os_linux.cpp revision 8555:c30414cbbd88
1/* 2 * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25// no precompiled headers 26#include "classfile/classLoader.hpp" 27#include "classfile/systemDictionary.hpp" 28#include "classfile/vmSymbols.hpp" 29#include "code/icBuffer.hpp" 30#include "code/vtableStubs.hpp" 31#include "compiler/compileBroker.hpp" 32#include "compiler/disassembler.hpp" 33#include "interpreter/interpreter.hpp" 34#include "jvm_linux.h" 35#include "memory/allocation.inline.hpp" 36#include "memory/filemap.hpp" 37#include "mutex_linux.inline.hpp" 38#include "oops/oop.inline.hpp" 39#include "os_linux.inline.hpp" 40#include "os_share_linux.hpp" 41#include "prims/jniFastGetField.hpp" 42#include "prims/jvm.h" 43#include "prims/jvm_misc.hpp" 44#include "runtime/arguments.hpp" 45#include "runtime/atomic.inline.hpp" 46#include "runtime/extendedPC.hpp" 47#include "runtime/globals.hpp" 48#include "runtime/interfaceSupport.hpp" 49#include "runtime/init.hpp" 50#include "runtime/java.hpp" 51#include "runtime/javaCalls.hpp" 52#include "runtime/mutexLocker.hpp" 53#include "runtime/objectMonitor.hpp" 54#include "runtime/orderAccess.inline.hpp" 55#include "runtime/osThread.hpp" 56#include "runtime/perfMemory.hpp" 57#include "runtime/sharedRuntime.hpp" 58#include "runtime/statSampler.hpp" 59#include "runtime/stubRoutines.hpp" 60#include "runtime/thread.inline.hpp" 61#include "runtime/threadCritical.hpp" 62#include "runtime/timer.hpp" 63#include "services/attachListener.hpp" 64#include "services/memTracker.hpp" 65#include "services/runtimeService.hpp" 66#include "utilities/decoder.hpp" 67#include "utilities/defaultStream.hpp" 68#include "utilities/events.hpp" 69#include "utilities/elfFile.hpp" 70#include "utilities/growableArray.hpp" 71#include "utilities/macros.hpp" 72#include "utilities/vmError.hpp" 73 74// put OS-includes here 75# include <sys/types.h> 76# include <sys/mman.h> 77# include <sys/stat.h> 78# include <sys/select.h> 79# include <pthread.h> 80# include <signal.h> 81# include <errno.h> 82# include <dlfcn.h> 83# include <stdio.h> 84# include <unistd.h> 85# include <sys/resource.h> 86# include <pthread.h> 87# include <sys/stat.h> 88# include <sys/time.h> 89# include <sys/times.h> 90# include <sys/utsname.h> 91# include <sys/socket.h> 92# include <sys/wait.h> 93# include <pwd.h> 94# include <poll.h> 95# include <semaphore.h> 96# include <fcntl.h> 97# include <string.h> 98# include <syscall.h> 99# include <sys/sysinfo.h> 100# include <gnu/libc-version.h> 101# include <sys/ipc.h> 102# include <sys/shm.h> 103# include <link.h> 104# include <stdint.h> 105# include <inttypes.h> 106# include <sys/ioctl.h> 107 108PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC 109 110// if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling 111// getrusage() is prepared to handle the associated failure. 112#ifndef RUSAGE_THREAD 113 #define RUSAGE_THREAD (1) /* only the calling thread */ 114#endif 115 116#define MAX_PATH (2 * K) 117 118#define MAX_SECS 100000000 119 120// for timer info max values which include all bits 121#define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF) 122 123#define LARGEPAGES_BIT (1 << 6) 124//////////////////////////////////////////////////////////////////////////////// 125// global variables 126julong os::Linux::_physical_memory = 0; 127 128address os::Linux::_initial_thread_stack_bottom = NULL; 129uintptr_t os::Linux::_initial_thread_stack_size = 0; 130 131int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL; 132int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL; 133int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL; 134Mutex* os::Linux::_createThread_lock = NULL; 135pthread_t os::Linux::_main_thread; 136int os::Linux::_page_size = -1; 137const int os::Linux::_vm_default_page_size = (8 * K); 138bool os::Linux::_is_floating_stack = false; 139bool os::Linux::_is_NPTL = false; 140bool os::Linux::_supports_fast_thread_cpu_time = false; 141const char * os::Linux::_glibc_version = NULL; 142const char * os::Linux::_libpthread_version = NULL; 143pthread_condattr_t os::Linux::_condattr[1]; 144 145static jlong initial_time_count=0; 146 147static int clock_tics_per_sec = 100; 148 149// For diagnostics to print a message once. see run_periodic_checks 150static sigset_t check_signal_done; 151static bool check_signals = true; 152 153static pid_t _initial_pid = 0; 154 155// Signal number used to suspend/resume a thread 156 157// do not use any signal number less than SIGSEGV, see 4355769 158static int SR_signum = SIGUSR2; 159sigset_t SR_sigset; 160 161// Declarations 162static void unpackTime(timespec* absTime, bool isAbsolute, jlong time); 163 164// utility functions 165 166static int SR_initialize(); 167 168julong os::available_memory() { 169 return Linux::available_memory(); 170} 171 172julong os::Linux::available_memory() { 173 // values in struct sysinfo are "unsigned long" 174 struct sysinfo si; 175 sysinfo(&si); 176 177 return (julong)si.freeram * si.mem_unit; 178} 179 180julong os::physical_memory() { 181 return Linux::physical_memory(); 182} 183 184// Return true if user is running as root. 185 186bool os::have_special_privileges() { 187 static bool init = false; 188 static bool privileges = false; 189 if (!init) { 190 privileges = (getuid() != geteuid()) || (getgid() != getegid()); 191 init = true; 192 } 193 return privileges; 194} 195 196 197#ifndef SYS_gettid 198// i386: 224, ia64: 1105, amd64: 186, sparc 143 199 #ifdef __ia64__ 200 #define SYS_gettid 1105 201 #else 202 #ifdef __i386__ 203 #define SYS_gettid 224 204 #else 205 #ifdef __amd64__ 206 #define SYS_gettid 186 207 #else 208 #ifdef __sparc__ 209 #define SYS_gettid 143 210 #else 211 #error define gettid for the arch 212 #endif 213 #endif 214 #endif 215 #endif 216#endif 217 218// Cpu architecture string 219static char cpu_arch[] = HOTSPOT_LIB_ARCH; 220 221 222// pid_t gettid() 223// 224// Returns the kernel thread id of the currently running thread. Kernel 225// thread id is used to access /proc. 226// 227// (Note that getpid() on LinuxThreads returns kernel thread id too; but 228// on NPTL, it returns the same pid for all threads, as required by POSIX.) 229// 230pid_t os::Linux::gettid() { 231 int rslt = syscall(SYS_gettid); 232 if (rslt == -1) { 233 // old kernel, no NPTL support 234 return getpid(); 235 } else { 236 return (pid_t)rslt; 237 } 238} 239 240// Most versions of linux have a bug where the number of processors are 241// determined by looking at the /proc file system. In a chroot environment, 242// the system call returns 1. This causes the VM to act as if it is 243// a single processor and elide locking (see is_MP() call). 