os_linux.cpp revision 8575:5916110131c4
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::_supports_fast_thread_cpu_time = false; 139const char * os::Linux::_glibc_version = NULL; 140const char * os::Linux::_libpthread_version = NULL; 141pthread_condattr_t os::Linux::_condattr[1]; 142 143static jlong initial_time_count=0; 144 145static int clock_tics_per_sec = 100; 146 147// For diagnostics to print a message once. see run_periodic_checks 148static sigset_t check_signal_done; 149static bool check_signals = true; 150 151// Signal number used to suspend/resume a thread 152 153// do not use any signal number less than SIGSEGV, see 4355769 154static int SR_signum = SIGUSR2; 155sigset_t SR_sigset; 156 157// Declarations 158static void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 159 160// utility functions 161 162static int SR_initialize(); 163 164julong os::available_memory() { 165 return Linux::available_memory(); 166} 167 168julong os::Linux::available_memory() { 169 // values in struct sysinfo are "unsigned long" 170 struct sysinfo si; 171 sysinfo(&si); 172 173 return (julong)si.freeram * si.mem_unit; 174} 175 176julong os::physical_memory() { 177 return Linux::physical_memory(); 178} 179 180// Return true if user is running as root. 181 182bool os::have_special_privileges() { 183 static bool init = false; 184 static bool privileges = false; 185 if (!init) { 186 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 187 init = true; 188 } 189 return privileges; 190} 191 192 193#ifndef SYS_gettid 194// i386: 224, ia64: 1105, amd64: 186, sparc 143 195 #ifdef __ia64__ 196 #define SYS_gettid 1105 197 #else 198 #ifdef __i386__ 199 #define SYS_gettid 224 200 #else 201 #ifdef __amd64__ 202 #define SYS_gettid 186 203 #else 204 #ifdef __sparc__ 205 #define SYS_gettid 143 206 #else 207 #error define gettid for the arch 208 #endif 209 #endif 210 #endif 211 #endif 212#endif 213 214// Cpu architecture string 215static char cpu_arch[] = HOTSPOT_LIB_ARCH; 216 217 218// pid_t gettid() 219// 220// Returns the kernel thread id of the currently running thread. Kernel 221// thread id is used to access /proc. 222pid_t os::Linux::gettid() { 223 int rslt = syscall(SYS_gettid); 224 assert(rslt != -1, "must be."); // old linuxthreads implementation? 225 return (pid_t)rslt; 226} 227 228// Most versions of linux have a bug where the number of processors are 229// determined by looking at the /proc file system. In a chroot environment, 230// the system call returns 1. This causes the VM to act as if it is 231// a single processor and elide locking (see is_MP() call). 232static bool unsafe_chroot_detected = false; 233static const char *unstable_chroot_error = "/proc file system not found.\n" 234 "Java may be unstable running multithreaded in a chroot " 235 "environment on Linux when /proc filesystem is not mounted."; 236 237void os::Linux::initialize_system_info() { 238 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 239 if (processor_count() == 1) { 240 pid_t pid = os::Linux::gettid(); 241 char fname[32]; 242 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 243 FILE *fp = fopen(fname, "r"); 244 if (fp == NULL) { 245 unsafe_chroot_detected = true; 246 } else { 247 fclose(fp); 248 } 249 } 250 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 251 assert(processor_count() > 0, "linux error"); 252} 253 254void os::init_system_properties_values() { 255 // The next steps are taken in the product version: 256 // 257 // Obtain the JAVA_HOME value from the location of libjvm.so. 258 // This library should be located at: 259 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 260 // 261 // If "/jre/lib/" appears at the right place in the path, then we 262 // assume libjvm.so is installed in a JDK and we use this path. 263 // 264 // Otherwise exit with message: "Could not create the Java virtual machine." 265 // 266 // The following extra steps are taken in the debugging version: 267 // 268 // If "/jre/lib/" does NOT appear at the right place in the path 269 // instead of exit check for $JAVA_HOME environment variable. 270 // 271 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 272 // then we append a fake suffix "hotspot/libjvm.so" to this path so 273 // it looks like libjvm.so is installed there 274 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 275 // 276 // Otherwise exit. 277 // 278 // Important note: if the location of libjvm.so changes this 279 // code needs to be changed accordingly. 280 281 // See ld(1): 282 // The linker uses the following search paths to locate required 283 // shared libraries: 284 // 1: ... 285 // ... 286 // 7: The default directories, normally /lib and /usr/lib. 287#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 288 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 289#else 290 #define DEFAULT_LIBPATH "/lib:/usr/lib" 291#endif 292 293// Base path of extensions installed on the system. 294#define SYS_EXT_DIR "/usr/java/packages" 295#define EXTENSIONS_DIR "/lib/ext" 296 297 // Buffer that fits several sprintfs. 298 // Note that the space for the colon and the trailing null are provided 299 // by the nulls included by the sizeof operator. 300 const size_t bufsize = 301 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 302 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 303 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 304 305 // sysclasspath, java_home, dll_dir 306 { 307 char *pslash; 308 os::jvm_path(buf, bufsize); 309 310 // Found the full path to libjvm.so. 311 // Now cut the path to <java_home>/jre if we can. 312 pslash = strrchr(buf, '/'); 313 if (pslash != NULL) { 314 *pslash = '\0'; // Get rid of /libjvm.so. 315 } 316 pslash = strrchr(buf, '/'); 317 if (pslash != NULL) { 318 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 319 } 320 Arguments::set_dll_dir(buf); 321 322 if (pslash != NULL) { 323 pslash = strrchr(buf, '/'); 324 if (pslash != NULL) { 325 *pslash = '\0'; // Get rid of /<arch>. 326 pslash = strrchr(buf, '/'); 327 if (pslash != NULL) { 328 *pslash = '\0'; // Get rid of /lib. 329 } 330 } 331 } 332 Arguments::set_java_home(buf); 333 set_boot_path('/', ':'); 334 } 335 336 // Where to look for native libraries. 337 // 338 // Note: Due to a legacy implementation, most of the library path 339 // is set in the launcher. This was to accomodate linking restrictions 340 // on legacy Linux implementations (which are no longer supported). 341 // Eventually, all the library path setting will be done here. 342 // 343 // However, to prevent the proliferation of improperly built native 344 // libraries, the new path component /usr/java/packages is added here. 345 // Eventually, all the library path setting will be done here. 346 { 347 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 348 // should always exist (until the legacy problem cited above is 349 // addressed). 350 const char *v = ::getenv("LD_LIBRARY_PATH"); 351 const char *v_colon = ":"; 352 if (v == NULL) { v = ""; v_colon = ""; } 353 // That's +1 for the colon and +1 for the trailing '\0'. 354 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 355 strlen(v) + 1 + 356 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1, 357 mtInternal); 358 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch); 359 Arguments::set_library_path(ld_library_path); 360 FREE_C_HEAP_ARRAY(char, ld_library_path); 361 } 362 363 // Extensions directories. 364 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 365 Arguments::set_ext_dirs(buf); 366 367 FREE_C_HEAP_ARRAY(char, buf); 368 369#undef DEFAULT_LIBPATH 370#undef SYS_EXT_DIR 371#undef EXTENSIONS_DIR 372} 373 374//////////////////////////////////////////////////////////////////////////////// 375// breakpoint support 376 377void os::breakpoint() { 378 BREAKPOINT; 379} 380 381extern "C" void breakpoint() { 382 // use debugger to set breakpoint here 383} 384 385//////////////////////////////////////////////////////////////////////////////// 386// signal support 387 388debug_only(static bool signal_sets_initialized = false); 389static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 390 391bool os::Linux::is_sig_ignored(int sig) { 392 struct sigaction oact; 393 sigaction(sig, (struct sigaction*)NULL, &oact); 394 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 395 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 396 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) { 397 return true; 398 } else { 399 return false; 400 } 401} 402 403void os::Linux::signal_sets_init() { 404 // Should also have an assertion stating we are still single-threaded. 405 assert(!signal_sets_initialized, "Already initialized"); 406 // Fill in signals that are necessarily unblocked for all threads in 407 // the VM. Currently, we unblock the following signals: 408 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 409 // by -Xrs (=ReduceSignalUsage)); 410 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 411 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 412 // the dispositions or masks wrt these signals. 413 // Programs embedding the VM that want to use the above signals for their 414 // own purposes must, at this time, use the "-Xrs" option to prevent 415 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 416 // (See bug 4345157, and other related bugs). 417 // In reality, though, unblocking these signals is really a nop, since 418 // these signals are not blocked by default. 419 sigemptyset(&unblocked_sigs); 420 sigemptyset(&allowdebug_blocked_sigs); 421 sigaddset(&unblocked_sigs, SIGILL); 422 sigaddset(&unblocked_sigs, SIGSEGV); 423 sigaddset(&unblocked_sigs, SIGBUS); 424 sigaddset(&unblocked_sigs, SIGFPE); 425#if defined(PPC64) 426 sigaddset(&unblocked_sigs, SIGTRAP); 427#endif 428 sigaddset(&unblocked_sigs, SR_signum); 429 430 if (!ReduceSignalUsage) { 431 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 432 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 433 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 434 } 435 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 436 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 437 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 438 } 439 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 440 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 441 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 442 } 443 } 444 // Fill in signals that are blocked by all but the VM thread. 445 sigemptyset(&vm_sigs); 446 if (!ReduceSignalUsage) { 447 sigaddset(&vm_sigs, BREAK_SIGNAL); 448 } 449 debug_only(signal_sets_initialized = true); 450 451} 452 453// These are signals that are unblocked while a thread is running Java. 454// (For some reason, they get blocked by default.) 455sigset_t* os::Linux::unblocked_signals() { 456 assert(signal_sets_initialized, "Not initialized"); 457 return &unblocked_sigs; 458} 459 460// These are the signals that are blocked while a (non-VM) thread is 461// running Java. Only the VM thread handles these signals. 462sigset_t* os::Linux::vm_signals() { 463 assert(signal_sets_initialized, "Not initialized"); 464 return &vm_sigs; 465} 466 467// These are signals that are blocked during cond_wait to allow debugger in 468sigset_t* os::Linux::allowdebug_blocked_signals() { 469 assert(signal_sets_initialized, "Not initialized"); 470 return &allowdebug_blocked_sigs; 471} 472 473void os::Linux::hotspot_sigmask(Thread* thread) { 474 475 //Save caller's signal mask before setting VM signal mask 476 sigset_t caller_sigmask; 477 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 478 479 OSThread* osthread = thread->osthread(); 480 osthread->set_caller_sigmask(caller_sigmask); 481 482 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 483 484 if (!ReduceSignalUsage) { 485 if (thread->is_VM_thread()) { 486 // Only the VM thread handles BREAK_SIGNAL ... 487 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 488 } else { 489 // ... all other threads block BREAK_SIGNAL 490 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 491 } 492 } 493} 494 495////////////////////////////////////////////////////////////////////////////// 496// detecting pthread library 497 498void os::Linux::libpthread_init() { 499 // Save glibc and pthread version strings. 500#if !defined(_CS_GNU_LIBC_VERSION) || \ 501 !defined(_CS_GNU_LIBPTHREAD_VERSION) 502 #error "glibc too old (< 2.3.2)" 503#endif 504 505 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 506 assert(n > 0, "cannot retrieve glibc version"); 507 char *str = (char *)malloc(n, mtInternal); 508 confstr(_CS_GNU_LIBC_VERSION, str, n); 509 os::Linux::set_glibc_version(str); 510 511 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 512 assert(n > 0, "cannot retrieve pthread version"); 513 str = (char *)malloc(n, mtInternal); 514 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 515 os::Linux::set_libpthread_version(str); 516} 517 518///////////////////////////////////////////////////////////////////////////// 519// thread stack expansion 520 521// os::Linux::manually_expand_stack() takes care of expanding the thread 522// stack. Note that this is normally not needed: pthread stacks allocate 523// thread stack using mmap() without MAP_NORESERVE, so the stack is already 524// committed. Therefore it is not necessary to expand the stack manually. 525// 526// Manually expanding the stack was historically needed on LinuxThreads 527// thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays 528// it is kept to deal with very rare corner cases: 529// 530// For one, user may run the VM on an own implementation of threads 531// whose stacks are - like the old LinuxThreads - implemented using 532// mmap(MAP_GROWSDOWN). 533// 534// Also, this coding may be needed if the VM is running on the primordial 535// thread. Normally we avoid running on the primordial thread; however, 536// user may still invoke the VM on the primordial thread. 537// 538// The following historical comment describes the details about running 539// on a thread stack allocated with mmap(MAP_GROWSDOWN): 540 541 542// Force Linux kernel to expand current thread stack. If "bottom" is close 543// to the stack guard, caller should block all signals. 544// 545// MAP_GROWSDOWN: 546// A special mmap() flag that is used to implement thread stacks. It tells 547// kernel that the memory region should extend downwards when needed. This 548// allows early versions of LinuxThreads to only mmap the first few pages 549// when creating a new thread. Linux kernel will automatically expand thread 550// stack as needed (on page faults). 551// 552// However, because the memory region of a MAP_GROWSDOWN stack can grow on 553// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 554// region, it's hard to tell if the fault is due to a legitimate stack 555// access or because of reading/writing non-exist memory (e.g. buffer 556// overrun). As a rule, if the fault happens below current stack pointer, 557// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 558// application (see Linux kernel fault.c). 559// 560// This Linux feature can cause SIGSEGV when VM bangs thread stack for 561// stack overflow detection. 562// 563// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 564// not use MAP_GROWSDOWN. 565// 566// To get around the problem and allow stack banging on Linux, we need to 567// manually expand thread stack after receiving the SIGSEGV. 568// 569// There are two ways to expand thread stack to address "bottom", we used 570// both of them in JVM before 1.5: 571// 1. adjust stack pointer first so that it is below "bottom", and then 572// touch "bottom" 573// 2. mmap() the page in question 574// 575// Now alternate signal stack is gone, it's harder to use 2. For instance, 576// if current sp is already near the lower end of page 101, and we need to 577// call mmap() to map page 100, it is possible that part of the mmap() frame 578// will be placed in page 100. When page 100 is mapped, it is zero-filled. 579// That will destroy the mmap() frame and cause VM to crash. 580// 581// The following code works by adjusting sp first, then accessing the "bottom" 582// page to force a page fault. Linux kernel will then automatically expand the 583// stack mapping. 584// 585// _expand_stack_to() assumes its frame size is less than page size, which 586// should always be true if the function is not inlined. 587 588#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 589 #define NOINLINE 590#else 591 #define NOINLINE __attribute__ ((noinline)) 592#endif 593 594static void _expand_stack_to(address bottom) NOINLINE; 595 596static void _expand_stack_to(address bottom) { 597 address sp; 598 size_t size; 599 volatile char *p; 600 601 // Adjust bottom to point to the largest address within the same page, it 602 // gives us a one-page buffer if alloca() allocates slightly more memory. 603 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 604 bottom += os::Linux::page_size() - 1; 605 606 // sp might be slightly above current stack pointer; if that's the case, we 607 // will alloca() a little more space than necessary, which is OK. Don't use 608 // os::current_stack_pointer(), as its result can be slightly below current 609 // stack pointer, causing us to not alloca enough to reach "bottom". 610 sp = (address)&sp; 611 612 if (sp > bottom) { 613 size = sp - bottom; 614 p = (volatile char *)alloca(size); 615 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 616 p[0] = '\0'; 617 } 618} 619 620bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 621 assert(t!=NULL, "just checking"); 622 assert(t->osthread()->expanding_stack(), "expand should be set"); 623 assert(t->stack_base() != NULL, "stack_base was not initialized"); 624 625 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 626 sigset_t mask_all, old_sigset; 627 sigfillset(&mask_all); 628 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 629 _expand_stack_to(addr); 630 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 631 return true; 632 } 633 return false; 634} 635 636////////////////////////////////////////////////////////////////////////////// 637// create new thread 638 639// Thread start routine for all newly created threads 640static void *java_start(Thread *thread) { 641 // Try to randomize the cache line index of hot stack frames. 642 // This helps when threads of the same stack traces evict each other's 643 // cache lines. The threads can be either from the same JVM instance, or 644 // from different JVM instances. The benefit is especially true for 645 // processors with hyperthreading technology. 646 static int counter = 0; 647 int pid = os::current_process_id(); 648 alloca(((pid ^ counter++) & 7) * 128); 649 650 ThreadLocalStorage::set_thread(thread); 651 652 OSThread* osthread = thread->osthread(); 653 Monitor* sync = osthread->startThread_lock(); 654 655 // thread_id is kernel thread id (similar to Solaris LWP id) 656 osthread->set_thread_id(os::Linux::gettid()); 657 658 if (UseNUMA) { 659 int lgrp_id = os::numa_get_group_id(); 660 if (lgrp_id != -1) { 661 thread->set_lgrp_id(lgrp_id); 662 } 663 } 664 // initialize signal mask for this thread 665 os::Linux::hotspot_sigmask(thread); 666 667 // initialize floating point control register 668 os::Linux::init_thread_fpu_state(); 669 670 // handshaking with parent thread 671 { 672 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 673 674 // notify parent thread 675 osthread->set_state(INITIALIZED); 676 sync->notify_all(); 677 678 // wait until os::start_thread() 679 while (osthread->get_state() == INITIALIZED) { 680 sync->wait(Mutex::_no_safepoint_check_flag); 681 } 682 } 683 684 // call one more level start routine 685 thread->run(); 686 687 return 0; 688} 689 690bool os::create_thread(Thread* thread, ThreadType thr_type, 691 size_t stack_size) { 692 assert(thread->osthread() == NULL, "caller responsible"); 693 694 // Allocate the OSThread object 695 OSThread* osthread = new OSThread(NULL, NULL); 696 if (osthread == NULL) { 697 return false; 698 } 699 700 // set the correct thread state 701 osthread->set_thread_type(thr_type); 702 703 // Initial state is ALLOCATED but not INITIALIZED 704 osthread->set_state(ALLOCATED); 705 706 thread->set_osthread(osthread); 707 708 // init thread attributes 709 pthread_attr_t attr; 710 pthread_attr_init(&attr); 711 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 712 713 // stack size 714 if (os::Linux::supports_variable_stack_size()) { 715 // calculate stack size if it's not specified by caller 716 if (stack_size == 0) { 717 stack_size = os::Linux::default_stack_size(thr_type); 718 719 switch (thr_type) { 720 case os::java_thread: 721 // Java threads use ThreadStackSize which default value can be 722 // changed with the flag -Xss 723 assert(JavaThread::stack_size_at_create() > 0, "this should be set"); 724 stack_size = JavaThread::stack_size_at_create(); 725 break; 726 case os::compiler_thread: 727 if (CompilerThreadStackSize > 0) { 728 stack_size = (size_t)(CompilerThreadStackSize * K); 729 break; 730 } // else fall through: 731 // use VMThreadStackSize if CompilerThreadStackSize is not defined 732 case os::vm_thread: 733 case os::pgc_thread: 734 case os::cgc_thread: 735 case os::watcher_thread: 736 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 737 break; 738 } 739 } 740 741 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 742 pthread_attr_setstacksize(&attr, stack_size); 743 } else { 744 // let pthread_create() pick the default value. 745 } 746 747 // glibc guard page 748 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 749 750 ThreadState state; 751 752 { 753 pthread_t tid; 754 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 755 756 pthread_attr_destroy(&attr); 757 758 if (ret != 0) { 759 if (PrintMiscellaneous && (Verbose || WizardMode)) { 760 perror("pthread_create()"); 761 } 762 // Need to clean up stuff we've allocated so far 763 thread->set_osthread(NULL); 764 delete osthread; 765 return false; 766 } 767 768 // Store pthread info into the OSThread 769 osthread->set_pthread_id(tid); 770 771 // Wait until child thread is either initialized or aborted 772 { 773 Monitor* sync_with_child = osthread->startThread_lock(); 774 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 775 while ((state = osthread->get_state()) == ALLOCATED) { 776 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 777 } 778 } 779 } 780 781 // Aborted due to thread limit being reached 782 if (state == ZOMBIE) { 783 thread->set_osthread(NULL); 784 delete osthread; 785 return false; 786 } 787 788 // The thread is returned suspended (in state INITIALIZED), 789 // and is started higher up in the call chain 790 assert(state == INITIALIZED, "race condition"); 791 return true; 792} 793 794///////////////////////////////////////////////////////////////////////////// 795// attach existing thread 796 797// bootstrap the main thread 798bool os::create_main_thread(JavaThread* thread) { 799 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 800 return create_attached_thread(thread); 801} 802 803bool os::create_attached_thread(JavaThread* thread) { 804#ifdef ASSERT 805 thread->verify_not_published(); 806#endif 807 808 // Allocate the OSThread object 809 OSThread* osthread = new OSThread(NULL, NULL); 810 811 if (osthread == NULL) { 812 return false; 813 } 814 815 // Store pthread info into the OSThread 816 osthread->set_thread_id(os::Linux::gettid()); 817 osthread->set_pthread_id(::pthread_self()); 818 819 // initialize floating point control register 820 os::Linux::init_thread_fpu_state(); 821 822 // Initial thread state is RUNNABLE 823 osthread->set_state(RUNNABLE); 824 825 thread->set_osthread(osthread); 826 827 if (UseNUMA) { 828 int lgrp_id = os::numa_get_group_id(); 829 if (lgrp_id != -1) { 830 thread->set_lgrp_id(lgrp_id); 831 } 832 } 833 834 if (os::Linux::is_initial_thread()) { 835 // If current thread is initial thread, its stack is mapped on demand, 836 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 837 // the entire stack region to avoid SEGV in stack banging. 838 // It is also useful to get around the heap-stack-gap problem on SuSE 839 // kernel (see 4821821 for details). We first expand stack to the top 840 // of yellow zone, then enable stack yellow zone (order is significant, 841 // enabling yellow zone first will crash JVM on SuSE Linux), so there 842 // is no gap between the last two virtual memory regions. 