this 285 // code needs to be changed accordingly. 286 287 // See ld(1): 288 // The linker uses the following search paths to locate required 289 // shared libraries: 290 // 1: ... 291 // ... 292 // 7: The default directories, normally /lib and /usr/lib. 293#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 294 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 295#else 296 #define DEFAULT_LIBPATH "/lib:/usr/lib" 297#endif 298 299// Base path of extensions installed on the system. 300#define SYS_EXT_DIR "/usr/java/packages" 301#define EXTENSIONS_DIR "/lib/ext" 302 303 // Buffer that fits several sprintfs. 304 // Note that the space for the colon and the trailing null are provided 305 // by the nulls included by the sizeof operator. 306 const size_t bufsize = 307 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 308 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 309 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 310 311 // sysclasspath, java_home, dll_dir 312 { 313 char *pslash; 314 os::jvm_path(buf, bufsize); 315 316 // Found the full path to libjvm.so. 317 // Now cut the path to <java_home>/jre if we can. 318 pslash = strrchr(buf, '/'); 319 if (pslash != NULL) { 320 *pslash = '\0'; // Get rid of /libjvm.so. 321 } 322 pslash = strrchr(buf, '/'); 323 if (pslash != NULL) { 324 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 325 } 326 Arguments::set_dll_dir(buf); 327 328 if (pslash != NULL) { 329 pslash = strrchr(buf, '/'); 330 if (pslash != NULL) { 331 *pslash = '\0'; // Get rid of /<arch>. 332 pslash = strrchr(buf, '/'); 333 if (pslash != NULL) { 334 *pslash = '\0'; // Get rid of /lib. 335 } 336 } 337 } 338 Arguments::set_java_home(buf); 339 set_boot_path('/', ':'); 340 } 341 342 // Where to look for native libraries. 343 // 344 // Note: Due to a legacy implementation, most of the library path 345 // is set in the launcher. This was to accomodate linking restrictions 346 // on legacy Linux implementations (which are no longer supported). 347 // Eventually, all the library path setting will be done here. 348 // 349 // However, to prevent the proliferation of improperly built native 350 // libraries, the new path component /usr/java/packages is added here. 351 // Eventually, all the library path setting will be done here. 352 { 353 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 354 // should always exist (until the legacy problem cited above is 355 // addressed). 356 const char *v = ::getenv("LD_LIBRARY_PATH"); 357 const char *v_colon = ":"; 358 if (v == NULL) { v = ""; v_colon = ""; } 359 // That's +1 for the colon and +1 for the trailing '\0'. 360 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 361 strlen(v) + 1 + 362 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1, 363 mtInternal); 364 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch); 365 Arguments::set_library_path(ld_library_path); 366 FREE_C_HEAP_ARRAY(char, ld_library_path); 367 } 368 369 // Extensions directories. 370 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 371 Arguments::set_ext_dirs(buf); 372 373 FREE_C_HEAP_ARRAY(char, buf); 374 375#undef DEFAULT_LIBPATH 376#undef SYS_EXT_DIR 377#undef EXTENSIONS_DIR 378} 379 380//////////////////////////////////////////////////////////////////////////////// 381// breakpoint support 382 383void os::breakpoint() { 384 BREAKPOINT; 385} 386 387extern "C" void breakpoint() { 388 // use debugger to set breakpoint here 389} 390 391//////////////////////////////////////////////////////////////////////////////// 392// signal support 393 394debug_only(static bool signal_sets_initialized = false); 395static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 396 397bool os::Linux::is_sig_ignored(int sig) { 398 struct sigaction oact; 399 sigaction(sig, (struct sigaction*)NULL, &oact); 400 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 401 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 402 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) { 403 return true; 404 } else { 405 return false; 406 } 407} 408 409void os::Linux::signal_sets_init() { 410 // Should also have an assertion stating we are still single-threaded. 411 assert(!signal_sets_initialized, "Already initialized"); 412 // Fill in signals that are necessarily unblocked for all threads in 413 // the VM. Currently, we unblock the following signals: 414 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 415 // by -Xrs (=ReduceSignalUsage)); 416 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 417 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 418 // the dispositions or masks wrt these signals. 419 // Programs embedding the VM that want to use the above signals for their 420 // own purposes must, at this time, use the "-Xrs" option to prevent 421 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 422 // (See bug 4345157, and other related bugs). 423 // In reality, though, unblocking these signals is really a nop, since 424 // these signals are not blocked by default. 425 sigemptyset(&unblocked_sigs); 426 sigemptyset(&allowdebug_blocked_sigs); 427 sigaddset(&unblocked_sigs, SIGILL); 428 sigaddset(&unblocked_sigs, SIGSEGV); 429 sigaddset(&unblocked_sigs, SIGBUS); 430 sigaddset(&unblocked_sigs, SIGFPE); 431#if defined(PPC64) 432 sigaddset(&unblocked_sigs, SIGTRAP); 433#endif 434 sigaddset(&unblocked_sigs, SR_signum); 435 436 if (!ReduceSignalUsage) { 437 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 438 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 439 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 440 } 441 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 442 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 443 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 444 } 445 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 446 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 447 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 448 } 449 } 450 // Fill in signals that are blocked by all but the VM thread. 451 sigemptyset(&vm_sigs); 452 if (!ReduceSignalUsage) { 453 sigaddset(&vm_sigs, BREAK_SIGNAL); 454 } 455 debug_only(signal_sets_initialized = true); 456 457} 458 459// These are signals that are unblocked while a thread is running Java. 460// (For some reason, they get blocked by default.) 461sigset_t* os::Linux::unblocked_signals() { 462 assert(signal_sets_initialized, "Not initialized"); 463 return &unblocked_sigs; 464} 465 466// These are the signals that are blocked while a (non-VM) thread is 467// running Java. Only the VM thread handles these signals. 468sigset_t* os::Linux::vm_signals() { 469 assert(signal_sets_initialized, "Not initialized"); 470 return &vm_sigs; 471} 472 473// These are signals that are blocked during cond_wait to allow debugger in 474sigset_t* os::Linux::allowdebug_blocked_signals() { 475 assert(signal_sets_initialized, "Not initialized"); 476 return &allowdebug_blocked_sigs; 477} 478 479void os::Linux::hotspot_sigmask(Thread* thread) { 480 481 //Save caller's signal mask before setting VM signal mask 482 sigset_t caller_sigmask; 483 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 484 485 OSThread* osthread = thread->osthread(); 486 osthread->set_caller_sigmask(caller_sigmask); 487 488 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 489 490 if (!ReduceSignalUsage) { 491 if (thread->is_VM_thread()) { 492 // Only the VM thread handles BREAK_SIGNAL ... 493 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 494 } else { 495 // ... all other threads block BREAK_SIGNAL 496 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 497 } 498 } 499} 500 501////////////////////////////////////////////////////////////////////////////// 502// detecting pthread library 503 504void os::Linux::libpthread_init() { 505 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 506 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 507 // generic name for earlier versions. 508 // Define macros here so we can build HotSpot on old systems. 509#ifndef _CS_GNU_LIBC_VERSION 510 #define _CS_GNU_LIBC_VERSION 2 511#endif 512#ifndef _CS_GNU_LIBPTHREAD_VERSION 513 #define _CS_GNU_LIBPTHREAD_VERSION 3 514#endif 515 516 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 517 if (n > 0) { 518 char *str = (char *)malloc(n, mtInternal); 519 confstr(_CS_GNU_LIBC_VERSION, str, n); 520 os::Linux::set_glibc_version(str); 521 } else { 522 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 523 static char _gnu_libc_version[32]; 524 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 525 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 526 os::Linux::set_glibc_version(_gnu_libc_version); 527 } 528 529 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 530 if (n > 0) { 531 char *str = (char *)malloc(n, mtInternal); 532 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 533 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 534 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 535 // is the case. LinuxThreads has a hard limit on max number of threads. 536 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 537 // On the other hand, NPTL does not have such a limit, sysconf() 538 // will return -1 and errno is not changed. Check if it is really NPTL. 539 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 540 strstr(str, "NPTL") && 541 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 542 free(str); 543 os::Linux::set_libpthread_version("linuxthreads"); 544 } else { 545 os::Linux::set_libpthread_version(str); 546 } 547 } else { 548 // glibc before 2.3.2 only has LinuxThreads. 549 os::Linux::set_libpthread_version("linuxthreads"); 550 } 551 552 if (strstr(libpthread_version(), "NPTL")) { 553 os::Linux::set_is_NPTL(); 554 } else { 555 os::Linux::set_is_LinuxThreads(); 556 } 557 558 // LinuxThreads have two flavors: floating-stack mode, which allows variable 559 // stack size; and fixed-stack mode. NPTL is always floating-stack. 560 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 561 os::Linux::set_is_floating_stack(); 562 } 563} 564 565///////////////////////////////////////////////////////////////////////////// 566// thread stack 567 568// Force Linux kernel to expand current thread stack. If "bottom" is close 569// to the stack guard, caller should block all signals. 570// 571// MAP_GROWSDOWN: 572// A special mmap() flag that is used to implement thread stacks. It tells 573// kernel that the memory region should extend downwards when needed. This 574// allows early versions of LinuxThreads to only mmap the first few pages 575// when creating a new thread. Linux kernel will automatically expand thread 576// stack as needed (on page faults). 577// 578// However, because the memory region of a MAP_GROWSDOWN stack can grow on 579// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 580// region, it's hard to tell if the fault is due to a legitimate stack 581// access or because of reading/writing non-exist memory (e.g. buffer 582// overrun). As a rule, if the fault happens below current stack pointer, 583// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 584// application (see Linux kernel fault.c). 585// 586// This Linux feature can cause SIGSEGV when VM bangs thread stack for 587// stack overflow detection. 588// 589// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 590// not use this flag. However, the stack of initial thread is not created 591// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 592// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 593// and then attach the thread to JVM. 594// 595// To get around the problem and allow stack banging on Linux, we need to 596// manually expand thread stack after receiving the SIGSEGV. 597// 598// There are two ways to expand thread stack to address "bottom", we used 599// both of them in JVM before 1.5: 600// 1. adjust stack pointer first so that it is below "bottom", and then 601// touch "bottom" 602// 2. mmap() the page in question 603// 604// Now alternate signal stack is gone, it's harder to use 2. For instance, 605// if current sp is already near the lower end of page 101, and we need to 606// call mmap() to map page 100, it is possible that part of the mmap() frame 607// will be placed in page 100. When page 100 is mapped, it is zero-filled. 608// That will destroy the mmap() frame and cause VM to crash. 609// 610// The following code works by adjusting sp first, then accessing the "bottom" 611// page to force a page fault. Linux kernel will then automatically expand the 612// stack mapping. 613// 614// _expand_stack_to() assumes its frame size is less than page size, which 615// should always be true if the function is not inlined. 616 617#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 618 #define NOINLINE 619#else 620 #define NOINLINE __attribute__ ((noinline)) 621#endif 622 623static void _expand_stack_to(address bottom) NOINLINE; 624 625static void _expand_stack_to(address bottom) { 626 address sp; 627 size_t size; 628 volatile char *p; 629 630 // Adjust bottom to point to the largest address within the same page, it 631 // gives us a one-page buffer if alloca() allocates slightly more memory. 632 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 633 bottom += os::Linux::page_size() - 1; 634 635 // sp might be slightly above current stack pointer; if that's the case, we 636 // will alloca() a little more space than necessary, which is OK. Don't use 637 // os::current_stack_pointer(), as its result can be slightly below current 638 // stack pointer, causing us to not alloca enough to reach "bottom". 639 sp = (address)&sp; 640 641 if (sp > bottom) { 642 size = sp - bottom; 643 p = (volatile char *)alloca(size); 644 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 645 p[0] = '\0'; 646 } 647} 648 649bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 650 assert(t!=NULL, "just checking"); 651 assert(t->osthread()->expanding_stack(), "expand should be set"); 652 assert(t->stack_base() != NULL, "stack_base was not initialized"); 653 654 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 655 sigset_t mask_all, old_sigset; 656 sigfillset(&mask_all); 657 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 658 _expand_stack_to(addr); 659 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 660 return true; 661 } 662 return false; 663} 664 665////////////////////////////////////////////////////////////////////////////// 666// create new thread 667 668static address highest_vm_reserved_address(); 669 670// check if it's safe to start a new thread 671static bool _thread_safety_check(Thread* thread) { 672 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 673 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 674 // Heap is mmap'ed at lower end of memory space. Thread stacks are 675 // allocated (MAP_FIXED) from high address space. Every thread stack 676 // occupies a fixed size slot (usually 2Mbytes, but user can change 677 // it to other values if they rebuild LinuxThreads). 678 // 679 // Problem with MAP_FIXED is that mmap() can still succeed even part of 680 // the memory region has already been mmap'ed. That means if we have too 681 // many threads and/or very large heap, eventually thread stack will 682 // collide with heap. 683 // 684 // Here we try to prevent heap/stack collision by comparing current 685 // stack bottom with the highest address that has been mmap'ed by JVM 686 // plus a safety margin for memory maps created by native code. 687 // 688 // This feature can be disabled by setting ThreadSafetyMargin to 0 689 // 690 if (ThreadSafetyMargin > 0) { 691 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 692 693 // not safe if our stack extends below the safety margin 694 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 695 } else { 696 return true; 697 } 698 } else { 699 // Floating stack LinuxThreads or NPTL: 700 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 701 // there's not enough space left, pthread_create() will fail. If we come 702 // here, that means enough space has been reserved for stack. 703 return true; 704 } 705} 706 707// Thread start routine for all newly created threads 708static void *java_start(Thread *thread) { 709 // Try to randomize the cache line index of hot stack frames. 710 // This helps when threads of the same stack traces evict each other's 711 // cache lines. The threads can be either from the same JVM instance, or 712 // from different JVM instances. The benefit is especially true for 713 // processors with hyperthreading technology. 714 static int counter = 0; 715 int pid = os::current_process_id(); 716 alloca(((pid ^ counter++) & 7) * 128); 717 718 ThreadLocalStorage::set_thread(thread); 719 720 OSThread* osthread = thread->osthread(); 721 Monitor* sync = osthread->startThread_lock(); 722 723 // non floating stack LinuxThreads needs extra check, see above 724 if (!_thread_safety_check(thread)) { 725 // notify parent thread 726 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 727 osthread->set_state(ZOMBIE); 728 sync->notify_all(); 729 return NULL; 730 } 731 732 // thread_id is kernel thread id (similar to Solaris LWP id) 733 osthread->set_thread_id(os::Linux::gettid()); 734 735 if (UseNUMA) { 736 int lgrp_id = os::numa_get_group_id(); 737 if (lgrp_id != -1) { 738 thread->set_lgrp_id(lgrp_id); 739 } 740 } 741 // initialize signal mask for this thread 742 os::Linux::hotspot_sigmask(thread); 743 744 // initialize floating point control register 745 os::Linux::init_thread_fpu_state(); 746 747 // handshaking with parent thread 748 { 749 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 750 751 // notify parent thread 752 osthread->set_state(INITIALIZED); 753 sync->notify_all(); 754 755 // wait until os::start_thread() 756 while (osthread->get_state() == INITIALIZED) { 757 sync->wait(Mutex::_no_safepoint_check_flag); 758 } 759 } 760 761 // call one more level start routine 762 thread->run(); 763 764 return 0; 765} 766 767bool os::create_thread(Thread* thread, ThreadType thr_type, 768 size_t stack_size) { 769 assert(thread->osthread() == NULL, "caller responsible"); 770 771 // Allocate the OSThread object 772 OSThread* osthread = new OSThread(NULL, NULL); 773 if (osthread == NULL) { 774 return false; 775 } 776 777 // set the correct thread state 778 osthread->set_thread_type(thr_type); 779 780 // Initial state is ALLOCATED but not INITIALIZED 781 osthread->set_state(ALLOCATED); 782 783 thread->set_osthread(osthread); 784 785 // init thread attributes 786 pthread_attr_t attr; 787 pthread_attr_init(&attr); 788 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 789 790 // stack size 791 if (os::Linux::supports_variable_stack_size()) { 792 // calculate stack size if it's not specified by caller 793 if (stack_size == 0) { 794 stack_size = os::Linux::default_stack_size(thr_type); 795 796 switch (thr_type) { 797 case os::java_thread: 798 // Java threads use ThreadStackSize which default value can be 799 // changed with the flag -Xss 800 assert(JavaThread::stack_size_at_create() > 0, "this should be set"); 801 stack_size = JavaThread::stack_size_at_create(); 802 break; 803 case os::compiler_thread: 804 if (CompilerThreadStackSize > 0) { 805 stack_size = (size_t)(CompilerThreadStackSize * K); 806 break; 807 } // else fall through: 808 // use VMThreadStackSize if CompilerThreadStackSize is not defined 809 case os::vm_thread: 810 case os::pgc_thread: 811 case os::cgc_thread: 812 case os::watcher_thread: 813 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 814 break; 815 } 816 } 817 818 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 819 pthread_attr_setstacksize(&attr, stack_size); 820 } else { 821 // let pthread_create() pick the default value. 822 } 823 824 // glibc guard page 825 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 826 827 ThreadState state; 828 829 { 830 // Serialize thread creation if we are running with fixed stack LinuxThreads 831 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 832 if (lock) { 833 os::Linux::createThread_lock()->lock_without_safepoint_check(); 834 } 835 836 pthread_t tid; 837 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 838 839 pthread_attr_destroy(&attr); 840 841 if (ret != 0) { 842 if (PrintMiscellaneous && (Verbose || WizardMode)) { 843 perror("pthread_create()"); 844 } 845 // Need to clean up stuff we've allocated so far 846 thread->set_osthread(NULL); 847 delete osthread; 848 if (lock) os::Linux::createThread_lock()->unlock(); 849 return false; 850 } 851 852 // Store pthread info into the OSThread 853 osthread->set_pthread_id(tid); 854 855 // Wait until child thread is either initialized or aborted 856 { 857 Monitor* sync_with_child = osthread->startThread_lock(); 858 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 859 while ((state = osthread->get_state()) == ALLOCATED) { 860 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 861 } 862 } 863 864 if (lock) { 865 os::Linux::createThread_lock()->unlock(); 866 } 867 } 868 869 // Aborted due to thread limit being reached 870 if (state == ZOMBIE) { 871 thread->set_osthread(NULL); 872 delete osthread; 873 return false; 874 } 875 876 // The thread is returned suspended (in state INITIALIZED), 877 // and is started higher up in the call chain 878 assert(state == INITIALIZED, "race condition"); 879 return true; 880} 881 882///////////////////////////////////////////////////////////////////////////// 883// attach existing thread 884 885// bootstrap the main thread 886bool os::create_main_thread(JavaThread* thread) { 887 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 888 return create_attached_thread(thread); 889} 890 891bool os::create_attached_thread(JavaThread* thread) { 892#ifdef ASSERT 893 thread->verify_not_published(); 894#endif 895 896 // Allocate the OSThread object 897 OSThread* osthread = new OSThread(NULL, NULL); 898 899 if (osthread == NULL) { 900 return false; 901 } 902 903 // Store pthread info into the OSThread 904 osthread->set_thread_id(os::Linux::gettid()); 905 osthread->set_pthread_id(::pthread_self()); 906 907 // initialize floating point control register 908 os::Linux::init_thread_fpu_state(); 909 910 // Initial thread state is RUNNABLE 911 osthread->set_state(RUNNABLE); 912 913 thread->set_osthread(osthread); 914 915 if (UseNUMA) { 916 int lgrp_id = os::numa_get_group_id(); 917 if (lgrp_id != -1) { 918 thread->set_lgrp_id(lgrp_id); 919 } 920 } 921 922 if (os::Linux::is_initial_thread()) { 923 // If current thread is initial thread, its stack is mapped on demand, 924 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 925 // the entire stack region to avoid SEGV in stack banging. 926 // It is also useful to get around the heap-stack-gap problem on SuSE 927 // kernel (see 4821821 for details). We first expand stack to the top 928 // of yellow zone, then enable stack yellow zone (order is significant, 929 // enabling yellow zone first will crash JVM on SuSE Linux), so there 930 // is no gap between the last two virtual memory regions. 931 932 JavaThread *jt = (JavaThread *)thread; 933 address addr = jt->stack_yellow_zone_base(); 934 assert(addr != NULL, "initialization problem?"); 935 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 936 937 osthread->set_expanding_stack(); 938 os::Linux::manually_expand_stack(jt, addr); 939 osthread->clear_expanding_stack(); 940 } 941 942 // initialize signal mask for this thread 943 // and save the caller's signal mask 944 os::Linux::hotspot_sigmask(thread); 945 946 return true; 947} 948 949void os::pd_start_thread(Thread* thread) { 950 OSThread * osthread = thread->osthread(); 951 assert(osthread->get_state() != INITIALIZED, "just checking"); 952 Monitor* sync_with_child = osthread->startThread_lock(); 953 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 954 sync_with_child->notify(); 955} 956 957// Free Linux resources related to the OSThread 958void os::free_thread(OSThread* osthread) { 959 assert(osthread != NULL, "osthread not set"); 960 961 if (Thread::current()->osthread() == osthread) { 962 // Restore caller's signal mask 963 sigset_t sigmask = osthread->caller_sigmask(); 964 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 965 } 966 967 delete osthread; 968} 969 970////////////////////////////////////////////////////////////////////////////// 971// thread local storage 972 973// Restore the thread pointer if the destructor is called. This is in case 974// someone from JNI code sets up a destructor with pthread_key_create to run 975// detachCurrentThread on thread death. Unless we restore the thread pointer we 976// will hang or crash. When detachCurrentThread is called the key will be set 977// to null and we will not be called again. If detachCurrentThread is never 978// called we could loop forever depending on the pthread implementation. 