244static bool unsafe_chroot_detected = false; 245static const char *unstable_chroot_error = "/proc file system not found.\n" 246 "Java may be unstable running multithreaded in a chroot " 247 "environment on Linux when /proc filesystem is not mounted."; 248 249void os::Linux::initialize_system_info() { 250 set_processor_count(sysconf(_SC_NPROCESSORS_CONF)); 251 if (processor_count() == 1) { 252 pid_t pid = os::Linux::gettid(); 253 char fname[32]; 254 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid); 255 FILE *fp = fopen(fname, "r"); 256 if (fp == NULL) { 257 unsafe_chroot_detected = true; 258 } else { 259 fclose(fp); 260 } 261 } 262 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE); 263 assert(processor_count() > 0, "linux error"); 264} 265 266void os::init_system_properties_values() { 267 // The next steps are taken in the product version: 268 // 269 // Obtain the JAVA_HOME value from the location of libjvm.so. 270 // This library should be located at: 271 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so. 272 // 273 // If "/jre/lib/" appears at the right place in the path, then we 274 // assume libjvm.so is installed in a JDK and we use this path. 275 // 276 // Otherwise exit with message: "Could not create the Java virtual machine." 277 // 278 // The following extra steps are taken in the debugging version: 279 // 280 // If "/jre/lib/" does NOT appear at the right place in the path 281 // instead of exit check for $JAVA_HOME environment variable. 282 // 283 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>, 284 // then we append a fake suffix "hotspot/libjvm.so" to this path so 285 // it looks like libjvm.so is installed there 286 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so. 287 // 288 // Otherwise exit. 289 // 290 // Important note: if the location of libjvm.so changes this 291 // code needs to be changed accordingly. 292 293 // See ld(1): 294 // The linker uses the following search paths to locate required 295 // shared libraries: 296 // 1: ... 297 // ... 298 // 7: The default directories, normally /lib and /usr/lib. 299#if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390)) 300 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib" 301#else 302 #define DEFAULT_LIBPATH "/lib:/usr/lib" 303#endif 304 305// Base path of extensions installed on the system. 306#define SYS_EXT_DIR "/usr/java/packages" 307#define EXTENSIONS_DIR "/lib/ext" 308 309 // Buffer that fits several sprintfs. 310 // Note that the space for the colon and the trailing null are provided 311 // by the nulls included by the sizeof operator. 312 const size_t bufsize = 313 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends. 314 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir 315 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal); 316 317 // sysclasspath, java_home, dll_dir 318 { 319 char *pslash; 320 os::jvm_path(buf, bufsize); 321 322 // Found the full path to libjvm.so. 323 // Now cut the path to <java_home>/jre if we can. 324 pslash = strrchr(buf, '/'); 325 if (pslash != NULL) { 326 *pslash = '\0'; // Get rid of /libjvm.so. 327 } 328 pslash = strrchr(buf, '/'); 329 if (pslash != NULL) { 330 *pslash = '\0'; // Get rid of /{client|server|hotspot}. 331 } 332 Arguments::set_dll_dir(buf); 333 334 if (pslash != NULL) { 335 pslash = strrchr(buf, '/'); 336 if (pslash != NULL) { 337 *pslash = '\0'; // Get rid of /<arch>. 338 pslash = strrchr(buf, '/'); 339 if (pslash != NULL) { 340 *pslash = '\0'; // Get rid of /lib. 341 } 342 } 343 } 344 Arguments::set_java_home(buf); 345 set_boot_path('/', ':'); 346 } 347 348 // Where to look for native libraries. 349 // 350 // Note: Due to a legacy implementation, most of the library path 351 // is set in the launcher. This was to accomodate linking restrictions 352 // on legacy Linux implementations (which are no longer supported). 353 // Eventually, all the library path setting will be done here. 354 // 355 // However, to prevent the proliferation of improperly built native 356 // libraries, the new path component /usr/java/packages is added here. 357 // Eventually, all the library path setting will be done here. 358 { 359 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It 360 // should always exist (until the legacy problem cited above is 361 // addressed). 362 const char *v = ::getenv("LD_LIBRARY_PATH"); 363 const char *v_colon = ":"; 364 if (v == NULL) { v = ""; v_colon = ""; } 365 // That's +1 for the colon and +1 for the trailing '\0'. 366 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, 367 strlen(v) + 1 + 368 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1, 369 mtInternal); 370 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch); 371 Arguments::set_library_path(ld_library_path); 372 FREE_C_HEAP_ARRAY(char, ld_library_path); 373 } 374 375 // Extensions directories. 376 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home()); 377 Arguments::set_ext_dirs(buf); 378 379 FREE_C_HEAP_ARRAY(char, buf); 380 381#undef DEFAULT_LIBPATH 382#undef SYS_EXT_DIR 383#undef EXTENSIONS_DIR 384} 385 386//////////////////////////////////////////////////////////////////////////////// 387// breakpoint support 388 389void os::breakpoint() { 390 BREAKPOINT; 391} 392 393extern "C" void breakpoint() { 394 // use debugger to set breakpoint here 395} 396 397//////////////////////////////////////////////////////////////////////////////// 398// signal support 399 400debug_only(static bool signal_sets_initialized = false); 401static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs; 402 403bool os::Linux::is_sig_ignored(int sig) { 404 struct sigaction oact; 405 sigaction(sig, (struct sigaction*)NULL, &oact); 406 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction) 407 : CAST_FROM_FN_PTR(void*, oact.sa_handler); 408 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) { 409 return true; 410 } else { 411 return false; 412 } 413} 414 415void os::Linux::signal_sets_init() { 416 // Should also have an assertion stating we are still single-threaded. 