843 844 JavaThread *jt = (JavaThread *)thread; 845 address addr = jt->stack_yellow_zone_base(); 846 assert(addr != NULL, "initialization problem?"); 847 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 848 849 osthread->set_expanding_stack(); 850 os::Linux::manually_expand_stack(jt, addr); 851 osthread->clear_expanding_stack(); 852 } 853 854 // initialize signal mask for this thread 855 // and save the caller's signal mask 856 os::Linux::hotspot_sigmask(thread); 857 858 return true; 859} 860 861void os::pd_start_thread(Thread* thread) { 862 OSThread * osthread = thread->osthread(); 863 assert(osthread->get_state() != INITIALIZED, "just checking"); 864 Monitor* sync_with_child = osthread->startThread_lock(); 865 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 866 sync_with_child->notify(); 867} 868 869// Free Linux resources related to the OSThread 870void os::free_thread(OSThread* osthread) { 871 assert(osthread != NULL, "osthread not set"); 872 873 if (Thread::current()->osthread() == osthread) { 874 // Restore caller's signal mask 875 sigset_t sigmask = osthread->caller_sigmask(); 876 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 877 } 878 879 delete osthread; 880} 881 882////////////////////////////////////////////////////////////////////////////// 883// thread local storage 884 885// Restore the thread pointer if the destructor is called. This is in case 886// someone from JNI code sets up a destructor with pthread_key_create to run 887// detachCurrentThread on thread death. Unless we restore the thread pointer we 888// will hang or crash. When detachCurrentThread is called the key will be set 889// to null and we will not be called again. If detachCurrentThread is never 890// called we could loop forever depending on the pthread implementation. 891static void restore_thread_pointer(void* p) { 892 Thread* thread = (Thread*) p; 893 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread); 894} 895 896int os::allocate_thread_local_storage() { 897 pthread_key_t key; 898 int rslt = pthread_key_create(&key, restore_thread_pointer); 899 assert(rslt == 0, "cannot allocate thread local storage"); 900 return (int)key; 901} 902 903// Note: This is currently not used by VM, as we don't destroy TLS key 904// on VM exit. 905void os::free_thread_local_storage(int index) { 906 int rslt = pthread_key_delete((pthread_key_t)index); 907 assert(rslt == 0, "invalid index"); 908} 909 910void os::thread_local_storage_at_put(int index, void* value) { 911 int rslt = pthread_setspecific((pthread_key_t)index, value); 912 assert(rslt == 0, "pthread_setspecific failed"); 913} 914 915extern "C" Thread* get_thread() { 916 return ThreadLocalStorage::thread(); 917} 918 919////////////////////////////////////////////////////////////////////////////// 920// initial thread 921 922// Check if current thread is the initial thread, similar to Solaris thr_main. 923bool os::Linux::is_initial_thread(void) { 924 char dummy; 925 // If called before init complete, thread stack bottom will be null. 926 // Can be called if fatal error occurs before initialization. 927 if (initial_thread_stack_bottom() == NULL) return false; 928 assert(initial_thread_stack_bottom() != NULL && 929 initial_thread_stack_size() != 0, 930 "os::init did not locate initial thread's stack region"); 931 if ((address)&dummy >= initial_thread_stack_bottom() && 932 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) { 933 return true; 934 } else { 935 return false; 936 } 937} 938 939// Find the virtual memory area that contains addr 940static bool find_vma(address addr, address* vma_low, address* vma_high) { 941 FILE *fp = fopen("/proc/self/maps", "r"); 942 if (fp) { 943 address low, high; 944 while (!feof(fp)) { 945 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 946 if (low <= addr && addr < high) { 947 if (vma_low) *vma_low = low; 948 if (vma_high) *vma_high = high; 949 fclose(fp); 950 return true; 951 } 952 } 953 for (;;) { 954 int ch = fgetc(fp); 955 if (ch == EOF || ch == (int)'\n') break; 956 } 957 } 958 fclose(fp); 959 } 960 return false; 961} 962 963// Locate initial thread stack. This special handling of initial thread stack 964// is needed because pthread_getattr_np() on most (all?) Linux distros returns 965// bogus value for initial thread. 966void os::Linux::capture_initial_stack(size_t max_size) { 967 // stack size is the easy part, get it from RLIMIT_STACK 968 size_t stack_size; 969 struct rlimit rlim; 970 getrlimit(RLIMIT_STACK, &rlim); 971 stack_size = rlim.rlim_cur; 972 973 // 6308388: a bug in ld.so will relocate its own .data section to the 974 // lower end of primordial stack; reduce ulimit -s value a little bit 975 // so we won't install guard page on ld.so's data section. 976 stack_size -= 2 * page_size(); 977 978 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 979 // 7.1, in both cases we will get 2G in return value. 980 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 981 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 982 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 983 // in case other parts in glibc still assumes 2M max stack size. 984 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 985 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 986 if (stack_size > 2 * K * K IA64_ONLY(*2)) { 987 stack_size = 2 * K * K IA64_ONLY(*2); 988 } 989 // Try to figure out where the stack base (top) is. This is harder. 990 // 991 // When an application is started, glibc saves the initial stack pointer in 992 // a global variable "__libc_stack_end", which is then used by system 993 // libraries. __libc_stack_end should be pretty close to stack top. The 994 // variable is available since the very early days. However, because it is 995 // a private interface, it could disappear in the future. 996 // 997 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 998 // to __libc_stack_end, it is very close to stack top, but isn't the real 999 // stack top. Note that /proc may not exist if VM is running as a chroot 1000 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1001 // /proc/<pid>/stat could change in the future (though unlikely). 1002 // 1003 // We try __libc_stack_end first. If that doesn't work, look for 1004 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1005 // as a hint, which should work well in most cases. 1006 1007 uintptr_t stack_start; 1008 1009 // try __libc_stack_end first 1010 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1011 if (p && *p) { 1012 stack_start = *p; 1013 } else { 1014 // see if we can get the start_stack field from /proc/self/stat 1015 FILE *fp; 1016 int pid; 1017 char state; 1018 int ppid; 1019 int pgrp; 1020 int session; 1021 int nr; 1022 int tpgrp; 1023 unsigned long flags; 1024 unsigned long minflt; 1025 unsigned long cminflt; 1026 unsigned long majflt; 1027 unsigned long cmajflt; 1028 unsigned long utime; 1029 unsigned long stime; 1030 long cutime; 1031 long cstime; 1032 long prio; 1033 long nice; 1034 long junk; 1035 long it_real; 1036 uintptr_t start; 1037 uintptr_t vsize; 1038 intptr_t rss; 1039 uintptr_t rsslim; 1040 uintptr_t scodes; 1041 uintptr_t ecode; 1042 int i; 1043 1044 // Figure what the primordial thread stack base is. Code is inspired 1045 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1046 // followed by command name surrounded by parentheses, state, etc. 1047 char stat[2048]; 1048 int statlen; 1049 1050 fp = fopen("/proc/self/stat", "r"); 1051 if (fp) { 1052 statlen = fread(stat, 1, 2047, fp); 1053 stat[statlen] = '\0'; 1054 fclose(fp); 1055 1056 // Skip pid and the command string. Note that we could be dealing with 1057 // weird command names, e.g. user could decide to rename java launcher 1058 // to "java 1.4.2 :)", then the stat file would look like 1059 // 1234 (java 1.4.2 :)) R ... ... 1060 // We don't really need to know the command string, just find the last 1061 // occurrence of ")" and then start parsing from there. See bug 4726580. 1062 char * s = strrchr(stat, ')'); 1063 1064 i = 0; 1065 if (s) { 1066 // Skip blank chars 1067 do { s++; } while (s && isspace(*s)); 1068 1069#define _UFM UINTX_FORMAT 1070#define _DFM INTX_FORMAT 1071 1072 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1073 // 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 1074 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, 1075 &state, // 3 %c 1076 &ppid, // 4 %d 1077 &pgrp, // 5 %d 1078 &session, // 6 %d 1079 &nr, // 7 %d 1080 &tpgrp, // 8 %d 1081 &flags, // 9 %lu 1082 &minflt, // 10 %lu 1083 &cminflt, // 11 %lu 1084 &majflt, // 12 %lu 1085 &cmajflt, // 13 %lu 1086 &utime, // 14 %lu 1087 &stime, // 15 %lu 1088 &cutime, // 16 %ld 1089 &cstime, // 17 %ld 1090 &prio, // 18 %ld 1091 &nice, // 19 %ld 1092 &junk, // 20 %ld 1093 &it_real, // 21 %ld 1094 &start, // 22 UINTX_FORMAT 1095 &vsize, // 23 UINTX_FORMAT 1096 &rss, // 24 INTX_FORMAT 1097 &rsslim, // 25 UINTX_FORMAT 1098 &scodes, // 26 UINTX_FORMAT 1099 &ecode, // 27 UINTX_FORMAT 1100 &stack_start); // 28 UINTX_FORMAT 1101 } 1102 1103#undef _UFM 1104#undef _DFM 1105 1106 if (i != 28 - 2) { 1107 assert(false, "Bad conversion from /proc/self/stat"); 1108 // product mode - assume we are the initial thread, good luck in the 1109 // embedded case. 1110 warning("Can't detect initial thread stack location - bad conversion"); 1111 stack_start = (uintptr_t) &rlim; 1112 } 1113 } else { 1114 // For some reason we can't open /proc/self/stat (for example, running on 1115 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1116 // most cases, so don't abort: 1117 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1118 stack_start = (uintptr_t) &rlim; 1119 } 1120 } 1121 1122 // Now we have a pointer (stack_start) very close to the stack top, the 1123 // next thing to do is to figure out the exact location of stack top. We 1124 // can find out the virtual memory area that contains stack_start by 1125 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1126 // and its upper limit is the real stack top. (again, this would fail if 1127 // running inside chroot, because /proc may not exist.) 1128 1129 uintptr_t stack_top; 1130 address low, high; 1131 if (find_vma((address)stack_start, &low, &high)) { 1132 // success, "high" is the true stack top. (ignore "low", because initial 1133 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1134 stack_top = (uintptr_t)high; 1135 } else { 1136 // failed, likely because /proc/self/maps does not exist 1137 warning("Can't detect initial thread stack location - find_vma failed"); 1138 // best effort: stack_start is normally within a few pages below the real 1139 // stack top, use it as stack top, and reduce stack size so we won't put 1140 // guard page outside stack. 1141 stack_top = stack_start; 1142 stack_size -= 16 * page_size(); 1143 } 1144 1145 // stack_top could be partially down the page so align it 1146 stack_top = align_size_up(stack_top, page_size()); 1147 1148 if (max_size && stack_size > max_size) { 1149 _initial_thread_stack_size = max_size; 1150 } else { 1151 _initial_thread_stack_size = stack_size; 1152 } 1153 1154 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1155 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1156} 1157 1158//////////////////////////////////////////////////////////////////////////////// 1159// time support 1160 1161// Time since start-up in seconds to a fine granularity. 1162// Used by VMSelfDestructTimer and the MemProfiler. 1163double os::elapsedTime() { 1164 1165 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1166} 1167 1168jlong os::elapsed_counter() { 1169 return javaTimeNanos() - initial_time_count; 1170} 1171 1172jlong os::elapsed_frequency() { 1173 return NANOSECS_PER_SEC; // nanosecond resolution 1174} 1175 1176bool os::supports_vtime() { return true; } 1177bool os::enable_vtime() { return false; } 1178bool os::vtime_enabled() { return false; } 1179 1180double os::elapsedVTime() { 1181 struct rusage usage; 1182 int retval = getrusage(RUSAGE_THREAD, &usage); 1183 if (retval == 0) { 1184 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); 1185 } else { 1186 // better than nothing, but not much 1187 return elapsedTime(); 1188 } 1189} 1190 1191jlong os::javaTimeMillis() { 1192 timeval time; 1193 int status = gettimeofday(&time, NULL); 1194 assert(status != -1, "linux error"); 1195 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1196} 1197 1198void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) { 1199 timeval time; 1200 int status = gettimeofday(&time, NULL); 1201 assert(status != -1, "linux error"); 1202 seconds = jlong(time.tv_sec); 1203 nanos = jlong(time.tv_usec) * 1000; 1204} 1205 1206 1207#ifndef CLOCK_MONOTONIC 1208 #define CLOCK_MONOTONIC (1) 1209#endif 1210 1211void os::Linux::clock_init() { 1212 // we do dlopen's in this particular order due to bug in linux 1213 // dynamical loader (see 6348968) leading to crash on exit 1214 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1215 if (handle == NULL) { 1216 handle = dlopen("librt.so", RTLD_LAZY); 1217 } 1218 1219 if (handle) { 1220 int (*clock_getres_func)(clockid_t, struct timespec*) = 1221 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1222 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1223 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1224 if (clock_getres_func && clock_gettime_func) { 1225 // See if monotonic clock is supported by the kernel. Note that some 1226 // early implementations simply return kernel jiffies (updated every 1227 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1228 // for nano time (though the monotonic property is still nice to have). 1229 // It's fixed in newer kernels, however clock_getres() still returns 1230 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1231 // resolution for now. Hopefully as people move to new kernels, this 1232 // won't be a problem. 1233 struct timespec res; 1234 struct timespec tp; 1235 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1236 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1237 // yes, monotonic clock is supported 1238 _clock_gettime = clock_gettime_func; 1239 return; 1240 } else { 1241 // close librt if there is no monotonic clock 1242 dlclose(handle); 1243 } 1244 } 1245 } 1246 warning("No monotonic clock was available - timed services may " \ 1247 "be adversely affected if the time-of-day clock changes"); 1248} 1249 1250#ifndef SYS_clock_getres 1251 #if defined(IA32) || defined(AMD64) 1252 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1253 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1254 #else 1255 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1256 #define sys_clock_getres(x,y) -1 1257 #endif 1258#else 1259 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1260#endif 1261 1262void os::Linux::fast_thread_clock_init() { 1263 if (!UseLinuxPosixThreadCPUClocks) { 1264 return; 1265 } 1266 clockid_t clockid; 1267 struct timespec tp; 1268 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1269 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1270 1271 // Switch to using fast clocks for thread cpu time if 1272 // the sys_clock_getres() returns 0 error code. 1273 // Note, that some kernels may support the current thread 1274 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1275 // returned by the pthread_getcpuclockid(). 1276 // If the fast Posix clocks are supported then the sys_clock_getres() 1277 // must return at least tp.tv_sec == 0 which means a resolution 1278 // better than 1 sec. This is extra check for reliability. 1279 1280 if (pthread_getcpuclockid_func && 1281 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1282 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1283 _supports_fast_thread_cpu_time = true; 1284 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1285 } 1286} 1287 1288jlong os::javaTimeNanos() { 1289 if (os::supports_monotonic_clock()) { 1290 struct timespec tp; 1291 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1292 assert(status == 0, "gettime error"); 1293 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1294 return result; 1295 } else { 1296 timeval time; 1297 int status = gettimeofday(&time, NULL); 1298 assert(status != -1, "linux error"); 1299 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1300 return 1000 * usecs; 1301 } 1302} 1303 1304void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1305 if (os::supports_monotonic_clock()) { 1306 info_ptr->max_value = ALL_64_BITS; 1307 1308 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1309 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1310 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1311 } else { 1312 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1313 info_ptr->max_value = ALL_64_BITS; 1314 1315 // gettimeofday is a real time clock so it skips 1316 info_ptr->may_skip_backward = true; 1317 info_ptr->may_skip_forward = true; 1318 } 1319 1320 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1321} 1322 1323// Return the real, user, and system times in seconds from an 1324// arbitrary fixed point in the past. 1325bool os::getTimesSecs(double* process_real_time, 1326 double* process_user_time, 1327 double* process_system_time) { 1328 struct tms ticks; 1329 clock_t real_ticks = times(&ticks); 1330 1331 if (real_ticks == (clock_t) (-1)) { 1332 return false; 1333 } else { 1334 double ticks_per_second = (double) clock_tics_per_sec; 1335 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1336 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1337 *process_real_time = ((double) real_ticks) / ticks_per_second; 1338 1339 return true; 1340 } 1341} 1342 1343 1344char * os::local_time_string(char *buf, size_t buflen) { 1345 struct tm t; 1346 time_t long_time; 1347 time(&long_time); 1348 localtime_r(&long_time, &t); 1349 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1350 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1351 t.tm_hour, t.tm_min, t.tm_sec); 1352 return buf; 1353} 1354 1355struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1356 return localtime_r(clock, res); 1357} 1358 1359//////////////////////////////////////////////////////////////////////////////// 1360// runtime exit support 1361 1362// Note: os::shutdown() might be called very early during initialization, or 1363// called from signal handler. Before adding something to os::shutdown(), make 1364// sure it is async-safe and can handle partially initialized VM. 1365void os::shutdown() { 1366 1367 // allow PerfMemory to attempt cleanup of any persistent resources 1368 perfMemory_exit(); 1369 1370 // needs to remove object in file system 1371 AttachListener::abort(); 1372 1373 // flush buffered output, finish log files 1374 ostream_abort(); 1375 1376 // Check for abort hook 1377 abort_hook_t abort_hook = Arguments::abort_hook(); 1378 if (abort_hook != NULL) { 1379 abort_hook(); 1380 } 1381 1382} 1383 1384// Note: os::abort() might be called very early during initialization, or 1385// called from signal handler. Before adding something to os::abort(), make 1386// sure it is async-safe and can handle partially initialized VM. 1387void os::abort(bool dump_core, void* siginfo, void* context) { 1388 os::shutdown(); 1389 if (dump_core) { 1390#ifndef PRODUCT 1391 fdStream out(defaultStream::output_fd()); 1392 out.print_raw("Current thread is "); 1393 char buf[16]; 1394 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1395 out.print_raw_cr(buf); 1396 out.print_raw_cr("Dumping core ..."); 1397#endif 1398 ::abort(); // dump core 1399 } 1400 1401 ::exit(1); 1402} 1403 1404// Die immediately, no exit hook, no abort hook, no cleanup. 1405void os::die() { 1406 ::abort(); 1407} 1408 1409 1410// This method is a copy of JDK's sysGetLastErrorString 1411// from src/solaris/hpi/src/system_md.c 1412 1413size_t os::lasterror(char *buf, size_t len) { 1414 if (errno == 0) return 0; 1415 1416 const char *s = ::strerror(errno); 1417 size_t n = ::strlen(s); 1418 if (n >= len) { 1419 n = len - 1; 1420 } 1421 ::strncpy(buf, s, n); 1422 buf[n] = '\0'; 1423 return n; 1424} 1425 1426intx os::current_thread_id() { return (intx)pthread_self(); } 1427int os::current_process_id() { 1428 return ::getpid(); 1429} 1430 1431// DLL functions 1432 1433const char* os::dll_file_extension() { return ".so"; } 1434 1435// This must be hard coded because it's the system's temporary 1436// directory not the java application's temp directory, ala java.io.tmpdir. 1437const char* os::get_temp_directory() { return "/tmp"; } 1438 1439static bool file_exists(const char* filename) { 1440 struct stat statbuf; 1441 if (filename == NULL || strlen(filename) == 0) { 1442 return false; 1443 } 1444 return os::stat(filename, &statbuf) == 0; 1445} 1446 1447bool os::dll_build_name(char* buffer, size_t buflen, 1448 const char* pname, const char* fname) { 1449 bool retval = false; 1450 // Copied from libhpi 1451 const size_t pnamelen = pname ? strlen(pname) : 0; 1452 1453 // Return error on buffer overflow. 1454 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1455 return retval; 1456 } 1457 1458 if (pnamelen == 0) { 1459 snprintf(buffer, buflen, "lib%s.so", fname); 1460 retval = true; 1461 } else if (strchr(pname, *os::path_separator()) != NULL) { 1462 int n; 1463 char** pelements = split_path(pname, &n); 1464 if (pelements == NULL) { 1465 return false; 1466 } 1467 for (int i = 0; i < n; i++) { 1468 // Really shouldn't be NULL, but check can't hurt 1469 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1470 continue; // skip the empty path values 1471 } 1472 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1473 if (file_exists(buffer)) { 1474 retval = true; 1475 break; 1476 } 1477 } 1478 // release the storage 1479 for (int i = 0; i < n; i++) { 1480 if (pelements[i] != NULL) { 1481 FREE_C_HEAP_ARRAY(char, pelements[i]); 1482 } 1483 } 1484 if (pelements != NULL) { 1485 FREE_C_HEAP_ARRAY(char*, pelements); 1486 } 1487 } else { 1488 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1489 retval = true; 1490 } 1491 return retval; 1492} 1493 1494// check if addr is inside libjvm.so 1495bool os::address_is_in_vm(address addr) { 1496 static address libjvm_base_addr; 1497 Dl_info dlinfo; 1498 1499 if (libjvm_base_addr == NULL) { 1500 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1501 libjvm_base_addr = (address)dlinfo.dli_fbase; 1502 } 1503 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1504 } 1505 1506 if (dladdr((void *)addr, &dlinfo) != 0) { 1507 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1508 } 1509 1510 return false; 1511} 1512 1513bool os::dll_address_to_function_name(address addr, char *buf, 1514 int buflen, int *offset, 1515 bool demangle) { 1516 // buf is not optional, but offset is optional 1517 assert(buf != NULL, "sanity check"); 1518 1519 Dl_info dlinfo; 1520 1521 if (dladdr((void*)addr, &dlinfo) != 0) { 1522 // see if we have a matching symbol 1523 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1524 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) { 1525 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1526 } 1527 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1528 return true; 1529 } 1530 // no matching symbol so try for just file info 1531 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1532 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1533 buf, buflen, offset, dlinfo.dli_fname, demangle)) { 1534 return true; 1535 } 1536 } 1537 } 1538 1539 buf[0] = '\0'; 1540 if (offset != NULL) *offset = -1; 1541 return false; 1542} 1543 1544struct _address_to_library_name { 1545 address addr; // input : memory address 1546 size_t buflen; // size of fname 1547 char* fname; // output: library name 1548 address base; // library base addr 1549}; 1550 1551static int address_to_library_name_callback(struct dl_phdr_info *info, 1552 size_t size, void *data) { 1553 int i; 1554 bool found = false; 1555 address libbase = NULL; 1556 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1557 1558 // iterate through all loadable segments 1559 for (i = 0; i < info->dlpi_phnum; i++) { 1560 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1561 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1562 // base address of a library is the lowest address of its loaded 1563 // segments. 1564 if (libbase == NULL || libbase > segbase) { 1565 libbase = segbase; 1566 } 1567 // see if 'addr' is within current segment 1568 if (segbase <= d->addr && 1569 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1570 found = true; 1571 } 1572 } 1573 } 1574 1575 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1576 // so dll_address_to_library_name() can fall through to use dladdr() which 1577 // can figure out executable name from argv[0]. 