979static void restore_thread_pointer(void* p) { 980 Thread* thread = (Thread*) p; 981 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread); 982} 983 984int os::allocate_thread_local_storage() { 985 pthread_key_t key; 986 int rslt = pthread_key_create(&key, restore_thread_pointer); 987 assert(rslt == 0, "cannot allocate thread local storage"); 988 return (int)key; 989} 990 991// Note: This is currently not used by VM, as we don't destroy TLS key 992// on VM exit. 993void os::free_thread_local_storage(int index) { 994 int rslt = pthread_key_delete((pthread_key_t)index); 995 assert(rslt == 0, "invalid index"); 996} 997 998void os::thread_local_storage_at_put(int index, void* value) { 999 int rslt = pthread_setspecific((pthread_key_t)index, value); 1000 assert(rslt == 0, "pthread_setspecific failed"); 1001} 1002 1003extern "C" Thread* get_thread() { 1004 return ThreadLocalStorage::thread(); 1005} 1006 1007////////////////////////////////////////////////////////////////////////////// 1008// initial thread 1009 1010// Check if current thread is the initial thread, similar to Solaris thr_main. 1011bool os::Linux::is_initial_thread(void) { 1012 char dummy; 1013 // If called before init complete, thread stack bottom will be null. 1014 // Can be called if fatal error occurs before initialization. 1015 if (initial_thread_stack_bottom() == NULL) return false; 1016 assert(initial_thread_stack_bottom() != NULL && 1017 initial_thread_stack_size() != 0, 1018 "os::init did not locate initial thread's stack region"); 1019 if ((address)&dummy >= initial_thread_stack_bottom() && 1020 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) { 1021 return true; 1022 } else { 1023 return false; 1024 } 1025} 1026 1027// Find the virtual memory area that contains addr 1028static bool find_vma(address addr, address* vma_low, address* vma_high) { 1029 FILE *fp = fopen("/proc/self/maps", "r"); 1030 if (fp) { 1031 address low, high; 1032 while (!feof(fp)) { 1033 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1034 if (low <= addr && addr < high) { 1035 if (vma_low) *vma_low = low; 1036 if (vma_high) *vma_high = high; 1037 fclose(fp); 1038 return true; 1039 } 1040 } 1041 for (;;) { 1042 int ch = fgetc(fp); 1043 if (ch == EOF || ch == (int)'\n') break; 1044 } 1045 } 1046 fclose(fp); 1047 } 1048 return false; 1049} 1050 1051// Locate initial thread stack. This special handling of initial thread stack 1052// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1053// bogus value for initial thread. 1054void os::Linux::capture_initial_stack(size_t max_size) { 1055 // stack size is the easy part, get it from RLIMIT_STACK 1056 size_t stack_size; 1057 struct rlimit rlim; 1058 getrlimit(RLIMIT_STACK, &rlim); 1059 stack_size = rlim.rlim_cur; 1060 1061 // 6308388: a bug in ld.so will relocate its own .data section to the 1062 // lower end of primordial stack; reduce ulimit -s value a little bit 1063 // so we won't install guard page on ld.so's data section. 1064 stack_size -= 2 * page_size(); 1065 1066 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1067 // 7.1, in both cases we will get 2G in return value. 1068 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1069 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1070 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1071 // in case other parts in glibc still assumes 2M max stack size. 1072 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1073 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1074 if (stack_size > 2 * K * K IA64_ONLY(*2)) { 1075 stack_size = 2 * K * K IA64_ONLY(*2); 1076 } 1077 // Try to figure out where the stack base (top) is. This is harder. 1078 // 1079 // When an application is started, glibc saves the initial stack pointer in 1080 // a global variable "__libc_stack_end", which is then used by system 1081 // libraries. __libc_stack_end should be pretty close to stack top. The 1082 // variable is available since the very early days. However, because it is 1083 // a private interface, it could disappear in the future. 1084 // 1085 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1086 // to __libc_stack_end, it is very close to stack top, but isn't the real 1087 // stack top. Note that /proc may not exist if VM is running as a chroot 1088 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1089 // /proc/<pid>/stat could change in the future (though unlikely). 1090 // 1091 // We try __libc_stack_end first. If that doesn't work, look for 1092 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1093 // as a hint, which should work well in most cases. 1094 1095 uintptr_t stack_start; 1096 1097 // try __libc_stack_end first 1098 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1099 if (p && *p) { 1100 stack_start = *p; 1101 } else { 1102 // see if we can get the start_stack field from /proc/self/stat 1103 FILE *fp; 1104 int pid; 1105 char state; 1106 int ppid; 1107 int pgrp; 1108 int session; 1109 int nr; 1110 int tpgrp; 1111 unsigned long flags; 1112 unsigned long minflt; 1113 unsigned long cminflt; 1114 unsigned long majflt; 1115 unsigned long cmajflt; 1116 unsigned long utime; 1117 unsigned long stime; 1118 long cutime; 1119 long cstime; 1120 long prio; 1121 long nice; 1122 long junk; 1123 long it_real; 1124 uintptr_t start; 1125 uintptr_t vsize; 1126 intptr_t rss; 1127 uintptr_t rsslim; 1128 uintptr_t scodes; 1129 uintptr_t ecode; 1130 int i; 1131 1132 // Figure what the primordial thread stack base is. Code is inspired 1133 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1134 // followed by command name surrounded by parentheses, state, etc. 1135 char stat[2048]; 1136 int statlen; 1137 1138 fp = fopen("/proc/self/stat", "r"); 1139 if (fp) { 1140 statlen = fread(stat, 1, 2047, fp); 1141 stat[statlen] = '\0'; 1142 fclose(fp); 1143 1144 // Skip pid and the command string. Note that we could be dealing with 1145 // weird command names, e.g. user could decide to rename java launcher 1146 // to "java 1.4.2 :)", then the stat file would look like 1147 // 1234 (java 1.4.2 :)) R ... ... 1148 // We don't really need to know the command string, just find the last 1149 // occurrence of ")" and then start parsing from there. See bug 4726580. 1150 char * s = strrchr(stat, ')'); 1151 1152 i = 0; 1153 if (s) { 1154 // Skip blank chars 1155 do { s++; } while (s && isspace(*s)); 1156 1157#define _UFM UINTX_FORMAT 1158#define _DFM INTX_FORMAT 1159 1160 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1161 // 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 1162 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, 1163 &state, // 3 %c 1164 &ppid, // 4 %d 1165 &pgrp, // 5 %d 1166 &session, // 6 %d 1167 &nr, // 7 %d 1168 &tpgrp, // 8 %d 1169 &flags, // 9 %lu 1170 &minflt, // 10 %lu 1171 &cminflt, // 11 %lu 1172 &majflt, // 12 %lu 1173 &cmajflt, // 13 %lu 1174 &utime, // 14 %lu 1175 &stime, // 15 %lu 1176 &cutime, // 16 %ld 1177 &cstime, // 17 %ld 1178 &prio, // 18 %ld 1179 &nice, // 19 %ld 1180 &junk, // 20 %ld 1181 &it_real, // 21 %ld 1182 &start, // 22 UINTX_FORMAT 1183 &vsize, // 23 UINTX_FORMAT 1184 &rss, // 24 INTX_FORMAT 1185 &rsslim, // 25 UINTX_FORMAT 1186 &scodes, // 26 UINTX_FORMAT 1187 &ecode, // 27 UINTX_FORMAT 1188 &stack_start); // 28 UINTX_FORMAT 1189 } 1190 1191#undef _UFM 1192#undef _DFM 1193 1194 if (i != 28 - 2) { 1195 assert(false, "Bad conversion from /proc/self/stat"); 1196 // product mode - assume we are the initial thread, good luck in the 1197 // embedded case. 1198 warning("Can't detect initial thread stack location - bad conversion"); 1199 stack_start = (uintptr_t) &rlim; 1200 } 1201 } else { 1202 // For some reason we can't open /proc/self/stat (for example, running on 1203 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1204 // most cases, so don't abort: 1205 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1206 stack_start = (uintptr_t) &rlim; 1207 } 1208 } 1209 1210 // Now we have a pointer (stack_start) very close to the stack top, the 1211 // next thing to do is to figure out the exact location of stack top. We 1212 // can find out the virtual memory area that contains stack_start by 1213 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1214 // and its upper limit is the real stack top. (again, this would fail if 1215 // running inside chroot, because /proc may not exist.) 1216 1217 uintptr_t stack_top; 1218 address low, high; 1219 if (find_vma((address)stack_start, &low, &high)) { 1220 // success, "high" is the true stack top. (ignore "low", because initial 1221 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1222 stack_top = (uintptr_t)high; 1223 } else { 1224 // failed, likely because /proc/self/maps does not exist 1225 warning("Can't detect initial thread stack location - find_vma failed"); 1226 // best effort: stack_start is normally within a few pages below the real 1227 // stack top, use it as stack top, and reduce stack size so we won't put 1228 // guard page outside stack. 1229 stack_top = stack_start; 1230 stack_size -= 16 * page_size(); 1231 } 1232 1233 // stack_top could be partially down the page so align it 1234 stack_top = align_size_up(stack_top, page_size()); 1235 1236 if (max_size && stack_size > max_size) { 1237 _initial_thread_stack_size = max_size; 1238 } else { 1239 _initial_thread_stack_size = stack_size; 1240 } 1241 1242 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1243 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1244} 1245 1246//////////////////////////////////////////////////////////////////////////////// 1247// time support 1248 1249// Time since start-up in seconds to a fine granularity. 1250// Used by VMSelfDestructTimer and the MemProfiler. 1251double os::elapsedTime() { 1252 1253 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1254} 1255 1256jlong os::elapsed_counter() { 1257 return javaTimeNanos() - initial_time_count; 1258} 1259 1260jlong os::elapsed_frequency() { 1261 return NANOSECS_PER_SEC; // nanosecond resolution 1262} 1263 1264bool os::supports_vtime() { return true; } 1265bool os::enable_vtime() { return false; } 1266bool os::vtime_enabled() { return false; } 1267 1268double os::elapsedVTime() { 1269 struct rusage usage; 1270 int retval = getrusage(RUSAGE_THREAD, &usage); 1271 if (retval == 0) { 1272 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); 1273 } else { 1274 // better than nothing, but not much 1275 return elapsedTime(); 1276 } 1277} 1278 1279jlong os::javaTimeMillis() { 1280 timeval time; 1281 int status = gettimeofday(&time, NULL); 1282 assert(status != -1, "linux error"); 1283 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1284} 1285 1286void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) { 1287 timeval time; 1288 int status = gettimeofday(&time, NULL); 1289 assert(status != -1, "linux error"); 1290 seconds = jlong(time.tv_sec); 1291 nanos = jlong(time.tv_usec) * 1000; 1292} 1293 1294 1295#ifndef CLOCK_MONOTONIC 1296 #define CLOCK_MONOTONIC (1) 1297#endif 1298 1299void os::Linux::clock_init() { 1300 // we do dlopen's in this particular order due to bug in linux 1301 // dynamical loader (see 6348968) leading to crash on exit 1302 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1303 if (handle == NULL) { 1304 handle = dlopen("librt.so", RTLD_LAZY); 1305 } 1306 1307 if (handle) { 1308 int (*clock_getres_func)(clockid_t, struct timespec*) = 1309 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1310 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1311 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1312 if (clock_getres_func && clock_gettime_func) { 1313 // See if monotonic clock is supported by the kernel. Note that some 1314 // early implementations simply return kernel jiffies (updated every 1315 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1316 // for nano time (though the monotonic property is still nice to have). 1317 // It's fixed in newer kernels, however clock_getres() still returns 1318 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1319 // resolution for now. Hopefully as people move to new kernels, this 1320 // won't be a problem. 1321 struct timespec res; 1322 struct timespec tp; 1323 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1324 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1325 // yes, monotonic clock is supported 1326 _clock_gettime = clock_gettime_func; 1327 return; 1328 } else { 1329 // close librt if there is no monotonic clock 1330 dlclose(handle); 1331 } 1332 } 1333 } 1334 warning("No monotonic clock was available - timed services may " \ 1335 "be adversely affected if the time-of-day clock changes"); 1336} 1337 1338#ifndef SYS_clock_getres 1339 #if defined(IA32) || defined(AMD64) 1340 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1341 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1342 #else 1343 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1344 #define sys_clock_getres(x,y) -1 1345 #endif 1346#else 1347 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1348#endif 1349 1350void os::Linux::fast_thread_clock_init() { 1351 if (!UseLinuxPosixThreadCPUClocks) { 1352 return; 1353 } 1354 clockid_t clockid; 1355 struct timespec tp; 1356 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1357 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1358 1359 // Switch to using fast clocks for thread cpu time if 1360 // the sys_clock_getres() returns 0 error code. 1361 // Note, that some kernels may support the current thread 1362 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1363 // returned by the pthread_getcpuclockid(). 1364 // If the fast Posix clocks are supported then the sys_clock_getres() 1365 // must return at least tp.tv_sec == 0 which means a resolution 1366 // better than 1 sec. This is extra check for reliability. 1367 1368 if (pthread_getcpuclockid_func && 1369 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1370 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1371 _supports_fast_thread_cpu_time = true; 1372 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1373 } 1374} 1375 1376jlong os::javaTimeNanos() { 1377 if (os::supports_monotonic_clock()) { 1378 struct timespec tp; 1379 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1380 assert(status == 0, "gettime error"); 1381 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1382 return result; 1383 } else { 1384 timeval time; 1385 int status = gettimeofday(&time, NULL); 1386 assert(status != -1, "linux error"); 1387 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1388 return 1000 * usecs; 1389 } 1390} 1391 1392void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1393 if (os::supports_monotonic_clock()) { 1394 info_ptr->max_value = ALL_64_BITS; 1395 1396 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1397 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1398 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1399 } else { 1400 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1401 info_ptr->max_value = ALL_64_BITS; 1402 1403 // gettimeofday is a real time clock so it skips 1404 info_ptr->may_skip_backward = true; 1405 info_ptr->may_skip_forward = true; 1406 } 1407 1408 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1409} 1410 1411// Return the real, user, and system times in seconds from an 1412// arbitrary fixed point in the past. 1413bool os::getTimesSecs(double* process_real_time, 1414 double* process_user_time, 1415 double* process_system_time) { 1416 struct tms ticks; 1417 clock_t real_ticks = times(&ticks); 1418 1419 if (real_ticks == (clock_t) (-1)) { 1420 return false; 1421 } else { 1422 double ticks_per_second = (double) clock_tics_per_sec; 1423 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1424 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1425 *process_real_time = ((double) real_ticks) / ticks_per_second; 1426 1427 return true; 1428 } 1429} 1430 1431 1432char * os::local_time_string(char *buf, size_t buflen) { 1433 struct tm t; 1434 time_t long_time; 1435 time(&long_time); 1436 localtime_r(&long_time, &t); 1437 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1438 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1439 t.tm_hour, t.tm_min, t.tm_sec); 1440 return buf; 1441} 1442 1443struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1444 return localtime_r(clock, res); 1445} 1446 1447//////////////////////////////////////////////////////////////////////////////// 1448// runtime exit support 1449 1450// Note: os::shutdown() might be called very early during initialization, or 1451// called from signal handler. Before adding something to os::shutdown(), make 1452// sure it is async-safe and can handle partially initialized VM. 1453void os::shutdown() { 1454 1455 // allow PerfMemory to attempt cleanup of any persistent resources 1456 perfMemory_exit(); 1457 1458 // needs to remove object in file system 1459 AttachListener::abort(); 1460 1461 // flush buffered output, finish log files 1462 ostream_abort(); 1463 1464 // Check for abort hook 1465 abort_hook_t abort_hook = Arguments::abort_hook(); 1466 if (abort_hook != NULL) { 1467 abort_hook(); 1468 } 1469 1470} 1471 1472// Note: os::abort() might be called very early during initialization, or 1473// called from signal handler. Before adding something to os::abort(), make 1474// sure it is async-safe and can handle partially initialized VM. 1475void os::abort(bool dump_core) { 1476 abort(dump_core, NULL, NULL); 1477} 1478 1479void os::abort(bool dump_core, void* siginfo, void* context) { 1480 os::shutdown(); 1481 if (dump_core) { 1482#ifndef PRODUCT 1483 fdStream out(defaultStream::output_fd()); 1484 out.print_raw("Current thread is "); 1485 char buf[16]; 1486 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1487 out.print_raw_cr(buf); 1488 out.print_raw_cr("Dumping core ..."); 1489#endif 1490 ::abort(); // dump core 1491 } 1492 1493 ::exit(1); 1494} 1495 1496// Die immediately, no exit hook, no abort hook, no cleanup. 1497void os::die() { 1498 // _exit() on LinuxThreads only kills current thread 1499 ::abort(); 1500} 1501 1502 1503// This method is a copy of JDK's sysGetLastErrorString 1504// from src/solaris/hpi/src/system_md.c 1505 1506size_t os::lasterror(char *buf, size_t len) { 1507 if (errno == 0) return 0; 1508 1509 const char *s = ::strerror(errno); 1510 size_t n = ::strlen(s); 1511 if (n >= len) { 1512 n = len - 1; 1513 } 1514 ::strncpy(buf, s, n); 1515 buf[n] = '\0'; 1516 return n; 1517} 1518 1519intx os::current_thread_id() { return (intx)pthread_self(); } 1520int os::current_process_id() { 1521 1522 // Under the old linux thread library, linux gives each thread 1523 // its own process id. Because of this each thread will return 1524 // a different pid if this method were to return the result 1525 // of getpid(2). Linux provides no api that returns the pid 1526 // of the launcher thread for the vm. This implementation 1527 // returns a unique pid, the pid of the launcher thread 1528 // that starts the vm 'process'. 1529 1530 // Under the NPTL, getpid() returns the same pid as the 1531 // launcher thread rather than a unique pid per thread. 1532 // Use gettid() if you want the old pre NPTL behaviour. 1533 1534 // if you are looking for the result of a call to getpid() that 1535 // returns a unique pid for the calling thread, then look at the 1536 // OSThread::thread_id() method in osThread_linux.hpp file 1537 1538 return (int)(_initial_pid ? _initial_pid : getpid()); 1539} 1540 1541// DLL functions 1542 1543const char* os::dll_file_extension() { return ".so"; } 1544 1545// This must be hard coded because it's the system's temporary 1546// directory not the java application's temp directory, ala java.io.tmpdir. 1547const char* os::get_temp_directory() { return "/tmp"; } 1548 1549static bool file_exists(const char* filename) { 1550 struct stat statbuf; 1551 if (filename == NULL || strlen(filename) == 0) { 1552 return false; 1553 } 1554 return os::stat(filename, &statbuf) == 0; 1555} 1556 1557bool os::dll_build_name(char* buffer, size_t buflen, 1558 const char* pname, const char* fname) { 1559 bool retval = false; 1560 // Copied from libhpi 1561 const size_t pnamelen = pname ? strlen(pname) : 0; 1562 1563 // Return error on buffer overflow. 1564 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1565 return retval; 1566 } 1567 1568 if (pnamelen == 0) { 1569 snprintf(buffer, buflen, "lib%s.so", fname); 1570 retval = true; 1571 } else if (strchr(pname, *os::path_separator()) != NULL) { 1572 int n; 1573 char** pelements = split_path(pname, &n); 1574 if (pelements == NULL) { 1575 return false; 1576 } 1577 for (int i = 0; i < n; i++) { 1578 // Really shouldn't be NULL, but check can't hurt 1579 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1580 continue; // skip the empty path values 1581 } 1582 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1583 if (file_exists(buffer)) { 1584 retval = true; 1585 break; 1586 } 1587 } 1588 // release the storage 1589 for (int i = 0; i < n; i++) { 1590 if (pelements[i] != NULL) { 1591 FREE_C_HEAP_ARRAY(char, pelements[i]); 1592 } 1593 } 1594 if (pelements != NULL) { 1595 FREE_C_HEAP_ARRAY(char*, pelements); 1596 } 1597 } else { 1598 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1599 retval = true; 1600 } 1601 return retval; 1602} 1603 1604// check if addr is inside libjvm.so 1605bool os::address_is_in_vm(address addr) { 1606 static address libjvm_base_addr; 1607 Dl_info dlinfo; 1608 1609 if (libjvm_base_addr == NULL) { 1610 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1611 libjvm_base_addr = (address)dlinfo.dli_fbase; 1612 } 1613 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1614 } 1615 1616 if (dladdr((void *)addr, &dlinfo) != 0) { 1617 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1618 } 1619 1620 return false; 1621} 1622 1623bool os::dll_address_to_function_name(address addr, char *buf, 1624 int buflen, int *offset) { 1625 // buf is not optional, but offset is optional 1626 assert(buf != NULL, "sanity check"); 1627 1628 Dl_info dlinfo; 1629 1630 if (dladdr((void*)addr, &dlinfo) != 0) { 1631 // see if we have a matching symbol 1632 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1633 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) { 1634 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1635 } 1636 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1637 return true; 1638 } 1639 // no matching symbol so try for just file info 1640 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1641 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1642 buf, buflen, offset, dlinfo.dli_fname)) { 1643 return true; 1644 } 1645 } 1646 } 1647 1648 buf[0] = '\0'; 1649 if (offset != NULL) *offset = -1; 1650 return false; 1651} 1652 1653struct _address_to_library_name { 1654 address addr; // input : memory address 1655 size_t buflen; // size of fname 1656 char* fname; // output: library name 1657 address base; // library base addr 1658}; 1659 1660static int address_to_library_name_callback(struct dl_phdr_info *info, 1661 size_t size, void *data) { 1662 int i; 1663 bool found = false; 1664 address libbase = NULL; 1665 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1666 1667 // iterate through all loadable segments 1668 for (i = 0; i < info->dlpi_phnum; i++) { 1669 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1670 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1671 // base address of a library is the lowest address of its loaded 1672 // segments. 1673 if (libbase == NULL || libbase > segbase) { 1674 libbase = segbase; 1675 } 1676 // see if 'addr' is within current segment 1677 if (segbase <= d->addr && 1678 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1679 found = true; 1680 } 1681 } 1682 } 1683 1684 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1685 // so dll_address_to_library_name() can fall through to use dladdr() which 1686 // can figure out executable name from argv[0]. 