417 assert(!signal_sets_initialized, "Already initialized"); 418 // Fill in signals that are necessarily unblocked for all threads in 419 // the VM. Currently, we unblock the following signals: 420 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden 421 // by -Xrs (=ReduceSignalUsage)); 422 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all 423 // other threads. The "ReduceSignalUsage" boolean tells us not to alter 424 // the dispositions or masks wrt these signals. 425 // Programs embedding the VM that want to use the above signals for their 426 // own purposes must, at this time, use the "-Xrs" option to prevent 427 // interference with shutdown hooks and BREAK_SIGNAL thread dumping. 428 // (See bug 4345157, and other related bugs). 429 // In reality, though, unblocking these signals is really a nop, since 430 // these signals are not blocked by default. 431 sigemptyset(&unblocked_sigs); 432 sigemptyset(&allowdebug_blocked_sigs); 433 sigaddset(&unblocked_sigs, SIGILL); 434 sigaddset(&unblocked_sigs, SIGSEGV); 435 sigaddset(&unblocked_sigs, SIGBUS); 436 sigaddset(&unblocked_sigs, SIGFPE); 437#if defined(PPC64) 438 sigaddset(&unblocked_sigs, SIGTRAP); 439#endif 440 sigaddset(&unblocked_sigs, SR_signum); 441 442 if (!ReduceSignalUsage) { 443 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) { 444 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL); 445 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL); 446 } 447 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) { 448 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL); 449 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL); 450 } 451 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) { 452 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL); 453 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL); 454 } 455 } 456 // Fill in signals that are blocked by all but the VM thread. 457 sigemptyset(&vm_sigs); 458 if (!ReduceSignalUsage) { 459 sigaddset(&vm_sigs, BREAK_SIGNAL); 460 } 461 debug_only(signal_sets_initialized = true); 462 463} 464 465// These are signals that are unblocked while a thread is running Java. 466// (For some reason, they get blocked by default.) 467sigset_t* os::Linux::unblocked_signals() { 468 assert(signal_sets_initialized, "Not initialized"); 469 return &unblocked_sigs; 470} 471 472// These are the signals that are blocked while a (non-VM) thread is 473// running Java. Only the VM thread handles these signals. 474sigset_t* os::Linux::vm_signals() { 475 assert(signal_sets_initialized, "Not initialized"); 476 return &vm_sigs; 477} 478 479// These are signals that are blocked during cond_wait to allow debugger in 480sigset_t* os::Linux::allowdebug_blocked_signals() { 481 assert(signal_sets_initialized, "Not initialized"); 482 return &allowdebug_blocked_sigs; 483} 484 485void os::Linux::hotspot_sigmask(Thread* thread) { 486 487 //Save caller's signal mask before setting VM signal mask 488 sigset_t caller_sigmask; 489 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask); 490 491 OSThread* osthread = thread->osthread(); 492 osthread->set_caller_sigmask(caller_sigmask); 493 494 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL); 495 496 if (!ReduceSignalUsage) { 497 if (thread->is_VM_thread()) { 498 // Only the VM thread handles BREAK_SIGNAL ... 499 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL); 500 } else { 501 // ... all other threads block BREAK_SIGNAL 502 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL); 503 } 504 } 505} 506 507////////////////////////////////////////////////////////////////////////////// 508// detecting pthread library 509 510void os::Linux::libpthread_init() { 511 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION 512 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a 513 // generic name for earlier versions. 514 // Define macros here so we can build HotSpot on old systems. 515#ifndef _CS_GNU_LIBC_VERSION 516 #define _CS_GNU_LIBC_VERSION 2 517#endif 518#ifndef _CS_GNU_LIBPTHREAD_VERSION 519 #define _CS_GNU_LIBPTHREAD_VERSION 3 520#endif 521 522 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0); 523 if (n > 0) { 524 char *str = (char *)malloc(n, mtInternal); 525 confstr(_CS_GNU_LIBC_VERSION, str, n); 526 os::Linux::set_glibc_version(str); 527 } else { 528 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version() 529 static char _gnu_libc_version[32]; 530 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version), 531 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release()); 532 os::Linux::set_glibc_version(_gnu_libc_version); 533 } 534 535 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0); 536 if (n > 0) { 537 char *str = (char *)malloc(n, mtInternal); 538 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n); 539 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells 540 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this 541 // is the case. LinuxThreads has a hard limit on max number of threads. 542 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value. 543 // On the other hand, NPTL does not have such a limit, sysconf() 544 // will return -1 and errno is not changed. Check if it is really NPTL. 545 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 && 546 strstr(str, "NPTL") && 547 sysconf(_SC_THREAD_THREADS_MAX) > 0) { 548 free(str); 549 os::Linux::set_libpthread_version("linuxthreads"); 550 } else { 551 os::Linux::set_libpthread_version(str); 552 } 553 } else { 554 // glibc before 2.3.2 only has LinuxThreads. 555 os::Linux::set_libpthread_version("linuxthreads"); 556 } 557 558 if (strstr(libpthread_version(), "NPTL")) { 559 os::Linux::set_is_NPTL(); 560 } else { 561 os::Linux::set_is_LinuxThreads(); 562 } 563 564 // LinuxThreads have two flavors: floating-stack mode, which allows variable 565 // stack size; and fixed-stack mode. NPTL is always floating-stack. 566 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) { 567 os::Linux::set_is_floating_stack(); 568 } 569} 570 571///////////////////////////////////////////////////////////////////////////// 572// thread stack 573 574// Force Linux kernel to expand current thread stack. If "bottom" is close 575// to the stack guard, caller should block all signals. 576// 577// MAP_GROWSDOWN: 578// A special mmap() flag that is used to implement thread stacks. It tells 579// kernel that the memory region should extend downwards when needed. This 580// allows early versions of LinuxThreads to only mmap the first few pages 581// when creating a new thread. Linux kernel will automatically expand thread 582// stack as needed (on page faults). 