1578 if (found && info->dlpi_name && info->dlpi_name[0]) { 1579 d->base = libbase; 1580 if (d->fname) { 1581 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1582 } 1583 return 1; 1584 } 1585 return 0; 1586} 1587 1588bool os::dll_address_to_library_name(address addr, char* buf, 1589 int buflen, int* offset) { 1590 // buf is not optional, but offset is optional 1591 assert(buf != NULL, "sanity check"); 1592 1593 Dl_info dlinfo; 1594 struct _address_to_library_name data; 1595 1596 // There is a bug in old glibc dladdr() implementation that it could resolve 1597 // to wrong library name if the .so file has a base address != NULL. Here 1598 // we iterate through the program headers of all loaded libraries to find 1599 // out which library 'addr' really belongs to. This workaround can be 1600 // removed once the minimum requirement for glibc is moved to 2.3.x. 1601 data.addr = addr; 1602 data.fname = buf; 1603 data.buflen = buflen; 1604 data.base = NULL; 1605 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1606 1607 if (rslt) { 1608 // buf already contains library name 1609 if (offset) *offset = addr - data.base; 1610 return true; 1611 } 1612 if (dladdr((void*)addr, &dlinfo) != 0) { 1613 if (dlinfo.dli_fname != NULL) { 1614 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1615 } 1616 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1617 *offset = addr - (address)dlinfo.dli_fbase; 1618 } 1619 return true; 1620 } 1621 1622 buf[0] = '\0'; 1623 if (offset) *offset = -1; 1624 return false; 1625} 1626 1627// Loads .dll/.so and 1628// in case of error it checks if .dll/.so was built for the 1629// same architecture as Hotspot is running on 1630 1631 1632// Remember the stack's state. The Linux dynamic linker will change 1633// the stack to 'executable' at most once, so we must safepoint only once. 1634bool os::Linux::_stack_is_executable = false; 1635 1636// VM operation that loads a library. This is necessary if stack protection 1637// of the Java stacks can be lost during loading the library. If we 1638// do not stop the Java threads, they can stack overflow before the stacks 1639// are protected again. 1640class VM_LinuxDllLoad: public VM_Operation { 1641 private: 1642 const char *_filename; 1643 char *_ebuf; 1644 int _ebuflen; 1645 void *_lib; 1646 public: 1647 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1648 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1649 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1650 void doit() { 1651 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1652 os::Linux::_stack_is_executable = true; 1653 } 1654 void* loaded_library() { return _lib; } 1655}; 1656 1657void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1658 void * result = NULL; 1659 bool load_attempted = false; 1660 1661 // Check whether the library to load might change execution rights 1662 // of the stack. If they are changed, the protection of the stack 1663 // guard pages will be lost. We need a safepoint to fix this. 1664 // 1665 // See Linux man page execstack(8) for more info. 1666 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1667 ElfFile ef(filename); 1668 if (!ef.specifies_noexecstack()) { 1669 if (!is_init_completed()) { 1670 os::Linux::_stack_is_executable = true; 1671 // This is OK - No Java threads have been created yet, and hence no 1672 // stack guard pages to fix. 1673 // 1674 // This should happen only when you are building JDK7 using a very 1675 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1676 // 1677 // Dynamic loader will make all stacks executable after 1678 // this function returns, and will not do that again. 1679 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1680 } else { 1681 warning("You have loaded library %s which might have disabled stack guard. " 1682 "The VM will try to fix the stack guard now.\n" 1683 "It's highly recommended that you fix the library with " 1684 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1685 filename); 1686 1687 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1688 JavaThread *jt = JavaThread::current(); 1689 if (jt->thread_state() != _thread_in_native) { 1690 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1691 // that requires ExecStack. Cannot enter safe point. Let's give up. 1692 warning("Unable to fix stack guard. Giving up."); 1693 } else { 1694 if (!LoadExecStackDllInVMThread) { 1695 // This is for the case where the DLL has an static 1696 // constructor function that executes JNI code. We cannot 1697 // load such DLLs in the VMThread. 1698 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1699 } 1700 1701 ThreadInVMfromNative tiv(jt); 1702 debug_only(VMNativeEntryWrapper vew;) 1703 1704 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1705 VMThread::execute(&op); 1706 if (LoadExecStackDllInVMThread) { 1707 result = op.loaded_library(); 1708 } 1709 load_attempted = true; 1710 } 1711 } 1712 } 1713 } 1714 1715 if (!load_attempted) { 1716 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1717 } 1718 1719 if (result != NULL) { 1720 // Successful loading 1721 return result; 1722 } 1723 1724 Elf32_Ehdr elf_head; 1725 int diag_msg_max_length=ebuflen-strlen(ebuf); 1726 char* diag_msg_buf=ebuf+strlen(ebuf); 1727 1728 if (diag_msg_max_length==0) { 1729 // No more space in ebuf for additional diagnostics message 1730 return NULL; 1731 } 1732 1733 1734 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1735 1736 if (file_descriptor < 0) { 1737 // Can't open library, report dlerror() message 1738 return NULL; 1739 } 1740 1741 bool failed_to_read_elf_head= 1742 (sizeof(elf_head)!= 1743 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1744 1745 ::close(file_descriptor); 1746 if (failed_to_read_elf_head) { 1747 // file i/o error - report dlerror() msg 1748 return NULL; 1749 } 1750 1751 typedef struct { 1752 Elf32_Half code; // Actual value as defined in elf.h 1753 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1754 char elf_class; // 32 or 64 bit 1755 char endianess; // MSB or LSB 1756 char* name; // String representation 1757 } arch_t; 1758 1759#ifndef EM_486 1760 #define EM_486 6 /* Intel 80486 */ 1761#endif 1762#ifndef EM_AARCH64 1763 #define EM_AARCH64 183 /* ARM AARCH64 */ 1764#endif 1765 1766 static const arch_t arch_array[]={ 1767 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1768 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1769 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1770 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1771 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1772 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1773 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1774 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1775#if defined(VM_LITTLE_ENDIAN) 1776 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"}, 1777#else 1778 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1779#endif 1780 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1781 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1782 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1783 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1784 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1785 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1786 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1787 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1788 }; 1789 1790#if (defined IA32) 1791 static Elf32_Half running_arch_code=EM_386; 1792#elif (defined AMD64) 1793 static Elf32_Half running_arch_code=EM_X86_64; 1794#elif (defined IA64) 1795 static Elf32_Half running_arch_code=EM_IA_64; 1796#elif (defined __sparc) && (defined _LP64) 1797 static Elf32_Half running_arch_code=EM_SPARCV9; 1798#elif (defined __sparc) && (!defined _LP64) 1799 static Elf32_Half running_arch_code=EM_SPARC; 1800#elif (defined __powerpc64__) 1801 static Elf32_Half running_arch_code=EM_PPC64; 1802#elif (defined __powerpc__) 1803 static Elf32_Half running_arch_code=EM_PPC; 1804#elif (defined ARM) 1805 static Elf32_Half running_arch_code=EM_ARM; 1806#elif (defined S390) 1807 static Elf32_Half running_arch_code=EM_S390; 1808#elif (defined ALPHA) 1809 static Elf32_Half running_arch_code=EM_ALPHA; 1810#elif (defined MIPSEL) 1811 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1812#elif (defined PARISC) 1813 static Elf32_Half running_arch_code=EM_PARISC; 1814#elif (defined MIPS) 1815 static Elf32_Half running_arch_code=EM_MIPS; 1816#elif (defined M68K) 1817 static Elf32_Half running_arch_code=EM_68K; 1818#elif (defined AARCH64) 1819 static Elf32_Half running_arch_code=EM_AARCH64; 1820#else 1821 #error Method os::dll_load requires that one of following is defined:\ 1822 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64 1823#endif 1824 1825 // Identify compatability class for VM's architecture and library's architecture 1826 // Obtain string descriptions for architectures 1827 1828 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1829 int running_arch_index=-1; 1830 1831 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1832 if (running_arch_code == arch_array[i].code) { 1833 running_arch_index = i; 1834 } 1835 if (lib_arch.code == arch_array[i].code) { 1836 lib_arch.compat_class = arch_array[i].compat_class; 1837 lib_arch.name = arch_array[i].name; 1838 } 1839 } 1840 1841 assert(running_arch_index != -1, 1842 "Didn't find running architecture code (running_arch_code) in arch_array"); 1843 if (running_arch_index == -1) { 1844 // Even though running architecture detection failed 1845 // we may still continue with reporting dlerror() message 1846 return NULL; 1847 } 1848 1849 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1850 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1851 return NULL; 1852 } 1853 1854#ifndef S390 1855 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1856 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1857 return NULL; 1858 } 1859#endif // !S390 1860 1861 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1862 if (lib_arch.name!=NULL) { 1863 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1864 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1865 lib_arch.name, arch_array[running_arch_index].name); 1866 } else { 1867 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1868 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1869 lib_arch.code, 1870 arch_array[running_arch_index].name); 1871 } 1872 } 1873 1874 return NULL; 1875} 1876 1877void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 1878 int ebuflen) { 1879 void * result = ::dlopen(filename, RTLD_LAZY); 1880 if (result == NULL) { 1881 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 1882 ebuf[ebuflen-1] = '\0'; 1883 } 1884 return result; 1885} 1886 1887void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 1888 int ebuflen) { 1889 void * result = NULL; 1890 if (LoadExecStackDllInVMThread) { 1891 result = dlopen_helper(filename, ebuf, ebuflen); 1892 } 1893 1894 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 1895 // library that requires an executable stack, or which does not have this 1896 // stack attribute set, dlopen changes the stack attribute to executable. The 1897 // read protection of the guard pages gets lost. 1898 // 1899 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 1900 // may have been queued at the same time. 1901 1902 if (!_stack_is_executable) { 1903 JavaThread *jt = Threads::first(); 1904 1905 while (jt) { 1906 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 1907 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 1908 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 1909 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 1910 warning("Attempt to reguard stack yellow zone failed."); 1911 } 1912 } 1913 jt = jt->next(); 1914 } 1915 } 1916 1917 return result; 1918} 1919 1920void* os::dll_lookup(void* handle, const char* name) { 1921 void* res = dlsym(handle, name); 1922 return res; 1923} 1924 1925void* os::get_default_process_handle() { 1926 return (void*)::dlopen(NULL, RTLD_LAZY); 1927} 1928 1929static bool _print_ascii_file(const char* filename, outputStream* st) { 1930 int fd = ::open(filename, O_RDONLY); 1931 if (fd == -1) { 1932 return false; 1933 } 1934 1935 char buf[32]; 1936 int bytes; 1937 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 1938 st->print_raw(buf, bytes); 1939 } 1940 1941 ::close(fd); 1942 1943 return true; 1944} 1945 1946void os::print_dll_info(outputStream *st) { 1947 st->print_cr("Dynamic libraries:"); 1948 1949 char fname[32]; 1950 pid_t pid = os::Linux::gettid(); 1951 1952 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 1953 1954 if (!_print_ascii_file(fname, st)) { 1955 st->print("Can not get library information for pid = %d\n", pid); 1956 } 1957} 1958 1959int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 1960 FILE *procmapsFile = NULL; 1961 1962 // Open the procfs maps file for the current process 1963 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 1964 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 1965 char line[PATH_MAX + 100]; 1966 1967 // Read line by line from 'file' 1968 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 1969 u8 base, top, offset, inode; 1970 char permissions[5]; 1971 char device[6]; 1972 char name[PATH_MAX + 1]; 1973 1974 // Parse fields from line 1975 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name); 1976 1977 // Filter by device id '00:00' so that we only get file system mapped files. 1978 if (strcmp(device, "00:00") != 0) { 1979 1980 // Call callback with the fields of interest 1981 if(callback(name, (address)base, (address)top, param)) { 1982 // Oops abort, callback aborted 1983 fclose(procmapsFile); 1984 return 1; 1985 } 1986 } 1987 } 1988 fclose(procmapsFile); 1989 } 1990 return 0; 1991} 1992 1993void os::print_os_info_brief(outputStream* st) { 1994 os::Linux::print_distro_info(st); 1995 1996 os::Posix::print_uname_info(st); 1997 1998 os::Linux::print_libversion_info(st); 1999 2000} 2001 2002void os::print_os_info(outputStream* st) { 2003 st->print("OS:"); 2004 2005 os::Linux::print_distro_info(st); 2006 2007 os::Posix::print_uname_info(st); 2008 2009 // Print warning if unsafe chroot environment detected 2010 if (unsafe_chroot_detected) { 2011 st->print("WARNING!! "); 2012 st->print_cr("%s", unstable_chroot_error); 2013 } 2014 2015 os::Linux::print_libversion_info(st); 2016 2017 os::Posix::print_rlimit_info(st); 2018 2019 os::Posix::print_load_average(st); 2020 2021 os::Linux::print_full_memory_info(st); 2022} 2023 2024// Try to identify popular distros. 2025// Most Linux distributions have a /etc/XXX-release file, which contains 2026// the OS version string. Newer Linux distributions have a /etc/lsb-release 2027// file that also contains the OS version string. Some have more than one 2028// /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2029// /etc/redhat-release.), so the order is important. 2030// Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2031// their own specific XXX-release file as well as a redhat-release file. 2032// Because of this the XXX-release file needs to be searched for before the 2033// redhat-release file. 2034// Since Red Hat has a lsb-release file that is not very descriptive the 2035// search for redhat-release needs to be before lsb-release. 2036// Since the lsb-release file is the new standard it needs to be searched 2037// before the older style release files. 2038// Searching system-release (Red Hat) and os-release (other Linuxes) are a 2039// next to last resort. The os-release file is a new standard that contains 2040// distribution information and the system-release file seems to be an old 2041// standard that has been replaced by the lsb-release and os-release files. 2042// Searching for the debian_version file is the last resort. It contains 2043// an informative string like "6.0.6" or "wheezy/sid". Because of this 2044// "Debian " is printed before the contents of the debian_version file. 2045void os::Linux::print_distro_info(outputStream* st) { 2046 if (!_print_ascii_file("/etc/oracle-release", st) && 2047 !_print_ascii_file("/etc/mandriva-release", st) && 2048 !_print_ascii_file("/etc/mandrake-release", st) && 2049 !_print_ascii_file("/etc/sun-release", st) && 2050 !_print_ascii_file("/etc/redhat-release", st) && 2051 !_print_ascii_file("/etc/lsb-release", st) && 2052 !_print_ascii_file("/etc/SuSE-release", st) && 2053 !_print_ascii_file("/etc/turbolinux-release", st) && 2054 !_print_ascii_file("/etc/gentoo-release", st) && 2055 !_print_ascii_file("/etc/ltib-release", st) && 2056 !_print_ascii_file("/etc/angstrom-version", st) && 2057 !_print_ascii_file("/etc/system-release", st) && 2058 !_print_ascii_file("/etc/os-release", st)) { 2059 2060 if (file_exists("/etc/debian_version")) { 2061 st->print("Debian "); 2062 _print_ascii_file("/etc/debian_version", st); 2063 } else { 2064 st->print("Linux"); 2065 } 2066 } 2067 st->cr(); 2068} 2069 2070void os::Linux::print_libversion_info(outputStream* st) { 2071 // libc, pthread 2072 st->print("libc:"); 2073 st->print("%s ", os::Linux::glibc_version()); 2074 st->print("%s ", os::Linux::libpthread_version()); 2075 st->cr(); 2076} 2077 2078void os::Linux::print_full_memory_info(outputStream* st) { 2079 st->print("\n/proc/meminfo:\n"); 2080 _print_ascii_file("/proc/meminfo", st); 2081 st->cr(); 2082} 2083 2084void os::print_memory_info(outputStream* st) { 2085 2086 st->print("Memory:"); 2087 st->print(" %dk page", os::vm_page_size()>>10); 2088 2089 // values in struct sysinfo are "unsigned long" 2090 struct sysinfo si; 2091 sysinfo(&si); 2092 2093 st->print(", physical " UINT64_FORMAT "k", 2094 os::physical_memory() >> 10); 2095 st->print("(" UINT64_FORMAT "k free)", 2096 os::available_memory() >> 10); 2097 st->print(", swap " UINT64_FORMAT "k", 2098 ((jlong)si.totalswap * si.mem_unit) >> 10); 2099 st->print("(" UINT64_FORMAT "k free)", 2100 ((jlong)si.freeswap * si.mem_unit) >> 10); 2101 st->cr(); 2102} 2103 2104// Print the first "model name" line and the first "flags" line 2105// that we find and nothing more. We assume "model name" comes 2106// before "flags" so if we find a second "model name", then the 2107// "flags" field is considered missing. 2108static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) { 2109#if defined(IA32) || defined(AMD64) 2110 // Other platforms have less repetitive cpuinfo files 2111 FILE *fp = fopen("/proc/cpuinfo", "r"); 2112 if (fp) { 2113 while (!feof(fp)) { 2114 if (fgets(buf, buflen, fp)) { 2115 // Assume model name comes before flags 2116 bool model_name_printed = false; 2117 if (strstr(buf, "model name") != NULL) { 2118 if (!model_name_printed) { 2119 st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n"); 2120 st->print_raw(buf); 2121 model_name_printed = true; 2122 } else { 2123 // model name printed but not flags? Odd, just return 2124 fclose(fp); 2125 return true; 2126 } 2127 } 2128 // print the flags line too 2129 if (strstr(buf, "flags") != NULL) { 2130 st->print_raw(buf); 2131 fclose(fp); 2132 return true; 2133 } 2134 } 2135 } 2136 fclose(fp); 2137 } 2138#endif // x86 platforms 2139 return false; 2140} 2141 2142void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) { 2143 // Only print the model name if the platform provides this as a summary 2144 if (!print_model_name_and_flags(st, buf, buflen)) { 2145 st->print("\n/proc/cpuinfo:\n"); 2146 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2147 st->print_cr(" <Not Available>"); 2148 } 2149 } 2150} 2151 2152void os::print_siginfo(outputStream* st, void* siginfo) { 2153 const siginfo_t* si = (const siginfo_t*)siginfo; 2154 2155 os::Posix::print_siginfo_brief(st, si); 2156#if INCLUDE_CDS 2157 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2158 UseSharedSpaces) { 2159 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2160 if (mapinfo->is_in_shared_space(si->si_addr)) { 2161 st->print("\n\nError accessing class data sharing archive." \ 2162 " Mapped file inaccessible during execution, " \ 2163 " possible disk/network problem."); 2164 } 2165 } 2166#endif 2167 st->cr(); 2168} 2169 2170 2171static void print_signal_handler(outputStream* st, int sig, 2172 char* buf, size_t buflen); 2173 2174void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2175 st->print_cr("Signal Handlers:"); 2176 print_signal_handler(st, SIGSEGV, buf, buflen); 2177 print_signal_handler(st, SIGBUS , buf, buflen); 2178 print_signal_handler(st, SIGFPE , buf, buflen); 2179 print_signal_handler(st, SIGPIPE, buf, buflen); 2180 print_signal_handler(st, SIGXFSZ, buf, buflen); 2181 print_signal_handler(st, SIGILL , buf, buflen); 2182 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2183 print_signal_handler(st, SR_signum, buf, buflen); 2184 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2185 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2186 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2187 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2188#if defined(PPC64) 2189 print_signal_handler(st, SIGTRAP, buf, buflen); 2190#endif 2191} 2192 2193static char saved_jvm_path[MAXPATHLEN] = {0}; 2194 2195// Find the full path to the current module, libjvm.so 2196void os::jvm_path(char *buf, jint buflen) { 2197 // Error checking. 2198 if (buflen < MAXPATHLEN) { 2199 assert(false, "must use a large-enough buffer"); 2200 buf[0] = '\0'; 2201 return; 2202 } 2203 // Lazy resolve the path to current module. 2204 if (saved_jvm_path[0] != 0) { 2205 strcpy(buf, saved_jvm_path); 2206 return; 2207 } 2208 2209 char dli_fname[MAXPATHLEN]; 2210 bool ret = dll_address_to_library_name( 2211 CAST_FROM_FN_PTR(address, os::jvm_path), 2212 dli_fname, sizeof(dli_fname), NULL); 2213 assert(ret, "cannot locate libjvm"); 2214 char *rp = NULL; 2215 if (ret && dli_fname[0] != '\0') { 2216 rp = realpath(dli_fname, buf); 2217 } 2218 if (rp == NULL) { 2219 return; 2220 } 2221 2222 if (Arguments::sun_java_launcher_is_altjvm()) { 2223 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2224 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2225 // If "/jre/lib/" appears at the right place in the string, then 2226 // assume we are installed in a JDK and we're done. Otherwise, check 2227 // for a JAVA_HOME environment variable and fix up the path so it 2228 // looks like libjvm.so is installed there (append a fake suffix 2229 // hotspot/libjvm.so). 2230 const char *p = buf + strlen(buf) - 1; 2231 for (int count = 0; p > buf && count < 5; ++count) { 2232 for (--p; p > buf && *p != '/'; --p) 2233 /* empty */ ; 2234 } 2235 2236 if (strncmp(p, "/jre/lib/", 9) != 0) { 2237 // Look for JAVA_HOME in the environment. 2238 char* java_home_var = ::getenv("JAVA_HOME"); 2239 if (java_home_var != NULL && java_home_var[0] != 0) { 2240 char* jrelib_p; 2241 int len; 2242 2243 // Check the current module name "libjvm.so". 2244 p = strrchr(buf, '/'); 2245 if (p == NULL) { 2246 return; 2247 } 2248 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2249 2250 rp = realpath(java_home_var, buf); 2251 if (rp == NULL) { 2252 return; 2253 } 2254 2255 // determine if this is a legacy image or modules image 2256 // modules image doesn't have "jre" subdirectory 2257 len = strlen(buf); 2258 assert(len < buflen, "Ran out of buffer room"); 2259 jrelib_p = buf + len; 2260 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2261 if (0 != access(buf, F_OK)) { 2262 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2263 } 2264 2265 if (0 == access(buf, F_OK)) { 2266 // Use current module name "libjvm.so" 2267 len = strlen(buf); 2268 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2269 } else { 2270 // Go back to path of .so 2271 rp = realpath(dli_fname, buf); 2272 if (rp == NULL) { 2273 return; 2274 } 2275 } 2276 } 2277 } 2278 } 2279 2280 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2281 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2282} 2283 2284void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2285 // no prefix required, not even "_" 2286} 2287 2288void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2289 // no suffix required 2290} 2291 2292//////////////////////////////////////////////////////////////////////////////// 2293// sun.misc.Signal support 2294 2295static volatile jint sigint_count = 0; 2296 2297static void UserHandler(int sig, void *siginfo, void *context) { 2298 // 4511530 - sem_post is serialized and handled by the manager thread. When 2299 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2300 // don't want to flood the manager thread with sem_post requests. 2301 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2302 return; 2303 } 2304 2305 // Ctrl-C is pressed during error reporting, likely because the error 2306 // handler fails to abort. Let VM die immediately. 2307 if (sig == SIGINT && is_error_reported()) { 2308 os::die(); 2309 } 2310 2311 os::signal_notify(sig); 2312} 2313 2314void* os::user_handler() { 2315 return CAST_FROM_FN_PTR(void*, UserHandler); 2316} 2317 2318class Semaphore : public StackObj { 2319 public: 2320 Semaphore(); 2321 ~Semaphore(); 2322 void signal(); 2323 void wait(); 2324 bool trywait(); 2325 bool timedwait(unsigned int sec, int nsec); 2326 private: 2327 sem_t _semaphore; 2328}; 2329 2330Semaphore::Semaphore() { 2331 sem_init(&_semaphore, 0, 0); 2332} 2333 2334Semaphore::~Semaphore() { 2335 sem_destroy(&_semaphore); 2336} 2337 2338void Semaphore::signal() { 2339 sem_post(&_semaphore); 2340} 2341 2342void Semaphore::wait() { 2343 sem_wait(&_semaphore); 2344} 2345 2346bool Semaphore::trywait() { 2347 return sem_trywait(&_semaphore) == 0; 2348} 2349 2350bool Semaphore::timedwait(unsigned int sec, int nsec) { 2351 2352 struct timespec ts; 2353 // Semaphore's are always associated with CLOCK_REALTIME 2354 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2355 // see unpackTime for discussion on overflow checking 2356 if (sec >= MAX_SECS) { 2357 ts.tv_sec += MAX_SECS; 2358 ts.tv_nsec = 0; 2359 } else { 2360 ts.tv_sec += sec; 2361 ts.tv_nsec += nsec; 2362 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2363 ts.tv_nsec -= NANOSECS_PER_SEC; 2364 ++ts.tv_sec; // note: this must be <= max_secs 2365 } 2366 } 2367 2368 while (1) { 2369 int result = sem_timedwait(&_semaphore, &ts); 2370 if (result == 0) { 2371 return true; 2372 } else if (errno == EINTR) { 2373 continue; 2374 } else if (errno == ETIMEDOUT) { 2375 return false; 2376 } else { 2377 return false; 2378 } 2379 } 2380} 2381 2382extern "C" { 2383 typedef void (*sa_handler_t)(int); 2384 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2385} 2386 2387void* os::signal(int signal_number, void* handler) { 2388 struct sigaction sigAct, oldSigAct; 2389 2390 sigfillset(&(sigAct.sa_mask)); 2391 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2392 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2393 2394 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2395 // -1 means registration failed 2396 return (void *)-1; 2397 } 2398 2399 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2400} 2401 2402void os::signal_raise(int signal_number) { 2403 ::raise(signal_number); 2404} 2405 2406// The following code is moved from os.cpp for making this 2407// code platform specific, which it is by its very nature. 