1687 if (found && info->dlpi_name && info->dlpi_name[0]) { 1688 d->base = libbase; 1689 if (d->fname) { 1690 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1691 } 1692 return 1; 1693 } 1694 return 0; 1695} 1696 1697bool os::dll_address_to_library_name(address addr, char* buf, 1698 int buflen, int* offset) { 1699 // buf is not optional, but offset is optional 1700 assert(buf != NULL, "sanity check"); 1701 1702 Dl_info dlinfo; 1703 struct _address_to_library_name data; 1704 1705 // There is a bug in old glibc dladdr() implementation that it could resolve 1706 // to wrong library name if the .so file has a base address != NULL. Here 1707 // we iterate through the program headers of all loaded libraries to find 1708 // out which library 'addr' really belongs to. This workaround can be 1709 // removed once the minimum requirement for glibc is moved to 2.3.x. 1710 data.addr = addr; 1711 data.fname = buf; 1712 data.buflen = buflen; 1713 data.base = NULL; 1714 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1715 1716 if (rslt) { 1717 // buf already contains library name 1718 if (offset) *offset = addr - data.base; 1719 return true; 1720 } 1721 if (dladdr((void*)addr, &dlinfo) != 0) { 1722 if (dlinfo.dli_fname != NULL) { 1723 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1724 } 1725 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1726 *offset = addr - (address)dlinfo.dli_fbase; 1727 } 1728 return true; 1729 } 1730 1731 buf[0] = '\0'; 1732 if (offset) *offset = -1; 1733 return false; 1734} 1735 1736// Loads .dll/.so and 1737// in case of error it checks if .dll/.so was built for the 1738// same architecture as Hotspot is running on 1739 1740 1741// Remember the stack's state. The Linux dynamic linker will change 1742// the stack to 'executable' at most once, so we must safepoint only once. 1743bool os::Linux::_stack_is_executable = false; 1744 1745// VM operation that loads a library. This is necessary if stack protection 1746// of the Java stacks can be lost during loading the library. If we 1747// do not stop the Java threads, they can stack overflow before the stacks 1748// are protected again. 1749class VM_LinuxDllLoad: public VM_Operation { 1750 private: 1751 const char *_filename; 1752 char *_ebuf; 1753 int _ebuflen; 1754 void *_lib; 1755 public: 1756 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1757 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1758 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1759 void doit() { 1760 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1761 os::Linux::_stack_is_executable = true; 1762 } 1763 void* loaded_library() { return _lib; } 1764}; 1765 1766void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1767 void * result = NULL; 1768 bool load_attempted = false; 1769 1770 // Check whether the library to load might change execution rights 1771 // of the stack. If they are changed, the protection of the stack 1772 // guard pages will be lost. We need a safepoint to fix this. 1773 // 1774 // See Linux man page execstack(8) for more info. 1775 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1776 ElfFile ef(filename); 1777 if (!ef.specifies_noexecstack()) { 1778 if (!is_init_completed()) { 1779 os::Linux::_stack_is_executable = true; 1780 // This is OK - No Java threads have been created yet, and hence no 1781 // stack guard pages to fix. 1782 // 1783 // This should happen only when you are building JDK7 using a very 1784 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1785 // 1786 // Dynamic loader will make all stacks executable after 1787 // this function returns, and will not do that again. 1788 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1789 } else { 1790 warning("You have loaded library %s which might have disabled stack guard. " 1791 "The VM will try to fix the stack guard now.\n" 1792 "It's highly recommended that you fix the library with " 1793 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1794 filename); 1795 1796 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1797 JavaThread *jt = JavaThread::current(); 1798 if (jt->thread_state() != _thread_in_native) { 1799 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1800 // that requires ExecStack. Cannot enter safe point. Let's give up. 1801 warning("Unable to fix stack guard. Giving up."); 1802 } else { 1803 if (!LoadExecStackDllInVMThread) { 1804 // This is for the case where the DLL has an static 1805 // constructor function that executes JNI code. We cannot 1806 // load such DLLs in the VMThread. 1807 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1808 } 1809 1810 ThreadInVMfromNative tiv(jt); 1811 debug_only(VMNativeEntryWrapper vew;) 1812 1813 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1814 VMThread::execute(&op); 1815 if (LoadExecStackDllInVMThread) { 1816 result = op.loaded_library(); 1817 } 1818 load_attempted = true; 1819 } 1820 } 1821 } 1822 } 1823 1824 if (!load_attempted) { 1825 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1826 } 1827 1828 if (result != NULL) { 1829 // Successful loading 1830 return result; 1831 } 1832 1833 Elf32_Ehdr elf_head; 1834 int diag_msg_max_length=ebuflen-strlen(ebuf); 1835 char* diag_msg_buf=ebuf+strlen(ebuf); 1836 1837 if (diag_msg_max_length==0) { 1838 // No more space in ebuf for additional diagnostics message 1839 return NULL; 1840 } 1841 1842 1843 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1844 1845 if (file_descriptor < 0) { 1846 // Can't open library, report dlerror() message 1847 return NULL; 1848 } 1849 1850 bool failed_to_read_elf_head= 1851 (sizeof(elf_head)!= 1852 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1853 1854 ::close(file_descriptor); 1855 if (failed_to_read_elf_head) { 1856 // file i/o error - report dlerror() msg 1857 return NULL; 1858 } 1859 1860 typedef struct { 1861 Elf32_Half code; // Actual value as defined in elf.h 1862 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1863 char elf_class; // 32 or 64 bit 1864 char endianess; // MSB or LSB 1865 char* name; // String representation 1866 } arch_t; 1867 1868#ifndef EM_486 1869 #define EM_486 6 /* Intel 80486 */ 1870#endif 1871#ifndef EM_AARCH64 1872 #define EM_AARCH64 183 /* ARM AARCH64 */ 1873#endif 1874 1875 static const arch_t arch_array[]={ 1876 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1877 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1878 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1879 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1880 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1881 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1882 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1883 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1884#if defined(VM_LITTLE_ENDIAN) 1885 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"}, 1886#else 1887 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1888#endif 1889 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1890 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1891 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1892 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1893 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1894 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1895 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1896 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1897 }; 1898 1899#if (defined IA32) 1900 static Elf32_Half running_arch_code=EM_386; 1901#elif (defined AMD64) 1902 static Elf32_Half running_arch_code=EM_X86_64; 1903#elif (defined IA64) 1904 static Elf32_Half running_arch_code=EM_IA_64; 1905#elif (defined __sparc) && (defined _LP64) 1906 static Elf32_Half running_arch_code=EM_SPARCV9; 1907#elif (defined __sparc) && (!defined _LP64) 1908 static Elf32_Half running_arch_code=EM_SPARC; 1909#elif (defined __powerpc64__) 1910 static Elf32_Half running_arch_code=EM_PPC64; 1911#elif (defined __powerpc__) 1912 static Elf32_Half running_arch_code=EM_PPC; 1913#elif (defined ARM) 1914 static Elf32_Half running_arch_code=EM_ARM; 1915#elif (defined S390) 1916 static Elf32_Half running_arch_code=EM_S390; 1917#elif (defined ALPHA) 1918 static Elf32_Half running_arch_code=EM_ALPHA; 1919#elif (defined MIPSEL) 1920 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1921#elif (defined PARISC) 1922 static Elf32_Half running_arch_code=EM_PARISC; 1923#elif (defined MIPS) 1924 static Elf32_Half running_arch_code=EM_MIPS; 1925#elif (defined M68K) 1926 static Elf32_Half running_arch_code=EM_68K; 1927#elif (defined AARCH64) 1928 static Elf32_Half running_arch_code=EM_AARCH64; 1929#else 1930 #error Method os::dll_load requires that one of following is defined:\ 1931 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64 1932#endif 1933 1934 // Identify compatability class for VM's architecture and library's architecture 1935 // Obtain string descriptions for architectures 1936 1937 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1938 int running_arch_index=-1; 1939 1940 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1941 if (running_arch_code == arch_array[i].code) { 1942 running_arch_index = i; 1943 } 1944 if (lib_arch.code == arch_array[i].code) { 1945 lib_arch.compat_class = arch_array[i].compat_class; 1946 lib_arch.name = arch_array[i].name; 1947 } 1948 } 1949 1950 assert(running_arch_index != -1, 1951 "Didn't find running architecture code (running_arch_code) in arch_array"); 1952 if (running_arch_index == -1) { 1953 // Even though running architecture detection failed 1954 // we may still continue with reporting dlerror() message 1955 return NULL; 1956 } 1957 1958 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1959 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1960 return NULL; 1961 } 1962 1963#ifndef S390 1964 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1965 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1966 return NULL; 1967 } 1968#endif // !S390 1969 1970 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1971 if (lib_arch.name!=NULL) { 1972 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1973 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1974 lib_arch.name, arch_array[running_arch_index].name); 1975 } else { 1976 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1977 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1978 lib_arch.code, 1979 arch_array[running_arch_index].name); 1980 } 1981 } 1982 1983 return NULL; 1984} 1985 1986void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 1987 int ebuflen) { 1988 void * result = ::dlopen(filename, RTLD_LAZY); 1989 if (result == NULL) { 1990 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 1991 ebuf[ebuflen-1] = '\0'; 1992 } 1993 return result; 1994} 1995 1996void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 1997 int ebuflen) { 1998 void * result = NULL; 1999 if (LoadExecStackDllInVMThread) { 2000 result = dlopen_helper(filename, ebuf, ebuflen); 2001 } 2002 2003 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2004 // library that requires an executable stack, or which does not have this 2005 // stack attribute set, dlopen changes the stack attribute to executable. The 2006 // read protection of the guard pages gets lost. 2007 // 2008 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2009 // may have been queued at the same time. 2010 2011 if (!_stack_is_executable) { 2012 JavaThread *jt = Threads::first(); 2013 2014 while (jt) { 2015 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2016 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2017 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2018 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2019 warning("Attempt to reguard stack yellow zone failed."); 2020 } 2021 } 2022 jt = jt->next(); 2023 } 2024 } 2025 2026 return result; 2027} 2028 2029void* os::dll_lookup(void* handle, const char* name) { 2030 void* res = dlsym(handle, name); 2031 return res; 2032} 2033 2034void* os::get_default_process_handle() { 2035 return (void*)::dlopen(NULL, RTLD_LAZY); 2036} 2037 2038static bool _print_ascii_file(const char* filename, outputStream* st) { 2039 int fd = ::open(filename, O_RDONLY); 2040 if (fd == -1) { 2041 return false; 2042 } 2043 2044 char buf[32]; 2045 int bytes; 2046 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2047 st->print_raw(buf, bytes); 2048 } 2049 2050 ::close(fd); 2051 2052 return true; 2053} 2054 2055void os::print_dll_info(outputStream *st) { 2056 st->print_cr("Dynamic libraries:"); 2057 2058 char fname[32]; 2059 pid_t pid = os::Linux::gettid(); 2060 2061 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2062 2063 if (!_print_ascii_file(fname, st)) { 2064 st->print("Can not get library information for pid = %d\n", pid); 2065 } 2066} 2067 2068int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 2069 FILE *procmapsFile = NULL; 2070 2071 // Open the procfs maps file for the current process 2072 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 2073 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 2074 char line[PATH_MAX + 100]; 2075 2076 // Read line by line from 'file' 2077 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 2078 u8 base, top, offset, inode; 2079 char permissions[5]; 2080 char device[6]; 2081 char name[PATH_MAX + 1]; 2082 2083 // Parse fields from line 2084 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name); 2085 2086 // Filter by device id '00:00' so that we only get file system mapped files. 2087 if (strcmp(device, "00:00") != 0) { 2088 2089 // Call callback with the fields of interest 2090 if(callback(name, (address)base, (address)top, param)) { 2091 // Oops abort, callback aborted 2092 fclose(procmapsFile); 2093 return 1; 2094 } 2095 } 2096 } 2097 fclose(procmapsFile); 2098 } 2099 return 0; 2100} 2101 2102void os::print_os_info_brief(outputStream* st) { 2103 os::Linux::print_distro_info(st); 2104 2105 os::Posix::print_uname_info(st); 2106 2107 os::Linux::print_libversion_info(st); 2108 2109} 2110 2111void os::print_os_info(outputStream* st) { 2112 st->print("OS:"); 2113 2114 os::Linux::print_distro_info(st); 2115 2116 os::Posix::print_uname_info(st); 2117 2118 // Print warning if unsafe chroot environment detected 2119 if (unsafe_chroot_detected) { 2120 st->print("WARNING!! "); 2121 st->print_cr("%s", unstable_chroot_error); 2122 } 2123 2124 os::Linux::print_libversion_info(st); 2125 2126 os::Posix::print_rlimit_info(st); 2127 2128 os::Posix::print_load_average(st); 2129 2130 os::Linux::print_full_memory_info(st); 2131} 2132 2133// Try to identify popular distros. 2134// Most Linux distributions have a /etc/XXX-release file, which contains 2135// the OS version string. Newer Linux distributions have a /etc/lsb-release 2136// file that also contains the OS version string. Some have more than one 2137// /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2138// /etc/redhat-release.), so the order is important. 2139// Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2140// their own specific XXX-release file as well as a redhat-release file. 2141// Because of this the XXX-release file needs to be searched for before the 2142// redhat-release file. 2143// Since Red Hat has a lsb-release file that is not very descriptive the 2144// search for redhat-release needs to be before lsb-release. 2145// Since the lsb-release file is the new standard it needs to be searched 2146// before the older style release files. 2147// Searching system-release (Red Hat) and os-release (other Linuxes) are a 2148// next to last resort. The os-release file is a new standard that contains 2149// distribution information and the system-release file seems to be an old 2150// standard that has been replaced by the lsb-release and os-release files. 2151// Searching for the debian_version file is the last resort. It contains 2152// an informative string like "6.0.6" or "wheezy/sid". Because of this 2153// "Debian " is printed before the contents of the debian_version file. 2154void os::Linux::print_distro_info(outputStream* st) { 2155 if (!_print_ascii_file("/etc/oracle-release", st) && 2156 !_print_ascii_file("/etc/mandriva-release", st) && 2157 !_print_ascii_file("/etc/mandrake-release", st) && 2158 !_print_ascii_file("/etc/sun-release", st) && 2159 !_print_ascii_file("/etc/redhat-release", st) && 2160 !_print_ascii_file("/etc/lsb-release", st) && 2161 !_print_ascii_file("/etc/SuSE-release", st) && 2162 !_print_ascii_file("/etc/turbolinux-release", st) && 2163 !_print_ascii_file("/etc/gentoo-release", st) && 2164 !_print_ascii_file("/etc/ltib-release", st) && 2165 !_print_ascii_file("/etc/angstrom-version", st) && 2166 !_print_ascii_file("/etc/system-release", st) && 2167 !_print_ascii_file("/etc/os-release", st)) { 2168 2169 if (file_exists("/etc/debian_version")) { 2170 st->print("Debian "); 2171 _print_ascii_file("/etc/debian_version", st); 2172 } else { 2173 st->print("Linux"); 2174 } 2175 } 2176 st->cr(); 2177} 2178 2179void os::Linux::print_libversion_info(outputStream* st) { 2180 // libc, pthread 2181 st->print("libc:"); 2182 st->print("%s ", os::Linux::glibc_version()); 2183 st->print("%s ", os::Linux::libpthread_version()); 2184 if (os::Linux::is_LinuxThreads()) { 2185 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2186 } 2187 st->cr(); 2188} 2189 2190void os::Linux::print_full_memory_info(outputStream* st) { 2191 st->print("\n/proc/meminfo:\n"); 2192 _print_ascii_file("/proc/meminfo", st); 2193 st->cr(); 2194} 2195 2196void os::print_memory_info(outputStream* st) { 2197 2198 st->print("Memory:"); 2199 st->print(" %dk page", os::vm_page_size()>>10); 2200 2201 // values in struct sysinfo are "unsigned long" 2202 struct sysinfo si; 2203 sysinfo(&si); 2204 2205 st->print(", physical " UINT64_FORMAT "k", 2206 os::physical_memory() >> 10); 2207 st->print("(" UINT64_FORMAT "k free)", 2208 os::available_memory() >> 10); 2209 st->print(", swap " UINT64_FORMAT "k", 2210 ((jlong)si.totalswap * si.mem_unit) >> 10); 2211 st->print("(" UINT64_FORMAT "k free)", 2212 ((jlong)si.freeswap * si.mem_unit) >> 10); 2213 st->cr(); 2214} 2215 2216void os::pd_print_cpu_info(outputStream* st) { 2217 st->print("\n/proc/cpuinfo:\n"); 2218 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2219 st->print(" <Not Available>"); 2220 } 2221 st->cr(); 2222} 2223 2224void os::print_siginfo(outputStream* st, void* siginfo) { 2225 const siginfo_t* si = (const siginfo_t*)siginfo; 2226 2227 os::Posix::print_siginfo_brief(st, si); 2228#if INCLUDE_CDS 2229 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2230 UseSharedSpaces) { 2231 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2232 if (mapinfo->is_in_shared_space(si->si_addr)) { 2233 st->print("\n\nError accessing class data sharing archive." \ 2234 " Mapped file inaccessible during execution, " \ 2235 " possible disk/network problem."); 2236 } 2237 } 2238#endif 2239 st->cr(); 2240} 2241 2242 2243static void print_signal_handler(outputStream* st, int sig, 2244 char* buf, size_t buflen); 2245 2246void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2247 st->print_cr("Signal Handlers:"); 2248 print_signal_handler(st, SIGSEGV, buf, buflen); 2249 print_signal_handler(st, SIGBUS , buf, buflen); 2250 print_signal_handler(st, SIGFPE , buf, buflen); 2251 print_signal_handler(st, SIGPIPE, buf, buflen); 2252 print_signal_handler(st, SIGXFSZ, buf, buflen); 2253 print_signal_handler(st, SIGILL , buf, buflen); 2254 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2255 print_signal_handler(st, SR_signum, buf, buflen); 2256 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2257 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2258 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2259 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2260#if defined(PPC64) 2261 print_signal_handler(st, SIGTRAP, buf, buflen); 2262#endif 2263} 2264 2265static char saved_jvm_path[MAXPATHLEN] = {0}; 2266 2267// Find the full path to the current module, libjvm.so 2268void os::jvm_path(char *buf, jint buflen) { 2269 // Error checking. 2270 if (buflen < MAXPATHLEN) { 2271 assert(false, "must use a large-enough buffer"); 2272 buf[0] = '\0'; 2273 return; 2274 } 2275 // Lazy resolve the path to current module. 2276 if (saved_jvm_path[0] != 0) { 2277 strcpy(buf, saved_jvm_path); 2278 return; 2279 } 2280 2281 char dli_fname[MAXPATHLEN]; 2282 bool ret = dll_address_to_library_name( 2283 CAST_FROM_FN_PTR(address, os::jvm_path), 2284 dli_fname, sizeof(dli_fname), NULL); 2285 assert(ret, "cannot locate libjvm"); 2286 char *rp = NULL; 2287 if (ret && dli_fname[0] != '\0') { 2288 rp = realpath(dli_fname, buf); 2289 } 2290 if (rp == NULL) { 2291 return; 2292 } 2293 2294 if (Arguments::sun_java_launcher_is_altjvm()) { 2295 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2296 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2297 // If "/jre/lib/" appears at the right place in the string, then 2298 // assume we are installed in a JDK and we're done. Otherwise, check 2299 // for a JAVA_HOME environment variable and fix up the path so it 2300 // looks like libjvm.so is installed there (append a fake suffix 2301 // hotspot/libjvm.so). 2302 const char *p = buf + strlen(buf) - 1; 2303 for (int count = 0; p > buf && count < 5; ++count) { 2304 for (--p; p > buf && *p != '/'; --p) 2305 /* empty */ ; 2306 } 2307 2308 if (strncmp(p, "/jre/lib/", 9) != 0) { 2309 // Look for JAVA_HOME in the environment. 2310 char* java_home_var = ::getenv("JAVA_HOME"); 2311 if (java_home_var != NULL && java_home_var[0] != 0) { 2312 char* jrelib_p; 2313 int len; 2314 2315 // Check the current module name "libjvm.so". 2316 p = strrchr(buf, '/'); 2317 if (p == NULL) { 2318 return; 2319 } 2320 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2321 2322 rp = realpath(java_home_var, buf); 2323 if (rp == NULL) { 2324 return; 2325 } 2326 2327 // determine if this is a legacy image or modules image 2328 // modules image doesn't have "jre" subdirectory 2329 len = strlen(buf); 2330 assert(len < buflen, "Ran out of buffer room"); 2331 jrelib_p = buf + len; 2332 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2333 if (0 != access(buf, F_OK)) { 2334 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2335 } 2336 2337 if (0 == access(buf, F_OK)) { 2338 // Use current module name "libjvm.so" 2339 len = strlen(buf); 2340 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2341 } else { 2342 // Go back to path of .so 2343 rp = realpath(dli_fname, buf); 2344 if (rp == NULL) { 2345 return; 2346 } 2347 } 2348 } 2349 } 2350 } 2351 2352 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2353 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2354} 2355 2356void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2357 // no prefix required, not even "_" 2358} 2359 2360void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2361 // no suffix required 2362} 2363 2364//////////////////////////////////////////////////////////////////////////////// 2365// sun.misc.Signal support 2366 2367static volatile jint sigint_count = 0; 2368 2369static void UserHandler(int sig, void *siginfo, void *context) { 2370 // 4511530 - sem_post is serialized and handled by the manager thread. When 2371 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2372 // don't want to flood the manager thread with sem_post requests. 2373 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2374 return; 2375 } 2376 2377 // Ctrl-C is pressed during error reporting, likely because the error 2378 // handler fails to abort. Let VM die immediately. 2379 if (sig == SIGINT && is_error_reported()) { 2380 os::die(); 2381 } 2382 2383 os::signal_notify(sig); 2384} 2385 2386void* os::user_handler() { 2387 return CAST_FROM_FN_PTR(void*, UserHandler); 2388} 2389 2390class Semaphore : public StackObj { 2391 public: 2392 Semaphore(); 2393 ~Semaphore(); 2394 void signal(); 2395 void wait(); 2396 bool trywait(); 2397 bool timedwait(unsigned int sec, int nsec); 2398 private: 2399 sem_t _semaphore; 2400}; 2401 2402Semaphore::Semaphore() { 2403 sem_init(&_semaphore, 0, 0); 2404} 2405 2406Semaphore::~Semaphore() { 2407 sem_destroy(&_semaphore); 2408} 2409 2410void Semaphore::signal() { 2411 sem_post(&_semaphore); 2412} 2413 2414void Semaphore::wait() { 2415 sem_wait(&_semaphore); 2416} 2417 2418bool Semaphore::trywait() { 2419 return sem_trywait(&_semaphore) == 0; 2420} 2421 2422bool Semaphore::timedwait(unsigned int sec, int nsec) { 2423 2424 struct timespec ts; 2425 // Semaphore's are always associated with CLOCK_REALTIME 2426 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2427 // see unpackTime for discussion on overflow checking 2428 if (sec >= MAX_SECS) { 2429 ts.tv_sec += MAX_SECS; 2430 ts.tv_nsec = 0; 2431 } else { 2432 ts.tv_sec += sec; 2433 ts.tv_nsec += nsec; 2434 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2435 ts.tv_nsec -= NANOSECS_PER_SEC; 2436 ++ts.tv_sec; // note: this must be <= max_secs 2437 } 2438 } 2439 2440 while (1) { 2441 int result = sem_timedwait(&_semaphore, &ts); 2442 if (result == 0) { 2443 return true; 2444 } else if (errno == EINTR) { 2445 continue; 2446 } else if (errno == ETIMEDOUT) { 2447 return false; 2448 } else { 2449 return false; 2450 } 2451 } 2452} 2453 2454extern "C" { 2455 typedef void (*sa_handler_t)(int); 2456 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2457} 2458 2459void* os::signal(int signal_number, void* handler) { 2460 struct sigaction sigAct, oldSigAct; 2461 2462 sigfillset(&(sigAct.sa_mask)); 2463 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2464 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2465 2466 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2467 // -1 means registration failed 2468 return (void *)-1; 2469 } 2470 2471 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2472} 2473 2474void os::signal_raise(int signal_number) { 2475 ::raise(signal_number); 2476} 2477 2478// The following code is moved from os.cpp for making this 2479// code platform specific, which it is by its very nature. 