583// 584// However, because the memory region of a MAP_GROWSDOWN stack can grow on 585// demand, if a page fault happens outside an already mapped MAP_GROWSDOWN 586// region, it's hard to tell if the fault is due to a legitimate stack 587// access or because of reading/writing non-exist memory (e.g. buffer 588// overrun). As a rule, if the fault happens below current stack pointer, 589// Linux kernel does not expand stack, instead a SIGSEGV is sent to the 590// application (see Linux kernel fault.c). 591// 592// This Linux feature can cause SIGSEGV when VM bangs thread stack for 593// stack overflow detection. 594// 595// Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do 596// not use this flag. However, the stack of initial thread is not created 597// by pthread, it is still MAP_GROWSDOWN. Also it's possible (though 598// unlikely) that user code can create a thread with MAP_GROWSDOWN stack 599// and then attach the thread to JVM. 600// 601// To get around the problem and allow stack banging on Linux, we need to 602// manually expand thread stack after receiving the SIGSEGV. 603// 604// There are two ways to expand thread stack to address "bottom", we used 605// both of them in JVM before 1.5: 606// 1. adjust stack pointer first so that it is below "bottom", and then 607// touch "bottom" 608// 2. mmap() the page in question 609// 610// Now alternate signal stack is gone, it's harder to use 2. For instance, 611// if current sp is already near the lower end of page 101, and we need to 612// call mmap() to map page 100, it is possible that part of the mmap() frame 613// will be placed in page 100. When page 100 is mapped, it is zero-filled. 614// That will destroy the mmap() frame and cause VM to crash. 615// 616// The following code works by adjusting sp first, then accessing the "bottom" 617// page to force a page fault. Linux kernel will then automatically expand the 618// stack mapping. 619// 620// _expand_stack_to() assumes its frame size is less than page size, which 621// should always be true if the function is not inlined. 622 623#if __GNUC__ < 3 // gcc 2.x does not support noinline attribute 624 #define NOINLINE 625#else 626 #define NOINLINE __attribute__ ((noinline)) 627#endif 628 629static void _expand_stack_to(address bottom) NOINLINE; 630 631static void _expand_stack_to(address bottom) { 632 address sp; 633 size_t size; 634 volatile char *p; 635 636 // Adjust bottom to point to the largest address within the same page, it 637 // gives us a one-page buffer if alloca() allocates slightly more memory. 638 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size()); 639 bottom += os::Linux::page_size() - 1; 640 641 // sp might be slightly above current stack pointer; if that's the case, we 642 // will alloca() a little more space than necessary, which is OK. Don't use 643 // os::current_stack_pointer(), as its result can be slightly below current 644 // stack pointer, causing us to not alloca enough to reach "bottom". 645 sp = (address)&sp; 646 647 if (sp > bottom) { 648 size = sp - bottom; 649 p = (volatile char *)alloca(size); 650 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?"); 651 p[0] = '\0'; 652 } 653} 654 655bool os::Linux::manually_expand_stack(JavaThread * t, address addr) { 656 assert(t!=NULL, "just checking"); 657 assert(t->osthread()->expanding_stack(), "expand should be set"); 658 assert(t->stack_base() != NULL, "stack_base was not initialized"); 659 660 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) { 661 sigset_t mask_all, old_sigset; 662 sigfillset(&mask_all); 663 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset); 664 _expand_stack_to(addr); 665 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL); 666 return true; 667 } 668 return false; 669} 670 671////////////////////////////////////////////////////////////////////////////// 672// create new thread 673 674static address highest_vm_reserved_address(); 675 676// check if it's safe to start a new thread 677static bool _thread_safety_check(Thread* thread) { 678 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) { 679 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat) 680 // Heap is mmap'ed at lower end of memory space. Thread stacks are 681 // allocated (MAP_FIXED) from high address space. Every thread stack 682 // occupies a fixed size slot (usually 2Mbytes, but user can change 683 // it to other values if they rebuild LinuxThreads). 684 // 685 // Problem with MAP_FIXED is that mmap() can still succeed even part of 686 // the memory region has already been mmap'ed. That means if we have too 687 // many threads and/or very large heap, eventually thread stack will 688 // collide with heap. 689 // 690 // Here we try to prevent heap/stack collision by comparing current 691 // stack bottom with the highest address that has been mmap'ed by JVM 692 // plus a safety margin for memory maps created by native code. 693 // 694 // This feature can be disabled by setting ThreadSafetyMargin to 0 695 // 696 if (ThreadSafetyMargin > 0) { 697 address stack_bottom = os::current_stack_base() - os::current_stack_size(); 698 699 // not safe if our stack extends below the safety margin 700 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address(); 701 } else { 702 return true; 703 } 704 } else { 705 // Floating stack LinuxThreads or NPTL: 706 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When 707 // there's not enough space left, pthread_create() will fail. If we come 708 // here, that means enough space has been reserved for stack. 709 return true; 710 } 711} 712 713// Thread start routine for all newly created threads 714static void *java_start(Thread *thread) { 715 // Try to randomize the cache line index of hot stack frames. 716 // This helps when threads of the same stack traces evict each other's 717 // cache lines. The threads can be either from the same JVM instance, or 718 // from different JVM instances. The benefit is especially true for 719 // processors with hyperthreading technology. 720 static int counter = 0; 721 int pid = os::current_process_id(); 722 alloca(((pid ^ counter++) & 7) * 128); 723 724 ThreadLocalStorage::set_thread(thread); 725 726 OSThread* osthread = thread->osthread(); 727 Monitor* sync = osthread->startThread_lock(); 728 729 // non floating stack LinuxThreads needs extra check, see above 730 if (!