2408 2409// Will be modified when max signal is changed to be dynamic 2410int os::sigexitnum_pd() { 2411 return NSIG; 2412} 2413 2414// a counter for each possible signal value 2415static volatile jint pending_signals[NSIG+1] = { 0 }; 2416 2417// Linux(POSIX) specific hand shaking semaphore. 2418static sem_t sig_sem; 2419static Semaphore sr_semaphore; 2420 2421void os::signal_init_pd() { 2422 // Initialize signal structures 2423 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2424 2425 // Initialize signal semaphore 2426 ::sem_init(&sig_sem, 0, 0); 2427} 2428 2429void os::signal_notify(int sig) { 2430 Atomic::inc(&pending_signals[sig]); 2431 ::sem_post(&sig_sem); 2432} 2433 2434static int check_pending_signals(bool wait) { 2435 Atomic::store(0, &sigint_count); 2436 for (;;) { 2437 for (int i = 0; i < NSIG + 1; i++) { 2438 jint n = pending_signals[i]; 2439 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2440 return i; 2441 } 2442 } 2443 if (!wait) { 2444 return -1; 2445 } 2446 JavaThread *thread = JavaThread::current(); 2447 ThreadBlockInVM tbivm(thread); 2448 2449 bool threadIsSuspended; 2450 do { 2451 thread->set_suspend_equivalent(); 2452 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2453 ::sem_wait(&sig_sem); 2454 2455 // were we externally suspended while we were waiting? 2456 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2457 if (threadIsSuspended) { 2458 // The semaphore has been incremented, but while we were waiting 2459 // another thread suspended us. We don't want to continue running 2460 // while suspended because that would surprise the thread that 2461 // suspended us. 2462 ::sem_post(&sig_sem); 2463 2464 thread->java_suspend_self(); 2465 } 2466 } while (threadIsSuspended); 2467 } 2468} 2469 2470int os::signal_lookup() { 2471 return check_pending_signals(false); 2472} 2473 2474int os::signal_wait() { 2475 return check_pending_signals(true); 2476} 2477 2478//////////////////////////////////////////////////////////////////////////////// 2479// Virtual Memory 2480 2481int os::vm_page_size() { 2482 // Seems redundant as all get out 2483 assert(os::Linux::page_size() != -1, "must call os::init"); 2484 return os::Linux::page_size(); 2485} 2486 2487// Solaris allocates memory by pages. 2488int os::vm_allocation_granularity() { 2489 assert(os::Linux::page_size() != -1, "must call os::init"); 2490 return os::Linux::page_size(); 2491} 2492 2493// Rationale behind this function: 2494// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2495// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2496// samples for JITted code. Here we create private executable mapping over the code cache 2497// and then we can use standard (well, almost, as mapping can change) way to provide 2498// info for the reporting script by storing timestamp and location of symbol 2499void linux_wrap_code(char* base, size_t size) { 2500 static volatile jint cnt = 0; 2501 2502 if (!UseOprofile) { 2503 return; 2504 } 2505 2506 char buf[PATH_MAX+1]; 2507 int num = Atomic::add(1, &cnt); 2508 2509 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2510 os::get_temp_directory(), os::current_process_id(), num); 2511 unlink(buf); 2512 2513 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2514 2515 if (fd != -1) { 2516 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2517 if (rv != (off_t)-1) { 2518 if (::write(fd, "", 1) == 1) { 2519 mmap(base, size, 2520 PROT_READ|PROT_WRITE|PROT_EXEC, 2521 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2522 } 2523 } 2524 ::close(fd); 2525 unlink(buf); 2526 } 2527} 2528 2529static bool recoverable_mmap_error(int err) { 2530 // See if the error is one we can let the caller handle. This 2531 // list of errno values comes from JBS-6843484. I can't find a 2532 // Linux man page that documents this specific set of errno 2533 // values so while this list currently matches Solaris, it may 2534 // change as we gain experience with this failure mode. 2535 switch (err) { 2536 case EBADF: 2537 case EINVAL: 2538 case ENOTSUP: 2539 // let the caller deal with these errors 2540 return true; 2541 2542 default: 2543 // Any remaining errors on this OS can cause our reserved mapping 2544 // to be lost. That can cause confusion where different data 2545 // structures think they have the same memory mapped. The worst 2546 // scenario is if both the VM and a library think they have the 2547 // same memory mapped. 2548 return false; 2549 } 2550} 2551 2552static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2553 int err) { 2554 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2555 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2556 strerror(err), err); 2557} 2558 2559static void warn_fail_commit_memory(char* addr, size_t size, 2560 size_t alignment_hint, bool exec, 2561 int err) { 2562 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2563 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2564 alignment_hint, exec, strerror(err), err); 2565} 2566 2567// NOTE: Linux kernel does not really reserve the pages for us. 2568// All it does is to check if there are enough free pages 2569// left at the time of mmap(). This could be a potential 2570// problem. 2571int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2572 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2573 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2574 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2575 if (res != (uintptr_t) MAP_FAILED) { 2576 if (UseNUMAInterleaving) { 2577 numa_make_global(addr, size); 2578 } 2579 return 0; 2580 } 2581 2582 int err = errno; // save errno from mmap() call above 2583 2584 if (!recoverable_mmap_error(err)) { 2585 warn_fail_commit_memory(addr, size, exec, err); 2586 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2587 } 2588 2589 return err; 2590} 2591 2592bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2593 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2594} 2595 2596void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2597 const char* mesg) { 2598 assert(mesg != NULL, "mesg must be specified"); 2599 int err = os::Linux::commit_memory_impl(addr, size, exec); 2600 if (err != 0) { 2601 // the caller wants all commit errors to exit with the specified mesg: 2602 warn_fail_commit_memory(addr, size, exec, err); 2603 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2604 } 2605} 2606 2607// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2608#ifndef MAP_HUGETLB 2609 #define MAP_HUGETLB 0x40000 2610#endif 2611 2612// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2613#ifndef MADV_HUGEPAGE 2614 #define MADV_HUGEPAGE 14 2615#endif 2616 2617int os::Linux::commit_memory_impl(char* addr, size_t size, 2618 size_t alignment_hint, bool exec) { 2619 int err = os::Linux::commit_memory_impl(addr, size, exec); 2620 if (err == 0) { 2621 realign_memory(addr, size, alignment_hint); 2622 } 2623 return err; 2624} 2625 2626bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2627 bool exec) { 2628 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2629} 2630 2631void os::pd_commit_memory_or_exit(char* addr, size_t size, 2632 size_t alignment_hint, bool exec, 2633 const char* mesg) { 2634 assert(mesg != NULL, "mesg must be specified"); 2635 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2636 if (err != 0) { 2637 // the caller wants all commit errors to exit with the specified mesg: 2638 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2639 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2640 } 2641} 2642 2643void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2644 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2645 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2646 // be supported or the memory may already be backed by huge pages. 2647 ::madvise(addr, bytes, MADV_HUGEPAGE); 2648 } 2649} 2650 2651void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2652 // This method works by doing an mmap over an existing mmaping and effectively discarding 2653 // the existing pages. However it won't work for SHM-based large pages that cannot be 2654 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2655 // small pages on top of the SHM segment. This method always works for small pages, so we 2656 // allow that in any case. 2657 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2658 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2659 } 2660} 2661 2662void os::numa_make_global(char *addr, size_t bytes) { 2663 Linux::numa_interleave_memory(addr, bytes); 2664} 2665 2666// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2667// bind policy to MPOL_PREFERRED for the current thread. 2668#define USE_MPOL_PREFERRED 0 2669 2670void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2671 // To make NUMA and large pages more robust when both enabled, we need to ease 2672 // the requirements on where the memory should be allocated. MPOL_BIND is the 2673 // default policy and it will force memory to be allocated on the specified 2674 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2675 // the specified node, but will not force it. Using this policy will prevent 2676 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2677 // free large pages. 2678 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2679 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2680} 2681 2682bool os::numa_topology_changed() { return false; } 2683 2684size_t os::numa_get_groups_num() { 2685 int max_node = Linux::numa_max_node(); 2686 return max_node > 0 ? max_node + 1 : 1; 2687} 2688 2689int os::numa_get_group_id() { 2690 int cpu_id = Linux::sched_getcpu(); 2691 if (cpu_id != -1) { 2692 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2693 if (lgrp_id != -1) { 2694 return lgrp_id; 2695 } 2696 } 2697 return 0; 2698} 2699 2700size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2701 for (size_t i = 0; i < size; i++) { 2702 ids[i] = i; 2703 } 2704 return size; 2705} 2706 2707bool os::get_page_info(char *start, page_info* info) { 2708 return false; 2709} 2710 2711char *os::scan_pages(char *start, char* end, page_info* page_expected, 2712 page_info* page_found) { 2713 return end; 2714} 2715 2716 2717int os::Linux::sched_getcpu_syscall(void) { 2718 unsigned int cpu; 2719 int retval = -1; 2720 2721#if defined(IA32) 2722 #ifndef SYS_getcpu 2723 #define SYS_getcpu 318 2724 #endif 2725 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2726#elif defined(AMD64) 2727// Unfortunately we have to bring all these macros here from vsyscall.h 2728// to be able to compile on old linuxes. 2729 #define __NR_vgetcpu 2 2730 #define VSYSCALL_START (-10UL << 20) 2731 #define VSYSCALL_SIZE 1024 2732 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2733 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2734 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2735 retval = vgetcpu(&cpu, NULL, NULL); 2736#endif 2737 2738 return (retval == -1) ? retval : cpu; 2739} 2740 2741// Something to do with the numa-aware allocator needs these symbols 2742extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2743extern "C" JNIEXPORT void numa_error(char *where) { } 2744extern "C" JNIEXPORT int fork1() { return fork(); } 2745 2746 2747// If we are running with libnuma version > 2, then we should 2748// be trying to use symbols with versions 1.1 2749// If we are running with earlier version, which did not have symbol versions, 2750// we should use the base version. 2751void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2752 void *f = dlvsym(handle, name, "libnuma_1.1"); 2753 if (f == NULL) { 2754 f = dlsym(handle, name); 2755 } 2756 return f; 2757} 2758 2759bool os::Linux::libnuma_init() { 2760 // sched_getcpu() should be in libc. 2761 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2762 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2763 2764 // If it's not, try a direct syscall. 2765 if (sched_getcpu() == -1) { 2766 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2767 (void*)&sched_getcpu_syscall)); 2768 } 2769 2770 if (sched_getcpu() != -1) { // Does it work? 2771 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2772 if (handle != NULL) { 2773 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2774 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2775 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2776 libnuma_dlsym(handle, "numa_max_node"))); 2777 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2778 libnuma_dlsym(handle, "numa_available"))); 2779 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2780 libnuma_dlsym(handle, "numa_tonode_memory"))); 2781 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2782 libnuma_dlsym(handle, "numa_interleave_memory"))); 2783 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2784 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2785 2786 2787 if (numa_available() != -1) { 2788 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2789 // Create a cpu -> node mapping 2790 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2791 rebuild_cpu_to_node_map(); 2792 return true; 2793 } 2794 } 2795 } 2796 return false; 2797} 2798 2799// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2800// The table is later used in get_node_by_cpu(). 2801void os::Linux::rebuild_cpu_to_node_map() { 2802 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2803 // in libnuma (possible values are starting from 16, 2804 // and continuing up with every other power of 2, but less 2805 // than the maximum number of CPUs supported by kernel), and 2806 // is a subject to change (in libnuma version 2 the requirements 2807 // are more reasonable) we'll just hardcode the number they use 2808 // in the library. 2809 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2810 2811 size_t cpu_num = os::active_processor_count(); 2812 size_t cpu_map_size = NCPUS / BitsPerCLong; 2813 size_t cpu_map_valid_size = 2814 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2815 2816 cpu_to_node()->clear(); 2817 cpu_to_node()->at_grow(cpu_num - 1); 2818 size_t node_num = numa_get_groups_num(); 2819 2820 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2821 for (size_t i = 0; i < node_num; i++) { 2822 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2823 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2824 if (cpu_map[j] != 0) { 2825 for (size_t k = 0; k < BitsPerCLong; k++) { 2826 if (cpu_map[j] & (1UL << k)) { 2827 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2828 } 2829 } 2830 } 2831 } 2832 } 2833 } 2834 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2835} 2836 2837int os::Linux::get_node_by_cpu(int cpu_id) { 2838 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2839 return cpu_to_node()->at(cpu_id); 2840 } 2841 return -1; 2842} 2843 2844GrowableArray<int>* os::Linux::_cpu_to_node; 2845os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2846os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2847os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2848os::Linux::numa_available_func_t os::Linux::_numa_available; 2849os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2850os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2851os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2852unsigned long* os::Linux::_numa_all_nodes; 2853 2854bool os::pd_uncommit_memory(char* addr, size_t size) { 2855 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2856 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2857 return res != (uintptr_t) MAP_FAILED; 2858} 2859 2860static address get_stack_commited_bottom(address bottom, size_t size) { 2861 address nbot = bottom; 2862 address ntop = bottom + size; 2863 2864 size_t page_sz = os::vm_page_size(); 2865 unsigned pages = size / page_sz; 2866 2867 unsigned char vec[1]; 2868 unsigned imin = 1, imax = pages + 1, imid; 2869 int mincore_return_value = 0; 2870 2871 assert(imin <= imax, "Unexpected page size"); 2872 2873 while (imin < imax) { 2874 imid = (imax + imin) / 2; 2875 nbot = ntop - (imid * page_sz); 2876 2877 // Use a trick with mincore to check whether the page is mapped or not. 2878 // mincore sets vec to 1 if page resides in memory and to 0 if page 2879 // is swapped output but if page we are asking for is unmapped 2880 // it returns -1,ENOMEM 2881 mincore_return_value = mincore(nbot, page_sz, vec); 2882 2883 if (mincore_return_value == -1) { 2884 // Page is not mapped go up 2885 // to find first mapped page 2886 if (errno != EAGAIN) { 2887 assert(errno == ENOMEM, "Unexpected mincore errno"); 2888 imax = imid; 2889 } 2890 } else { 2891 // Page is mapped go down 2892 // to find first not mapped page 2893 imin = imid + 1; 2894 } 2895 } 2896 2897 nbot = nbot + page_sz; 2898 2899 // Adjust stack bottom one page up if last checked page is not mapped 2900 if (mincore_return_value == -1) { 2901 nbot = nbot + page_sz; 2902 } 2903 2904 return nbot; 2905} 2906 2907 2908// Linux uses a growable mapping for the stack, and if the mapping for 2909// the stack guard pages is not removed when we detach a thread the 2910// stack cannot grow beyond the pages where the stack guard was 2911// mapped. If at some point later in the process the stack expands to 2912// that point, the Linux kernel cannot expand the stack any further 2913// because the guard pages are in the way, and a segfault occurs. 2914// 2915// However, it's essential not to split the stack region by unmapping 2916// a region (leaving a hole) that's already part of the stack mapping, 2917// so if the stack mapping has already grown beyond the guard pages at 2918// the time we create them, we have to truncate the stack mapping. 2919// So, we need to know the extent of the stack mapping when 2920// create_stack_guard_pages() is called. 2921 2922// We only need this for stacks that are growable: at the time of 2923// writing thread stacks don't use growable mappings (i.e. those 2924// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 2925// only applies to the main thread. 2926 2927// If the (growable) stack mapping already extends beyond the point 2928// where we're going to put our guard pages, truncate the mapping at 2929// that point by munmap()ping it. This ensures that when we later 2930// munmap() the guard pages we don't leave a hole in the stack 2931// mapping. This only affects the main/initial thread 2932 2933bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 2934 if (os::Linux::is_initial_thread()) { 2935 // As we manually grow stack up to bottom inside create_attached_thread(), 2936 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 2937 // we don't need to do anything special. 2938 // Check it first, before calling heavy function. 2939 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 2940 unsigned char vec[1]; 2941 2942 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 2943 // Fallback to slow path on all errors, including EAGAIN 2944 stack_extent = (uintptr_t) get_stack_commited_bottom( 2945 os::Linux::initial_thread_stack_bottom(), 2946 (size_t)addr - stack_extent); 2947 } 2948 2949 if (stack_extent < (uintptr_t)addr) { 2950 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 2951 } 2952 } 2953 2954 return os::commit_memory(addr, size, !ExecMem); 2955} 2956 2957// If this is a growable mapping, remove the guard pages entirely by 2958// munmap()ping them. If not, just call uncommit_memory(). This only 2959// affects the main/initial thread, but guard against future OS changes 2960// It's safe to always unmap guard pages for initial thread because we 2961// always place it right after end of the mapped region 2962 2963bool os::remove_stack_guard_pages(char* addr, size_t size) { 2964 uintptr_t stack_extent, stack_base; 2965 2966 if (os::Linux::is_initial_thread()) { 2967 return ::munmap(addr, size) == 0; 2968 } 2969 2970 return os::uncommit_memory(addr, size); 2971} 2972 2973// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 2974// at 'requested_addr'. If there are existing memory mappings at the same 2975// location, however, they will be overwritten. If 'fixed' is false, 2976// 'requested_addr' is only treated as a hint, the return value may or 2977// may not start from the requested address. Unlike Linux mmap(), this 2978// function returns NULL to indicate failure. 2979static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 2980 char * addr; 2981 int flags; 2982 2983 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 2984 if (fixed) { 2985 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 2986 flags |= MAP_FIXED; 2987 } 2988 2989 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 2990 // touch an uncommitted page. Otherwise, the read/write might 2991 // succeed if we have enough swap space to back the physical page. 2992 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 2993 flags, -1, 0); 2994 2995 return addr == MAP_FAILED ? NULL : addr; 2996} 2997 2998static int anon_munmap(char * addr, size_t size) { 2999 return ::munmap(addr, size) == 0; 3000} 3001 3002char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3003 size_t alignment_hint) { 3004 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3005} 3006 3007bool os::pd_release_memory(char* addr, size_t size) { 3008 return anon_munmap(addr, size); 3009} 3010 3011static bool linux_mprotect(char* addr, size_t size, int prot) { 3012 // Linux wants the mprotect address argument to be page aligned. 3013 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3014 3015 // According to SUSv3, mprotect() should only be used with mappings 3016 // established by mmap(), and mmap() always maps whole pages. Unaligned 3017 // 'addr' likely indicates problem in the VM (e.g. trying to change 3018 // protection of malloc'ed or statically allocated memory). Check the 3019 // caller if you hit this assert. 3020 assert(addr == bottom, "sanity check"); 3021 3022 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3023 return ::mprotect(bottom, size, prot) == 0; 3024} 3025 3026// Set protections specified 3027bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3028 bool is_committed) { 3029 unsigned int p = 0; 3030 switch (prot) { 3031 case MEM_PROT_NONE: p = PROT_NONE; break; 3032 case MEM_PROT_READ: p = PROT_READ; break; 3033 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3034 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3035 default: 3036 ShouldNotReachHere(); 3037 } 3038 // is_committed is unused. 