2480 2481// Will be modified when max signal is changed to be dynamic 2482int os::sigexitnum_pd() { 2483 return NSIG; 2484} 2485 2486// a counter for each possible signal value 2487static volatile jint pending_signals[NSIG+1] = { 0 }; 2488 2489// Linux(POSIX) specific hand shaking semaphore. 2490static sem_t sig_sem; 2491static Semaphore sr_semaphore; 2492 2493void os::signal_init_pd() { 2494 // Initialize signal structures 2495 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2496 2497 // Initialize signal semaphore 2498 ::sem_init(&sig_sem, 0, 0); 2499} 2500 2501void os::signal_notify(int sig) { 2502 Atomic::inc(&pending_signals[sig]); 2503 ::sem_post(&sig_sem); 2504} 2505 2506static int check_pending_signals(bool wait) { 2507 Atomic::store(0, &sigint_count); 2508 for (;;) { 2509 for (int i = 0; i < NSIG + 1; i++) { 2510 jint n = pending_signals[i]; 2511 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2512 return i; 2513 } 2514 } 2515 if (!wait) { 2516 return -1; 2517 } 2518 JavaThread *thread = JavaThread::current(); 2519 ThreadBlockInVM tbivm(thread); 2520 2521 bool threadIsSuspended; 2522 do { 2523 thread->set_suspend_equivalent(); 2524 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2525 ::sem_wait(&sig_sem); 2526 2527 // were we externally suspended while we were waiting? 2528 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2529 if (threadIsSuspended) { 2530 // The semaphore has been incremented, but while we were waiting 2531 // another thread suspended us. We don't want to continue running 2532 // while suspended because that would surprise the thread that 2533 // suspended us. 2534 ::sem_post(&sig_sem); 2535 2536 thread->java_suspend_self(); 2537 } 2538 } while (threadIsSuspended); 2539 } 2540} 2541 2542int os::signal_lookup() { 2543 return check_pending_signals(false); 2544} 2545 2546int os::signal_wait() { 2547 return check_pending_signals(true); 2548} 2549 2550//////////////////////////////////////////////////////////////////////////////// 2551// Virtual Memory 2552 2553int os::vm_page_size() { 2554 // Seems redundant as all get out 2555 assert(os::Linux::page_size() != -1, "must call os::init"); 2556 return os::Linux::page_size(); 2557} 2558 2559// Solaris allocates memory by pages. 2560int os::vm_allocation_granularity() { 2561 assert(os::Linux::page_size() != -1, "must call os::init"); 2562 return os::Linux::page_size(); 2563} 2564 2565// Rationale behind this function: 2566// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2567// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2568// samples for JITted code. Here we create private executable mapping over the code cache 2569// and then we can use standard (well, almost, as mapping can change) way to provide 2570// info for the reporting script by storing timestamp and location of symbol 2571void linux_wrap_code(char* base, size_t size) { 2572 static volatile jint cnt = 0; 2573 2574 if (!UseOprofile) { 2575 return; 2576 } 2577 2578 char buf[PATH_MAX+1]; 2579 int num = Atomic::add(1, &cnt); 2580 2581 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2582 os::get_temp_directory(), os::current_process_id(), num); 2583 unlink(buf); 2584 2585 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2586 2587 if (fd != -1) { 2588 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2589 if (rv != (off_t)-1) { 2590 if (::write(fd, "", 1) == 1) { 2591 mmap(base, size, 2592 PROT_READ|PROT_WRITE|PROT_EXEC, 2593 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2594 } 2595 } 2596 ::close(fd); 2597 unlink(buf); 2598 } 2599} 2600 2601static bool recoverable_mmap_error(int err) { 2602 // See if the error is one we can let the caller handle. This 2603 // list of errno values comes from JBS-6843484. I can't find a 2604 // Linux man page that documents this specific set of errno 2605 // values so while this list currently matches Solaris, it may 2606 // change as we gain experience with this failure mode. 2607 switch (err) { 2608 case EBADF: 2609 case EINVAL: 2610 case ENOTSUP: 2611 // let the caller deal with these errors 2612 return true; 2613 2614 default: 2615 // Any remaining errors on this OS can cause our reserved mapping 2616 // to be lost. That can cause confusion where different data 2617 // structures think they have the same memory mapped. The worst 2618 // scenario is if both the VM and a library think they have the 2619 // same memory mapped. 2620 return false; 2621 } 2622} 2623 2624static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2625 int err) { 2626 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2627 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2628 strerror(err), err); 2629} 2630 2631static void warn_fail_commit_memory(char* addr, size_t size, 2632 size_t alignment_hint, bool exec, 2633 int err) { 2634 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2635 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2636 alignment_hint, exec, strerror(err), err); 2637} 2638 2639// NOTE: Linux kernel does not really reserve the pages for us. 2640// All it does is to check if there are enough free pages 2641// left at the time of mmap(). This could be a potential 2642// problem. 2643int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2644 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2645 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2646 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2647 if (res != (uintptr_t) MAP_FAILED) { 2648 if (UseNUMAInterleaving) { 2649 numa_make_global(addr, size); 2650 } 2651 return 0; 2652 } 2653 2654 int err = errno; // save errno from mmap() call above 2655 2656 if (!recoverable_mmap_error(err)) { 2657 warn_fail_commit_memory(addr, size, exec, err); 2658 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2659 } 2660 2661 return err; 2662} 2663 2664bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2665 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2666} 2667 2668void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2669 const char* mesg) { 2670 assert(mesg != NULL, "mesg must be specified"); 2671 int err = os::Linux::commit_memory_impl(addr, size, exec); 2672 if (err != 0) { 2673 // the caller wants all commit errors to exit with the specified mesg: 2674 warn_fail_commit_memory(addr, size, exec, err); 2675 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2676 } 2677} 2678 2679// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2680#ifndef MAP_HUGETLB 2681 #define MAP_HUGETLB 0x40000 2682#endif 2683 2684// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2685#ifndef MADV_HUGEPAGE 2686 #define MADV_HUGEPAGE 14 2687#endif 2688 2689int os::Linux::commit_memory_impl(char* addr, size_t size, 2690 size_t alignment_hint, bool exec) { 2691 int err = os::Linux::commit_memory_impl(addr, size, exec); 2692 if (err == 0) { 2693 realign_memory(addr, size, alignment_hint); 2694 } 2695 return err; 2696} 2697 2698bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2699 bool exec) { 2700 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2701} 2702 2703void os::pd_commit_memory_or_exit(char* addr, size_t size, 2704 size_t alignment_hint, bool exec, 2705 const char* mesg) { 2706 assert(mesg != NULL, "mesg must be specified"); 2707 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2708 if (err != 0) { 2709 // the caller wants all commit errors to exit with the specified mesg: 2710 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2711 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2712 } 2713} 2714 2715void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2716 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2717 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2718 // be supported or the memory may already be backed by huge pages. 2719 ::madvise(addr, bytes, MADV_HUGEPAGE); 2720 } 2721} 2722 2723void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2724 // This method works by doing an mmap over an existing mmaping and effectively discarding 2725 // the existing pages. However it won't work for SHM-based large pages that cannot be 2726 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2727 // small pages on top of the SHM segment. This method always works for small pages, so we 2728 // allow that in any case. 2729 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2730 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2731 } 2732} 2733 2734void os::numa_make_global(char *addr, size_t bytes) { 2735 Linux::numa_interleave_memory(addr, bytes); 2736} 2737 2738// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2739// bind policy to MPOL_PREFERRED for the current thread. 2740#define USE_MPOL_PREFERRED 0 2741 2742void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2743 // To make NUMA and large pages more robust when both enabled, we need to ease 2744 // the requirements on where the memory should be allocated. MPOL_BIND is the 2745 // default policy and it will force memory to be allocated on the specified 2746 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2747 // the specified node, but will not force it. Using this policy will prevent 2748 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2749 // free large pages. 2750 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2751 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2752} 2753 2754bool os::numa_topology_changed() { return false; } 2755 2756size_t os::numa_get_groups_num() { 2757 int max_node = Linux::numa_max_node(); 2758 return max_node > 0 ? max_node + 1 : 1; 2759} 2760 2761int os::numa_get_group_id() { 2762 int cpu_id = Linux::sched_getcpu(); 2763 if (cpu_id != -1) { 2764 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2765 if (lgrp_id != -1) { 2766 return lgrp_id; 2767 } 2768 } 2769 return 0; 2770} 2771 2772size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2773 for (size_t i = 0; i < size; i++) { 2774 ids[i] = i; 2775 } 2776 return size; 2777} 2778 2779bool os::get_page_info(char *start, page_info* info) { 2780 return false; 2781} 2782 2783char *os::scan_pages(char *start, char* end, page_info* page_expected, 2784 page_info* page_found) { 2785 return end; 2786} 2787 2788 2789int os::Linux::sched_getcpu_syscall(void) { 2790 unsigned int cpu; 2791 int retval = -1; 2792 2793#if defined(IA32) 2794 #ifndef SYS_getcpu 2795 #define SYS_getcpu 318 2796 #endif 2797 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2798#elif defined(AMD64) 2799// Unfortunately we have to bring all these macros here from vsyscall.h 2800// to be able to compile on old linuxes. 2801 #define __NR_vgetcpu 2 2802 #define VSYSCALL_START (-10UL << 20) 2803 #define VSYSCALL_SIZE 1024 2804 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2805 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2806 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2807 retval = vgetcpu(&cpu, NULL, NULL); 2808#endif 2809 2810 return (retval == -1) ? retval : cpu; 2811} 2812 2813// Something to do with the numa-aware allocator needs these symbols 2814extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2815extern "C" JNIEXPORT void numa_error(char *where) { } 2816extern "C" JNIEXPORT int fork1() { return fork(); } 2817 2818 2819// If we are running with libnuma version > 2, then we should 2820// be trying to use symbols with versions 1.1 2821// If we are running with earlier version, which did not have symbol versions, 2822// we should use the base version. 2823void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2824 void *f = dlvsym(handle, name, "libnuma_1.1"); 2825 if (f == NULL) { 2826 f = dlsym(handle, name); 2827 } 2828 return f; 2829} 2830 2831bool os::Linux::libnuma_init() { 2832 // sched_getcpu() should be in libc. 2833 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2834 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2835 2836 // If it's not, try a direct syscall. 2837 if (sched_getcpu() == -1) { 2838 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2839 (void*)&sched_getcpu_syscall)); 2840 } 2841 2842 if (sched_getcpu() != -1) { // Does it work? 2843 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2844 if (handle != NULL) { 2845 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2846 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2847 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2848 libnuma_dlsym(handle, "numa_max_node"))); 2849 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2850 libnuma_dlsym(handle, "numa_available"))); 2851 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2852 libnuma_dlsym(handle, "numa_tonode_memory"))); 2853 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2854 libnuma_dlsym(handle, "numa_interleave_memory"))); 2855 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2856 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2857 2858 2859 if (numa_available() != -1) { 2860 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2861 // Create a cpu -> node mapping 2862 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2863 rebuild_cpu_to_node_map(); 2864 return true; 2865 } 2866 } 2867 } 2868 return false; 2869} 2870 2871// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2872// The table is later used in get_node_by_cpu(). 2873void os::Linux::rebuild_cpu_to_node_map() { 2874 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2875 // in libnuma (possible values are starting from 16, 2876 // and continuing up with every other power of 2, but less 2877 // than the maximum number of CPUs supported by kernel), and 2878 // is a subject to change (in libnuma version 2 the requirements 2879 // are more reasonable) we'll just hardcode the number they use 2880 // in the library. 2881 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2882 2883 size_t cpu_num = os::active_processor_count(); 2884 size_t cpu_map_size = NCPUS / BitsPerCLong; 2885 size_t cpu_map_valid_size = 2886 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2887 2888 cpu_to_node()->clear(); 2889 cpu_to_node()->at_grow(cpu_num - 1); 2890 size_t node_num = numa_get_groups_num(); 2891 2892 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2893 for (size_t i = 0; i < node_num; i++) { 2894 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2895 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2896 if (cpu_map[j] != 0) { 2897 for (size_t k = 0; k < BitsPerCLong; k++) { 2898 if (cpu_map[j] & (1UL << k)) { 2899 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2900 } 2901 } 2902 } 2903 } 2904 } 2905 } 2906 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2907} 2908 2909int os::Linux::get_node_by_cpu(int cpu_id) { 2910 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2911 return cpu_to_node()->at(cpu_id); 2912 } 2913 return -1; 2914} 2915 2916GrowableArray<int>* os::Linux::_cpu_to_node; 2917os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2918os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2919os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2920os::Linux::numa_available_func_t os::Linux::_numa_available; 2921os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2922os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2923os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2924unsigned long* os::Linux::_numa_all_nodes; 2925 2926bool os::pd_uncommit_memory(char* addr, size_t size) { 2927 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2928 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2929 return res != (uintptr_t) MAP_FAILED; 2930} 2931 2932static address get_stack_commited_bottom(address bottom, size_t size) { 2933 address nbot = bottom; 2934 address ntop = bottom + size; 2935 2936 size_t page_sz = os::vm_page_size(); 2937 unsigned pages = size / page_sz; 2938 2939 unsigned char vec[1]; 2940 unsigned imin = 1, imax = pages + 1, imid; 2941 int mincore_return_value = 0; 2942 2943 assert(imin <= imax, "Unexpected page size"); 2944 2945 while (imin < imax) { 2946 imid = (imax + imin) / 2; 2947 nbot = ntop - (imid * page_sz); 2948 2949 // Use a trick with mincore to check whether the page is mapped or not. 2950 // mincore sets vec to 1 if page resides in memory and to 0 if page 2951 // is swapped output but if page we are asking for is unmapped 2952 // it returns -1,ENOMEM 2953 mincore_return_value = mincore(nbot, page_sz, vec); 2954 2955 if (mincore_return_value == -1) { 2956 // Page is not mapped go up 2957 // to find first mapped page 2958 if (errno != EAGAIN) { 2959 assert(errno == ENOMEM, "Unexpected mincore errno"); 2960 imax = imid; 2961 } 2962 } else { 2963 // Page is mapped go down 2964 // to find first not mapped page 2965 imin = imid + 1; 2966 } 2967 } 2968 2969 nbot = nbot + page_sz; 2970 2971 // Adjust stack bottom one page up if last checked page is not mapped 2972 if (mincore_return_value == -1) { 2973 nbot = nbot + page_sz; 2974 } 2975 2976 return nbot; 2977} 2978 2979 2980// Linux uses a growable mapping for the stack, and if the mapping for 2981// the stack guard pages is not removed when we detach a thread the 2982// stack cannot grow beyond the pages where the stack guard was 2983// mapped. If at some point later in the process the stack expands to 2984// that point, the Linux kernel cannot expand the stack any further 2985// because the guard pages are in the way, and a segfault occurs. 2986// 2987// However, it's essential not to split the stack region by unmapping 2988// a region (leaving a hole) that's already part of the stack mapping, 2989// so if the stack mapping has already grown beyond the guard pages at 2990// the time we create them, we have to truncate the stack mapping. 2991// So, we need to know the extent of the stack mapping when 2992// create_stack_guard_pages() is called. 2993 2994// We only need this for stacks that are growable: at the time of 2995// writing thread stacks don't use growable mappings (i.e. those 2996// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 2997// only applies to the main thread. 2998 2999// If the (growable) stack mapping already extends beyond the point 3000// where we're going to put our guard pages, truncate the mapping at 3001// that point by munmap()ping it. This ensures that when we later 3002// munmap() the guard pages we don't leave a hole in the stack 3003// mapping. This only affects the main/initial thread 3004 3005bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3006 if (os::Linux::is_initial_thread()) { 3007 // As we manually grow stack up to bottom inside create_attached_thread(), 3008 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3009 // we don't need to do anything special. 3010 // Check it first, before calling heavy function. 3011 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3012 unsigned char vec[1]; 3013 3014 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3015 // Fallback to slow path on all errors, including EAGAIN 3016 stack_extent = (uintptr_t) get_stack_commited_bottom( 3017 os::Linux::initial_thread_stack_bottom(), 3018 (size_t)addr - stack_extent); 3019 } 3020 3021 if (stack_extent < (uintptr_t)addr) { 3022 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3023 } 3024 } 3025 3026 return os::commit_memory(addr, size, !ExecMem); 3027} 3028 3029// If this is a growable mapping, remove the guard pages entirely by 3030// munmap()ping them. If not, just call uncommit_memory(). This only 3031// affects the main/initial thread, but guard against future OS changes 3032// It's safe to always unmap guard pages for initial thread because we 3033// always place it right after end of the mapped region 3034 3035bool os::remove_stack_guard_pages(char* addr, size_t size) { 3036 uintptr_t stack_extent, stack_base; 3037 3038 if (os::Linux::is_initial_thread()) { 3039 return ::munmap(addr, size) == 0; 3040 } 3041 3042 return os::uncommit_memory(addr, size); 3043} 3044 3045static address _highest_vm_reserved_address = NULL; 3046 3047// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3048// at 'requested_addr'. If there are existing memory mappings at the same 3049// location, however, they will be overwritten. If 'fixed' is false, 3050// 'requested_addr' is only treated as a hint, the return value may or 3051// may not start from the requested address. Unlike Linux mmap(), this 3052// function returns NULL to indicate failure. 3053static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3054 char * addr; 3055 int flags; 3056 3057 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3058 if (fixed) { 3059 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3060 flags |= MAP_FIXED; 3061 } 3062 3063 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3064 // touch an uncommitted page. Otherwise, the read/write might 3065 // succeed if we have enough swap space to back the physical page. 3066 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3067 flags, -1, 0); 3068 3069 if (addr != MAP_FAILED) { 3070 // anon_mmap() should only get called during VM initialization, 3071 // don't need lock (actually we can skip locking even it can be called 3072 // from multiple threads, because _highest_vm_reserved_address is just a 3073 // hint about the upper limit of non-stack memory regions.) 3074 if ((address)addr + bytes > _highest_vm_reserved_address) { 3075 _highest_vm_reserved_address = (address)addr + bytes; 3076 } 3077 } 3078 3079 return addr == MAP_FAILED ? NULL : addr; 3080} 3081 3082// Don't update _highest_vm_reserved_address, because there might be memory 3083// regions above addr + size. If so, releasing a memory region only creates 3084// a hole in the address space, it doesn't help prevent heap-stack collision. 3085// 3086static int anon_munmap(char * addr, size_t size) { 3087 return ::munmap(addr, size) == 0; 3088} 3089 3090char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3091 size_t alignment_hint) { 3092 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3093} 3094 3095bool os::pd_release_memory(char* addr, size_t size) { 3096 return anon_munmap(addr, size); 3097} 3098 3099static address highest_vm_reserved_address() { 3100 return _highest_vm_reserved_address; 3101} 3102 3103static bool linux_mprotect(char* addr, size_t size, int prot) { 3104 // Linux wants the mprotect address argument to be page aligned. 3105 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3106 3107 // According to SUSv3, mprotect() should only be used with mappings 3108 // established by mmap(), and mmap() always maps whole pages. Unaligned 3109 // 'addr' likely indicates problem in the VM (e.g. trying to change 3110 // protection of malloc'ed or statically allocated memory). Check the 3111 // caller if you hit this assert. 3112 assert(addr == bottom, "sanity check"); 3113 3114 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3115 return ::mprotect(bottom, size, prot) == 0; 3116} 3117 3118// Set protections specified 3119bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3120 bool is_committed) { 3121 unsigned int p = 0; 3122 switch (prot) { 3123 case MEM_PROT_NONE: p = PROT_NONE; break; 3124 case MEM_PROT_READ: p = PROT_READ; break; 3125 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3126 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3127 default: 3128 ShouldNotReachHere(); 3129 } 3130 // is_committed is unused. 