_thread_safety_check(thread)) { 731 // notify parent thread 732 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 733 osthread->set_state(ZOMBIE); 734 sync->notify_all(); 735 return NULL; 736 } 737 738 // thread_id is kernel thread id (similar to Solaris LWP id) 739 osthread->set_thread_id(os::Linux::gettid()); 740 741 if (UseNUMA) { 742 int lgrp_id = os::numa_get_group_id(); 743 if (lgrp_id != -1) { 744 thread->set_lgrp_id(lgrp_id); 745 } 746 } 747 // initialize signal mask for this thread 748 os::Linux::hotspot_sigmask(thread); 749 750 // initialize floating point control register 751 os::Linux::init_thread_fpu_state(); 752 753 // handshaking with parent thread 754 { 755 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag); 756 757 // notify parent thread 758 osthread->set_state(INITIALIZED); 759 sync->notify_all(); 760 761 // wait until os::start_thread() 762 while (osthread->get_state() == INITIALIZED) { 763 sync->wait(Mutex::_no_safepoint_check_flag); 764 } 765 } 766 767 // call one more level start routine 768 thread->run(); 769 770 return 0; 771} 772 773bool os::create_thread(Thread* thread, ThreadType thr_type, 774 size_t stack_size) { 775 assert(thread->osthread() == NULL, "caller responsible"); 776 777 // Allocate the OSThread object 778 OSThread* osthread = new OSThread(NULL, NULL); 779 if (osthread == NULL) { 780 return false; 781 } 782 783 // set the correct thread state 784 osthread->set_thread_type(thr_type); 785 786 // Initial state is ALLOCATED but not INITIALIZED 787 osthread->set_state(ALLOCATED); 788 789 thread->set_osthread(osthread); 790 791 // init thread attributes 792 pthread_attr_t attr; 793 pthread_attr_init(&attr); 794 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); 795 796 // stack size 797 if (os::Linux::supports_variable_stack_size()) { 798 // calculate stack size if it's not specified by caller 799 if (stack_size == 0) { 800 stack_size = os::Linux::default_stack_size(thr_type); 801 802 switch (thr_type) { 803 case os::java_thread: 804 // Java threads use ThreadStackSize which default value can be 805 // changed with the flag -Xss 806 assert(JavaThread::stack_size_at_create() > 0, "this should be set"); 807 stack_size = JavaThread::stack_size_at_create(); 808 break; 809 case os::compiler_thread: 810 if (CompilerThreadStackSize > 0) { 811 stack_size = (size_t)(CompilerThreadStackSize * K); 812 break; 813 } // else fall through: 814 // use VMThreadStackSize if CompilerThreadStackSize is not defined 815 case os::vm_thread: 816 case os::pgc_thread: 817 case os::cgc_thread: 818 case os::watcher_thread: 819 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K); 820 break; 821 } 822 } 823 824 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed); 825 pthread_attr_setstacksize(&attr, stack_size); 826 } else { 827 // let pthread_create() pick the default value. 828 } 829 830 // glibc guard page 831 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type)); 832 833 ThreadState state; 834 835 { 836 // Serialize thread creation if we are running with fixed stack LinuxThreads 837 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack(); 838 if (lock) { 839 os::Linux::createThread_lock()->lock_without_safepoint_check(); 840 } 841 842 pthread_t tid; 843 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread); 844 845 pthread_attr_destroy(&attr); 846 847 if (ret != 0) { 848 if (PrintMiscellaneous && (Verbose || WizardMode)) { 849 perror("pthread_create()"); 850 } 851 // Need to clean up stuff we've allocated so far 852 thread->set_osthread(NULL); 853 delete osthread; 854 if (lock) os::Linux::createThread_lock()->unlock(); 855 return false; 856 } 857 858 // Store pthread info into the OSThread 859 osthread->set_pthread_id(tid); 860 861 // Wait until child thread is either initialized or aborted 862 { 863 Monitor* sync_with_child = osthread->startThread_lock(); 864 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 865 while ((state = osthread->get_state()) == ALLOCATED) { 866 sync_with_child->wait(Mutex::_no_safepoint_check_flag); 867 } 868 } 869 870 if (lock) { 871 os::Linux::createThread_lock()->unlock(); 872 } 873 } 874 875 // Aborted due to thread limit being reached 876 if (state == ZOMBIE) { 877 thread->set_osthread(NULL); 878 delete osthread; 879 return false; 880 } 881 882 // The thread is returned suspended (in state INITIALIZED), 883 // and is started higher up in the call chain 884 assert(state == INITIALIZED, "race condition"); 885 return true; 886} 887 888///////////////////////////////////////////////////////////////////////////// 889// attach existing thread 890 891// bootstrap the main thread 892bool os::create_main_thread(JavaThread* thread) { 893 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread"); 894 return create_attached_thread(thread); 895} 896 897bool os::create_attached_thread(JavaThread* thread) { 898#ifdef ASSERT 899 thread->verify_not_published(); 900#endif 901 902 // Allocate the OSThread object 903 OSThread* osthread = new OSThread(NULL, NULL); 904 905 if (osthread == NULL) { 906 return false; 907 } 908 909 // Store pthread info into the OSThread 910 osthread->set_thread_id(os::Linux::gettid()); 911 osthread->set_pthread_id(::pthread_self()); 912 913 // initialize floating point control register 914 os::Linux::init_thread_fpu_state(); 915 916 // Initial thread state is RUNNABLE 917 osthread->set_state(RUNNABLE); 918 919 thread->set_osthread(osthread); 920 921 if (UseNUMA) { 922 int lgrp_id = os::numa_get_group_id(); 923 if (lgrp_id != -1) { 924 thread->set_lgrp_id(lgrp_id); 925 } 926 } 927 928 if (os::Linux::is_initial_thread()) { 929 // If current thread is initial thread, its stack is mapped on demand, 930 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map 931 // the entire stack region to avoid SEGV in stack banging. 932 // It is also useful to get around the heap-stack-gap problem on SuSE 933 // kernel (see 4821821 for details). We first expand stack to the top 934 // of yellow zone, then enable stack yellow zone (order is significant, 935 // enabling yellow zone first will crash JVM on SuSE Linux), so there 936 // is no gap between the last two virtual memory regions. 