3039 return linux_mprotect(addr, bytes, p); 3040} 3041 3042bool os::guard_memory(char* addr, size_t size) { 3043 return linux_mprotect(addr, size, PROT_NONE); 3044} 3045 3046bool os::unguard_memory(char* addr, size_t size) { 3047 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3048} 3049 3050bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3051 size_t page_size) { 3052 bool result = false; 3053 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3054 MAP_ANONYMOUS|MAP_PRIVATE, 3055 -1, 0); 3056 if (p != MAP_FAILED) { 3057 void *aligned_p = align_ptr_up(p, page_size); 3058 3059 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3060 3061 munmap(p, page_size * 2); 3062 } 3063 3064 if (warn && !result) { 3065 warning("TransparentHugePages is not supported by the operating system."); 3066 } 3067 3068 return result; 3069} 3070 3071bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3072 bool result = false; 3073 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3074 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3075 -1, 0); 3076 3077 if (p != MAP_FAILED) { 3078 // We don't know if this really is a huge page or not. 3079 FILE *fp = fopen("/proc/self/maps", "r"); 3080 if (fp) { 3081 while (!feof(fp)) { 3082 char chars[257]; 3083 long x = 0; 3084 if (fgets(chars, sizeof(chars), fp)) { 3085 if (sscanf(chars, "%lx-%*x", &x) == 1 3086 && x == (long)p) { 3087 if (strstr (chars, "hugepage")) { 3088 result = true; 3089 break; 3090 } 3091 } 3092 } 3093 } 3094 fclose(fp); 3095 } 3096 munmap(p, page_size); 3097 } 3098 3099 if (warn && !result) { 3100 warning("HugeTLBFS is not supported by the operating system."); 3101 } 3102 3103 return result; 3104} 3105 3106// Set the coredump_filter bits to include largepages in core dump (bit 6) 3107// 3108// From the coredump_filter documentation: 3109// 3110// - (bit 0) anonymous private memory 3111// - (bit 1) anonymous shared memory 3112// - (bit 2) file-backed private memory 3113// - (bit 3) file-backed shared memory 3114// - (bit 4) ELF header pages in file-backed private memory areas (it is 3115// effective only if the bit 2 is cleared) 3116// - (bit 5) hugetlb private memory 3117// - (bit 6) hugetlb shared memory 3118// 3119static void set_coredump_filter(void) { 3120 FILE *f; 3121 long cdm; 3122 3123 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3124 return; 3125 } 3126 3127 if (fscanf(f, "%lx", &cdm) != 1) { 3128 fclose(f); 3129 return; 3130 } 3131 3132 rewind(f); 3133 3134 if ((cdm & LARGEPAGES_BIT) == 0) { 3135 cdm |= LARGEPAGES_BIT; 3136 fprintf(f, "%#lx", cdm); 3137 } 3138 3139 fclose(f); 3140} 3141 3142// Large page support 3143 3144static size_t _large_page_size = 0; 3145 3146size_t os::Linux::find_large_page_size() { 3147 size_t large_page_size = 0; 3148 3149 // large_page_size on Linux is used to round up heap size. x86 uses either 3150 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3151 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3152 // page as large as 256M. 3153 // 3154 // Here we try to figure out page size by parsing /proc/meminfo and looking 3155 // for a line with the following format: 3156 // Hugepagesize: 2048 kB 3157 // 3158 // If we can't determine the value (e.g. /proc is not mounted, or the text 3159 // format has been changed), we'll use the largest page size supported by 3160 // the processor. 3161 3162#ifndef ZERO 3163 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3164 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M); 3165#endif // ZERO 3166 3167 FILE *fp = fopen("/proc/meminfo", "r"); 3168 if (fp) { 3169 while (!feof(fp)) { 3170 int x = 0; 3171 char buf[16]; 3172 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3173 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3174 large_page_size = x * K; 3175 break; 3176 } 3177 } else { 3178 // skip to next line 3179 for (;;) { 3180 int ch = fgetc(fp); 3181 if (ch == EOF || ch == (int)'\n') break; 3182 } 3183 } 3184 } 3185 fclose(fp); 3186 } 3187 3188 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3189 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3190 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3191 proper_unit_for_byte_size(large_page_size)); 3192 } 3193 3194 return large_page_size; 3195} 3196 3197size_t os::Linux::setup_large_page_size() { 3198 _large_page_size = Linux::find_large_page_size(); 3199 const size_t default_page_size = (size_t)Linux::page_size(); 3200 if (_large_page_size > default_page_size) { 3201 _page_sizes[0] = _large_page_size; 3202 _page_sizes[1] = default_page_size; 3203 _page_sizes[2] = 0; 3204 } 3205 3206 return _large_page_size; 3207} 3208 3209bool os::Linux::setup_large_page_type(size_t page_size) { 3210 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3211 FLAG_IS_DEFAULT(UseSHM) && 3212 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3213 3214 // The type of large pages has not been specified by the user. 3215 3216 // Try UseHugeTLBFS and then UseSHM. 3217 UseHugeTLBFS = UseSHM = true; 3218 3219 // Don't try UseTransparentHugePages since there are known 3220 // performance issues with it turned on. This might change in the future. 3221 UseTransparentHugePages = false; 3222 } 3223 3224 if (UseTransparentHugePages) { 3225 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3226 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3227 UseHugeTLBFS = false; 3228 UseSHM = false; 3229 return true; 3230 } 3231 UseTransparentHugePages = false; 3232 } 3233 3234 if (UseHugeTLBFS) { 3235 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3236 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3237 UseSHM = false; 3238 return true; 3239 } 3240 UseHugeTLBFS = false; 3241 } 3242 3243 return UseSHM; 3244} 3245 3246void os::large_page_init() { 3247 if (!UseLargePages && 3248 !UseTransparentHugePages && 3249 !UseHugeTLBFS && 3250 !UseSHM) { 3251 // Not using large pages. 3252 return; 3253 } 3254 3255 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3256 // The user explicitly turned off large pages. 3257 // Ignore the rest of the large pages flags. 3258 UseTransparentHugePages = false; 3259 UseHugeTLBFS = false; 3260 UseSHM = false; 3261 return; 3262 } 3263 3264 size_t large_page_size = Linux::setup_large_page_size(); 3265 UseLargePages = Linux::setup_large_page_type(large_page_size); 3266 3267 set_coredump_filter(); 3268} 3269 3270#ifndef SHM_HUGETLB 3271 #define SHM_HUGETLB 04000 3272#endif 3273 3274char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3275 char* req_addr, bool exec) { 3276 // "exec" is passed in but not used. Creating the shared image for 3277 // the code cache doesn't have an SHM_X executable permission to check. 3278 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3279 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3280 3281 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3282 return NULL; // Fallback to small pages. 3283 } 3284 3285 key_t key = IPC_PRIVATE; 3286 char *addr; 3287 3288 bool warn_on_failure = UseLargePages && 3289 (!FLAG_IS_DEFAULT(UseLargePages) || 3290 !FLAG_IS_DEFAULT(UseSHM) || 3291 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3292 char msg[128]; 3293 3294 // Create a large shared memory region to attach to based on size. 3295 // Currently, size is the total size of the heap 3296 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3297 if (shmid == -1) { 3298 // Possible reasons for shmget failure: 3299 // 1. shmmax is too small for Java heap. 3300 // > check shmmax value: cat /proc/sys/kernel/shmmax 3301 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3302 // 2. not enough large page memory. 3303 // > check available large pages: cat /proc/meminfo 3304 // > increase amount of large pages: 3305 // echo new_value > /proc/sys/vm/nr_hugepages 3306 // Note 1: different Linux may use different name for this property, 3307 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3308 // Note 2: it's possible there's enough physical memory available but 3309 // they are so fragmented after a long run that they can't 3310 // coalesce into large pages. Try to reserve large pages when 3311 // the system is still "fresh". 3312 if (warn_on_failure) { 3313 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3314 warning("%s", msg); 3315 } 3316 return NULL; 3317 } 3318 3319 // attach to the region 3320 addr = (char*)shmat(shmid, req_addr, 0); 3321 int err = errno; 3322 3323 // Remove shmid. If shmat() is successful, the actual shared memory segment 3324 // will be deleted when it's detached by shmdt() or when the process 3325 // terminates. If shmat() is not successful this will remove the shared 3326 // segment immediately. 3327 shmctl(shmid, IPC_RMID, NULL); 3328 3329 if ((intptr_t)addr == -1) { 3330 if (warn_on_failure) { 3331 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3332 warning("%s", msg); 3333 } 3334 return NULL; 3335 } 3336 3337 return addr; 3338} 3339 3340static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3341 int error) { 3342 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3343 3344 bool warn_on_failure = UseLargePages && 3345 (!FLAG_IS_DEFAULT(UseLargePages) || 3346 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3347 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3348 3349 if (warn_on_failure) { 3350 char msg[128]; 3351 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3352 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3353 warning("%s", msg); 3354 } 3355} 3356 3357char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3358 char* req_addr, 3359 bool exec) { 3360 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3361 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3362 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3363 3364 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3365 char* addr = (char*)::mmap(req_addr, bytes, prot, 3366 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3367 -1, 0); 3368 3369 if (addr == MAP_FAILED) { 3370 warn_on_large_pages_failure(req_addr, bytes, errno); 3371 return NULL; 3372 } 3373 3374 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3375 3376 return addr; 3377} 3378 3379// Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed(). 3380// Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address 3381// (req_addr != NULL) or with a given alignment. 3382// - bytes shall be a multiple of alignment. 3383// - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3384// - alignment sets the alignment at which memory shall be allocated. 3385// It must be a multiple of allocation granularity. 3386// Returns address of memory or NULL. If req_addr was not NULL, will only return 3387// req_addr or NULL. 3388static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) { 3389 3390 size_t extra_size = bytes; 3391 if (req_addr == NULL && alignment > 0) { 3392 extra_size += alignment; 3393 } 3394 3395 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE, 3396 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 3397 -1, 0); 3398 if (start == MAP_FAILED) { 3399 start = NULL; 3400 } else { 3401 if (req_addr != NULL) { 3402 if (start != req_addr) { 3403 ::munmap(start, extra_size); 3404 start = NULL; 3405 } 3406 } else { 3407 char* const start_aligned = (char*) align_ptr_up(start, alignment); 3408 char* const end_aligned = start_aligned + bytes; 3409 char* const end = start + extra_size; 3410 if (start_aligned > start) { 3411 ::munmap(start, start_aligned - start); 3412 } 3413 if (end_aligned < end) { 3414 ::munmap(end_aligned, end - end_aligned); 3415 } 3416 start = start_aligned; 3417 } 3418 } 3419 return start; 3420 3421} 3422 3423// Reserve memory using mmap(MAP_HUGETLB). 3424// - bytes shall be a multiple of alignment. 3425// - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3426// - alignment sets the alignment at which memory shall be allocated. 3427// It must be a multiple of allocation granularity. 3428// Returns address of memory or NULL. If req_addr was not NULL, will only return 3429// req_addr or NULL. 3430char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3431 size_t alignment, 3432 char* req_addr, 3433 bool exec) { 3434 size_t large_page_size = os::large_page_size(); 3435 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3436 3437 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3438 assert(is_size_aligned(bytes, alignment), "Must be"); 3439 3440 // First reserve - but not commit - the address range in small pages. 3441 char* const start = anon_mmap_aligned(bytes, alignment, req_addr); 3442 3443 if (start == NULL) { 3444 return NULL; 3445 } 3446 3447 assert(is_ptr_aligned(start, alignment), "Must be"); 3448 3449 char* end = start + bytes; 3450 3451 // Find the regions of the allocated chunk that can be promoted to large pages. 3452 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3453 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3454 3455 size_t lp_bytes = lp_end - lp_start; 3456 3457 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3458 3459 if (lp_bytes == 0) { 3460 // The mapped region doesn't even span the start and the end of a large page. 3461 // Fall back to allocate a non-special area. 3462 ::munmap(start, end - start); 3463 return NULL; 3464 } 3465 3466 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3467 3468 void* result; 3469 3470 // Commit small-paged leading area. 3471 if (start != lp_start) { 3472 result = ::mmap(start, lp_start - start, prot, 3473 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3474 -1, 0); 3475 if (result == MAP_FAILED) { 3476 ::munmap(lp_start, end - lp_start); 3477 return NULL; 3478 } 3479 } 3480 3481 // Commit large-paged area. 3482 result = ::mmap(lp_start, lp_bytes, prot, 3483 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3484 -1, 0); 3485 if (result == MAP_FAILED) { 3486 warn_on_large_pages_failure(lp_start, lp_bytes, errno); 3487 // If the mmap above fails, the large pages region will be unmapped and we 3488 // have regions before and after with small pages. Release these regions. 3489 // 3490 // | mapped | unmapped | mapped | 3491 // ^ ^ ^ ^ 3492 // start lp_start lp_end end 3493 // 3494 ::munmap(start, lp_start - start); 3495 ::munmap(lp_end, end - lp_end); 3496 return NULL; 3497 } 3498 3499 // Commit small-paged trailing area. 3500 if (lp_end != end) { 3501 result = ::mmap(lp_end, end - lp_end, prot, 3502 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3503 -1, 0); 3504 if (result == MAP_FAILED) { 3505 ::munmap(start, lp_end - start); 3506 return NULL; 3507 } 3508 } 3509 3510 return start; 3511} 3512 3513char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3514 size_t alignment, 3515 char* req_addr, 3516 bool exec) { 3517 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3518 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3519 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be"); 3520 assert(is_power_of_2(os::large_page_size()), "Must be"); 3521 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3522 3523 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3524 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3525 } else { 3526 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3527 } 3528} 3529 3530char* os::reserve_memory_special(size_t bytes, size_t alignment, 3531 char* req_addr, bool exec) { 3532 assert(UseLargePages, "only for large pages"); 3533 3534 char* addr; 3535 if (UseSHM) { 3536 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3537 } else { 3538 assert(UseHugeTLBFS, "must be"); 3539 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3540 } 3541 3542 if (addr != NULL) { 3543 if (UseNUMAInterleaving) { 3544 numa_make_global(addr, bytes); 3545 } 3546 3547 // The memory is committed 3548 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3549 } 3550 3551 return addr; 3552} 3553 3554bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3555 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3556 return shmdt(base) == 0; 3557} 3558 3559bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3560 return pd_release_memory(base, bytes); 3561} 3562 3563bool os::release_memory_special(char* base, size_t bytes) { 3564 bool res; 3565 if (MemTracker::tracking_level() > NMT_minimal) { 3566 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3567 res = os::Linux::release_memory_special_impl(base, bytes); 3568 if (res) { 3569 tkr.record((address)base, bytes); 3570 } 3571 3572 } else { 3573 res = os::Linux::release_memory_special_impl(base, bytes); 3574 } 3575 return res; 3576} 3577 3578bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3579 assert(UseLargePages, "only for large pages"); 3580 bool res; 3581 3582 if (UseSHM) { 3583 res = os::Linux::release_memory_special_shm(base, bytes); 3584 } else { 3585 assert(UseHugeTLBFS, "must be"); 3586 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3587 } 3588 return res; 3589} 3590 3591size_t os::large_page_size() { 3592 return _large_page_size; 3593} 3594 3595// With SysV SHM the entire memory region must be allocated as shared 3596// memory. 3597// HugeTLBFS allows application to commit large page memory on demand. 3598// However, when committing memory with HugeTLBFS fails, the region 3599// that was supposed to be committed will lose the old reservation 3600// and allow other threads to steal that memory region. Because of this 3601// behavior we can't commit HugeTLBFS memory. 3602bool os::can_commit_large_page_memory() { 3603 return UseTransparentHugePages; 3604} 3605 3606bool os::can_execute_large_page_memory() { 3607 return UseTransparentHugePages || UseHugeTLBFS; 3608} 3609 3610// Reserve memory at an arbitrary address, only if that area is 3611// available (and not reserved for something else). 3612 3613char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3614 const int max_tries = 10; 3615 char* base[max_tries]; 3616 size_t size[max_tries]; 3617 const size_t gap = 0x000000; 3618 3619 // Assert only that the size is a multiple of the page size, since 3620 // that's all that mmap requires, and since that's all we really know 3621 // about at this low abstraction level. If we need higher alignment, 3622 // we can either pass an alignment to this method or verify alignment 3623 // in one of the methods further up the call chain. See bug 5044738. 3624 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3625 3626 // Repeatedly allocate blocks until the block is allocated at the 3627 // right spot. 3628 3629 // Linux mmap allows caller to pass an address as hint; give it a try first, 3630 // if kernel honors the hint then we can return immediately. 3631 char * addr = anon_mmap(requested_addr, bytes, false); 3632 if (addr == requested_addr) { 3633 return requested_addr; 3634 } 3635 3636 if (addr != NULL) { 3637 // mmap() is successful but it fails to reserve at the requested address 3638 anon_munmap(addr, bytes); 3639 } 3640 3641 int i; 3642 for (i = 0; i < max_tries; ++i) { 3643 base[i] = reserve_memory(bytes); 3644 3645 if (base[i] != NULL) { 3646 // Is this the block we wanted? 3647 if (base[i] == requested_addr) { 3648 size[i] = bytes; 3649 break; 3650 } 3651 3652 // Does this overlap the block we wanted? Give back the overlapped 3653 // parts and try again. 3654 3655 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3656 if (top_overlap >= 0 && (size_t)top_overlap < bytes) { 3657 unmap_memory(base[i], top_overlap); 3658 base[i] += top_overlap; 3659 size[i] = bytes - top_overlap; 3660 } else { 3661 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr; 3662 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) { 3663 unmap_memory(requested_addr, bottom_overlap); 3664 size[i] = bytes - bottom_overlap; 3665 } else { 3666 size[i] = bytes; 3667 } 3668 } 3669 } 3670 } 3671 3672 // Give back the unused reserved pieces. 3673 3674 for (int j = 0; j < i; ++j) { 3675 if (base[j] != NULL) { 3676 unmap_memory(base[j], size[j]); 3677 } 3678 } 3679 3680 if (i < max_tries) { 3681 return requested_addr; 3682 } else { 3683 return NULL; 3684 } 3685} 3686 3687size_t os::read(int fd, void *buf, unsigned int nBytes) { 3688 return ::read(fd, buf, nBytes); 3689} 3690 3691size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) { 3692 return ::pread(fd, buf, nBytes, offset); 3693} 3694 3695// Short sleep, direct OS call. 3696// 3697// Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3698// sched_yield(2) will actually give up the CPU: 3699// 3700// * Alone on this pariticular CPU, keeps running. 3701// * Before the introduction of "skip_buddy" with "compat_yield" disabled 3702// (pre 2.6.39). 3703// 3704// So calling this with 0 is an alternative. 3705// 3706void os::naked_short_sleep(jlong ms) { 3707 struct timespec req; 3708 3709 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3710 req.tv_sec = 0; 3711 if (ms > 0) { 3712 req.tv_nsec = (ms % 1000) * 1000000; 3713 } else { 3714 req.tv_nsec = 1; 3715 } 3716 3717 nanosleep(&req, NULL); 3718 3719 return; 3720} 3721 3722// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3723void os::infinite_sleep() { 3724 while (true) { // sleep forever ... 3725 ::sleep(100); // ... 100 seconds at a time 3726 } 3727} 3728 3729// Used to convert frequent JVM_Yield() to nops 3730bool os::dont_yield() { 3731 return DontYieldALot; 3732} 3733 3734void os::naked_yield() { 3735 sched_yield(); 3736} 3737 3738//////////////////////////////////////////////////////////////////////////////// 3739// thread priority support 3740 3741// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3742// only supports dynamic priority, static priority must be zero. For real-time 3743// applications, Linux supports SCHED_RR which allows static priority (1-99). 3744// However, for large multi-threaded applications, SCHED_RR is not only slower 3745// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3746// of 5 runs - Sep 2005). 3747// 3748// The following code actually changes the niceness of kernel-thread/LWP. It 3749// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3750// not the entire user process, and user level threads are 1:1 mapped to kernel 3751// threads. It has always been the case, but could change in the future. For 3752// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3753// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3754 3755int os::java_to_os_priority[CriticalPriority + 1] = { 3756 19, // 0 Entry should never be used 3757 3758 4, // 1 MinPriority 3759 3, // 2 3760 2, // 3 3761 3762 1, // 4 3763 0, // 5 NormPriority 3764 -1, // 6 3765 3766 -2, // 7 3767 -3, // 8 3768 -4, // 9 NearMaxPriority 3769 3770 -5, // 10 MaxPriority 3771 3772 -5 // 11 CriticalPriority 3773}; 3774 3775static int prio_init() { 3776 if (ThreadPriorityPolicy == 1) { 3777 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3778 // if effective uid is not root. Perhaps, a more elegant way of doing 3779 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3780 if (geteuid() != 0) { 3781 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3782 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3783 } 3784 ThreadPriorityPolicy = 0; 3785 } 3786 } 3787 if (UseCriticalJavaThreadPriority) { 3788 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3789 } 3790 return 0; 3791} 3792 3793OSReturn os::set_native_priority(Thread* thread, int newpri) { 3794 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 3795 3796 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3797 return (ret == 0) ? OS_OK : OS_ERR; 3798} 3799 3800OSReturn os::get_native_priority(const Thread* const thread, 3801 int *priority_ptr) { 3802 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 3803 *priority_ptr = java_to_os_priority[NormPriority]; 3804 return OS_OK; 3805 } 3806 3807 errno = 0; 3808 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3809 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3810} 3811 3812// Hint to the underlying OS that a task switch would not be good. 3813// Void return because it's a hint and can fail. 3814void os::hint_no_preempt() {} 3815 3816//////////////////////////////////////////////////////////////////////////////// 3817// suspend/resume support 3818 3819// the low-level signal-based suspend/resume support is a remnant from the 3820// old VM-suspension that used to be for java-suspension, safepoints etc, 3821// within hotspot. Now there is a single use-case for this: 3822// - calling get_thread_pc() on the VMThread by the flat-profiler task 3823// that runs in the watcher thread. 3824// The remaining code is greatly simplified from the more general suspension 3825// code that used to be used. 3826// 3827// The protocol is quite simple: 3828// - suspend: 3829// - sends a signal to the target thread 3830// - polls the suspend state of the osthread using a yield loop 3831// - target thread signal handler (SR_handler) sets suspend state 3832// and blocks in sigsuspend until continued 3833// - resume: 3834// - sets target osthread state to continue 3835// - sends signal to end the sigsuspend loop in the SR_handler 3836// 3837// Note that the SR_lock plays no role in this suspend/resume protocol. 3838 3839static void resume_clear_context(OSThread *osthread) { 3840 osthread->set_ucontext(NULL); 3841 osthread->set_siginfo(NULL); 3842} 3843 3844static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 3845 ucontext_t* context) { 3846 osthread->set_ucontext(context); 3847 osthread->set_siginfo(siginfo); 3848} 3849 3850// Handler function invoked when a thread's execution is suspended or 3851// resumed. We have to be careful that only async-safe functions are 3852// called here (Note: most pthread functions are not async safe and 3853// should be avoided.) 3854// 3855// Note: sigwait() is a more natural fit than sigsuspend() from an 3856// interface point of view, but sigwait() prevents the signal hander 3857// from being run. libpthread would get very confused by not having 3858// its signal handlers run and prevents sigwait()'s use with the 3859// mutex granting granting signal. 3860// 3861// Currently only ever called on the VMThread and JavaThreads (PC sampling) 3862// 3863static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 3864 // Save and restore errno to avoid confusing native code with EINTR 3865 // after sigsuspend. 