3131 return linux_mprotect(addr, bytes, p); 3132} 3133 3134bool os::guard_memory(char* addr, size_t size) { 3135 return linux_mprotect(addr, size, PROT_NONE); 3136} 3137 3138bool os::unguard_memory(char* addr, size_t size) { 3139 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3140} 3141 3142bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3143 size_t page_size) { 3144 bool result = false; 3145 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3146 MAP_ANONYMOUS|MAP_PRIVATE, 3147 -1, 0); 3148 if (p != MAP_FAILED) { 3149 void *aligned_p = align_ptr_up(p, page_size); 3150 3151 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3152 3153 munmap(p, page_size * 2); 3154 } 3155 3156 if (warn && !result) { 3157 warning("TransparentHugePages is not supported by the operating system."); 3158 } 3159 3160 return result; 3161} 3162 3163bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3164 bool result = false; 3165 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3166 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3167 -1, 0); 3168 3169 if (p != MAP_FAILED) { 3170 // We don't know if this really is a huge page or not. 3171 FILE *fp = fopen("/proc/self/maps", "r"); 3172 if (fp) { 3173 while (!feof(fp)) { 3174 char chars[257]; 3175 long x = 0; 3176 if (fgets(chars, sizeof(chars), fp)) { 3177 if (sscanf(chars, "%lx-%*x", &x) == 1 3178 && x == (long)p) { 3179 if (strstr (chars, "hugepage")) { 3180 result = true; 3181 break; 3182 } 3183 } 3184 } 3185 } 3186 fclose(fp); 3187 } 3188 munmap(p, page_size); 3189 } 3190 3191 if (warn && !result) { 3192 warning("HugeTLBFS is not supported by the operating system."); 3193 } 3194 3195 return result; 3196} 3197 3198// Set the coredump_filter bits to include largepages in core dump (bit 6) 3199// 3200// From the coredump_filter documentation: 3201// 3202// - (bit 0) anonymous private memory 3203// - (bit 1) anonymous shared memory 3204// - (bit 2) file-backed private memory 3205// - (bit 3) file-backed shared memory 3206// - (bit 4) ELF header pages in file-backed private memory areas (it is 3207// effective only if the bit 2 is cleared) 3208// - (bit 5) hugetlb private memory 3209// - (bit 6) hugetlb shared memory 3210// 3211static void set_coredump_filter(void) { 3212 FILE *f; 3213 long cdm; 3214 3215 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3216 return; 3217 } 3218 3219 if (fscanf(f, "%lx", &cdm) != 1) { 3220 fclose(f); 3221 return; 3222 } 3223 3224 rewind(f); 3225 3226 if ((cdm & LARGEPAGES_BIT) == 0) { 3227 cdm |= LARGEPAGES_BIT; 3228 fprintf(f, "%#lx", cdm); 3229 } 3230 3231 fclose(f); 3232} 3233 3234// Large page support 3235 3236static size_t _large_page_size = 0; 3237 3238size_t os::Linux::find_large_page_size() { 3239 size_t large_page_size = 0; 3240 3241 // large_page_size on Linux is used to round up heap size. x86 uses either 3242 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3243 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3244 // page as large as 256M. 3245 // 3246 // Here we try to figure out page size by parsing /proc/meminfo and looking 3247 // for a line with the following format: 3248 // Hugepagesize: 2048 kB 3249 // 3250 // If we can't determine the value (e.g. /proc is not mounted, or the text 3251 // format has been changed), we'll use the largest page size supported by 3252 // the processor. 3253 3254#ifndef ZERO 3255 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3256 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M); 3257#endif // ZERO 3258 3259 FILE *fp = fopen("/proc/meminfo", "r"); 3260 if (fp) { 3261 while (!feof(fp)) { 3262 int x = 0; 3263 char buf[16]; 3264 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3265 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3266 large_page_size = x * K; 3267 break; 3268 } 3269 } else { 3270 // skip to next line 3271 for (;;) { 3272 int ch = fgetc(fp); 3273 if (ch == EOF || ch == (int)'\n') break; 3274 } 3275 } 3276 } 3277 fclose(fp); 3278 } 3279 3280 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3281 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3282 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3283 proper_unit_for_byte_size(large_page_size)); 3284 } 3285 3286 return large_page_size; 3287} 3288 3289size_t os::Linux::setup_large_page_size() { 3290 _large_page_size = Linux::find_large_page_size(); 3291 const size_t default_page_size = (size_t)Linux::page_size(); 3292 if (_large_page_size > default_page_size) { 3293 _page_sizes[0] = _large_page_size; 3294 _page_sizes[1] = default_page_size; 3295 _page_sizes[2] = 0; 3296 } 3297 3298 return _large_page_size; 3299} 3300 3301bool os::Linux::setup_large_page_type(size_t page_size) { 3302 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3303 FLAG_IS_DEFAULT(UseSHM) && 3304 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3305 3306 // The type of large pages has not been specified by the user. 3307 3308 // Try UseHugeTLBFS and then UseSHM. 3309 UseHugeTLBFS = UseSHM = true; 3310 3311 // Don't try UseTransparentHugePages since there are known 3312 // performance issues with it turned on. This might change in the future. 3313 UseTransparentHugePages = false; 3314 } 3315 3316 if (UseTransparentHugePages) { 3317 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3318 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3319 UseHugeTLBFS = false; 3320 UseSHM = false; 3321 return true; 3322 } 3323 UseTransparentHugePages = false; 3324 } 3325 3326 if (UseHugeTLBFS) { 3327 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3328 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3329 UseSHM = false; 3330 return true; 3331 } 3332 UseHugeTLBFS = false; 3333 } 3334 3335 return UseSHM; 3336} 3337 3338void os::large_page_init() { 3339 if (!UseLargePages && 3340 !UseTransparentHugePages && 3341 !UseHugeTLBFS && 3342 !UseSHM) { 3343 // Not using large pages. 3344 return; 3345 } 3346 3347 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3348 // The user explicitly turned off large pages. 3349 // Ignore the rest of the large pages flags. 3350 UseTransparentHugePages = false; 3351 UseHugeTLBFS = false; 3352 UseSHM = false; 3353 return; 3354 } 3355 3356 size_t large_page_size = Linux::setup_large_page_size(); 3357 UseLargePages = Linux::setup_large_page_type(large_page_size); 3358 3359 set_coredump_filter(); 3360} 3361 3362#ifndef SHM_HUGETLB 3363 #define SHM_HUGETLB 04000 3364#endif 3365 3366char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3367 char* req_addr, bool exec) { 3368 // "exec" is passed in but not used. Creating the shared image for 3369 // the code cache doesn't have an SHM_X executable permission to check. 3370 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3371 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3372 3373 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3374 return NULL; // Fallback to small pages. 3375 } 3376 3377 key_t key = IPC_PRIVATE; 3378 char *addr; 3379 3380 bool warn_on_failure = UseLargePages && 3381 (!FLAG_IS_DEFAULT(UseLargePages) || 3382 !FLAG_IS_DEFAULT(UseSHM) || 3383 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3384 char msg[128]; 3385 3386 // Create a large shared memory region to attach to based on size. 3387 // Currently, size is the total size of the heap 3388 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3389 if (shmid == -1) { 3390 // Possible reasons for shmget failure: 3391 // 1. shmmax is too small for Java heap. 3392 // > check shmmax value: cat /proc/sys/kernel/shmmax 3393 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3394 // 2. not enough large page memory. 3395 // > check available large pages: cat /proc/meminfo 3396 // > increase amount of large pages: 3397 // echo new_value > /proc/sys/vm/nr_hugepages 3398 // Note 1: different Linux may use different name for this property, 3399 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3400 // Note 2: it's possible there's enough physical memory available but 3401 // they are so fragmented after a long run that they can't 3402 // coalesce into large pages. Try to reserve large pages when 3403 // the system is still "fresh". 3404 if (warn_on_failure) { 3405 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3406 warning("%s", msg); 3407 } 3408 return NULL; 3409 } 3410 3411 // attach to the region 3412 addr = (char*)shmat(shmid, req_addr, 0); 3413 int err = errno; 3414 3415 // Remove shmid. If shmat() is successful, the actual shared memory segment 3416 // will be deleted when it's detached by shmdt() or when the process 3417 // terminates. If shmat() is not successful this will remove the shared 3418 // segment immediately. 3419 shmctl(shmid, IPC_RMID, NULL); 3420 3421 if ((intptr_t)addr == -1) { 3422 if (warn_on_failure) { 3423 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3424 warning("%s", msg); 3425 } 3426 return NULL; 3427 } 3428 3429 return addr; 3430} 3431 3432static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3433 int error) { 3434 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3435 3436 bool warn_on_failure = UseLargePages && 3437 (!FLAG_IS_DEFAULT(UseLargePages) || 3438 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3439 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3440 3441 if (warn_on_failure) { 3442 char msg[128]; 3443 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3444 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3445 warning("%s", msg); 3446 } 3447} 3448 3449char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3450 char* req_addr, 3451 bool exec) { 3452 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3453 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3454 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3455 3456 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3457 char* addr = (char*)::mmap(req_addr, bytes, prot, 3458 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3459 -1, 0); 3460 3461 if (addr == MAP_FAILED) { 3462 warn_on_large_pages_failure(req_addr, bytes, errno); 3463 return NULL; 3464 } 3465 3466 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3467 3468 return addr; 3469} 3470 3471char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3472 size_t alignment, 3473 char* req_addr, 3474 bool exec) { 3475 size_t large_page_size = os::large_page_size(); 3476 3477 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3478 3479 // Allocate small pages. 3480 3481 char* start; 3482 if (req_addr != NULL) { 3483 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3484 assert(is_size_aligned(bytes, alignment), "Must be"); 3485 start = os::reserve_memory(bytes, req_addr); 3486 assert(start == NULL || start == req_addr, "Must be"); 3487 } else { 3488 start = os::reserve_memory_aligned(bytes, alignment); 3489 } 3490 3491 if (start == NULL) { 3492 return NULL; 3493 } 3494 3495 assert(is_ptr_aligned(start, alignment), "Must be"); 3496 3497 if (MemTracker::tracking_level() > NMT_minimal) { 3498 // os::reserve_memory_special will record this memory area. 3499 // Need to release it here to prevent overlapping reservations. 3500 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3501 tkr.record((address)start, bytes); 3502 } 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, 3567 size_t alignment, 3568 char* req_addr, 3569 bool exec) { 3570 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3571 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3572 assert(is_power_of_2(alignment), "Must be"); 3573 assert(is_power_of_2(os::large_page_size()), "Must be"); 3574 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3575 3576 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3577 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3578 } else { 3579 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3580 } 3581} 3582 3583char* os::reserve_memory_special(size_t bytes, size_t alignment, 3584 char* req_addr, bool exec) { 3585 assert(UseLargePages, "only for large pages"); 3586 3587 char* addr; 3588 if (UseSHM) { 3589 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3590 } else { 3591 assert(UseHugeTLBFS, "must be"); 3592 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3593 } 3594 3595 if (addr != NULL) { 3596 if (UseNUMAInterleaving) { 3597 numa_make_global(addr, bytes); 3598 } 3599 3600 // The memory is committed 3601 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3602 } 3603 3604 return addr; 3605} 3606 3607bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3608 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3609 return shmdt(base) == 0; 3610} 3611 3612bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3613 return pd_release_memory(base, bytes); 3614} 3615 3616bool os::release_memory_special(char* base, size_t bytes) { 3617 bool res; 3618 if (MemTracker::tracking_level() > NMT_minimal) { 3619 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3620 res = os::Linux::release_memory_special_impl(base, bytes); 3621 if (res) { 3622 tkr.record((address)base, bytes); 3623 } 3624 3625 } else { 3626 res = os::Linux::release_memory_special_impl(base, bytes); 3627 } 3628 return res; 3629} 3630 3631bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3632 assert(UseLargePages, "only for large pages"); 3633 bool res; 3634 3635 if (UseSHM) { 3636 res = os::Linux::release_memory_special_shm(base, bytes); 3637 } else { 3638 assert(UseHugeTLBFS, "must be"); 3639 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3640 } 3641 return res; 3642} 3643 3644size_t os::large_page_size() { 3645 return _large_page_size; 3646} 3647 3648// With SysV SHM the entire memory region must be allocated as shared 3649// memory. 3650// HugeTLBFS allows application to commit large page memory on demand. 3651// However, when committing memory with HugeTLBFS fails, the region 3652// that was supposed to be committed will lose the old reservation 3653// and allow other threads to steal that memory region. Because of this 3654// behavior we can't commit HugeTLBFS memory. 3655bool os::can_commit_large_page_memory() { 3656 return UseTransparentHugePages; 3657} 3658 3659bool os::can_execute_large_page_memory() { 3660 return UseTransparentHugePages || UseHugeTLBFS; 3661} 3662 3663// Reserve memory at an arbitrary address, only if that area is 3664// available (and not reserved for something else). 3665 3666char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3667 const int max_tries = 10; 3668 char* base[max_tries]; 3669 size_t size[max_tries]; 3670 const size_t gap = 0x000000; 3671 3672 // Assert only that the size is a multiple of the page size, since 3673 // that's all that mmap requires, and since that's all we really know 3674 // about at this low abstraction level. If we need higher alignment, 3675 // we can either pass an alignment to this method or verify alignment 3676 // in one of the methods further up the call chain. See bug 5044738. 3677 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3678 3679 // Repeatedly allocate blocks until the block is allocated at the 3680 // right spot. Give up after max_tries. Note that reserve_memory() will 3681 // automatically update _highest_vm_reserved_address if the call is 3682 // successful. The variable tracks the highest memory address every reserved 3683 // by JVM. It is used to detect heap-stack collision if running with 3684 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3685 // space than needed, it could confuse the collision detecting code. To 3686 // solve the problem, save current _highest_vm_reserved_address and 3687 // calculate the correct value before return. 3688 address old_highest = _highest_vm_reserved_address; 3689 3690 // Linux mmap allows caller to pass an address as hint; give it a try first, 3691 // if kernel honors the hint then we can return immediately. 3692 char * addr = anon_mmap(requested_addr, bytes, false); 3693 if (addr == requested_addr) { 3694 return requested_addr; 3695 } 3696 3697 if (addr != NULL) { 3698 // mmap() is successful but it fails to reserve at the requested address 3699 anon_munmap(addr, bytes); 3700 } 3701 3702 int i; 3703 for (i = 0; i < max_tries; ++i) { 3704 base[i] = reserve_memory(bytes); 3705 3706 if (base[i] != NULL) { 3707 // Is this the block we wanted? 3708 if (base[i] == requested_addr) { 3709 size[i] = bytes; 3710 break; 3711 } 3712 3713 // Does this overlap the block we wanted? Give back the overlapped 3714 // parts and try again. 3715 3716 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3717 if (top_overlap >= 0 && (size_t)top_overlap < bytes) { 3718 unmap_memory(base[i], top_overlap); 3719 base[i] += top_overlap; 3720 size[i] = bytes - top_overlap; 3721 } else { 3722 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr; 3723 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) { 3724 unmap_memory(requested_addr, bottom_overlap); 3725 size[i] = bytes - bottom_overlap; 3726 } else { 3727 size[i] = bytes; 3728 } 3729 } 3730 } 3731 } 3732 3733 // Give back the unused reserved pieces. 3734 3735 for (int j = 0; j < i; ++j) { 3736 if (base[j] != NULL) { 3737 unmap_memory(base[j], size[j]); 3738 } 3739 } 3740 3741 if (i < max_tries) { 3742 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3743 return requested_addr; 3744 } else { 3745 _highest_vm_reserved_address = old_highest; 3746 return NULL; 3747 } 3748} 3749 3750size_t os::read(int fd, void *buf, unsigned int nBytes) { 3751 return ::read(fd, buf, nBytes); 3752} 3753 3754size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) { 3755 return ::pread(fd, buf, nBytes, offset); 3756} 3757 3758// Short sleep, direct OS call. 3759// 3760// Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3761// sched_yield(2) will actually give up the CPU: 3762// 3763// * Alone on this pariticular CPU, keeps running. 3764// * Before the introduction of "skip_buddy" with "compat_yield" disabled 3765// (pre 2.6.39). 3766// 3767// So calling this with 0 is an alternative. 3768// 3769void os::naked_short_sleep(jlong ms) { 3770 struct timespec req; 3771 3772 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3773 req.tv_sec = 0; 3774 if (ms > 0) { 3775 req.tv_nsec = (ms % 1000) * 1000000; 3776 } else { 3777 req.tv_nsec = 1; 3778 } 3779 3780 nanosleep(&req, NULL); 3781 3782 return; 3783} 3784 3785// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3786void os::infinite_sleep() { 3787 while (true) { // sleep forever ... 3788 ::sleep(100); // ... 100 seconds at a time 3789 } 3790} 3791 3792// Used to convert frequent JVM_Yield() to nops 3793bool os::dont_yield() { 3794 return DontYieldALot; 3795} 3796 3797void os::naked_yield() { 3798 sched_yield(); 3799} 3800 3801//////////////////////////////////////////////////////////////////////////////// 3802// thread priority support 3803 3804// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3805// only supports dynamic priority, static priority must be zero. For real-time 3806// applications, Linux supports SCHED_RR which allows static priority (1-99). 3807// However, for large multi-threaded applications, SCHED_RR is not only slower 3808// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3809// of 5 runs - Sep 2005). 3810// 3811// The following code actually changes the niceness of kernel-thread/LWP. It 3812// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3813// not the entire user process, and user level threads are 1:1 mapped to kernel 3814// threads. It has always been the case, but could change in the future. For 3815// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3816// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3817 3818int os::java_to_os_priority[CriticalPriority + 1] = { 3819 19, // 0 Entry should never be used 3820 3821 4, // 1 MinPriority 3822 3, // 2 3823 2, // 3 3824 3825 1, // 4 3826 0, // 5 NormPriority 3827 -1, // 6 3828 3829 -2, // 7 3830 -3, // 8 3831 -4, // 9 NearMaxPriority 3832 3833 -5, // 10 MaxPriority 3834 3835 -5 // 11 CriticalPriority 3836}; 3837 3838static int prio_init() { 3839 if (ThreadPriorityPolicy == 1) { 3840 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3841 // if effective uid is not root. Perhaps, a more elegant way of doing 3842 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3843 if (geteuid() != 0) { 3844 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3845 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3846 } 3847 ThreadPriorityPolicy = 0; 3848 } 3849 } 3850 if (UseCriticalJavaThreadPriority) { 3851 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3852 } 3853 return 0; 3854} 3855 3856OSReturn os::set_native_priority(Thread* thread, int newpri) { 3857 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 3858 3859 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3860 return (ret == 0) ? OS_OK : OS_ERR; 3861} 3862 3863OSReturn os::get_native_priority(const Thread* const thread, 3864 int *priority_ptr) { 3865 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 3866 *priority_ptr = java_to_os_priority[NormPriority]; 3867 return OS_OK; 3868 } 3869 3870 errno = 0; 3871 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3872 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3873} 3874 3875// Hint to the underlying OS that a task switch would not be good. 3876// Void return because it's a hint and can fail. 3877void os::hint_no_preempt() {} 3878 3879//////////////////////////////////////////////////////////////////////////////// 3880// suspend/resume support 3881 3882// the low-level signal-based suspend/resume support is a remnant from the 3883// old VM-suspension that used to be for java-suspension, safepoints etc, 3884// within hotspot. Now there is a single use-case for this: 3885// - calling get_thread_pc() on the VMThread by the flat-profiler task 3886// that runs in the watcher thread. 3887// The remaining code is greatly simplified from the more general suspension 3888// code that used to be used. 3889// 3890// The protocol is quite simple: 3891// - suspend: 3892// - sends a signal to the target thread 3893// - polls the suspend state of the osthread using a yield loop 3894// - target thread signal handler (SR_handler) sets suspend state 3895// and blocks in sigsuspend until continued 3896// - resume: 3897// - sets target osthread state to continue 3898// - sends signal to end the sigsuspend loop in the SR_handler 3899// 3900// Note that the SR_lock plays no role in this suspend/resume protocol. 3901 3902static void resume_clear_context(OSThread *osthread) { 3903 osthread->set_ucontext(NULL); 3904 osthread->set_siginfo(NULL); 3905} 3906 3907static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 3908 ucontext_t* context) { 3909 osthread->set_ucontext(context); 3910 osthread->set_siginfo(siginfo); 3911} 3912 3913// Handler function invoked when a thread's execution is suspended or 3914// resumed. We have to be careful that only async-safe functions are 3915// called here (Note: most pthread functions are not async safe and 3916// should be avoided.) 3917// 3918// Note: sigwait() is a more natural fit than sigsuspend() from an 3919// interface point of view, but sigwait() prevents the signal hander 3920// from being run. libpthread would get very confused by not having 3921// its signal handlers run and prevents sigwait()'s use with the 3922// mutex granting granting signal. 3923// 3924// Currently only ever called on the VMThread and JavaThreads (PC sampling) 3925// 3926static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3927 // Save and restore errno to avoid confusing native code with EINTR 3928 // after sigsuspend. 3929 int old_errno = errno; 3930 3931 Thread* thread = Thread::current(); 3932 OSThread* osthread = thread->osthread(); 3933 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3934 3935 os::SuspendResume::State current = osthread->sr.state(); 3936 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3937 suspend_save_context(osthread, siginfo, context); 3938 3939 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 3940 os::SuspendResume::State state = osthread->sr.suspended(); 3941 if (state == os::SuspendResume::SR_SUSPENDED) { 3942 sigset_t suspend_set; // signals for sigsuspend() 3943 3944 // get current set of blocked signals and unblock resume signal 3945 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3946 sigdelset(&suspend_set, SR_signum); 3947 3948 sr_semaphore.signal(); 3949 // wait here until we are resumed 3950 while (1) { 3951 sigsuspend(&suspend_set); 3952 3953 os::SuspendResume::State result = osthread->sr.running(); 3954 if (result == os::SuspendResume::SR_RUNNING) { 3955 sr_semaphore.signal(); 3956 break; 3957 } 3958 } 3959 3960 } else if (state == os::SuspendResume::SR_RUNNING) { 3961 // request was cancelled, continue 3962 } else { 3963 ShouldNotReachHere(); 3964 } 3965 3966 resume_clear_context(osthread); 3967 } else if (current == os::SuspendResume::SR_RUNNING) { 3968 // request was cancelled, continue 3969 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 3970 // ignore 3971 } else { 3972 // ignore 3973 } 3974 3975 errno = old_errno; 3976} 3977 3978 3979static int SR_initialize() { 3980 struct sigaction act; 3981 char *s; 3982 // Get signal number to use for suspend/resume 3983 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3984 int sig = ::strtol(s, 0, 10); 3985 if (sig > 0 || sig < _NSIG) { 3986 SR_signum = sig; 3987 } 3988 } 3989 3990 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3991 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3992 3993 sigemptyset(&SR_sigset); 3994 sigaddset(&SR_sigset, SR_signum); 3995 3996 // Set up signal handler for suspend/resume 3997 act.sa_flags = SA_RESTART|SA_SIGINFO; 3998 act.sa_handler = (void (*)(int)) SR_handler; 3999 4000 // SR_signum is blocked by default. 