937 938 JavaThread *jt = (JavaThread *)thread; 939 address addr = jt->stack_yellow_zone_base(); 940 assert(addr != NULL, "initialization problem?"); 941 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled"); 942 943 osthread->set_expanding_stack(); 944 os::Linux::manually_expand_stack(jt, addr); 945 osthread->clear_expanding_stack(); 946 } 947 948 // initialize signal mask for this thread 949 // and save the caller's signal mask 950 os::Linux::hotspot_sigmask(thread); 951 952 return true; 953} 954 955void os::pd_start_thread(Thread* thread) { 956 OSThread * osthread = thread->osthread(); 957 assert(osthread->get_state() != INITIALIZED, "just checking"); 958 Monitor* sync_with_child = osthread->startThread_lock(); 959 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag); 960 sync_with_child->notify(); 961} 962 963// Free Linux resources related to the OSThread 964void os::free_thread(OSThread* osthread) { 965 assert(osthread != NULL, "osthread not set"); 966 967 if (Thread::current()->osthread() == osthread) { 968 // Restore caller's signal mask 969 sigset_t sigmask = osthread->caller_sigmask(); 970 pthread_sigmask(SIG_SETMASK, &sigmask, NULL); 971 } 972 973 delete osthread; 974} 975 976////////////////////////////////////////////////////////////////////////////// 977// thread local storage 978 979// Restore the thread pointer if the destructor is called. This is in case 980// someone from JNI code sets up a destructor with pthread_key_create to run 981// detachCurrentThread on thread death. Unless we restore the thread pointer we 982// will hang or crash. When detachCurrentThread is called the key will be set 983// to null and we will not be called again. If detachCurrentThread is never 984// called we could loop forever depending on the pthread implementation. 985static void restore_thread_pointer(void* p) { 986 Thread* thread = (Thread*) p; 987 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread); 988} 989 990int os::allocate_thread_local_storage() { 991 pthread_key_t key; 992 int rslt = pthread_key_create(&key, restore_thread_pointer); 993 assert(rslt == 0, "cannot allocate thread local storage"); 994 return (int)key; 995} 996 997// Note: This is currently not used by VM, as we don't destroy TLS key 998// on VM exit. 999void os::free_thread_local_storage(int index) { 1000 int rslt = pthread_key_delete((pthread_key_t)index); 1001 assert(rslt == 0, "invalid index"); 1002} 1003 1004void os::thread_local_storage_at_put(int index, void* value) { 1005 int rslt = pthread_setspecific((pthread_key_t)index, value); 1006 assert(rslt == 0, "pthread_setspecific failed"); 1007} 1008 1009extern "C" Thread* get_thread() { 1010 return ThreadLocalStorage::thread(); 1011} 1012 1013////////////////////////////////////////////////////////////////////////////// 1014// initial thread 1015 1016// Check if current thread is the initial thread, similar to Solaris thr_main. 1017bool os::Linux::is_initial_thread(void) { 1018 char dummy; 1019 // If called before init complete, thread stack bottom will be null. 1020 // Can be called if fatal error occurs before initialization. 1021 if (initial_thread_stack_bottom() == NULL) return false; 1022 assert(initial_thread_stack_bottom() != NULL && 1023 initial_thread_stack_size() != 0, 1024 "os::init did not locate initial thread's stack region"); 1025 if ((address)&dummy >= initial_thread_stack_bottom() && 1026 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) { 1027 return true; 1028 } else { 1029 return false; 1030 } 1031} 1032 1033// Find the virtual memory area that contains addr 1034static bool find_vma(address addr, address* vma_low, address* vma_high) { 1035 FILE *fp = fopen("/proc/self/maps", "r"); 1036 if (fp) { 1037 address low, high; 1038 while (!feof(fp)) { 1039 if (fscanf(fp, "%p-%p", &low, &high) == 2) { 1040 if (low <= addr && addr < high) { 1041 if (vma_low) *vma_low = low; 1042 if (vma_high) *vma_high = high; 1043 fclose(fp); 1044 return true; 1045 } 1046 } 1047 for (;;) { 1048 int ch = fgetc(fp); 1049 if (ch == EOF || ch == (int)'\n') break; 1050 } 1051 } 1052 fclose(fp); 1053 } 1054 return false; 1055} 1056 1057// Locate initial thread stack. This special handling of initial thread stack 1058// is needed because pthread_getattr_np() on most (all?) Linux distros returns 1059// bogus value for initial thread. 1060void os::Linux::capture_initial_stack(size_t max_size) { 1061 // stack size is the easy part, get it from RLIMIT_STACK 1062 size_t stack_size; 1063 struct rlimit rlim; 1064 getrlimit(RLIMIT_STACK, &rlim); 1065 stack_size = rlim.rlim_cur; 1066 1067 // 6308388: a bug in ld.so will relocate its own .data section to the 1068 // lower end of primordial stack; reduce ulimit -s value a little bit 1069 // so we won't install guard page on ld.so's data section. 1070 stack_size -= 2 * page_size(); 1071 1072 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat 1073 // 7.1, in both cases we will get 2G in return value. 1074 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0, 1075 // SuSE 7.2, Debian) can not handle alternate signal stack correctly 1076 // for initial thread if its stack size exceeds 6M. Cap it at 2M, 1077 // in case other parts in glibc still assumes 2M max stack size. 1078 // FIXME: alt signal stack is gone, maybe we can relax this constraint? 1079 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small 1080 if (stack_size > 2 * K * K IA64_ONLY(*2)) { 1081 stack_size = 2 * K * K IA64_ONLY(*2); 1082 } 1083 // Try to figure out where the stack base (top) is. This is harder. 1084 // 1085 // When an application is started, glibc saves the initial stack pointer in 1086 // a global variable "__libc_stack_end", which is then used by system 1087 // libraries. __libc_stack_end should be pretty close to stack top. The 1088 // variable is available since the very early days. However, because it is 1089 // a private interface, it could disappear in the future. 1090 // 1091 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar 1092 // to __libc_stack_end, it is very close to stack top, but isn't the real 1093 // stack top. Note that /proc may not exist if VM is running as a chroot 1094 // program, so reading /proc/<pid>/stat could fail. Also the contents of 1095 // /proc/<pid>/stat could change in the future (though unlikely). 1096 // 1097 // We try __libc_stack_end first. If that doesn't work, look for 1098 // /proc/<pid>/stat. If neither of them works, we use current stack pointer 1099 // as a hint, which should work well in most cases. 