3866 int old_errno = errno; 3867 3868 Thread* thread = Thread::current(); 3869 OSThread* osthread = thread->osthread(); 3870 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 3871 3872 os::SuspendResume::State current = osthread->sr.state(); 3873 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 3874 suspend_save_context(osthread, siginfo, context); 3875 3876 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 3877 os::SuspendResume::State state = osthread->sr.suspended(); 3878 if (state == os::SuspendResume::SR_SUSPENDED) { 3879 sigset_t suspend_set; // signals for sigsuspend() 3880 3881 // get current set of blocked signals and unblock resume signal 3882 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 3883 sigdelset(&suspend_set, SR_signum); 3884 3885 sr_semaphore.signal(); 3886 // wait here until we are resumed 3887 while (1) { 3888 sigsuspend(&suspend_set); 3889 3890 os::SuspendResume::State result = osthread->sr.running(); 3891 if (result == os::SuspendResume::SR_RUNNING) { 3892 sr_semaphore.signal(); 3893 break; 3894 } 3895 } 3896 3897 } else if (state == os::SuspendResume::SR_RUNNING) { 3898 // request was cancelled, continue 3899 } else { 3900 ShouldNotReachHere(); 3901 } 3902 3903 resume_clear_context(osthread); 3904 } else if (current == os::SuspendResume::SR_RUNNING) { 3905 // request was cancelled, continue 3906 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 3907 // ignore 3908 } else { 3909 // ignore 3910 } 3911 3912 errno = old_errno; 3913} 3914 3915 3916static int SR_initialize() { 3917 struct sigaction act; 3918 char *s; 3919 // Get signal number to use for suspend/resume 3920 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 3921 int sig = ::strtol(s, 0, 10); 3922 if (sig > 0 || sig < _NSIG) { 3923 SR_signum = sig; 3924 } 3925 } 3926 3927 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 3928 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 3929 3930 sigemptyset(&SR_sigset); 3931 sigaddset(&SR_sigset, SR_signum); 3932 3933 // Set up signal handler for suspend/resume 3934 act.sa_flags = SA_RESTART|SA_SIGINFO; 3935 act.sa_handler = (void (*)(int)) SR_handler; 3936 3937 // SR_signum is blocked by default. 3938 // 4528190 - We also need to block pthread restart signal (32 on all 3939 // supported Linux platforms). Note that LinuxThreads need to block 3940 // this signal for all threads to work properly. So we don't have 3941 // to use hard-coded signal number when setting up the mask. 3942 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 3943 3944 if (sigaction(SR_signum, &act, 0) == -1) { 3945 return -1; 3946 } 3947 3948 // Save signal flag 3949 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 3950 return 0; 3951} 3952 3953static int sr_notify(OSThread* osthread) { 3954 int status = pthread_kill(osthread->pthread_id(), SR_signum); 3955 assert_status(status == 0, status, "pthread_kill"); 3956 return status; 3957} 3958 3959// "Randomly" selected value for how long we want to spin 3960// before bailing out on suspending a thread, also how often 3961// we send a signal to a thread we want to resume 3962static const int RANDOMLY_LARGE_INTEGER = 1000000; 3963static const int RANDOMLY_LARGE_INTEGER2 = 100; 3964 3965// returns true on success and false on error - really an error is fatal 3966// but this seems the normal response to library errors 3967static bool do_suspend(OSThread* osthread) { 3968 assert(osthread->sr.is_running(), "thread should be running"); 3969 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 3970 3971 // mark as suspended and send signal 3972 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 3973 // failed to switch, state wasn't running? 3974 ShouldNotReachHere(); 3975 return false; 3976 } 3977 3978 if (sr_notify(osthread) != 0) { 3979 ShouldNotReachHere(); 3980 } 3981 3982 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 3983 while (true) { 3984 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 3985 break; 3986 } else { 3987 // timeout 3988 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 3989 if (cancelled == os::SuspendResume::SR_RUNNING) { 3990 return false; 3991 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 3992 // make sure that we consume the signal on the semaphore as well 3993 sr_semaphore.wait(); 3994 break; 3995 } else { 3996 ShouldNotReachHere(); 3997 return false; 3998 } 3999 } 4000 } 4001 4002 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4003 return true; 4004} 4005 4006static void do_resume(OSThread* osthread) { 4007 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4008 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4009 4010 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4011 // failed to switch to WAKEUP_REQUEST 4012 ShouldNotReachHere(); 4013 return; 4014 } 4015 4016 while (true) { 4017 if (sr_notify(osthread) == 0) { 4018 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4019 if (osthread->sr.is_running()) { 4020 return; 4021 } 4022 } 4023 } else { 4024 ShouldNotReachHere(); 4025 } 4026 } 4027 4028 guarantee(osthread->sr.is_running(), "Must be running!"); 4029} 4030 4031/////////////////////////////////////////////////////////////////////////////////// 4032// signal handling (except suspend/resume) 4033 4034// This routine may be used by user applications as a "hook" to catch signals. 4035// The user-defined signal handler must pass unrecognized signals to this 4036// routine, and if it returns true (non-zero), then the signal handler must 4037// return immediately. If the flag "abort_if_unrecognized" is true, then this 4038// routine will never retun false (zero), but instead will execute a VM panic 4039// routine kill the process. 4040// 4041// If this routine returns false, it is OK to call it again. This allows 4042// the user-defined signal handler to perform checks either before or after 4043// the VM performs its own checks. Naturally, the user code would be making 4044// a serious error if it tried to handle an exception (such as a null check 4045// or breakpoint) that the VM was generating for its own correct operation. 4046// 4047// This routine may recognize any of the following kinds of signals: 4048// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4049// It should be consulted by handlers for any of those signals. 4050// 4051// The caller of this routine must pass in the three arguments supplied 4052// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4053// field of the structure passed to sigaction(). This routine assumes that 4054// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4055// 4056// Note that the VM will print warnings if it detects conflicting signal 4057// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4058// 4059extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4060 siginfo_t* siginfo, 4061 void* ucontext, 4062 int abort_if_unrecognized); 4063 4064void signalHandler(int sig, siginfo_t* info, void* uc) { 4065 assert(info != NULL && uc != NULL, "it must be old kernel"); 4066 int orig_errno = errno; // Preserve errno value over signal handler. 4067 JVM_handle_linux_signal(sig, info, uc, true); 4068 errno = orig_errno; 4069} 4070 4071 4072// This boolean allows users to forward their own non-matching signals 4073// to JVM_handle_linux_signal, harmlessly. 4074bool os::Linux::signal_handlers_are_installed = false; 4075 4076// For signal-chaining 4077struct sigaction os::Linux::sigact[MAXSIGNUM]; 4078unsigned int os::Linux::sigs = 0; 4079bool os::Linux::libjsig_is_loaded = false; 4080typedef struct sigaction *(*get_signal_t)(int); 4081get_signal_t os::Linux::get_signal_action = NULL; 4082 4083struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4084 struct sigaction *actp = NULL; 4085 4086 if (libjsig_is_loaded) { 4087 // Retrieve the old signal handler from libjsig 4088 actp = (*get_signal_action)(sig); 4089 } 4090 if (actp == NULL) { 4091 // Retrieve the preinstalled signal handler from jvm 4092 actp = get_preinstalled_handler(sig); 4093 } 4094 4095 return actp; 4096} 4097 4098static bool call_chained_handler(struct sigaction *actp, int sig, 4099 siginfo_t *siginfo, void *context) { 4100 // Call the old signal handler 4101 if (actp->sa_handler == SIG_DFL) { 4102 // It's more reasonable to let jvm treat it as an unexpected exception 4103 // instead of taking the default action. 4104 return false; 4105 } else if (actp->sa_handler != SIG_IGN) { 4106 if ((actp->sa_flags & SA_NODEFER) == 0) { 4107 // automaticlly block the signal 4108 sigaddset(&(actp->sa_mask), sig); 4109 } 4110 4111 sa_handler_t hand; 4112 sa_sigaction_t sa; 4113 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4114 // retrieve the chained handler 4115 if (siginfo_flag_set) { 4116 sa = actp->sa_sigaction; 4117 } else { 4118 hand = actp->sa_handler; 4119 } 4120 4121 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4122 actp->sa_handler = SIG_DFL; 4123 } 4124 4125 // try to honor the signal mask 4126 sigset_t oset; 4127 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4128 4129 // call into the chained handler 4130 if (siginfo_flag_set) { 4131 (*sa)(sig, siginfo, context); 4132 } else { 4133 (*hand)(sig); 4134 } 4135 4136 // restore the signal mask 4137 pthread_sigmask(SIG_SETMASK, &oset, 0); 4138 } 4139 // Tell jvm's signal handler the signal is taken care of. 4140 return true; 4141} 4142 4143bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4144 bool chained = false; 4145 // signal-chaining 4146 if (UseSignalChaining) { 4147 struct sigaction *actp = get_chained_signal_action(sig); 4148 if (actp != NULL) { 4149 chained = call_chained_handler(actp, sig, siginfo, context); 4150 } 4151 } 4152 return chained; 4153} 4154 4155struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4156 if ((((unsigned int)1 << sig) & sigs) != 0) { 4157 return &sigact[sig]; 4158 } 4159 return NULL; 4160} 4161 4162void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4163 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4164 sigact[sig] = oldAct; 4165 sigs |= (unsigned int)1 << sig; 4166} 4167 4168// for diagnostic 4169int os::Linux::sigflags[MAXSIGNUM]; 4170 4171int os::Linux::get_our_sigflags(int sig) { 4172 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4173 return sigflags[sig]; 4174} 4175 4176void os::Linux::set_our_sigflags(int sig, int flags) { 4177 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4178 sigflags[sig] = flags; 4179} 4180 4181void os::Linux::set_signal_handler(int sig, bool set_installed) { 4182 // Check for overwrite. 4183 struct sigaction oldAct; 4184 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4185 4186 void* oldhand = oldAct.sa_sigaction 4187 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4188 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4189 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4190 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4191 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4192 if (AllowUserSignalHandlers || !set_installed) { 4193 // Do not overwrite; user takes responsibility to forward to us. 4194 return; 4195 } else if (UseSignalChaining) { 4196 // save the old handler in jvm 4197 save_preinstalled_handler(sig, oldAct); 4198 // libjsig also interposes the sigaction() call below and saves the 4199 // old sigaction on it own. 4200 } else { 4201 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4202 "%#lx for signal %d.", (long)oldhand, sig)); 4203 } 4204 } 4205 4206 struct sigaction sigAct; 4207 sigfillset(&(sigAct.sa_mask)); 4208 sigAct.sa_handler = SIG_DFL; 4209 if (!set_installed) { 4210 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4211 } else { 4212 sigAct.sa_sigaction = signalHandler; 4213 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4214 } 4215 // Save flags, which are set by ours 4216 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4217 sigflags[sig] = sigAct.sa_flags; 4218 4219 int ret = sigaction(sig, &sigAct, &oldAct); 4220 assert(ret == 0, "check"); 4221 4222 void* oldhand2 = oldAct.sa_sigaction 4223 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4224 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4225 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4226} 4227 4228// install signal handlers for signals that HotSpot needs to 4229// handle in order to support Java-level exception handling. 4230 4231void os::Linux::install_signal_handlers() { 4232 if (!signal_handlers_are_installed) { 4233 signal_handlers_are_installed = true; 4234 4235 // signal-chaining 4236 typedef void (*signal_setting_t)(); 4237 signal_setting_t begin_signal_setting = NULL; 4238 signal_setting_t end_signal_setting = NULL; 4239 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4240 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4241 if (begin_signal_setting != NULL) { 4242 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4243 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4244 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4245 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4246 libjsig_is_loaded = true; 4247 assert(UseSignalChaining, "should enable signal-chaining"); 4248 } 4249 if (libjsig_is_loaded) { 4250 // Tell libjsig jvm is setting signal handlers 4251 (*begin_signal_setting)(); 4252 } 4253 4254 set_signal_handler(SIGSEGV, true); 4255 set_signal_handler(SIGPIPE, true); 4256 set_signal_handler(SIGBUS, true); 4257 set_signal_handler(SIGILL, true); 4258 set_signal_handler(SIGFPE, true); 4259#if defined(PPC64) 4260 set_signal_handler(SIGTRAP, true); 4261#endif 4262 set_signal_handler(SIGXFSZ, true); 4263 4264 if (libjsig_is_loaded) { 4265 // Tell libjsig jvm finishes setting signal handlers 4266 (*end_signal_setting)(); 4267 } 4268 4269 // We don't activate signal checker if libjsig is in place, we trust ourselves 4270 // and if UserSignalHandler is installed all bets are off. 4271 // Log that signal checking is off only if -verbose:jni is specified. 4272 if (CheckJNICalls) { 4273 if (libjsig_is_loaded) { 4274 if (PrintJNIResolving) { 4275 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4276 } 4277 check_signals = false; 4278 } 4279 if (AllowUserSignalHandlers) { 4280 if (PrintJNIResolving) { 4281 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4282 } 4283 check_signals = false; 4284 } 4285 } 4286 } 4287} 4288 4289// This is the fastest way to get thread cpu time on Linux. 4290// Returns cpu time (user+sys) for any thread, not only for current. 4291// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4292// It might work on 2.6.10+ with a special kernel/glibc patch. 4293// For reference, please, see IEEE Std 1003.1-2004: 4294// http://www.unix.org/single_unix_specification 4295 4296jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4297 struct timespec tp; 4298 int rc = os::Linux::clock_gettime(clockid, &tp); 4299 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4300 4301 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4302} 4303 4304///// 4305// glibc on Linux platform uses non-documented flag 4306// to indicate, that some special sort of signal 4307// trampoline is used. 4308// We will never set this flag, and we should 4309// ignore this flag in our diagnostic 4310#ifdef SIGNIFICANT_SIGNAL_MASK 4311 #undef SIGNIFICANT_SIGNAL_MASK 4312#endif 4313#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4314 4315static const char* get_signal_handler_name(address handler, 4316 char* buf, int buflen) { 4317 int offset; 4318 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4319 if (found) { 4320 // skip directory names 4321 const char *p1, *p2; 4322 p1 = buf; 4323 size_t len = strlen(os::file_separator()); 4324 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4325 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4326 } else { 4327 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4328 } 4329 return buf; 4330} 4331 4332static void print_signal_handler(outputStream* st, int sig, 4333 char* buf, size_t buflen) { 4334 struct sigaction sa; 4335 4336 sigaction(sig, NULL, &sa); 4337 4338 // See comment for SIGNIFICANT_SIGNAL_MASK define 4339 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4340 4341 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4342 4343 address handler = (sa.sa_flags & SA_SIGINFO) 4344 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4345 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4346 4347 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4348 st->print("SIG_DFL"); 4349 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4350 st->print("SIG_IGN"); 4351 } else { 4352 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4353 } 4354 4355 st->print(", sa_mask[0]="); 4356 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4357 4358 address rh = VMError::get_resetted_sighandler(sig); 4359 // May be, handler was resetted by VMError? 4360 if (rh != NULL) { 4361 handler = rh; 4362 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4363 } 4364 4365 st->print(", sa_flags="); 4366 os::Posix::print_sa_flags(st, sa.sa_flags); 4367 4368 // Check: is it our handler? 4369 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4370 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4371 // It is our signal handler 4372 // check for flags, reset system-used one! 4373 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4374 st->print( 4375 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4376 os::Linux::get_our_sigflags(sig)); 4377 } 4378 } 4379 st->cr(); 4380} 4381 4382 4383#define DO_SIGNAL_CHECK(sig) \ 4384 do { \ 4385 if (!sigismember(&check_signal_done, sig)) { \ 4386 os::Linux::check_signal_handler(sig); \ 4387 } \ 4388 } while (0) 4389 4390// This method is a periodic task to check for misbehaving JNI applications 4391// under CheckJNI, we can add any periodic checks here 4392 4393void os::run_periodic_checks() { 4394 if (check_signals == false) return; 4395 4396 // SEGV and BUS if overridden could potentially prevent 4397 // generation of hs*.log in the event of a crash, debugging 4398 // such a case can be very challenging, so we absolutely 4399 // check the following for a good measure: 4400 DO_SIGNAL_CHECK(SIGSEGV); 4401 DO_SIGNAL_CHECK(SIGILL); 4402 DO_SIGNAL_CHECK(SIGFPE); 4403 DO_SIGNAL_CHECK(SIGBUS); 4404 DO_SIGNAL_CHECK(SIGPIPE); 4405 DO_SIGNAL_CHECK(SIGXFSZ); 4406#if defined(PPC64) 4407 DO_SIGNAL_CHECK(SIGTRAP); 4408#endif 4409 4410 // ReduceSignalUsage allows the user to override these handlers 4411 // see comments at the very top and jvm_solaris.h 4412 if (!ReduceSignalUsage) { 4413 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4414 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4415 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4416 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4417 } 4418 4419 DO_SIGNAL_CHECK(SR_signum); 4420 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4421} 4422 4423typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4424 4425static os_sigaction_t os_sigaction = NULL; 4426 4427void os::Linux::check_signal_handler(int sig) { 4428 char buf[O_BUFLEN]; 4429 address jvmHandler = NULL; 4430 4431 4432 struct sigaction act; 4433 if (os_sigaction == NULL) { 4434 // only trust the default sigaction, in case it has been interposed 4435 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4436 if (os_sigaction == NULL) return; 4437 } 4438 4439 os_sigaction(sig, (struct sigaction*)NULL, &act); 4440 4441 4442 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4443 4444 address thisHandler = (act.sa_flags & SA_SIGINFO) 4445 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4446 : CAST_FROM_FN_PTR(address, act.sa_handler); 4447 4448 4449 switch (sig) { 4450 case SIGSEGV: 4451 case SIGBUS: 4452 case SIGFPE: 4453 case SIGPIPE: 4454 case SIGILL: 4455 case SIGXFSZ: 4456 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4457 break; 4458 4459 case SHUTDOWN1_SIGNAL: 4460 case SHUTDOWN2_SIGNAL: 4461 case SHUTDOWN3_SIGNAL: 4462 case BREAK_SIGNAL: 4463 jvmHandler = (address)user_handler(); 4464 break; 4465 4466 case INTERRUPT_SIGNAL: 4467 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4468 break; 4469 4470 default: 4471 if (sig == SR_signum) { 4472 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4473 } else { 4474 return; 4475 } 4476 break; 4477 } 4478 4479 if (thisHandler != jvmHandler) { 4480 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4481 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4482 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4483 // No need to check this sig any longer 4484 sigaddset(&check_signal_done, sig); 4485 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4486 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4487 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4488 exception_name(sig, buf, O_BUFLEN)); 4489 } 4490 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4491 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4492 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4493 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4494 // No need to check this sig any longer 4495 sigaddset(&check_signal_done, sig); 4496 } 4497 4498 // Dump all the signal 4499 if (sigismember(&check_signal_done, sig)) { 4500 print_signal_handlers(tty, buf, O_BUFLEN); 4501 } 4502} 4503 4504extern void report_error(char* file_name, int line_no, char* title, 4505 char* format, ...); 4506 4507extern bool signal_name(int signo, char* buf, size_t len); 4508 4509const char* os::exception_name(int exception_code, char* buf, size_t size) { 4510 if (0 < exception_code && exception_code <= SIGRTMAX) { 4511 // signal 4512 if (!signal_name(exception_code, buf, size)) { 4513 jio_snprintf(buf, size, "SIG%d", exception_code); 4514 } 4515 return buf; 4516 } else { 4517 return NULL; 4518 } 4519} 4520 4521// this is called _before_ the most of global arguments have been parsed 4522void os::init(void) { 4523 char dummy; // used to get a guess on initial stack address 4524// first_hrtime = gethrtime(); 4525 4526 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4527 4528 init_random(1234567); 4529 4530 ThreadCritical::initialize(); 4531 4532 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4533 if (Linux::page_size() == -1) { 4534 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4535 strerror(errno))); 4536 } 4537 init_page_sizes((size_t) Linux::page_size()); 4538 4539 Linux::initialize_system_info(); 4540 4541 // main_thread points to the aboriginal thread 4542 Linux::_main_thread = pthread_self(); 4543 4544 Linux::clock_init(); 4545 initial_time_count = javaTimeNanos(); 4546 4547 // pthread_condattr initialization for monotonic clock 4548 int status; 4549 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4550 if ((status = pthread_condattr_init(_condattr)) != 0) { 4551 fatal(err_msg("pthread_condattr_init: %s", strerror(status))); 4552 } 4553 // Only set the clock if CLOCK_MONOTONIC is available 4554 if (os::supports_monotonic_clock()) { 4555 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4556 if (status == EINVAL) { 4557 warning("Unable to use monotonic clock with relative timed-waits" \ 4558 " - changes to the time-of-day clock may have adverse affects"); 4559 } else { 4560 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status))); 4561 } 4562 } 4563 } 4564 // else it defaults to CLOCK_REALTIME 4565 4566 // If the pagesize of the VM is greater than 8K determine the appropriate 4567 // number of initial guard pages. The user can change this with the 4568 // command line arguments, if needed. 4569 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4570 StackYellowPages = 1; 4571 StackRedPages = 1; 4572 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4573 } 4574 4575 // retrieve entry point for pthread_setname_np 4576 Linux::_pthread_setname_np = 4577 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4578 4579} 4580 4581// To install functions for atexit system call 4582extern "C" { 4583 static void perfMemory_exit_helper() { 4584 perfMemory_exit(); 4585 } 4586} 4587 4588// this is called _after_ the global arguments have been parsed 4589jint os::init_2(void) { 4590 Linux::fast_thread_clock_init(); 4591 4592 // Allocate a single page and mark it as readable for safepoint polling 4593 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4594 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page"); 4595 4596 os::set_polling_page(polling_page); 4597 4598#ifndef PRODUCT 4599 if (Verbose && PrintMiscellaneous) { 4600 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", 4601 (intptr_t)polling_page); 4602 } 4603#endif 4604 4605 if (!UseMembar) { 4606 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4607 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4608 os::set_memory_serialize_page(mem_serialize_page); 4609 4610#ifndef PRODUCT 4611 if (Verbose && PrintMiscellaneous) { 4612 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", 4613 (intptr_t)mem_serialize_page); 4614 } 4615#endif 4616 } 4617 4618 // initialize suspend/resume support - must do this before signal_sets_init() 4619 if (SR_initialize() != 0) { 4620 perror("SR_initialize failed"); 4621 return JNI_ERR; 4622 } 4623 4624 Linux::signal_sets_init(); 4625 Linux::install_signal_handlers(); 4626 4627 // Check minimum allowable stack size for thread creation and to initialize 4628 // the java system classes, including StackOverflowError - depends on page 4629 // size. Add a page for compiler2 recursion in main thread. 