4001 // 4528190 - We also need to block pthread restart signal (32 on all 4002 // supported Linux platforms). Note that LinuxThreads need to block 4003 // this signal for all threads to work properly. So we don't have 4004 // to use hard-coded signal number when setting up the mask. 4005 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4006 4007 if (sigaction(SR_signum, &act, 0) == -1) { 4008 return -1; 4009 } 4010 4011 // Save signal flag 4012 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4013 return 0; 4014} 4015 4016static int sr_notify(OSThread* osthread) { 4017 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4018 assert_status(status == 0, status, "pthread_kill"); 4019 return status; 4020} 4021 4022// "Randomly" selected value for how long we want to spin 4023// before bailing out on suspending a thread, also how often 4024// we send a signal to a thread we want to resume 4025static const int RANDOMLY_LARGE_INTEGER = 1000000; 4026static const int RANDOMLY_LARGE_INTEGER2 = 100; 4027 4028// returns true on success and false on error - really an error is fatal 4029// but this seems the normal response to library errors 4030static bool do_suspend(OSThread* osthread) { 4031 assert(osthread->sr.is_running(), "thread should be running"); 4032 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4033 4034 // mark as suspended and send signal 4035 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4036 // failed to switch, state wasn't running? 4037 ShouldNotReachHere(); 4038 return false; 4039 } 4040 4041 if (sr_notify(osthread) != 0) { 4042 ShouldNotReachHere(); 4043 } 4044 4045 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4046 while (true) { 4047 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4048 break; 4049 } else { 4050 // timeout 4051 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4052 if (cancelled == os::SuspendResume::SR_RUNNING) { 4053 return false; 4054 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4055 // make sure that we consume the signal on the semaphore as well 4056 sr_semaphore.wait(); 4057 break; 4058 } else { 4059 ShouldNotReachHere(); 4060 return false; 4061 } 4062 } 4063 } 4064 4065 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4066 return true; 4067} 4068 4069static void do_resume(OSThread* osthread) { 4070 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4071 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4072 4073 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4074 // failed to switch to WAKEUP_REQUEST 4075 ShouldNotReachHere(); 4076 return; 4077 } 4078 4079 while (true) { 4080 if (sr_notify(osthread) == 0) { 4081 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4082 if (osthread->sr.is_running()) { 4083 return; 4084 } 4085 } 4086 } else { 4087 ShouldNotReachHere(); 4088 } 4089 } 4090 4091 guarantee(osthread->sr.is_running(), "Must be running!"); 4092} 4093 4094/////////////////////////////////////////////////////////////////////////////////// 4095// signal handling (except suspend/resume) 4096 4097// This routine may be used by user applications as a "hook" to catch signals. 4098// The user-defined signal handler must pass unrecognized signals to this 4099// routine, and if it returns true (non-zero), then the signal handler must 4100// return immediately. If the flag "abort_if_unrecognized" is true, then this 4101// routine will never retun false (zero), but instead will execute a VM panic 4102// routine kill the process. 4103// 4104// If this routine returns false, it is OK to call it again. This allows 4105// the user-defined signal handler to perform checks either before or after 4106// the VM performs its own checks. Naturally, the user code would be making 4107// a serious error if it tried to handle an exception (such as a null check 4108// or breakpoint) that the VM was generating for its own correct operation. 4109// 4110// This routine may recognize any of the following kinds of signals: 4111// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4112// It should be consulted by handlers for any of those signals. 4113// 4114// The caller of this routine must pass in the three arguments supplied 4115// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4116// field of the structure passed to sigaction(). This routine assumes that 4117// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4118// 4119// Note that the VM will print warnings if it detects conflicting signal 4120// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4121// 4122extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4123 siginfo_t* siginfo, 4124 void* ucontext, 4125 int abort_if_unrecognized); 4126 4127void signalHandler(int sig, siginfo_t* info, void* uc) { 4128 assert(info != NULL && uc != NULL, "it must be old kernel"); 4129 int orig_errno = errno; // Preserve errno value over signal handler. 4130 JVM_handle_linux_signal(sig, info, uc, true); 4131 errno = orig_errno; 4132} 4133 4134 4135// This boolean allows users to forward their own non-matching signals 4136// to JVM_handle_linux_signal, harmlessly. 4137bool os::Linux::signal_handlers_are_installed = false; 4138 4139// For signal-chaining 4140struct sigaction os::Linux::sigact[MAXSIGNUM]; 4141unsigned int os::Linux::sigs = 0; 4142bool os::Linux::libjsig_is_loaded = false; 4143typedef struct sigaction *(*get_signal_t)(int); 4144get_signal_t os::Linux::get_signal_action = NULL; 4145 4146struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4147 struct sigaction *actp = NULL; 4148 4149 if (libjsig_is_loaded) { 4150 // Retrieve the old signal handler from libjsig 4151 actp = (*get_signal_action)(sig); 4152 } 4153 if (actp == NULL) { 4154 // Retrieve the preinstalled signal handler from jvm 4155 actp = get_preinstalled_handler(sig); 4156 } 4157 4158 return actp; 4159} 4160 4161static bool call_chained_handler(struct sigaction *actp, int sig, 4162 siginfo_t *siginfo, void *context) { 4163 // Call the old signal handler 4164 if (actp->sa_handler == SIG_DFL) { 4165 // It's more reasonable to let jvm treat it as an unexpected exception 4166 // instead of taking the default action. 4167 return false; 4168 } else if (actp->sa_handler != SIG_IGN) { 4169 if ((actp->sa_flags & SA_NODEFER) == 0) { 4170 // automaticlly block the signal 4171 sigaddset(&(actp->sa_mask), sig); 4172 } 4173 4174 sa_handler_t hand; 4175 sa_sigaction_t sa; 4176 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4177 // retrieve the chained handler 4178 if (siginfo_flag_set) { 4179 sa = actp->sa_sigaction; 4180 } else { 4181 hand = actp->sa_handler; 4182 } 4183 4184 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4185 actp->sa_handler = SIG_DFL; 4186 } 4187 4188 // try to honor the signal mask 4189 sigset_t oset; 4190 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4191 4192 // call into the chained handler 4193 if (siginfo_flag_set) { 4194 (*sa)(sig, siginfo, context); 4195 } else { 4196 (*hand)(sig); 4197 } 4198 4199 // restore the signal mask 4200 pthread_sigmask(SIG_SETMASK, &oset, 0); 4201 } 4202 // Tell jvm's signal handler the signal is taken care of. 4203 return true; 4204} 4205 4206bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4207 bool chained = false; 4208 // signal-chaining 4209 if (UseSignalChaining) { 4210 struct sigaction *actp = get_chained_signal_action(sig); 4211 if (actp != NULL) { 4212 chained = call_chained_handler(actp, sig, siginfo, context); 4213 } 4214 } 4215 return chained; 4216} 4217 4218struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4219 if ((((unsigned int)1 << sig) & sigs) != 0) { 4220 return &sigact[sig]; 4221 } 4222 return NULL; 4223} 4224 4225void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4226 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4227 sigact[sig] = oldAct; 4228 sigs |= (unsigned int)1 << sig; 4229} 4230 4231// for diagnostic 4232int os::Linux::sigflags[MAXSIGNUM]; 4233 4234int os::Linux::get_our_sigflags(int sig) { 4235 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4236 return sigflags[sig]; 4237} 4238 4239void os::Linux::set_our_sigflags(int sig, int flags) { 4240 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4241 sigflags[sig] = flags; 4242} 4243 4244void os::Linux::set_signal_handler(int sig, bool set_installed) { 4245 // Check for overwrite. 4246 struct sigaction oldAct; 4247 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4248 4249 void* oldhand = oldAct.sa_sigaction 4250 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4251 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4252 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4253 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4254 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4255 if (AllowUserSignalHandlers || !set_installed) { 4256 // Do not overwrite; user takes responsibility to forward to us. 4257 return; 4258 } else if (UseSignalChaining) { 4259 // save the old handler in jvm 4260 save_preinstalled_handler(sig, oldAct); 4261 // libjsig also interposes the sigaction() call below and saves the 4262 // old sigaction on it own. 4263 } else { 4264 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4265 "%#lx for signal %d.", (long)oldhand, sig)); 4266 } 4267 } 4268 4269 struct sigaction sigAct; 4270 sigfillset(&(sigAct.sa_mask)); 4271 sigAct.sa_handler = SIG_DFL; 4272 if (!set_installed) { 4273 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4274 } else { 4275 sigAct.sa_sigaction = signalHandler; 4276 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4277 } 4278 // Save flags, which are set by ours 4279 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4280 sigflags[sig] = sigAct.sa_flags; 4281 4282 int ret = sigaction(sig, &sigAct, &oldAct); 4283 assert(ret == 0, "check"); 4284 4285 void* oldhand2 = oldAct.sa_sigaction 4286 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4287 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4288 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4289} 4290 4291// install signal handlers for signals that HotSpot needs to 4292// handle in order to support Java-level exception handling. 4293 4294void os::Linux::install_signal_handlers() { 4295 if (!signal_handlers_are_installed) { 4296 signal_handlers_are_installed = true; 4297 4298 // signal-chaining 4299 typedef void (*signal_setting_t)(); 4300 signal_setting_t begin_signal_setting = NULL; 4301 signal_setting_t end_signal_setting = NULL; 4302 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4303 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4304 if (begin_signal_setting != NULL) { 4305 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4306 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4307 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4308 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4309 libjsig_is_loaded = true; 4310 assert(UseSignalChaining, "should enable signal-chaining"); 4311 } 4312 if (libjsig_is_loaded) { 4313 // Tell libjsig jvm is setting signal handlers 4314 (*begin_signal_setting)(); 4315 } 4316 4317 set_signal_handler(SIGSEGV, true); 4318 set_signal_handler(SIGPIPE, true); 4319 set_signal_handler(SIGBUS, true); 4320 set_signal_handler(SIGILL, true); 4321 set_signal_handler(SIGFPE, true); 4322#if defined(PPC64) 4323 set_signal_handler(SIGTRAP, true); 4324#endif 4325 set_signal_handler(SIGXFSZ, true); 4326 4327 if (libjsig_is_loaded) { 4328 // Tell libjsig jvm finishes setting signal handlers 4329 (*end_signal_setting)(); 4330 } 4331 4332 // We don't activate signal checker if libjsig is in place, we trust ourselves 4333 // and if UserSignalHandler is installed all bets are off. 4334 // Log that signal checking is off only if -verbose:jni is specified. 4335 if (CheckJNICalls) { 4336 if (libjsig_is_loaded) { 4337 if (PrintJNIResolving) { 4338 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4339 } 4340 check_signals = false; 4341 } 4342 if (AllowUserSignalHandlers) { 4343 if (PrintJNIResolving) { 4344 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4345 } 4346 check_signals = false; 4347 } 4348 } 4349 } 4350} 4351 4352// This is the fastest way to get thread cpu time on Linux. 4353// Returns cpu time (user+sys) for any thread, not only for current. 4354// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4355// It might work on 2.6.10+ with a special kernel/glibc patch. 4356// For reference, please, see IEEE Std 1003.1-2004: 4357// http://www.unix.org/single_unix_specification 4358 4359jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4360 struct timespec tp; 4361 int rc = os::Linux::clock_gettime(clockid, &tp); 4362 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4363 4364 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4365} 4366 4367///// 4368// glibc on Linux platform uses non-documented flag 4369// to indicate, that some special sort of signal 4370// trampoline is used. 4371// We will never set this flag, and we should 4372// ignore this flag in our diagnostic 4373#ifdef SIGNIFICANT_SIGNAL_MASK 4374 #undef SIGNIFICANT_SIGNAL_MASK 4375#endif 4376#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4377 4378static const char* get_signal_handler_name(address handler, 4379 char* buf, int buflen) { 4380 int offset; 4381 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4382 if (found) { 4383 // skip directory names 4384 const char *p1, *p2; 4385 p1 = buf; 4386 size_t len = strlen(os::file_separator()); 4387 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4388 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4389 } else { 4390 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4391 } 4392 return buf; 4393} 4394 4395static void print_signal_handler(outputStream* st, int sig, 4396 char* buf, size_t buflen) { 4397 struct sigaction sa; 4398 4399 sigaction(sig, NULL, &sa); 4400 4401 // See comment for SIGNIFICANT_SIGNAL_MASK define 4402 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4403 4404 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4405 4406 address handler = (sa.sa_flags & SA_SIGINFO) 4407 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4408 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4409 4410 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4411 st->print("SIG_DFL"); 4412 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4413 st->print("SIG_IGN"); 4414 } else { 4415 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4416 } 4417 4418 st->print(", sa_mask[0]="); 4419 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4420 4421 address rh = VMError::get_resetted_sighandler(sig); 4422 // May be, handler was resetted by VMError? 4423 if (rh != NULL) { 4424 handler = rh; 4425 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4426 } 4427 4428 st->print(", sa_flags="); 4429 os::Posix::print_sa_flags(st, sa.sa_flags); 4430 4431 // Check: is it our handler? 4432 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4433 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4434 // It is our signal handler 4435 // check for flags, reset system-used one! 4436 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4437 st->print( 4438 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4439 os::Linux::get_our_sigflags(sig)); 4440 } 4441 } 4442 st->cr(); 4443} 4444 4445 4446#define DO_SIGNAL_CHECK(sig) \ 4447 do { \ 4448 if (!sigismember(&check_signal_done, sig)) { \ 4449 os::Linux::check_signal_handler(sig); \ 4450 } \ 4451 } while (0) 4452 4453// This method is a periodic task to check for misbehaving JNI applications 4454// under CheckJNI, we can add any periodic checks here 4455 4456void os::run_periodic_checks() { 4457 if (check_signals == false) return; 4458 4459 // SEGV and BUS if overridden could potentially prevent 4460 // generation of hs*.log in the event of a crash, debugging 4461 // such a case can be very challenging, so we absolutely 4462 // check the following for a good measure: 4463 DO_SIGNAL_CHECK(SIGSEGV); 4464 DO_SIGNAL_CHECK(SIGILL); 4465 DO_SIGNAL_CHECK(SIGFPE); 4466 DO_SIGNAL_CHECK(SIGBUS); 4467 DO_SIGNAL_CHECK(SIGPIPE); 4468 DO_SIGNAL_CHECK(SIGXFSZ); 4469#if defined(PPC64) 4470 DO_SIGNAL_CHECK(SIGTRAP); 4471#endif 4472 4473 // ReduceSignalUsage allows the user to override these handlers 4474 // see comments at the very top and jvm_solaris.h 4475 if (!ReduceSignalUsage) { 4476 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4477 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4478 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4479 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4480 } 4481 4482 DO_SIGNAL_CHECK(SR_signum); 4483 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4484} 4485 4486typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4487 4488static os_sigaction_t os_sigaction = NULL; 4489 4490void os::Linux::check_signal_handler(int sig) { 4491 char buf[O_BUFLEN]; 4492 address jvmHandler = NULL; 4493 4494 4495 struct sigaction act; 4496 if (os_sigaction == NULL) { 4497 // only trust the default sigaction, in case it has been interposed 4498 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4499 if (os_sigaction == NULL) return; 4500 } 4501 4502 os_sigaction(sig, (struct sigaction*)NULL, &act); 4503 4504 4505 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4506 4507 address thisHandler = (act.sa_flags & SA_SIGINFO) 4508 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4509 : CAST_FROM_FN_PTR(address, act.sa_handler); 4510 4511 4512 switch (sig) { 4513 case SIGSEGV: 4514 case SIGBUS: 4515 case SIGFPE: 4516 case SIGPIPE: 4517 case SIGILL: 4518 case SIGXFSZ: 4519 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4520 break; 4521 4522 case SHUTDOWN1_SIGNAL: 4523 case SHUTDOWN2_SIGNAL: 4524 case SHUTDOWN3_SIGNAL: 4525 case BREAK_SIGNAL: 4526 jvmHandler = (address)user_handler(); 4527 break; 4528 4529 case INTERRUPT_SIGNAL: 4530 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4531 break; 4532 4533 default: 4534 if (sig == SR_signum) { 4535 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4536 } else { 4537 return; 4538 } 4539 break; 4540 } 4541 4542 if (thisHandler != jvmHandler) { 4543 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4544 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4545 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4546 // No need to check this sig any longer 4547 sigaddset(&check_signal_done, sig); 4548 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4549 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4550 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4551 exception_name(sig, buf, O_BUFLEN)); 4552 } 4553 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4554 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4555 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4556 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4557 // No need to check this sig any longer 4558 sigaddset(&check_signal_done, sig); 4559 } 4560 4561 // Dump all the signal 4562 if (sigismember(&check_signal_done, sig)) { 4563 print_signal_handlers(tty, buf, O_BUFLEN); 4564 } 4565} 4566 4567extern void report_error(char* file_name, int line_no, char* title, 4568 char* format, ...); 4569 4570extern bool signal_name(int signo, char* buf, size_t len); 4571 4572const char* os::exception_name(int exception_code, char* buf, size_t size) { 4573 if (0 < exception_code && exception_code <= SIGRTMAX) { 4574 // signal 4575 if (!signal_name(exception_code, buf, size)) { 4576 jio_snprintf(buf, size, "SIG%d", exception_code); 4577 } 4578 return buf; 4579 } else { 4580 return NULL; 4581 } 4582} 4583 4584// this is called _before_ the most of global arguments have been parsed 4585void os::init(void) { 4586 char dummy; // used to get a guess on initial stack address 4587// first_hrtime = gethrtime(); 4588 4589 // With LinuxThreads the JavaMain thread pid (primordial thread) 4590 // is different than the pid of the java launcher thread. 4591 // So, on Linux, the launcher thread pid is passed to the VM 4592 // via the sun.java.launcher.pid property. 4593 // Use this property instead of getpid() if it was correctly passed. 4594 // See bug 6351349. 4595 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4596 4597 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4598 4599 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4600 4601 init_random(1234567); 4602 4603 ThreadCritical::initialize(); 4604 4605 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4606 if (Linux::page_size() == -1) { 4607 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4608 strerror(errno))); 4609 } 4610 init_page_sizes((size_t) Linux::page_size()); 4611 4612 Linux::initialize_system_info(); 4613 4614 // main_thread points to the aboriginal thread 4615 Linux::_main_thread = pthread_self(); 4616 4617 Linux::clock_init(); 4618 initial_time_count = javaTimeNanos(); 4619 4620 // pthread_condattr initialization for monotonic clock 4621 int status; 4622 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4623 if ((status = pthread_condattr_init(_condattr)) != 0) { 4624 fatal(err_msg("pthread_condattr_init: %s", strerror(status))); 4625 } 4626 // Only set the clock if CLOCK_MONOTONIC is available 4627 if (os::supports_monotonic_clock()) { 4628 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4629 if (status == EINVAL) { 4630 warning("Unable to use monotonic clock with relative timed-waits" \ 4631 " - changes to the time-of-day clock may have adverse affects"); 4632 } else { 4633 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status))); 4634 } 4635 } 4636 } 4637 // else it defaults to CLOCK_REALTIME 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 // retrieve entry point for pthread_setname_np 4649 Linux::_pthread_setname_np = 4650 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4651 4652} 4653 4654// To install functions for atexit system call 4655extern "C" { 4656 static void perfMemory_exit_helper() { 4657 perfMemory_exit(); 4658 } 4659} 4660 4661// this is called _after_ the global arguments have been parsed 4662jint os::init_2(void) { 4663 Linux::fast_thread_clock_init(); 4664 4665 // Allocate a single page and mark it as readable for safepoint polling 4666 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4667 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page"); 4668 4669 os::set_polling_page(polling_page); 4670 4671#ifndef PRODUCT 4672 if (Verbose && PrintMiscellaneous) { 4673 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", 4674 (intptr_t)polling_page); 4675 } 4676#endif 4677 4678 if (!UseMembar) { 4679 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4680 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4681 os::set_memory_serialize_page(mem_serialize_page); 4682 4683#ifndef PRODUCT 4684 if (Verbose && PrintMiscellaneous) { 4685 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", 4686 (intptr_t)mem_serialize_page); 4687 } 4688#endif 4689 } 4690 4691 // initialize suspend/resume support - must do this before signal_sets_init() 4692 if (SR_initialize() != 0) { 4693 perror("SR_initialize failed"); 4694 return JNI_ERR; 4695 } 4696 4697 Linux::signal_sets_init(); 4698 Linux::install_signal_handlers(); 4699 4700 // Check minimum allowable stack size for thread creation and to initialize 4701 // the java system classes, including StackOverflowError - depends on page 4702 // size. Add a page for compiler2 recursion in main thread. 