1100 1101 uintptr_t stack_start; 1102 1103 // try __libc_stack_end first 1104 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end"); 1105 if (p && *p) { 1106 stack_start = *p; 1107 } else { 1108 // see if we can get the start_stack field from /proc/self/stat 1109 FILE *fp; 1110 int pid; 1111 char state; 1112 int ppid; 1113 int pgrp; 1114 int session; 1115 int nr; 1116 int tpgrp; 1117 unsigned long flags; 1118 unsigned long minflt; 1119 unsigned long cminflt; 1120 unsigned long majflt; 1121 unsigned long cmajflt; 1122 unsigned long utime; 1123 unsigned long stime; 1124 long cutime; 1125 long cstime; 1126 long prio; 1127 long nice; 1128 long junk; 1129 long it_real; 1130 uintptr_t start; 1131 uintptr_t vsize; 1132 intptr_t rss; 1133 uintptr_t rsslim; 1134 uintptr_t scodes; 1135 uintptr_t ecode; 1136 int i; 1137 1138 // Figure what the primordial thread stack base is. Code is inspired 1139 // by email from Hans Boehm. /proc/self/stat begins with current pid, 1140 // followed by command name surrounded by parentheses, state, etc. 1141 char stat[2048]; 1142 int statlen; 1143 1144 fp = fopen("/proc/self/stat", "r"); 1145 if (fp) { 1146 statlen = fread(stat, 1, 2047, fp); 1147 stat[statlen] = '\0'; 1148 fclose(fp); 1149 1150 // Skip pid and the command string. Note that we could be dealing with 1151 // weird command names, e.g. user could decide to rename java launcher 1152 // to "java 1.4.2 :)", then the stat file would look like 1153 // 1234 (java 1.4.2 :)) R ... ... 1154 // We don't really need to know the command string, just find the last 1155 // occurrence of ")" and then start parsing from there. See bug 4726580. 1156 char * s = strrchr(stat, ')'); 1157 1158 i = 0; 1159 if (s) { 1160 // Skip blank chars 1161 do { s++; } while (s && isspace(*s)); 1162 1163#define _UFM UINTX_FORMAT 1164#define _DFM INTX_FORMAT 1165 1166 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1167 // 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 1168 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, 1169 &state, // 3 %c 1170 &ppid, // 4 %d 1171 &pgrp, // 5 %d 1172 &session, // 6 %d 1173 &nr, // 7 %d 1174 &tpgrp, // 8 %d 1175 &flags, // 9 %lu 1176 &minflt, // 10 %lu 1177 &cminflt, // 11 %lu 1178 &majflt, // 12 %lu 1179 &cmajflt, // 13 %lu 1180 &utime, // 14 %lu 1181 &stime, // 15 %lu 1182 &cutime, // 16 %ld 1183 &cstime, // 17 %ld 1184 &prio, // 18 %ld 1185 &nice, // 19 %ld 1186 &junk, // 20 %ld 1187 &it_real, // 21 %ld 1188 &start, // 22 UINTX_FORMAT 1189 &vsize, // 23 UINTX_FORMAT 1190 &rss, // 24 INTX_FORMAT 1191 &rsslim, // 25 UINTX_FORMAT 1192 &scodes, // 26 UINTX_FORMAT 1193 &ecode, // 27 UINTX_FORMAT 1194 &stack_start); // 28 UINTX_FORMAT 1195 } 1196 1197#undef _UFM 1198#undef _DFM 1199 1200 if (i != 28 - 2) { 1201 assert(false, "Bad conversion from /proc/self/stat"); 1202 // product mode - assume we are the initial thread, good luck in the 1203 // embedded case. 1204 warning("Can't detect initial thread stack location - bad conversion"); 1205 stack_start = (uintptr_t) &rlim; 1206 } 1207 } else { 1208 // For some reason we can't open /proc/self/stat (for example, running on 1209 // FreeBSD with a Linux emulator, or inside chroot), this should work for 1210 // most cases, so don't abort: 1211 warning("Can't detect initial thread stack location - no /proc/self/stat"); 1212 stack_start = (uintptr_t) &rlim; 1213 } 1214 } 1215 1216 // Now we have a pointer (stack_start) very close to the stack top, the 1217 // next thing to do is to figure out the exact location of stack top. We 1218 // can find out the virtual memory area that contains stack_start by 1219 // reading /proc/self/maps, it should be the last vma in /proc/self/maps, 1220 // and its upper limit is the real stack top. (again, this would fail if 1221 // running inside chroot, because /proc may not exist.) 1222 1223 uintptr_t stack_top; 1224 address low, high; 1225 if (find_vma((address)stack_start, &low, &high)) { 1226 // success, "high" is the true stack top. (ignore "low", because initial 1227 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.) 1228 stack_top = (uintptr_t)high; 1229 } else { 1230 // failed, likely because /proc/self/maps does not exist 1231 warning("Can't detect initial thread stack location - find_vma failed"); 1232 // best effort: stack_start is normally within a few pages below the real 1233 // stack top, use it as stack top, and reduce stack size so we won't put 1234 // guard page outside stack. 1235 stack_top = stack_start; 1236 stack_size -= 16 * page_size(); 1237 } 1238 1239 // stack_top could be partially down the page so align it 1240 stack_top = align_size_up(stack_top, page_size()); 1241 1242 if (max_size && stack_size > max_size) { 1243 _initial_thread_stack_size = max_size; 1244 } else { 1245 _initial_thread_stack_size = stack_size; 1246 } 1247 1248 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size()); 1249 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size; 1250} 1251 1252//////////////////////////////////////////////////////////////////////////////// 1253// time support 1254 1255// Time since start-up in seconds to a fine granularity. 1256// Used by VMSelfDestructTimer and the MemProfiler. 1257double os::elapsedTime() { 1258 1259 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution 1260} 1261 1262jlong os::elapsed_counter() { 1263 return javaTimeNanos() - initial_time_count; 1264} 1265 1266jlong os::elapsed_frequency() { 1267 return NANOSECS_PER_SEC; // nanosecond resolution 1268} 1269 1270bool os::supports_vtime() { return true; } 1271bool os::enable_vtime() { return false; } 1272bool os::vtime_enabled() { return false; } 1273 1274double os::elapsedVTime() { 1275 struct rusage usage; 1276 int retval = getrusage(RUSAGE_THREAD, &usage); 1277 if (retval == 0) { 1278 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); 1279 } else { 1280 // better than nothing, but not much 1281 return elapsedTime(); 1282 } 1283} 1284 1285jlong os::javaTimeMillis() { 1286 timeval time; 1287 int status = gettimeofday(&time, NULL); 1288 assert(status != -1, "linux error"); 1289 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000); 1290} 1291 1292void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) { 1293 timeval time; 1294 int status = gettimeofday(&time, NULL); 1295 assert(status != -1, "linux error"); 1296 seconds = jlong(time.tv_sec); 1297 nanos = jlong(time.tv_usec) * 1000; 1298} 1299 1300 1301#ifndef CLOCK_MONOTONIC 1302 #define CLOCK_MONOTONIC (1) 1303#endif 1304 1305void os::Linux::clock_init() { 1306 // we do dlopen's in this particular order due to bug in linux 1307 // dynamical loader (see 6348968) leading to crash on exit 1308 void* handle = dlopen("librt.so.1", RTLD_LAZY); 1309 if (handle == NULL) { 1310 handle = dlopen("librt.so", RTLD_LAZY); 1311 } 1312 1313 if (handle) { 1314 int (*clock_getres_func)(clockid_t, struct timespec*) = 1315 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres"); 1316 int (*clock_gettime_func)(clockid_t, struct timespec*) = 1317 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime"); 1318 if (clock_getres_func && clock_gettime_func) { 1319 // See if monotonic clock is supported by the kernel. Note that some 1320 // early implementations simply return kernel jiffies (updated every 1321 // 1/100 or 1/1000 second). It would be bad to use such a low res clock 1322 // for nano time (though the monotonic property is still nice to have). 