4630 // Add in 2*BytesPerWord times page size to account for VM stack during 4631 // class initialization depending on 32 or 64 bit VM. 4632 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4633 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4634 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4635 4636 size_t threadStackSizeInBytes = ThreadStackSize * K; 4637 if (threadStackSizeInBytes != 0 && 4638 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4639 tty->print_cr("\nThe stack size specified is too small, " 4640 "Specify at least %dk", 4641 os::Linux::min_stack_allowed/ K); 4642 return JNI_ERR; 4643 } 4644 4645 // Make the stack size a multiple of the page size so that 4646 // the yellow/red zones can be guarded. 4647 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4648 vm_page_size())); 4649 4650 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4651 4652#if defined(IA32) 4653 workaround_expand_exec_shield_cs_limit(); 4654#endif 4655 4656 Linux::libpthread_init(); 4657 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4658 tty->print_cr("[HotSpot is running with %s, %s]\n", 4659 Linux::glibc_version(), Linux::libpthread_version()); 4660 } 4661 4662 if (UseNUMA) { 4663 if (!Linux::libnuma_init()) { 4664 UseNUMA = false; 4665 } else { 4666 if ((Linux::numa_max_node() < 1)) { 4667 // There's only one node(they start from 0), disable NUMA. 4668 UseNUMA = false; 4669 } 4670 } 4671 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4672 // we can make the adaptive lgrp chunk resizing work. If the user specified 4673 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4674 // disable adaptive resizing. 4675 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4676 if (FLAG_IS_DEFAULT(UseNUMA)) { 4677 UseNUMA = false; 4678 } else { 4679 if (FLAG_IS_DEFAULT(UseLargePages) && 4680 FLAG_IS_DEFAULT(UseSHM) && 4681 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4682 UseLargePages = false; 4683 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) { 4684 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)"); 4685 UseAdaptiveSizePolicy = false; 4686 UseAdaptiveNUMAChunkSizing = false; 4687 } 4688 } 4689 } 4690 if (!UseNUMA && ForceNUMA) { 4691 UseNUMA = true; 4692 } 4693 } 4694 4695 if (MaxFDLimit) { 4696 // set the number of file descriptors to max. print out error 4697 // if getrlimit/setrlimit fails but continue regardless. 4698 struct rlimit nbr_files; 4699 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4700 if (status != 0) { 4701 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4702 perror("os::init_2 getrlimit failed"); 4703 } 4704 } else { 4705 nbr_files.rlim_cur = nbr_files.rlim_max; 4706 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4707 if (status != 0) { 4708 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4709 perror("os::init_2 setrlimit failed"); 4710 } 4711 } 4712 } 4713 } 4714 4715 // Initialize lock used to serialize thread creation (see os::create_thread) 4716 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4717 4718 // at-exit methods are called in the reverse order of their registration. 4719 // atexit functions are called on return from main or as a result of a 4720 // call to exit(3C). There can be only 32 of these functions registered 4721 // and atexit() does not set errno. 4722 4723 if (PerfAllowAtExitRegistration) { 4724 // only register atexit functions if PerfAllowAtExitRegistration is set. 4725 // atexit functions can be delayed until process exit time, which 4726 // can be problematic for embedded VM situations. Embedded VMs should 4727 // call DestroyJavaVM() to assure that VM resources are released. 4728 4729 // note: perfMemory_exit_helper atexit function may be removed in 4730 // the future if the appropriate cleanup code can be added to the 4731 // VM_Exit VMOperation's doit method. 4732 if (atexit(perfMemory_exit_helper) != 0) { 4733 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4734 } 4735 } 4736 4737 // initialize thread priority policy 4738 prio_init(); 4739 4740 return JNI_OK; 4741} 4742 4743// Mark the polling page as unreadable 4744void os::make_polling_page_unreadable(void) { 4745 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 4746 fatal("Could not disable polling page"); 4747 } 4748} 4749 4750// Mark the polling page as readable 4751void os::make_polling_page_readable(void) { 4752 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4753 fatal("Could not enable polling page"); 4754 } 4755} 4756 4757int os::active_processor_count() { 4758 // Linux doesn't yet have a (official) notion of processor sets, 4759 // so just return the number of online processors. 4760 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4761 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4762 return online_cpus; 4763} 4764 4765void os::set_native_thread_name(const char *name) { 4766 if (Linux::_pthread_setname_np) { 4767 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 4768 snprintf(buf, sizeof(buf), "%s", name); 4769 buf[sizeof(buf) - 1] = '\0'; 4770 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 4771 // ERANGE should not happen; all other errors should just be ignored. 4772 assert(rc != ERANGE, "pthread_setname_np failed"); 4773 } 4774} 4775 4776bool os::distribute_processes(uint length, uint* distribution) { 4777 // Not yet implemented. 4778 return false; 4779} 4780 4781bool os::bind_to_processor(uint processor_id) { 4782 // Not yet implemented. 4783 return false; 4784} 4785 4786/// 4787 4788void os::SuspendedThreadTask::internal_do_task() { 4789 if (do_suspend(_thread->osthread())) { 4790 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4791 do_task(context); 4792 do_resume(_thread->osthread()); 4793 } 4794} 4795 4796class PcFetcher : public os::SuspendedThreadTask { 4797 public: 4798 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4799 ExtendedPC result(); 4800 protected: 4801 void do_task(const os::SuspendedThreadTaskContext& context); 4802 private: 4803 ExtendedPC _epc; 4804}; 4805 4806ExtendedPC PcFetcher::result() { 4807 guarantee(is_done(), "task is not done yet."); 4808 return _epc; 4809} 4810 4811void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4812 Thread* thread = context.thread(); 4813 OSThread* osthread = thread->osthread(); 4814 if (osthread->ucontext() != NULL) { 4815 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4816 } else { 4817 // NULL context is unexpected, double-check this is the VMThread 4818 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4819 } 4820} 4821 4822// Suspends the target using the signal mechanism and then grabs the PC before 4823// resuming the target. Used by the flat-profiler only 4824ExtendedPC os::get_thread_pc(Thread* thread) { 4825 // Make sure that it is called by the watcher for the VMThread 4826 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4827 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4828 4829 PcFetcher fetcher(thread); 4830 fetcher.run(); 4831 return fetcher.result(); 4832} 4833 4834//////////////////////////////////////////////////////////////////////////////// 4835// debug support 4836 4837bool os::find(address addr, outputStream* st) { 4838 Dl_info dlinfo; 4839 memset(&dlinfo, 0, sizeof(dlinfo)); 4840 if (dladdr(addr, &dlinfo) != 0) { 4841 st->print(PTR_FORMAT ": ", addr); 4842 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 4843 st->print("%s+%#x", dlinfo.dli_sname, 4844 addr - (intptr_t)dlinfo.dli_saddr); 4845 } else if (dlinfo.dli_fbase != NULL) { 4846 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 4847 } else { 4848 st->print("<absolute address>"); 4849 } 4850 if (dlinfo.dli_fname != NULL) { 4851 st->print(" in %s", dlinfo.dli_fname); 4852 } 4853 if (dlinfo.dli_fbase != NULL) { 4854 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 4855 } 4856 st->cr(); 4857 4858 if (Verbose) { 4859 // decode some bytes around the PC 4860 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 4861 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 4862 address lowest = (address) dlinfo.dli_sname; 4863 if (!lowest) lowest = (address) dlinfo.dli_fbase; 4864 if (begin < lowest) begin = lowest; 4865 Dl_info dlinfo2; 4866 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 4867 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 4868 end = (address) dlinfo2.dli_saddr; 4869 } 4870 Disassembler::decode(begin, end, st); 4871 } 4872 return true; 4873 } 4874 return false; 4875} 4876 4877//////////////////////////////////////////////////////////////////////////////// 4878// misc 4879 4880// This does not do anything on Linux. This is basically a hook for being 4881// able to use structured exception handling (thread-local exception filters) 4882// on, e.g., Win32. 4883void 4884os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 4885 JavaCallArguments* args, Thread* thread) { 4886 f(value, method, args, thread); 4887} 4888 4889void os::print_statistics() { 4890} 4891 4892int os::message_box(const char* title, const char* message) { 4893 int i; 4894 fdStream err(defaultStream::error_fd()); 4895 for (i = 0; i < 78; i++) err.print_raw("="); 4896 err.cr(); 4897 err.print_raw_cr(title); 4898 for (i = 0; i < 78; i++) err.print_raw("-"); 4899 err.cr(); 4900 err.print_raw_cr(message); 4901 for (i = 0; i < 78; i++) err.print_raw("="); 4902 err.cr(); 4903 4904 char buf[16]; 4905 // Prevent process from exiting upon "read error" without consuming all CPU 4906 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 4907 4908 return buf[0] == 'y' || buf[0] == 'Y'; 4909} 4910 4911int os::stat(const char *path, struct stat *sbuf) { 4912 char pathbuf[MAX_PATH]; 4913 if (strlen(path) > MAX_PATH - 1) { 4914 errno = ENAMETOOLONG; 4915 return -1; 4916 } 4917 os::native_path(strcpy(pathbuf, path)); 4918 return ::stat(pathbuf, sbuf); 4919} 4920 4921bool os::check_heap(bool force) { 4922 return true; 4923} 4924 4925// Is a (classpath) directory empty? 4926bool os::dir_is_empty(const char* path) { 4927 DIR *dir = NULL; 4928 struct dirent *ptr; 4929 4930 dir = opendir(path); 4931 if (dir == NULL) return true; 4932 4933 // Scan the directory 4934 bool result = true; 4935 char buf[sizeof(struct dirent) + MAX_PATH]; 4936 while (result && (ptr = ::readdir(dir)) != NULL) { 4937 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 4938 result = false; 4939 } 4940 } 4941 closedir(dir); 4942 return result; 4943} 4944 4945// This code originates from JDK's sysOpen and open64_w 4946// from src/solaris/hpi/src/system_md.c 4947 4948int os::open(const char *path, int oflag, int mode) { 4949 if (strlen(path) > MAX_PATH - 1) { 4950 errno = ENAMETOOLONG; 4951 return -1; 4952 } 4953 4954 // All file descriptors that are opened in the Java process and not 4955 // specifically destined for a subprocess should have the close-on-exec 4956 // flag set. If we don't set it, then careless 3rd party native code 4957 // might fork and exec without closing all appropriate file descriptors 4958 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 4959 // turn might: 4960 // 4961 // - cause end-of-file to fail to be detected on some file 4962 // descriptors, resulting in mysterious hangs, or 4963 // 4964 // - might cause an fopen in the subprocess to fail on a system 4965 // suffering from bug 1085341. 4966 // 4967 // (Yes, the default setting of the close-on-exec flag is a Unix 4968 // design flaw) 4969 // 4970 // See: 4971 // 1085341: 32-bit stdio routines should support file descriptors >255 4972 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 4973 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 4974 // 4975 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 4976 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 4977 // because it saves a system call and removes a small window where the flag 4978 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 4979 // and we fall back to using FD_CLOEXEC (see below). 4980#ifdef O_CLOEXEC 4981 oflag |= O_CLOEXEC; 4982#endif 4983 4984 int fd = ::open64(path, oflag, mode); 4985 if (fd == -1) return -1; 4986 4987 //If the open succeeded, the file might still be a directory 4988 { 4989 struct stat64 buf64; 4990 int ret = ::fstat64(fd, &buf64); 4991 int st_mode = buf64.st_mode; 4992 4993 if (ret != -1) { 4994 if ((st_mode & S_IFMT) == S_IFDIR) { 4995 errno = EISDIR; 4996 ::close(fd); 4997 return -1; 4998 } 4999 } else { 5000 ::close(fd); 5001 return -1; 5002 } 5003 } 5004 5005#ifdef FD_CLOEXEC 5006 // Validate that the use of the O_CLOEXEC flag on open above worked. 5007 // With recent kernels, we will perform this check exactly once. 5008 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5009 if (!O_CLOEXEC_is_known_to_work) { 5010 int flags = ::fcntl(fd, F_GETFD); 5011 if (flags != -1) { 5012 if ((flags & FD_CLOEXEC) != 0) 5013 O_CLOEXEC_is_known_to_work = 1; 5014 else 5015 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5016 } 5017 } 5018#endif 5019 5020 return fd; 5021} 5022 5023 5024// create binary file, rewriting existing file if required 5025int os::create_binary_file(const char* path, bool rewrite_existing) { 5026 int oflags = O_WRONLY | O_CREAT; 5027 if (!rewrite_existing) { 5028 oflags |= O_EXCL; 5029 } 5030 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5031} 5032 5033// return current position of file pointer 5034jlong os::current_file_offset(int fd) { 5035 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5036} 5037 5038// move file pointer to the specified offset 5039jlong os::seek_to_file_offset(int fd, jlong offset) { 5040 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5041} 5042 5043// This code originates from JDK's sysAvailable 5044// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5045 5046int os::available(int fd, jlong *bytes) { 5047 jlong cur, end; 5048 int mode; 5049 struct stat64 buf64; 5050 5051 if (::fstat64(fd, &buf64) >= 0) { 5052 mode = buf64.st_mode; 5053 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5054 // XXX: is the following call interruptible? If so, this might 5055 // need to go through the INTERRUPT_IO() wrapper as for other 5056 // blocking, interruptible calls in this file. 5057 int n; 5058 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5059 *bytes = n; 5060 return 1; 5061 } 5062 } 5063 } 5064 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5065 return 0; 5066 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5067 return 0; 5068 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5069 return 0; 5070 } 5071 *bytes = end - cur; 5072 return 1; 5073} 5074 5075// Map a block of memory. 5076char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5077 char *addr, size_t bytes, bool read_only, 5078 bool allow_exec) { 5079 int prot; 5080 int flags = MAP_PRIVATE; 5081 5082 if (read_only) { 5083 prot = PROT_READ; 5084 } else { 5085 prot = PROT_READ | PROT_WRITE; 5086 } 5087 5088 if (allow_exec) { 5089 prot |= PROT_EXEC; 5090 } 5091 5092 if (addr != NULL) { 5093 flags |= MAP_FIXED; 5094 } 5095 5096 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5097 fd, file_offset); 5098 if (mapped_address == MAP_FAILED) { 5099 return NULL; 5100 } 5101 return mapped_address; 5102} 5103 5104 5105// Remap a block of memory. 5106char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5107 char *addr, size_t bytes, bool read_only, 5108 bool allow_exec) { 5109 // same as map_memory() on this OS 5110 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5111 allow_exec); 5112} 5113 5114 5115// Unmap a block of memory. 5116bool os::pd_unmap_memory(char* addr, size_t bytes) { 5117 return munmap(addr, bytes) == 0; 5118} 5119 5120static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5121 5122static clockid_t thread_cpu_clockid(Thread* thread) { 5123 pthread_t tid = thread->osthread()->pthread_id(); 5124 clockid_t clockid; 5125 5126 // Get thread clockid 5127 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5128 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5129 return clockid; 5130} 5131 5132// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5133// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5134// of a thread. 5135// 5136// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5137// the fast estimate available on the platform. 5138 5139jlong os::current_thread_cpu_time() { 5140 if (os::Linux::supports_fast_thread_cpu_time()) { 5141 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5142 } else { 5143 // return user + sys since the cost is the same 5144 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5145 } 5146} 5147 5148jlong os::thread_cpu_time(Thread* thread) { 5149 // consistent with what current_thread_cpu_time() returns 5150 if (os::Linux::supports_fast_thread_cpu_time()) { 5151 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5152 } else { 5153 return slow_thread_cpu_time(thread, true /* user + sys */); 5154 } 5155} 5156 5157jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5158 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5159 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5160 } else { 5161 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5162 } 5163} 5164 5165jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5166 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5167 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5168 } else { 5169 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5170 } 5171} 5172 5173// -1 on error. 5174static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5175 pid_t tid = thread->osthread()->thread_id(); 5176 char *s; 5177 char stat[2048]; 5178 int statlen; 5179 char proc_name[64]; 5180 int count; 5181 long sys_time, user_time; 5182 char cdummy; 5183 int idummy; 5184 long ldummy; 5185 FILE *fp; 5186 5187 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5188 fp = fopen(proc_name, "r"); 5189 if (fp == NULL) return -1; 5190 statlen = fread(stat, 1, 2047, fp); 5191 stat[statlen] = '\0'; 5192 fclose(fp); 5193 5194 // Skip pid and the command string. Note that we could be dealing with 5195 // weird command names, e.g. user could decide to rename java launcher 5196 // to "java 1.4.2 :)", then the stat file would look like 5197 // 1234 (java 1.4.2 :)) R ... ... 5198 // We don't really need to know the command string, just find the last 5199 // occurrence of ")" and then start parsing from there. See bug 4726580. 5200 s = strrchr(stat, ')'); 5201 if (s == NULL) return -1; 5202 5203 // Skip blank chars 5204 do { s++; } while (s && isspace(*s)); 5205 5206 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5207 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5208 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5209 &user_time, &sys_time); 5210 if (count != 13) return -1; 5211 if (user_sys_cpu_time) { 5212 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5213 } else { 5214 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5215 } 5216} 5217 5218void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5219 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5220 info_ptr->may_skip_backward = false; // elapsed time not wall time 5221 info_ptr->may_skip_forward = false; // elapsed time not wall time 5222 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5223} 5224 5225void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5226 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5227 info_ptr->may_skip_backward = false; // elapsed time not wall time 5228 info_ptr->may_skip_forward = false; // elapsed time not wall time 5229 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5230} 5231 5232bool os::is_thread_cpu_time_supported() { 5233 return true; 5234} 5235 5236// System loadavg support. Returns -1 if load average cannot be obtained. 5237// Linux doesn't yet have a (official) notion of processor sets, 5238// so just return the system wide load average. 5239int os::loadavg(double loadavg[], int nelem) { 5240 return ::getloadavg(loadavg, nelem); 5241} 5242 5243void os::pause() { 5244 char filename[MAX_PATH]; 5245 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5246 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5247 } else { 5248 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5249 } 5250 5251 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5252 if (fd != -1) { 5253 struct stat buf; 5254 ::close(fd); 5255 while (::stat(filename, &buf) == 0) { 5256 (void)::poll(NULL, 0, 100); 5257 } 5258 } else { 5259 jio_fprintf(stderr, 5260 "Could not open pause file '%s', continuing immediately.\n", filename); 5261 } 5262} 5263 5264 5265// Refer to the comments in os_solaris.cpp park-unpark. The next two 5266// comment paragraphs are worth repeating here: 5267// 5268// Assumption: 5269// Only one parker can exist on an event, which is why we allocate 5270// them per-thread. Multiple unparkers can coexist. 5271// 5272// _Event serves as a restricted-range semaphore. 5273// -1 : thread is blocked, i.e. there is a waiter 5274// 0 : neutral: thread is running or ready, 5275// could have been signaled after a wait started 5276// 1 : signaled - thread is running or ready 5277// 5278// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5279// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5280// For specifics regarding the bug see GLIBC BUGID 261237 : 5281// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5282// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5283// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5284// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5285// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5286// and monitorenter when we're using 1-0 locking. All those operations may result in 5287// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5288// of libpthread avoids the problem, but isn't practical. 5289// 5290// Possible remedies: 5291// 5292// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5293// This is palliative and probabilistic, however. If the thread is preempted 5294// between the call to compute_abstime() and pthread_cond_timedwait(), more 5295// than the minimum period may have passed, and the abstime may be stale (in the 5296// past) resultin in a hang. Using this technique reduces the odds of a hang 5297// but the JVM is still vulnerable, particularly on heavily loaded systems. 5298// 5299// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5300// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5301// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5302// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5303// thread. 5304// 5305// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5306// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5307// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5308// This also works well. In fact it avoids kernel-level scalability impediments 5309// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5310// timers in a graceful fashion. 5311// 5312// 4. When the abstime value is in the past it appears that control returns 5313// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5314// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5315// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5316// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5317// It may be possible to avoid reinitialization by checking the return 5318// value from pthread_cond_timedwait(). In addition to reinitializing the 5319// condvar we must establish the invariant that cond_signal() is only called 5320// within critical sections protected by the adjunct mutex. This prevents 5321// cond_signal() from "seeing" a condvar that's in the midst of being 5322// reinitialized or that is corrupt. Sadly, this invariant obviates the 5323// desirable signal-after-unlock optimization that avoids futile context switching. 5324// 5325// I'm also concerned that some versions of NTPL might allocate an auxilliary 5326// structure when a condvar is used or initialized. cond_destroy() would 5327// release the helper structure. Our reinitialize-after-timedwait fix 5328// put excessive stress on malloc/free and locks protecting the c-heap. 5329// 5330// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5331// It may be possible to refine (4) by checking the kernel and NTPL verisons 5332// and only enabling the work-around for vulnerable environments. 5333 5334// utility to compute the abstime argument to timedwait: 5335// millis is the relative timeout time 5336// abstime will be the absolute timeout time 5337// TODO: replace compute_abstime() with unpackTime() 5338 5339static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5340 if (millis < 0) millis = 0; 5341 5342 jlong seconds = millis / 1000; 5343 millis %= 1000; 5344 if (seconds > 50000000) { // see man cond_timedwait(3T) 5345 seconds = 50000000; 5346 } 5347 5348 if (os::supports_monotonic_clock()) { 5349 struct timespec now; 5350 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5351 assert_status(status == 0, status, "clock_gettime"); 5352 abstime->tv_sec = now.tv_sec + seconds; 5353 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5354 if (nanos >= NANOSECS_PER_SEC) { 5355 abstime->tv_sec += 1; 5356 nanos -= NANOSECS_PER_SEC; 5357 } 5358 abstime->tv_nsec = nanos; 5359 } else { 5360 struct timeval now; 5361 int status = gettimeofday(&now, NULL); 5362 assert(status == 0, "gettimeofday"); 5363 abstime->tv_sec = now.tv_sec + seconds; 5364 long usec = now.tv_usec + millis * 1000; 5365 if (usec >= 1000000) { 5366 abstime->tv_sec += 1; 5367 usec -= 1000000; 5368 } 5369 abstime->tv_nsec = usec * 1000; 5370 } 5371 return abstime; 5372} 5373 5374void os::PlatformEvent::park() { // AKA "down()" 5375 // Transitions for _Event: 5376 // -1 => -1 : illegal 5377 // 1 => 0 : pass - return immediately 5378 // 0 => -1 : block; then set _Event to 0 before returning 5379 5380 // Invariant: Only the thread associated with the Event/PlatformEvent 5381 // may call park(). 