4703 // Add in 2*BytesPerWord times page size to account for VM stack during 4704 // class initialization depending on 32 or 64 bit VM. 4705 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4706 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4707 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4708 4709 size_t threadStackSizeInBytes = ThreadStackSize * K; 4710 if (threadStackSizeInBytes != 0 && 4711 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4712 tty->print_cr("\nThe stack size specified is too small, " 4713 "Specify at least %dk", 4714 os::Linux::min_stack_allowed/ K); 4715 return JNI_ERR; 4716 } 4717 4718 // Make the stack size a multiple of the page size so that 4719 // the yellow/red zones can be guarded. 4720 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4721 vm_page_size())); 4722 4723 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4724 4725#if defined(IA32) 4726 workaround_expand_exec_shield_cs_limit(); 4727#endif 4728 4729 Linux::libpthread_init(); 4730 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4731 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4732 Linux::glibc_version(), Linux::libpthread_version(), 4733 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4734 } 4735 4736 if (UseNUMA) { 4737 if (!Linux::libnuma_init()) { 4738 UseNUMA = false; 4739 } else { 4740 if ((Linux::numa_max_node() < 1)) { 4741 // There's only one node(they start from 0), disable NUMA. 4742 UseNUMA = false; 4743 } 4744 } 4745 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4746 // we can make the adaptive lgrp chunk resizing work. If the user specified 4747 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4748 // disable adaptive resizing. 4749 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4750 if (FLAG_IS_DEFAULT(UseNUMA)) { 4751 UseNUMA = false; 4752 } else { 4753 if (FLAG_IS_DEFAULT(UseLargePages) && 4754 FLAG_IS_DEFAULT(UseSHM) && 4755 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4756 UseLargePages = false; 4757 } else { 4758 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing"); 4759 UseAdaptiveSizePolicy = false; 4760 UseAdaptiveNUMAChunkSizing = false; 4761 } 4762 } 4763 } 4764 if (!UseNUMA && ForceNUMA) { 4765 UseNUMA = true; 4766 } 4767 } 4768 4769 if (MaxFDLimit) { 4770 // set the number of file descriptors to max. print out error 4771 // if getrlimit/setrlimit fails but continue regardless. 4772 struct rlimit nbr_files; 4773 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4774 if (status != 0) { 4775 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4776 perror("os::init_2 getrlimit failed"); 4777 } 4778 } else { 4779 nbr_files.rlim_cur = nbr_files.rlim_max; 4780 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4781 if (status != 0) { 4782 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4783 perror("os::init_2 setrlimit failed"); 4784 } 4785 } 4786 } 4787 } 4788 4789 // Initialize lock used to serialize thread creation (see os::create_thread) 4790 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4791 4792 // at-exit methods are called in the reverse order of their registration. 4793 // atexit functions are called on return from main or as a result of a 4794 // call to exit(3C). There can be only 32 of these functions registered 4795 // and atexit() does not set errno. 4796 4797 if (PerfAllowAtExitRegistration) { 4798 // only register atexit functions if PerfAllowAtExitRegistration is set. 4799 // atexit functions can be delayed until process exit time, which 4800 // can be problematic for embedded VM situations. Embedded VMs should 4801 // call DestroyJavaVM() to assure that VM resources are released. 4802 4803 // note: perfMemory_exit_helper atexit function may be removed in 4804 // the future if the appropriate cleanup code can be added to the 4805 // VM_Exit VMOperation's doit method. 4806 if (atexit(perfMemory_exit_helper) != 0) { 4807 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4808 } 4809 } 4810 4811 // initialize thread priority policy 4812 prio_init(); 4813 4814 return JNI_OK; 4815} 4816 4817// Mark the polling page as unreadable 4818void os::make_polling_page_unreadable(void) { 4819 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 4820 fatal("Could not disable polling page"); 4821 } 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 if (Linux::_pthread_setname_np) { 4841 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 4842 snprintf(buf, sizeof(buf), "%s", name); 4843 buf[sizeof(buf) - 1] = '\0'; 4844 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 4845 // ERANGE should not happen; all other errors should just be ignored. 4846 assert(rc != ERANGE, "pthread_setname_np failed"); 4847 } 4848} 4849 4850bool os::distribute_processes(uint length, uint* distribution) { 4851 // Not yet implemented. 4852 return false; 4853} 4854 4855bool os::bind_to_processor(uint processor_id) { 4856 // Not yet implemented. 4857 return false; 4858} 4859 4860/// 4861 4862void os::SuspendedThreadTask::internal_do_task() { 4863 if (do_suspend(_thread->osthread())) { 4864 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4865 do_task(context); 4866 do_resume(_thread->osthread()); 4867 } 4868} 4869 4870class PcFetcher : public os::SuspendedThreadTask { 4871 public: 4872 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4873 ExtendedPC result(); 4874 protected: 4875 void do_task(const os::SuspendedThreadTaskContext& context); 4876 private: 4877 ExtendedPC _epc; 4878}; 4879 4880ExtendedPC PcFetcher::result() { 4881 guarantee(is_done(), "task is not done yet."); 4882 return _epc; 4883} 4884 4885void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4886 Thread* thread = context.thread(); 4887 OSThread* osthread = thread->osthread(); 4888 if (osthread->ucontext() != NULL) { 4889 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4890 } else { 4891 // NULL context is unexpected, double-check this is the VMThread 4892 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4893 } 4894} 4895 4896// Suspends the target using the signal mechanism and then grabs the PC before 4897// resuming the target. Used by the flat-profiler only 4898ExtendedPC os::get_thread_pc(Thread* thread) { 4899 // Make sure that it is called by the watcher for the VMThread 4900 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4901 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4902 4903 PcFetcher fetcher(thread); 4904 fetcher.run(); 4905 return fetcher.result(); 4906} 4907 4908int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, 4909 pthread_mutex_t *_mutex, 4910 const struct timespec *_abstime) { 4911 if (is_NPTL()) { 4912 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4913 } else { 4914 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4915 // word back to default 64bit precision if condvar is signaled. Java 4916 // wants 53bit precision. Save and restore current value. 4917 int fpu = get_fpu_control_word(); 4918 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 4919 set_fpu_control_word(fpu); 4920 return status; 4921 } 4922} 4923 4924//////////////////////////////////////////////////////////////////////////////// 4925// debug support 4926 4927bool os::find(address addr, outputStream* st) { 4928 Dl_info dlinfo; 4929 memset(&dlinfo, 0, sizeof(dlinfo)); 4930 if (dladdr(addr, &dlinfo) != 0) { 4931 st->print(PTR_FORMAT ": ", addr); 4932 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 4933 st->print("%s+%#x", dlinfo.dli_sname, 4934 addr - (intptr_t)dlinfo.dli_saddr); 4935 } else if (dlinfo.dli_fbase != NULL) { 4936 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4937 } else { 4938 st->print("<absolute address>"); 4939 } 4940 if (dlinfo.dli_fname != NULL) { 4941 st->print(" in %s", dlinfo.dli_fname); 4942 } 4943 if (dlinfo.dli_fbase != NULL) { 4944 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4945 } 4946 st->cr(); 4947 4948 if (Verbose) { 4949 // decode some bytes around the PC 4950 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 4951 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 4952 address lowest = (address) dlinfo.dli_sname; 4953 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4954 if (begin < lowest) begin = lowest; 4955 Dl_info dlinfo2; 4956 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 4957 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 4958 end = (address) dlinfo2.dli_saddr; 4959 } 4960 Disassembler::decode(begin, end, st); 4961 } 4962 return true; 4963 } 4964 return false; 4965} 4966 4967//////////////////////////////////////////////////////////////////////////////// 4968// misc 4969 4970// This does not do anything on Linux. This is basically a hook for being 4971// able to use structured exception handling (thread-local exception filters) 4972// on, e.g., Win32. 4973void 4974os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4975 JavaCallArguments* args, Thread* thread) { 4976 f(value, method, args, thread); 4977} 4978 4979void os::print_statistics() { 4980} 4981 4982int os::message_box(const char* title, const char* message) { 4983 int i; 4984 fdStream err(defaultStream::error_fd()); 4985 for (i = 0; i < 78; i++) err.print_raw("="); 4986 err.cr(); 4987 err.print_raw_cr(title); 4988 for (i = 0; i < 78; i++) err.print_raw("-"); 4989 err.cr(); 4990 err.print_raw_cr(message); 4991 for (i = 0; i < 78; i++) err.print_raw("="); 4992 err.cr(); 4993 4994 char buf[16]; 4995 // Prevent process from exiting upon "read error" without consuming all CPU 4996 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 4997 4998 return buf[0] == 'y' || buf[0] == 'Y'; 4999} 5000 5001int os::stat(const char *path, struct stat *sbuf) { 5002 char pathbuf[MAX_PATH]; 5003 if (strlen(path) > MAX_PATH - 1) { 5004 errno = ENAMETOOLONG; 5005 return -1; 5006 } 5007 os::native_path(strcpy(pathbuf, path)); 5008 return ::stat(pathbuf, sbuf); 5009} 5010 5011bool os::check_heap(bool force) { 5012 return true; 5013} 5014 5015// Is a (classpath) directory empty? 5016bool os::dir_is_empty(const char* path) { 5017 DIR *dir = NULL; 5018 struct dirent *ptr; 5019 5020 dir = opendir(path); 5021 if (dir == NULL) return true; 5022 5023 // Scan the directory 5024 bool result = true; 5025 char buf[sizeof(struct dirent) + MAX_PATH]; 5026 while (result && (ptr = ::readdir(dir)) != NULL) { 5027 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5028 result = false; 5029 } 5030 } 5031 closedir(dir); 5032 return result; 5033} 5034 5035// This code originates from JDK's sysOpen and open64_w 5036// from src/solaris/hpi/src/system_md.c 5037 5038int os::open(const char *path, int oflag, int mode) { 5039 if (strlen(path) > MAX_PATH - 1) { 5040 errno = ENAMETOOLONG; 5041 return -1; 5042 } 5043 5044 // All file descriptors that are opened in the Java process and not 5045 // specifically destined for a subprocess should have the close-on-exec 5046 // flag set. If we don't set it, then careless 3rd party native code 5047 // might fork and exec without closing all appropriate file descriptors 5048 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5049 // turn might: 5050 // 5051 // - cause end-of-file to fail to be detected on some file 5052 // descriptors, resulting in mysterious hangs, or 5053 // 5054 // - might cause an fopen in the subprocess to fail on a system 5055 // suffering from bug 1085341. 5056 // 5057 // (Yes, the default setting of the close-on-exec flag is a Unix 5058 // design flaw) 5059 // 5060 // See: 5061 // 1085341: 32-bit stdio routines should support file descriptors >255 5062 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5063 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5064 // 5065 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5066 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5067 // because it saves a system call and removes a small window where the flag 5068 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5069 // and we fall back to using FD_CLOEXEC (see below). 5070#ifdef O_CLOEXEC 5071 oflag |= O_CLOEXEC; 5072#endif 5073 5074 int fd = ::open64(path, oflag, mode); 5075 if (fd == -1) return -1; 5076 5077 //If the open succeeded, the file might still be a directory 5078 { 5079 struct stat64 buf64; 5080 int ret = ::fstat64(fd, &buf64); 5081 int st_mode = buf64.st_mode; 5082 5083 if (ret != -1) { 5084 if ((st_mode & S_IFMT) == S_IFDIR) { 5085 errno = EISDIR; 5086 ::close(fd); 5087 return -1; 5088 } 5089 } else { 5090 ::close(fd); 5091 return -1; 5092 } 5093 } 5094 5095#ifdef FD_CLOEXEC 5096 // Validate that the use of the O_CLOEXEC flag on open above worked. 5097 // With recent kernels, we will perform this check exactly once. 5098 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5099 if (!O_CLOEXEC_is_known_to_work) { 5100 int flags = ::fcntl(fd, F_GETFD); 5101 if (flags != -1) { 5102 if ((flags & FD_CLOEXEC) != 0) 5103 O_CLOEXEC_is_known_to_work = 1; 5104 else 5105 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5106 } 5107 } 5108#endif 5109 5110 return fd; 5111} 5112 5113 5114// create binary file, rewriting existing file if required 5115int os::create_binary_file(const char* path, bool rewrite_existing) { 5116 int oflags = O_WRONLY | O_CREAT; 5117 if (!rewrite_existing) { 5118 oflags |= O_EXCL; 5119 } 5120 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5121} 5122 5123// return current position of file pointer 5124jlong os::current_file_offset(int fd) { 5125 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5126} 5127 5128// move file pointer to the specified offset 5129jlong os::seek_to_file_offset(int fd, jlong offset) { 5130 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5131} 5132 5133// This code originates from JDK's sysAvailable 5134// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5135 5136int os::available(int fd, jlong *bytes) { 5137 jlong cur, end; 5138 int mode; 5139 struct stat64 buf64; 5140 5141 if (::fstat64(fd, &buf64) >= 0) { 5142 mode = buf64.st_mode; 5143 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5144 // XXX: is the following call interruptible? If so, this might 5145 // need to go through the INTERRUPT_IO() wrapper as for other 5146 // blocking, interruptible calls in this file. 5147 int n; 5148 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5149 *bytes = n; 5150 return 1; 5151 } 5152 } 5153 } 5154 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5155 return 0; 5156 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5157 return 0; 5158 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5159 return 0; 5160 } 5161 *bytes = end - cur; 5162 return 1; 5163} 5164 5165// Map a block of memory. 5166char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5167 char *addr, size_t bytes, bool read_only, 5168 bool allow_exec) { 5169 int prot; 5170 int flags = MAP_PRIVATE; 5171 5172 if (read_only) { 5173 prot = PROT_READ; 5174 } else { 5175 prot = PROT_READ | PROT_WRITE; 5176 } 5177 5178 if (allow_exec) { 5179 prot |= PROT_EXEC; 5180 } 5181 5182 if (addr != NULL) { 5183 flags |= MAP_FIXED; 5184 } 5185 5186 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5187 fd, file_offset); 5188 if (mapped_address == MAP_FAILED) { 5189 return NULL; 5190 } 5191 return mapped_address; 5192} 5193 5194 5195// Remap a block of memory. 5196char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5197 char *addr, size_t bytes, bool read_only, 5198 bool allow_exec) { 5199 // same as map_memory() on this OS 5200 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5201 allow_exec); 5202} 5203 5204 5205// Unmap a block of memory. 5206bool os::pd_unmap_memory(char* addr, size_t bytes) { 5207 return munmap(addr, bytes) == 0; 5208} 5209 5210static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5211 5212static clockid_t thread_cpu_clockid(Thread* thread) { 5213 pthread_t tid = thread->osthread()->pthread_id(); 5214 clockid_t clockid; 5215 5216 // Get thread clockid 5217 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5218 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5219 return clockid; 5220} 5221 5222// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5223// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5224// of a thread. 5225// 5226// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5227// the fast estimate available on the platform. 5228 5229jlong os::current_thread_cpu_time() { 5230 if (os::Linux::supports_fast_thread_cpu_time()) { 5231 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5232 } else { 5233 // return user + sys since the cost is the same 5234 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5235 } 5236} 5237 5238jlong os::thread_cpu_time(Thread* thread) { 5239 // consistent with what current_thread_cpu_time() returns 5240 if (os::Linux::supports_fast_thread_cpu_time()) { 5241 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5242 } else { 5243 return slow_thread_cpu_time(thread, true /* user + sys */); 5244 } 5245} 5246 5247jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5248 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5249 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5250 } else { 5251 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5252 } 5253} 5254 5255jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5256 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5257 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5258 } else { 5259 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5260 } 5261} 5262 5263// -1 on error. 5264static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5265 pid_t tid = thread->osthread()->thread_id(); 5266 char *s; 5267 char stat[2048]; 5268 int statlen; 5269 char proc_name[64]; 5270 int count; 5271 long sys_time, user_time; 5272 char cdummy; 5273 int idummy; 5274 long ldummy; 5275 FILE *fp; 5276 5277 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5278 fp = fopen(proc_name, "r"); 5279 if (fp == NULL) return -1; 5280 statlen = fread(stat, 1, 2047, fp); 5281 stat[statlen] = '\0'; 5282 fclose(fp); 5283 5284 // Skip pid and the command string. Note that we could be dealing with 5285 // weird command names, e.g. user could decide to rename java launcher 5286 // to "java 1.4.2 :)", then the stat file would look like 5287 // 1234 (java 1.4.2 :)) R ... ... 5288 // We don't really need to know the command string, just find the last 5289 // occurrence of ")" and then start parsing from there. See bug 4726580. 5290 s = strrchr(stat, ')'); 5291 if (s == NULL) return -1; 5292 5293 // Skip blank chars 5294 do { s++; } while (s && isspace(*s)); 5295 5296 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5297 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5298 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5299 &user_time, &sys_time); 5300 if (count != 13) return -1; 5301 if (user_sys_cpu_time) { 5302 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5303 } else { 5304 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5305 } 5306} 5307 5308void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5309 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5310 info_ptr->may_skip_backward = false; // elapsed time not wall time 5311 info_ptr->may_skip_forward = false; // elapsed time not wall time 5312 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5313} 5314 5315void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5316 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5317 info_ptr->may_skip_backward = false; // elapsed time not wall time 5318 info_ptr->may_skip_forward = false; // elapsed time not wall time 5319 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5320} 5321 5322bool os::is_thread_cpu_time_supported() { 5323 return true; 5324} 5325 5326// System loadavg support. Returns -1 if load average cannot be obtained. 5327// Linux doesn't yet have a (official) notion of processor sets, 5328// so just return the system wide load average. 5329int os::loadavg(double loadavg[], int nelem) { 5330 return ::getloadavg(loadavg, nelem); 5331} 5332 5333void os::pause() { 5334 char filename[MAX_PATH]; 5335 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5336 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5337 } else { 5338 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5339 } 5340 5341 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5342 if (fd != -1) { 5343 struct stat buf; 5344 ::close(fd); 5345 while (::stat(filename, &buf) == 0) { 5346 (void)::poll(NULL, 0, 100); 5347 } 5348 } else { 5349 jio_fprintf(stderr, 5350 "Could not open pause file '%s', continuing immediately.\n", filename); 5351 } 5352} 5353 5354 5355// Refer to the comments in os_solaris.cpp park-unpark. The next two 5356// comment paragraphs are worth repeating here: 5357// 5358// Assumption: 5359// Only one parker can exist on an event, which is why we allocate 5360// them per-thread. Multiple unparkers can coexist. 5361// 5362// _Event serves as a restricted-range semaphore. 5363// -1 : thread is blocked, i.e. there is a waiter 5364// 0 : neutral: thread is running or ready, 5365// could have been signaled after a wait started 5366// 1 : signaled - thread is running or ready 5367// 5368// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5369// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5370// For specifics regarding the bug see GLIBC BUGID 261237 : 5371// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5372// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5373// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5374// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5375// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5376// and monitorenter when we're using 1-0 locking. All those operations may result in 5377// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5378// of libpthread avoids the problem, but isn't practical. 5379// 5380// Possible remedies: 5381// 5382// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5383// This is palliative and probabilistic, however. If the thread is preempted 5384// between the call to compute_abstime() and pthread_cond_timedwait(), more 5385// than the minimum period may have passed, and the abstime may be stale (in the 5386// past) resultin in a hang. Using this technique reduces the odds of a hang 5387// but the JVM is still vulnerable, particularly on heavily loaded systems. 5388// 5389// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5390// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5391// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5392// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5393// thread. 5394// 5395// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5396// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5397// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5398// This also works well. In fact it avoids kernel-level scalability impediments 5399// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5400// timers in a graceful fashion. 5401// 5402// 4. When the abstime value is in the past it appears that control returns 5403// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5404// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5405// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5406// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5407// It may be possible to avoid reinitialization by checking the return 5408// value from pthread_cond_timedwait(). In addition to reinitializing the 5409// condvar we must establish the invariant that cond_signal() is only called 5410// within critical sections protected by the adjunct mutex. This prevents 5411// cond_signal() from "seeing" a condvar that's in the midst of being 5412// reinitialized or that is corrupt. Sadly, this invariant obviates the 5413// desirable signal-after-unlock optimization that avoids futile context switching. 5414// 5415// I'm also concerned that some versions of NTPL might allocate an auxilliary 5416// structure when a condvar is used or initialized. cond_destroy() would 5417// release the helper structure. Our reinitialize-after-timedwait fix 5418// put excessive stress on malloc/free and locks protecting the c-heap. 5419// 5420// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5421// It may be possible to refine (4) by checking the kernel and NTPL verisons 5422// and only enabling the work-around for vulnerable environments. 5423 5424// utility to compute the abstime argument to timedwait: 5425// millis is the relative timeout time 5426// abstime will be the absolute timeout time 5427// TODO: replace compute_abstime() with unpackTime() 5428 5429static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5430 if (millis < 0) millis = 0; 5431 5432 jlong seconds = millis / 1000; 5433 millis %= 1000; 5434 if (seconds > 50000000) { // see man cond_timedwait(3T) 5435 seconds = 50000000; 5436 } 5437 5438 if (os::supports_monotonic_clock()) { 5439 struct timespec now; 5440 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5441 assert_status(status == 0, status, "clock_gettime"); 5442 abstime->tv_sec = now.tv_sec + seconds; 5443 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5444 if (nanos >= NANOSECS_PER_SEC) { 5445 abstime->tv_sec += 1; 5446 nanos -= NANOSECS_PER_SEC; 5447 } 5448 abstime->tv_nsec = nanos; 5449 } else { 5450 struct timeval now; 5451 int status = gettimeofday(&now, NULL); 5452 assert(status == 0, "gettimeofday"); 5453 abstime->tv_sec = now.tv_sec + seconds; 5454 long usec = now.tv_usec + millis * 1000; 5455 if (usec >= 1000000) { 5456 abstime->tv_sec += 1; 5457 usec -= 1000000; 5458 } 5459 abstime->tv_nsec = usec * 1000; 5460 } 5461 return abstime; 5462} 5463 5464void os::PlatformEvent::park() { // AKA "down()" 5465 // Transitions for _Event: 5466 // -1 => -1 : illegal 5467 // 1 => 0 : pass - return immediately 5468 // 0 => -1 : block; then set _Event to 0 before returning 5469 5470 // Invariant: Only the thread associated with the Event/PlatformEvent 5471 // may call park(). 5472 // TODO: assert that _Assoc != NULL or _Assoc == Self 5473 assert(_nParked == 0, "invariant"); 5474 5475 int v; 5476 for (;;) { 5477 v = _Event; 5478 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5479 } 5480 guarantee(v >= 0, "invariant"); 5481 if (v == 0) { 5482 // Do this the hard way by blocking ... 5483 int status = pthread_mutex_lock(_mutex); 5484 assert_status(status == 0, status, "mutex_lock"); 5485 guarantee(_nParked == 0, "invariant"); 5486 ++_nParked; 5487 while (_Event < 0) { 5488 status = pthread_cond_wait(_cond, _mutex); 5489 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5490 // Treat this the same as if the wait was interrupted 5491 if (status == ETIME) { status = EINTR; } 5492 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5493 } 5494 --_nParked; 5495 5496 _Event = 0; 5497 status = pthread_mutex_unlock(_mutex); 5498 assert_status(status == 0, status, "mutex_unlock"); 5499 // Paranoia to ensure our locked and lock-free paths interact 5500 // correctly with each other. 5501 OrderAccess::fence(); 5502 } 5503 guarantee(_Event >= 0, "invariant"); 5504} 5505 5506int os::PlatformEvent::park(jlong millis) { 5507 // Transitions for _Event: 5508 // -1 => -1 : illegal 5509 // 1 => 0 : pass - return immediately 5510 // 0 => -1 : block; then set _Event to 0 before returning 5511 5512 guarantee(_nParked == 0, "invariant"); 5513 5514 int v; 5515 for (;;) { 5516 v = _Event; 5517 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5518 } 5519 guarantee(v >= 0, "invariant"); 5520 if (v != 0) return OS_OK; 5521 5522 // We do this the hard way, by blocking the thread. 5523 // Consider enforcing a minimum timeout value. 5524 struct timespec abst; 5525 compute_abstime(&abst, millis); 5526 5527 int ret = OS_TIMEOUT; 5528 int status = pthread_mutex_lock(_mutex); 5529 assert_status(status == 0, status, "mutex_lock"); 5530 guarantee(_nParked == 0, "invariant"); 5531 ++_nParked; 5532 5533 // Object.wait(timo) will return because of 5534 // (a) notification 5535 // (b) timeout 5536 // (c) thread.interrupt 5537 // 5538 // Thread.interrupt and object.notify{All} both call Event::set. 5539 // That is, we treat thread.interrupt as a special case of notification. 5540 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false. 5541 // We assume all ETIME returns are valid. 5542 // 5543 // TODO: properly differentiate simultaneous notify+interrupt. 5544 // In that case, we should propagate the notify to another waiter. 5545 5546 while (_Event < 0) { 5547 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5548 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5549 pthread_cond_destroy(_cond); 5550 pthread_cond_init(_cond, os::Linux::condAttr()); 5551 } 5552 assert_status(status == 0 || status == EINTR || 5553 status == ETIME || status == ETIMEDOUT, 5554 status, "cond_timedwait"); 5555 if (!FilterSpuriousWakeups) break; // previous semantics 5556 if (status == ETIME || status == ETIMEDOUT) break; 5557 // We consume and ignore EINTR and spurious wakeups. 5558 } 5559 --_nParked; 5560 if (_Event >= 0) { 5561 ret = OS_OK; 5562 } 5563 _Event = 0; 5564 status = pthread_mutex_unlock(_mutex); 5565 assert_status(status == 0, status, "mutex_unlock"); 5566 assert(_nParked == 0, "invariant"); 5567 // Paranoia to ensure our locked and lock-free paths interact 5568 // correctly with each other. 5569 OrderAccess::fence(); 5570 return ret; 5571} 5572 5573void os::PlatformEvent::unpark() { 5574 // Transitions for _Event: 5575 // 0 => 1 : just return 5576 // 1 => 1 : just return 5577 // -1 => either 0 or 1; must signal target thread 5578 // That is, we can safely transition _Event from -1 to either 5579 // 0 or 1. 5580 // See also: "Semaphores in Plan 9" by Mullender & Cox 5581 // 5582 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5583 // that it will take two back-to-back park() calls for the owning 5584 // thread to block. This has the benefit of forcing a spurious return 5585 // from the first park() call after an unpark() call which will help 5586 // shake out uses of park() and unpark() without condition variables. 5587 5588 if (Atomic::xchg(1, &_Event) >= 0) return; 5589 5590 // Wait for the thread associated with the event to vacate 5591 int status = pthread_mutex_lock(_mutex); 5592 assert_status(status == 0, status, "mutex_lock"); 5593 int AnyWaiters = _nParked; 5594 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5595 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5596 AnyWaiters = 0; 5597 pthread_cond_signal(_cond); 5598 } 5599 status = pthread_mutex_unlock(_mutex); 5600 assert_status(status == 0, status, "mutex_unlock"); 5601 if (AnyWaiters != 0) { 5602 // Note that we signal() *after* dropping the lock for "immortal" Events. 5603 // This is safe and avoids a common class of futile wakeups. In rare 5604 // circumstances this can cause a thread to return prematurely from 5605 // cond_{timed}wait() but the spurious wakeup is benign and the victim 5606 // will simply re-test the condition and re-park itself. 5607 // This provides particular benefit if the underlying platform does not 5608 // provide wait morphing. 5609 status = pthread_cond_signal(_cond); 5610 assert_status(status == 0, status, "cond_signal"); 5611 } 5612} 5613 5614 5615// JSR166 5616// ------------------------------------------------------- 5617 5618// The solaris and linux implementations of park/unpark are fairly 5619// conservative for now, but can be improved. They currently use a 5620// mutex/condvar pair, plus a a count. 5621// Park decrements count if > 0, else does a condvar wait. Unpark 5622// sets count to 1 and signals condvar. Only one thread ever waits 5623// on the condvar. Contention seen when trying to park implies that someone 5624// is unparking you, so don't wait. And spurious returns are fine, so there 5625// is no need to track notifications. 5626 5627// This code is common to linux and solaris and will be moved to a 5628// common place in dolphin. 5629// 5630// The passed in time value is either a relative time in nanoseconds 5631// or an absolute time in milliseconds. Either way it has to be unpacked 5632// into suitable seconds and nanoseconds components and stored in the 5633// given timespec structure. 5634// Given time is a 64-bit value and the time_t used in the timespec is only 5635// a signed-32-bit value (except on 64-bit Linux) we have to watch for 5636// overflow if times way in the future are given. Further on Solaris versions 5637// prior to 10 there is a restriction (see cond_timedwait) that the specified 5638// number of seconds, in abstime, is less than current_time + 100,000,000. 5639// As it will be 28 years before "now + 100000000" will overflow we can 5640// ignore overflow and just impose a hard-limit on seconds using the value 5641// of "now + 100,000,000". This places a limit on the timeout of about 3.17 5642// years from "now". 5643 5644static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5645 assert(time > 0, "convertTime"); 5646 time_t max_secs = 0; 5647 5648 if (!os::supports_monotonic_clock() || isAbsolute) { 5649 struct timeval now; 5650 int status = gettimeofday(&now, NULL); 5651 assert(status == 0, "gettimeofday"); 5652 5653 max_secs = now.tv_sec + MAX_SECS; 5654 5655 if (isAbsolute) { 5656 jlong secs = time / 1000; 5657 if (secs > max_secs) { 5658 absTime->tv_sec = max_secs; 5659 } else { 5660 absTime->tv_sec = secs; 5661 } 5662 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5663 } else { 5664 jlong secs = time / NANOSECS_PER_SEC; 5665 if (secs >= MAX_SECS) { 5666 absTime->tv_sec = max_secs; 5667 absTime->tv_nsec = 0; 5668 } else { 5669 absTime->tv_sec = now.tv_sec + secs; 5670 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5671 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5672 absTime->tv_nsec -= NANOSECS_PER_SEC; 5673 ++absTime->tv_sec; // note: this must be <= max_secs 5674 } 5675 } 5676 } 5677 } else { 5678 // must be relative using monotonic clock 5679 struct timespec now; 5680 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5681 assert_status(status == 0, status, "clock_gettime"); 5682 max_secs = now.tv_sec + MAX_SECS; 5683 jlong secs = time / NANOSECS_PER_SEC; 5684 if (secs >= MAX_SECS) { 5685 absTime->tv_sec = max_secs; 5686 absTime->tv_nsec = 0; 5687 } else { 5688 absTime->tv_sec = now.tv_sec + secs; 5689 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5690 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5691 absTime->tv_nsec -= NANOSECS_PER_SEC; 5692 ++absTime->tv_sec; // note: this must be <= max_secs 5693 } 5694 } 5695 } 5696 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5697 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5698 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5699 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5700} 5701 5702void Parker::park(bool isAbsolute, jlong time) { 5703 // Ideally we'd do something useful while spinning, such 5704 // as calling unpackTime(). 5705 5706 // Optional fast-path check: 5707 // Return immediately if a permit is available. 5708 // We depend on Atomic::xchg() having full barrier semantics 5709 // since we are doing a lock-free update to _counter. 5710 if (Atomic::xchg(0, &_counter) > 0) return; 5711 5712 Thread* thread = Thread::current(); 5713 assert(thread->is_Java_thread(), "Must be JavaThread"); 5714 JavaThread *jt = (JavaThread *)thread; 5715 5716 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5717 // Check interrupt before trying to wait 5718 if (Thread::is_interrupted(thread, false)) { 5719 return; 5720 } 5721 5722 // Next, demultiplex/decode time arguments 5723 timespec absTime; 5724 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all 5725 return; 5726 } 5727 if (time > 0) { 5728 unpackTime(&absTime, isAbsolute, time); 5729 } 5730 5731 5732 // Enter safepoint region 5733 // Beware of deadlocks such as 6317397. 5734 // The per-thread Parker:: mutex is a classic leaf-lock. 5735 // In particular a thread must never block on the Threads_lock while 5736 // holding the Parker:: mutex. If safepoints are pending both the 5737 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5738 ThreadBlockInVM tbivm(jt); 5739 5740 // Don't wait if cannot get lock since interference arises from 5741 // unblocking. Also. check interrupt before trying wait 5742 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5743 return; 5744 } 5745 5746 int status; 5747 if (_counter > 0) { // no wait needed 5748 _counter = 0; 5749 status = pthread_mutex_unlock(_mutex); 5750 assert(status == 0, "invariant"); 5751 // Paranoia to ensure our locked and lock-free paths interact 5752 // correctly with each other and Java-level accesses. 5753 OrderAccess::fence(); 5754 return; 5755 } 5756 5757#ifdef ASSERT 5758 // Don't catch signals while blocked; let the running threads have the signals. 5759 // (This allows a debugger to break into the running thread.) 5760 sigset_t oldsigs; 5761 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5762 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5763#endif 5764 5765 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5766 jt->set_suspend_equivalent(); 5767 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5768 5769 assert(_cur_index == -1, "invariant"); 5770 if (time == 0) { 5771 _cur_index = REL_INDEX; // arbitrary choice when not timed 5772 status = pthread_cond_wait(&_cond[_cur_index], _mutex); 5773 } else { 5774 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5775 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime); 5776 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5777 pthread_cond_destroy(&_cond[_cur_index]); 5778 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr()); 5779 } 5780 } 5781 _cur_index = -1; 5782 assert_status(status == 0 || status == EINTR || 5783 status == ETIME || status == ETIMEDOUT, 5784 status, "cond_timedwait"); 5785 5786#ifdef ASSERT 5787 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5788#endif 5789 5790 _counter = 0; 5791 status = pthread_mutex_unlock(_mutex); 5792 assert_status(status == 0, status, "invariant"); 5793 // Paranoia to ensure our locked and lock-free paths interact 5794 // correctly with each other and Java-level accesses. 5795 OrderAccess::fence(); 5796 5797 // If externally suspended while waiting, re-suspend 5798 if (jt->handle_special_suspend_equivalent_condition()) { 5799 jt->java_suspend_self(); 5800 } 5801} 5802 5803void Parker::unpark() { 5804 int status = pthread_mutex_lock(_mutex); 5805 assert(status == 0, "invariant"); 5806 const int s = _counter; 5807 _counter = 1; 5808 if (s < 1) { 5809 // thread might be parked 5810 if (_cur_index != -1) { 5811 // thread is definitely parked 5812 if (WorkAroundNPTLTimedWaitHang) { 5813 status = pthread_cond_signal(&_cond[_cur_index]); 5814 assert(status == 0, "invariant"); 5815 status = pthread_mutex_unlock(_mutex); 5816 assert(status == 0, "invariant"); 5817 } else { 5818 status = pthread_mutex_unlock(_mutex); 5819 assert(status == 0, "invariant"); 5820 status = pthread_cond_signal(&_cond[_cur_index]); 5821 assert(status == 0, "invariant"); 5822 } 5823 } else { 5824 pthread_mutex_unlock(_mutex); 5825 assert(status == 0, "invariant"); 5826 } 5827 } else { 5828 pthread_mutex_unlock(_mutex); 5829 assert(status == 0, "invariant"); 5830 } 5831} 5832 5833 5834extern char** environ; 5835 5836#ifndef __NR_fork 5837 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) AARCH64_ONLY(1079) 5838#endif 5839 5840#ifndef __NR_execve 5841 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) AARCH64_ONLY(221) 5842#endif 5843 5844// Run the specified command in a separate process. Return its exit value, 5845// or -1 on failure (e.g. can't fork a new process). 5846// Unlike system(), this function can be called from signal handler. It 5847// doesn't block SIGINT et al. 5848int os::fork_and_exec(char* cmd) { 5849 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5850 5851 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run 5852 // pthread_atfork handlers and reset pthread library. All we need is a 5853 // separate process to execve. Make a direct syscall to fork process. 5854 // On IA64 there's no fork syscall, we have to use fork() and hope for 5855 // the best... 5856 pid_t pid = NOT_IA64(syscall(__NR_fork);) 5857 IA64_ONLY(fork();) 5858 5859 if (pid < 0) { 5860 // fork failed 5861 return -1; 5862 5863 } else if (pid == 0) { 5864 // child process 5865 5866 // execve() in LinuxThreads will call pthread_kill_other_threads_np() 5867 // first to kill every thread on the thread list. Because this list is 5868 // not reset by fork() (see notes above), execve() will instead kill 5869 // every thread in the parent process. We know this is the only thread 5870 // in the new process, so make a system call directly. 5871 // IA64 should use normal execve() from glibc to match the glibc fork() 5872 // above. 5873 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);) 5874 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);) 5875 5876 // execve failed 5877 _exit(-1); 5878 5879 } else { 5880 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5881 // care about the actual exit code, for now. 5882 5883 int status; 5884 5885 // Wait for the child process to exit. This returns immediately if 5886 // the child has already exited. */ 5887 while (waitpid(pid, &status, 0) < 0) { 5888 switch (errno) { 5889 case ECHILD: return 0; 5890 case EINTR: break; 5891 default: return -1; 5892 } 5893 } 5894 5895 if (WIFEXITED(status)) { 5896 // The child exited normally; get its exit code. 5897 return WEXITSTATUS(status); 5898 } else if (WIFSIGNALED(status)) { 5899 // The child exited because of a signal 5900 // The best value to return is 0x80 + signal number, 5901 // because that is what all Unix shells do, and because 5902 // it allows callers to distinguish between process exit and 5903 // process death by signal. 5904 return 0x80 + WTERMSIG(status); 5905 } else { 5906 // Unknown exit code; pass it through 5907 return status; 5908 } 5909 } 5910} 5911 5912// is_headless_jre() 5913// 5914// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5915// in order to report if we are running in a headless jre 5916// 5917// Since JDK8 xawt/libmawt.so was moved into the same directory 5918// as libawt.so, and renamed libawt_xawt.so 5919// 5920bool os::is_headless_jre() { 5921 struct stat statbuf; 5922 char buf[MAXPATHLEN]; 5923 char libmawtpath[MAXPATHLEN]; 5924 const char *xawtstr = "/xawt/libmawt.so"; 5925 const char *new_xawtstr = "/libawt_xawt.so"; 5926 char *p; 5927 5928 // Get path to libjvm.so 5929 os::jvm_path(buf, sizeof(buf)); 5930 5931 // Get rid of libjvm.so 5932 p = strrchr(buf, '/'); 5933 if (p == NULL) { 5934 return false; 5935 } else { 5936 *p = '\0'; 5937 } 5938 5939 // Get rid of client or server 5940 p = strrchr(buf, '/'); 5941 if (p == NULL) { 5942 return false; 5943 } else { 5944 *p = '\0'; 5945 } 5946 5947 // check xawt/libmawt.so 5948 strcpy(libmawtpath, buf); 5949 strcat(libmawtpath, xawtstr); 5950 if (::stat(libmawtpath, &statbuf) == 0) return false; 5951 5952 // check libawt_xawt.so 5953 strcpy(libmawtpath, buf); 5954 strcat(libmawtpath, new_xawtstr); 5955 if (::stat(libmawtpath, &statbuf) == 0) return false; 5956 5957 return true; 5958} 5959 5960// Get the default path to the core file 5961// Returns the length of the string 5962int os::get_core_path(char* buffer, size_t bufferSize) { 5963 /* 5964 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 5965 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 5966 */ 5967 const int core_pattern_len = 129; 5968 char core_pattern[core_pattern_len] = {0}; 5969 5970 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 5971 if (core_pattern_file != -1) { 5972 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 5973 ::close(core_pattern_file); 5974 5975 if (ret > 0) { 5976 char *last_char = core_pattern + strlen(core_pattern) - 1; 5977 5978 if (*last_char == '\n') { 5979 *last_char = '\0'; 5980 } 5981 } 5982 } 5983 5984 if (strlen(core_pattern) == 0) { 5985 return -1; 5986 } 5987 5988 char *pid_pos = strstr(core_pattern, "%p"); 5989 int written; 5990 5991 if (core_pattern[0] == '/') { 5992 written = jio_snprintf(buffer, bufferSize, core_pattern); 5993 } else { 5994 char cwd[PATH_MAX]; 5995 5996 const char* p = get_current_directory(cwd, PATH_MAX); 5997 if (p == NULL) { 5998 return -1; 5999 } 6000 6001 if (core_pattern[0] == '|') { 6002 written = jio_snprintf(buffer, bufferSize, 6003 "\"%s\" (or dumping to %s/core.%d)", 6004 &core_pattern[1], p, current_process_id()); 6005 } else { 6006 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 6007 } 6008 } 6009 6010 if (written < 0) { 6011 return -1; 6012 } 6013 6014 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 6015 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 6016 6017 if (core_uses_pid_file != -1) { 6018 char core_uses_pid = 0; 6019 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 6020 ::close(core_uses_pid_file); 6021 6022 if (core_uses_pid == '1') { 6023 jio_snprintf(buffer + written, bufferSize - written, 6024 ".%d", current_process_id()); 6025 } 6026 } 6027 } 6028 6029 return strlen(buffer); 6030} 6031 6032/////////////// Unit tests /////////////// 6033 6034#ifndef PRODUCT 6035 6036#define test_log(...) \ 6037 do { \ 6038 if (VerboseInternalVMTests) { \ 6039 tty->print_cr(__VA_ARGS__); \ 6040 tty->flush(); \ 6041 } \ 6042 } while (false) 6043 6044class TestReserveMemorySpecial : AllStatic { 6045 public: 6046 static void small_page_write(void* addr, size_t size) { 6047 size_t page_size = os::vm_page_size(); 6048 6049 char* end = (char*)addr + size; 6050 for (char* p = (char*)addr; p < end; p += page_size) { 6051 *p = 1; 6052 } 6053 } 6054 6055 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6056 if (!UseHugeTLBFS) { 6057 return; 6058 } 6059 6060 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6061 6062 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6063 6064 if (addr != NULL) { 6065 small_page_write(addr, size); 6066 6067 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6068 } 6069 } 6070 6071 static void test_reserve_memory_special_huge_tlbfs_only() { 6072 if (!UseHugeTLBFS) { 6073 return; 6074 } 6075 6076 size_t lp = os::large_page_size(); 6077 6078 for (size_t size = lp; size <= lp * 10; size += lp) { 6079 test_reserve_memory_special_huge_tlbfs_only(size); 6080 } 6081 } 6082 6083 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) { 6084 if (!UseHugeTLBFS) { 6085 return; 6086 } 6087 6088 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")", 6089 size, alignment); 6090 6091 assert(size >= os::large_page_size(), "Incorrect input to test"); 6092 6093 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6094 6095 if (addr != NULL) { 6096 small_page_write(addr, size); 6097 6098 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6099 } 6100 } 6101 6102 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) { 6103 size_t lp = os::large_page_size(); 6104 size_t ag = os::vm_allocation_granularity(); 6105 6106 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6107 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment); 6108 } 6109 } 6110 6111 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6112 size_t lp = os::large_page_size(); 6113 size_t ag = os::vm_allocation_granularity(); 6114 6115 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp); 6116 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag); 6117 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2); 6118 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2); 6119 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag); 6120 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag); 6121 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2); 6122 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10); 6123 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2); 6124 } 6125 6126 static void test_reserve_memory_special_huge_tlbfs() { 6127 if (!UseHugeTLBFS) { 6128 return; 6129 } 6130 6131 test_reserve_memory_special_huge_tlbfs_only(); 6132 test_reserve_memory_special_huge_tlbfs_mixed(); 6133 } 6134 6135 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6136 if (!UseSHM) { 6137 return; 6138 } 6139 6140 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6141 6142 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6143 6144 if (addr != NULL) { 6145 assert(is_ptr_aligned(addr, alignment), "Check"); 6146 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6147 6148 small_page_write(addr, size); 6149 6150 os::Linux::release_memory_special_shm(addr, size); 6151 } 6152 } 6153 6154 static void test_reserve_memory_special_shm() { 6155 size_t lp = os::large_page_size(); 6156 size_t ag = os::vm_allocation_granularity(); 6157 6158 for (size_t size = ag; size < lp * 3; size += ag) { 6159 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6160 test_reserve_memory_special_shm(size, alignment); 6161 } 6162 } 6163 } 6164 6165 static void test() { 6166 test_reserve_memory_special_huge_tlbfs(); 6167 test_reserve_memory_special_shm(); 6168 } 6169}; 6170 6171void TestReserveMemorySpecial_test() { 6172 TestReserveMemorySpecial::test(); 6173} 6174 6175#endif 6176