5382 // TODO: assert that _Assoc != NULL or _Assoc == Self 5383 assert(_nParked == 0, "invariant"); 5384 5385 int v; 5386 for (;;) { 5387 v = _Event; 5388 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5389 } 5390 guarantee(v >= 0, "invariant"); 5391 if (v == 0) { 5392 // Do this the hard way by blocking ... 5393 int status = pthread_mutex_lock(_mutex); 5394 assert_status(status == 0, status, "mutex_lock"); 5395 guarantee(_nParked == 0, "invariant"); 5396 ++_nParked; 5397 while (_Event < 0) { 5398 status = pthread_cond_wait(_cond, _mutex); 5399 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5400 // Treat this the same as if the wait was interrupted 5401 if (status == ETIME) { status = EINTR; } 5402 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5403 } 5404 --_nParked; 5405 5406 _Event = 0; 5407 status = pthread_mutex_unlock(_mutex); 5408 assert_status(status == 0, status, "mutex_unlock"); 5409 // Paranoia to ensure our locked and lock-free paths interact 5410 // correctly with each other. 5411 OrderAccess::fence(); 5412 } 5413 guarantee(_Event >= 0, "invariant"); 5414} 5415 5416int os::PlatformEvent::park(jlong millis) { 5417 // Transitions for _Event: 5418 // -1 => -1 : illegal 5419 // 1 => 0 : pass - return immediately 5420 // 0 => -1 : block; then set _Event to 0 before returning 5421 5422 guarantee(_nParked == 0, "invariant"); 5423 5424 int v; 5425 for (;;) { 5426 v = _Event; 5427 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5428 } 5429 guarantee(v >= 0, "invariant"); 5430 if (v != 0) return OS_OK; 5431 5432 // We do this the hard way, by blocking the thread. 5433 // Consider enforcing a minimum timeout value. 5434 struct timespec abst; 5435 compute_abstime(&abst, millis); 5436 5437 int ret = OS_TIMEOUT; 5438 int status = pthread_mutex_lock(_mutex); 5439 assert_status(status == 0, status, "mutex_lock"); 5440 guarantee(_nParked == 0, "invariant"); 5441 ++_nParked; 5442 5443 // Object.wait(timo) will return because of 5444 // (a) notification 5445 // (b) timeout 5446 // (c) thread.interrupt 5447 // 5448 // Thread.interrupt and object.notify{All} both call Event::set. 5449 // That is, we treat thread.interrupt as a special case of notification. 5450 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false. 5451 // We assume all ETIME returns are valid. 5452 // 5453 // TODO: properly differentiate simultaneous notify+interrupt. 5454 // In that case, we should propagate the notify to another waiter. 5455 5456 while (_Event < 0) { 5457 status = pthread_cond_timedwait(_cond, _mutex, &abst); 5458 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5459 pthread_cond_destroy(_cond); 5460 pthread_cond_init(_cond, os::Linux::condAttr()); 5461 } 5462 assert_status(status == 0 || status == EINTR || 5463 status == ETIME || status == ETIMEDOUT, 5464 status, "cond_timedwait"); 5465 if (!FilterSpuriousWakeups) break; // previous semantics 5466 if (status == ETIME || status == ETIMEDOUT) break; 5467 // We consume and ignore EINTR and spurious wakeups. 5468 } 5469 --_nParked; 5470 if (_Event >= 0) { 5471 ret = OS_OK; 5472 } 5473 _Event = 0; 5474 status = pthread_mutex_unlock(_mutex); 5475 assert_status(status == 0, status, "mutex_unlock"); 5476 assert(_nParked == 0, "invariant"); 5477 // Paranoia to ensure our locked and lock-free paths interact 5478 // correctly with each other. 5479 OrderAccess::fence(); 5480 return ret; 5481} 5482 5483void os::PlatformEvent::unpark() { 5484 // Transitions for _Event: 5485 // 0 => 1 : just return 5486 // 1 => 1 : just return 5487 // -1 => either 0 or 1; must signal target thread 5488 // That is, we can safely transition _Event from -1 to either 5489 // 0 or 1. 5490 // See also: "Semaphores in Plan 9" by Mullender & Cox 5491 // 5492 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5493 // that it will take two back-to-back park() calls for the owning 5494 // thread to block. This has the benefit of forcing a spurious return 5495 // from the first park() call after an unpark() call which will help 5496 // shake out uses of park() and unpark() without condition variables. 5497 5498 if (Atomic::xchg(1, &_Event) >= 0) return; 5499 5500 // Wait for the thread associated with the event to vacate 5501 int status = pthread_mutex_lock(_mutex); 5502 assert_status(status == 0, status, "mutex_lock"); 5503 int AnyWaiters = _nParked; 5504 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5505 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5506 AnyWaiters = 0; 5507 pthread_cond_signal(_cond); 5508 } 5509 status = pthread_mutex_unlock(_mutex); 5510 assert_status(status == 0, status, "mutex_unlock"); 5511 if (AnyWaiters != 0) { 5512 // Note that we signal() *after* dropping the lock for "immortal" Events. 5513 // This is safe and avoids a common class of futile wakeups. In rare 5514 // circumstances this can cause a thread to return prematurely from 5515 // cond_{timed}wait() but the spurious wakeup is benign and the victim 5516 // will simply re-test the condition and re-park itself. 5517 // This provides particular benefit if the underlying platform does not 5518 // provide wait morphing. 5519 status = pthread_cond_signal(_cond); 5520 assert_status(status == 0, status, "cond_signal"); 5521 } 5522} 5523 5524 5525// JSR166 5526// ------------------------------------------------------- 5527 5528// The solaris and linux implementations of park/unpark are fairly 5529// conservative for now, but can be improved. They currently use a 5530// mutex/condvar pair, plus a a count. 5531// Park decrements count if > 0, else does a condvar wait. Unpark 5532// sets count to 1 and signals condvar. Only one thread ever waits 5533// on the condvar. Contention seen when trying to park implies that someone 5534// is unparking you, so don't wait. And spurious returns are fine, so there 5535// is no need to track notifications. 5536 5537// This code is common to linux and solaris and will be moved to a 5538// common place in dolphin. 5539// 5540// The passed in time value is either a relative time in nanoseconds 5541// or an absolute time in milliseconds. Either way it has to be unpacked 5542// into suitable seconds and nanoseconds components and stored in the 5543// given timespec structure. 5544// Given time is a 64-bit value and the time_t used in the timespec is only 5545// a signed-32-bit value (except on 64-bit Linux) we have to watch for 5546// overflow if times way in the future are given. Further on Solaris versions 5547// prior to 10 there is a restriction (see cond_timedwait) that the specified 5548// number of seconds, in abstime, is less than current_time + 100,000,000. 5549// As it will be 28 years before "now + 100000000" will overflow we can 5550// ignore overflow and just impose a hard-limit on seconds using the value 5551// of "now + 100,000,000". This places a limit on the timeout of about 3.17 5552// years from "now". 5553 5554static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5555 assert(time > 0, "convertTime"); 5556 time_t max_secs = 0; 5557 5558 if (!os::supports_monotonic_clock() || isAbsolute) { 5559 struct timeval now; 5560 int status = gettimeofday(&now, NULL); 5561 assert(status == 0, "gettimeofday"); 5562 5563 max_secs = now.tv_sec + MAX_SECS; 5564 5565 if (isAbsolute) { 5566 jlong secs = time / 1000; 5567 if (secs > max_secs) { 5568 absTime->tv_sec = max_secs; 5569 } else { 5570 absTime->tv_sec = secs; 5571 } 5572 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5573 } else { 5574 jlong secs = time / NANOSECS_PER_SEC; 5575 if (secs >= MAX_SECS) { 5576 absTime->tv_sec = max_secs; 5577 absTime->tv_nsec = 0; 5578 } else { 5579 absTime->tv_sec = now.tv_sec + secs; 5580 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5581 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5582 absTime->tv_nsec -= NANOSECS_PER_SEC; 5583 ++absTime->tv_sec; // note: this must be <= max_secs 5584 } 5585 } 5586 } 5587 } else { 5588 // must be relative using monotonic clock 5589 struct timespec now; 5590 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5591 assert_status(status == 0, status, "clock_gettime"); 5592 max_secs = now.tv_sec + MAX_SECS; 5593 jlong secs = time / NANOSECS_PER_SEC; 5594 if (secs >= MAX_SECS) { 5595 absTime->tv_sec = max_secs; 5596 absTime->tv_nsec = 0; 5597 } else { 5598 absTime->tv_sec = now.tv_sec + secs; 5599 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5600 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5601 absTime->tv_nsec -= NANOSECS_PER_SEC; 5602 ++absTime->tv_sec; // note: this must be <= max_secs 5603 } 5604 } 5605 } 5606 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5607 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5608 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5609 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5610} 5611 5612void Parker::park(bool isAbsolute, jlong time) { 5613 // Ideally we'd do something useful while spinning, such 5614 // as calling unpackTime(). 5615 5616 // Optional fast-path check: 5617 // Return immediately if a permit is available. 5618 // We depend on Atomic::xchg() having full barrier semantics 5619 // since we are doing a lock-free update to _counter. 5620 if (Atomic::xchg(0, &_counter) > 0) return; 5621 5622 Thread* thread = Thread::current(); 5623 assert(thread->is_Java_thread(), "Must be JavaThread"); 5624 JavaThread *jt = (JavaThread *)thread; 5625 5626 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5627 // Check interrupt before trying to wait 5628 if (Thread::is_interrupted(thread, false)) { 5629 return; 5630 } 5631 5632 // Next, demultiplex/decode time arguments 5633 timespec absTime; 5634 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all 5635 return; 5636 } 5637 if (time > 0) { 5638 unpackTime(&absTime, isAbsolute, time); 5639 } 5640 5641 5642 // Enter safepoint region 5643 // Beware of deadlocks such as 6317397. 5644 // The per-thread Parker:: mutex is a classic leaf-lock. 5645 // In particular a thread must never block on the Threads_lock while 5646 // holding the Parker:: mutex. If safepoints are pending both the 5647 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5648 ThreadBlockInVM tbivm(jt); 5649 5650 // Don't wait if cannot get lock since interference arises from 5651 // unblocking. Also. check interrupt before trying wait 5652 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5653 return; 5654 } 5655 5656 int status; 5657 if (_counter > 0) { // no wait needed 5658 _counter = 0; 5659 status = pthread_mutex_unlock(_mutex); 5660 assert(status == 0, "invariant"); 5661 // Paranoia to ensure our locked and lock-free paths interact 5662 // correctly with each other and Java-level accesses. 5663 OrderAccess::fence(); 5664 return; 5665 } 5666 5667#ifdef ASSERT 5668 // Don't catch signals while blocked; let the running threads have the signals. 5669 // (This allows a debugger to break into the running thread.) 5670 sigset_t oldsigs; 5671 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5672 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5673#endif 5674 5675 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5676 jt->set_suspend_equivalent(); 5677 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5678 5679 assert(_cur_index == -1, "invariant"); 5680 if (time == 0) { 5681 _cur_index = REL_INDEX; // arbitrary choice when not timed 5682 status = pthread_cond_wait(&_cond[_cur_index], _mutex); 5683 } else { 5684 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5685 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime); 5686 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5687 pthread_cond_destroy(&_cond[_cur_index]); 5688 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr()); 5689 } 5690 } 5691 _cur_index = -1; 5692 assert_status(status == 0 || status == EINTR || 5693 status == ETIME || status == ETIMEDOUT, 5694 status, "cond_timedwait"); 5695 5696#ifdef ASSERT 5697 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5698#endif 5699 5700 _counter = 0; 5701 status = pthread_mutex_unlock(_mutex); 5702 assert_status(status == 0, status, "invariant"); 5703 // Paranoia to ensure our locked and lock-free paths interact 5704 // correctly with each other and Java-level accesses. 5705 OrderAccess::fence(); 5706 5707 // If externally suspended while waiting, re-suspend 5708 if (jt->handle_special_suspend_equivalent_condition()) { 5709 jt->java_suspend_self(); 5710 } 5711} 5712 5713void Parker::unpark() { 5714 int status = pthread_mutex_lock(_mutex); 5715 assert(status == 0, "invariant"); 5716 const int s = _counter; 5717 _counter = 1; 5718 if (s < 1) { 5719 // thread might be parked 5720 if (_cur_index != -1) { 5721 // thread is definitely parked 5722 if (WorkAroundNPTLTimedWaitHang) { 5723 status = pthread_cond_signal(&_cond[_cur_index]); 5724 assert(status == 0, "invariant"); 5725 status = pthread_mutex_unlock(_mutex); 5726 assert(status == 0, "invariant"); 5727 } else { 5728 status = pthread_mutex_unlock(_mutex); 5729 assert(status == 0, "invariant"); 5730 status = pthread_cond_signal(&_cond[_cur_index]); 5731 assert(status == 0, "invariant"); 5732 } 5733 } else { 5734 pthread_mutex_unlock(_mutex); 5735 assert(status == 0, "invariant"); 5736 } 5737 } else { 5738 pthread_mutex_unlock(_mutex); 5739 assert(status == 0, "invariant"); 5740 } 5741} 5742 5743 5744extern char** environ; 5745 5746// Run the specified command in a separate process. Return its exit value, 5747// or -1 on failure (e.g. can't fork a new process). 5748// Unlike system(), this function can be called from signal handler. It 5749// doesn't block SIGINT et al. 5750int os::fork_and_exec(char* cmd) { 5751 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5752 5753 pid_t pid = fork(); 5754 5755 if (pid < 0) { 5756 // fork failed 5757 return -1; 5758 5759 } else if (pid == 0) { 5760 // child process 5761 5762 execve("/bin/sh", (char* const*)argv, environ); 5763 5764 // execve failed 5765 _exit(-1); 5766 5767 } else { 5768 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5769 // care about the actual exit code, for now. 5770 5771 int status; 5772 5773 // Wait for the child process to exit. This returns immediately if 5774 // the child has already exited. */ 5775 while (waitpid(pid, &status, 0) < 0) { 5776 switch (errno) { 5777 case ECHILD: return 0; 5778 case EINTR: break; 5779 default: return -1; 5780 } 5781 } 5782 5783 if (WIFEXITED(status)) { 5784 // The child exited normally; get its exit code. 5785 return WEXITSTATUS(status); 5786 } else if (WIFSIGNALED(status)) { 5787 // The child exited because of a signal 5788 // The best value to return is 0x80 + signal number, 5789 // because that is what all Unix shells do, and because 5790 // it allows callers to distinguish between process exit and 5791 // process death by signal. 5792 return 0x80 + WTERMSIG(status); 5793 } else { 5794 // Unknown exit code; pass it through 5795 return status; 5796 } 5797 } 5798} 5799 5800// is_headless_jre() 5801// 5802// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5803// in order to report if we are running in a headless jre 5804// 5805// Since JDK8 xawt/libmawt.so was moved into the same directory 5806// as libawt.so, and renamed libawt_xawt.so 5807// 5808bool os::is_headless_jre() { 5809 struct stat statbuf; 5810 char buf[MAXPATHLEN]; 5811 char libmawtpath[MAXPATHLEN]; 5812 const char *xawtstr = "/xawt/libmawt.so"; 5813 const char *new_xawtstr = "/libawt_xawt.so"; 5814 char *p; 5815 5816 // Get path to libjvm.so 5817 os::jvm_path(buf, sizeof(buf)); 5818 5819 // Get rid of libjvm.so 5820 p = strrchr(buf, '/'); 5821 if (p == NULL) { 5822 return false; 5823 } else { 5824 *p = '\0'; 5825 } 5826 5827 // Get rid of client or server 5828 p = strrchr(buf, '/'); 5829 if (p == NULL) { 5830 return false; 5831 } else { 5832 *p = '\0'; 5833 } 5834 5835 // check xawt/libmawt.so 5836 strcpy(libmawtpath, buf); 5837 strcat(libmawtpath, xawtstr); 5838 if (::stat(libmawtpath, &statbuf) == 0) return false; 5839 5840 // check libawt_xawt.so 5841 strcpy(libmawtpath, buf); 5842 strcat(libmawtpath, new_xawtstr); 5843 if (::stat(libmawtpath, &statbuf) == 0) return false; 5844 5845 return true; 5846} 5847 5848// Get the default path to the core file 5849// Returns the length of the string 5850int os::get_core_path(char* buffer, size_t bufferSize) { 5851 /* 5852 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 5853 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 5854 */ 5855 const int core_pattern_len = 129; 5856 char core_pattern[core_pattern_len] = {0}; 5857 5858 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 5859 if (core_pattern_file != -1) { 5860 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 5861 ::close(core_pattern_file); 5862 5863 if (ret > 0) { 5864 char *last_char = core_pattern + strlen(core_pattern) - 1; 5865 5866 if (*last_char == '\n') { 5867 *last_char = '\0'; 5868 } 5869 } 5870 } 5871 5872 if (strlen(core_pattern) == 0) { 5873 return -1; 5874 } 5875 5876 char *pid_pos = strstr(core_pattern, "%p"); 5877 int written; 5878 5879 if (core_pattern[0] == '/') { 5880 written = jio_snprintf(buffer, bufferSize, core_pattern); 5881 } else { 5882 char cwd[PATH_MAX]; 5883 5884 const char* p = get_current_directory(cwd, PATH_MAX); 5885 if (p == NULL) { 5886 return -1; 5887 } 5888 5889 if (core_pattern[0] == '|') { 5890 written = jio_snprintf(buffer, bufferSize, 5891 "\"%s\" (or dumping to %s/core.%d)", 5892 &core_pattern[1], p, current_process_id()); 5893 } else { 5894 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 5895 } 5896 } 5897 5898 if (written < 0) { 5899 return -1; 5900 } 5901 5902 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 5903 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 5904 5905 if (core_uses_pid_file != -1) { 5906 char core_uses_pid = 0; 5907 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 5908 ::close(core_uses_pid_file); 5909 5910 if (core_uses_pid == '1') { 5911 jio_snprintf(buffer + written, bufferSize - written, 5912 ".%d", current_process_id()); 5913 } 5914 } 5915 } 5916 5917 return strlen(buffer); 5918} 5919 5920/////////////// Unit tests /////////////// 5921 5922#ifndef PRODUCT 5923 5924#define test_log(...) \ 5925 do { \ 5926 if (VerboseInternalVMTests) { \ 5927 tty->print_cr(__VA_ARGS__); \ 5928 tty->flush(); \ 5929 } \ 5930 } while (false) 5931 5932class TestReserveMemorySpecial : AllStatic { 5933 public: 5934 static void small_page_write(void* addr, size_t size) { 5935 size_t page_size = os::vm_page_size(); 5936 5937 char* end = (char*)addr + size; 5938 for (char* p = (char*)addr; p < end; p += page_size) { 5939 *p = 1; 5940 } 5941 } 5942 5943 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 5944 if (!UseHugeTLBFS) { 5945 return; 5946 } 5947 5948 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 5949 5950 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 5951 5952 if (addr != NULL) { 5953 small_page_write(addr, size); 5954 5955 os::Linux::release_memory_special_huge_tlbfs(addr, size); 5956 } 5957 } 5958 5959 static void test_reserve_memory_special_huge_tlbfs_only() { 5960 if (!UseHugeTLBFS) { 5961 return; 5962 } 5963 5964 size_t lp = os::large_page_size(); 5965 5966 for (size_t size = lp; size <= lp * 10; size += lp) { 5967 test_reserve_memory_special_huge_tlbfs_only(size); 5968 } 5969 } 5970 5971 static void test_reserve_memory_special_huge_tlbfs_mixed() { 5972 size_t lp = os::large_page_size(); 5973 size_t ag = os::vm_allocation_granularity(); 5974 5975 // sizes to test 5976 const size_t sizes[] = { 5977 lp, lp + ag, lp + lp / 2, lp * 2, 5978 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2, 5979 lp * 10, lp * 10 + lp / 2 5980 }; 5981 const int num_sizes = sizeof(sizes) / sizeof(size_t); 5982 5983 // For each size/alignment combination, we test three scenarios: 5984 // 1) with req_addr == NULL 5985 // 2) with a non-null req_addr at which we expect to successfully allocate 5986 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we 5987 // expect the allocation to either fail or to ignore req_addr 5988 5989 // Pre-allocate two areas; they shall be as large as the largest allocation 5990 // and aligned to the largest alignment we will be testing. 5991 const size_t mapping_size = sizes[num_sizes - 1] * 2; 5992 char* const mapping1 = (char*) ::mmap(NULL, mapping_size, 5993 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 5994 -1, 0); 5995 assert(mapping1 != MAP_FAILED, "should work"); 5996 5997 char* const mapping2 = (char*) ::mmap(NULL, mapping_size, 5998 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 5999 -1, 0); 6000 assert(mapping2 != MAP_FAILED, "should work"); 6001 6002 // Unmap the first mapping, but leave the second mapping intact: the first 6003 // mapping will serve as a value for a "good" req_addr (case 2). The second 6004 // mapping, still intact, as "bad" req_addr (case 3). 6005 ::munmap(mapping1, mapping_size); 6006 6007 // Case 1 6008 test_log("%s, req_addr NULL:", __FUNCTION__); 6009 test_log("size align result"); 6010 6011 for (int i = 0; i < num_sizes; i++) { 6012 const size_t size = sizes[i]; 6013 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6014 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6015 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s", 6016 size, alignment, p, (p != NULL ? "" : "(failed)")); 6017 if (p != NULL) { 6018 assert(is_ptr_aligned(p, alignment), "must be"); 6019 small_page_write(p, size); 6020 os::Linux::release_memory_special_huge_tlbfs(p, size); 6021 } 6022 } 6023 } 6024 6025 // Case 2 6026 test_log("%s, req_addr non-NULL:", __FUNCTION__); 6027 test_log("size align req_addr result"); 6028 6029 for (int i = 0; i < num_sizes; i++) { 6030 const size_t size = sizes[i]; 6031 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6032 char* const req_addr = (char*) align_ptr_up(mapping1, alignment); 6033 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6034 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6035 size, alignment, req_addr, p, 6036 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)"))); 6037 if (p != NULL) { 6038 assert(p == req_addr, "must be"); 6039 small_page_write(p, size); 6040 os::Linux::release_memory_special_huge_tlbfs(p, size); 6041 } 6042 } 6043 } 6044 6045 // Case 3 6046 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__); 6047 test_log("size align req_addr result"); 6048 6049 for (int i = 0; i < num_sizes; i++) { 6050 const size_t size = sizes[i]; 6051 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6052 char* const req_addr = (char*) align_ptr_up(mapping2, alignment); 6053 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6054 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6055 size, alignment, req_addr, p, 6056 ((p != NULL ? "" : "(failed)"))); 6057 // as the area around req_addr contains already existing mappings, the API should always 6058 // return NULL (as per contract, it cannot return another address) 6059 assert(p == NULL, "must be"); 6060 } 6061 } 6062 6063 ::munmap(mapping2, mapping_size); 6064 6065 } 6066 6067 static void test_reserve_memory_special_huge_tlbfs() { 6068 if (!UseHugeTLBFS) { 6069 return; 6070 } 6071 6072 test_reserve_memory_special_huge_tlbfs_only(); 6073 test_reserve_memory_special_huge_tlbfs_mixed(); 6074 } 6075 6076 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6077 if (!UseSHM) { 6078 return; 6079 } 6080 6081 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6082 6083 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6084 6085 if (addr != NULL) { 6086 assert(is_ptr_aligned(addr, alignment), "Check"); 6087 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6088 6089 small_page_write(addr, size); 6090 6091 os::Linux::release_memory_special_shm(addr, size); 6092 } 6093 } 6094 6095 static void test_reserve_memory_special_shm() { 6096 size_t lp = os::large_page_size(); 6097 size_t ag = os::vm_allocation_granularity(); 6098 6099 for (size_t size = ag; size < lp * 3; size += ag) { 6100 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6101 test_reserve_memory_special_shm(size, alignment); 6102 } 6103 } 6104 } 6105 6106 static void test() { 6107 test_reserve_memory_special_huge_tlbfs(); 6108 test_reserve_memory_special_shm(); 6109 } 6110}; 6111 6112void TestReserveMemorySpecial_test() { 6113 TestReserveMemorySpecial::test(); 6114} 6115 6116#endif 6117