1323 // It's fixed in newer kernels, however clock_getres() still returns 1324 // 1/HZ. We check if clock_getres() works, but will ignore its reported 1325 // resolution for now. Hopefully as people move to new kernels, this 1326 // won't be a problem. 1327 struct timespec res; 1328 struct timespec tp; 1329 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 && 1330 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) { 1331 // yes, monotonic clock is supported 1332 _clock_gettime = clock_gettime_func; 1333 return; 1334 } else { 1335 // close librt if there is no monotonic clock 1336 dlclose(handle); 1337 } 1338 } 1339 } 1340 warning("No monotonic clock was available - timed services may " \ 1341 "be adversely affected if the time-of-day clock changes"); 1342} 1343 1344#ifndef SYS_clock_getres 1345 #if defined(IA32) || defined(AMD64) 1346 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229) 1347 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1348 #else 1349 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time" 1350 #define sys_clock_getres(x,y) -1 1351 #endif 1352#else 1353 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y) 1354#endif 1355 1356void os::Linux::fast_thread_clock_init() { 1357 if (!UseLinuxPosixThreadCPUClocks) { 1358 return; 1359 } 1360 clockid_t clockid; 1361 struct timespec tp; 1362 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) = 1363 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid"); 1364 1365 // Switch to using fast clocks for thread cpu time if 1366 // the sys_clock_getres() returns 0 error code. 1367 // Note, that some kernels may support the current thread 1368 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks 1369 // returned by the pthread_getcpuclockid(). 1370 // If the fast Posix clocks are supported then the sys_clock_getres() 1371 // must return at least tp.tv_sec == 0 which means a resolution 1372 // better than 1 sec. This is extra check for reliability. 1373 1374 if (pthread_getcpuclockid_func && 1375 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 && 1376 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) { 1377 _supports_fast_thread_cpu_time = true; 1378 _pthread_getcpuclockid = pthread_getcpuclockid_func; 1379 } 1380} 1381 1382jlong os::javaTimeNanos() { 1383 if (os::supports_monotonic_clock()) { 1384 struct timespec tp; 1385 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp); 1386 assert(status == 0, "gettime error"); 1387 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec); 1388 return result; 1389 } else { 1390 timeval time; 1391 int status = gettimeofday(&time, NULL); 1392 assert(status != -1, "linux error"); 1393 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec); 1394 return 1000 * usecs; 1395 } 1396} 1397 1398void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) { 1399 if (os::supports_monotonic_clock()) { 1400 info_ptr->max_value = ALL_64_BITS; 1401 1402 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past 1403 info_ptr->may_skip_backward = false; // not subject to resetting or drifting 1404 info_ptr->may_skip_forward = false; // not subject to resetting or drifting 1405 } else { 1406 // gettimeofday - based on time in seconds since the Epoch thus does not wrap 1407 info_ptr->max_value = ALL_64_BITS; 1408 1409 // gettimeofday is a real time clock so it skips 1410 info_ptr->may_skip_backward = true; 1411 info_ptr->may_skip_forward = true; 1412 } 1413 1414 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time 1415} 1416 1417// Return the real, user, and system times in seconds from an 1418// arbitrary fixed point in the past. 1419bool os::getTimesSecs(double* process_real_time, 1420 double* process_user_time, 1421 double* process_system_time) { 1422 struct tms ticks; 1423 clock_t real_ticks = times(&ticks); 1424 1425 if (real_ticks == (clock_t) (-1)) { 1426 return false; 1427 } else { 1428 double ticks_per_second = (double) clock_tics_per_sec; 1429 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second; 1430 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second; 1431 *process_real_time = ((double) real_ticks) / ticks_per_second; 1432 1433 return true; 1434 } 1435} 1436 1437 1438char * os::local_time_string(char *buf, size_t buflen) { 1439 struct tm t; 1440 time_t long_time; 1441 time(&long_time); 1442 localtime_r(&long_time, &t); 1443 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d", 1444 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday, 1445 t.tm_hour, t.tm_min, t.tm_sec); 1446 return buf; 1447} 1448 1449struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { 1450 return localtime_r(clock, res); 1451} 1452 1453//////////////////////////////////////////////////////////////////////////////// 1454// runtime exit support 1455 1456// Note: os::shutdown() might be called very early during initialization, or 1457// called from signal handler. Before adding something to os::shutdown(), make 1458// sure it is async-safe and can handle partially initialized VM. 1459void os::shutdown() { 1460 1461 // allow PerfMemory to attempt cleanup of any persistent resources 1462 perfMemory_exit(); 1463 1464 // needs to remove object in file system 1465 AttachListener::abort(); 1466 1467 // flush buffered output, finish log files 1468 ostream_abort(); 1469 1470 // Check for abort hook 1471 abort_hook_t abort_hook = Arguments::abort_hook(); 1472 if (abort_hook != NULL) { 1473 abort_hook(); 1474 } 1475 1476} 1477 1478// Note: os::abort() might be called very early during initialization, or 1479// called from signal handler. Before adding something to os::abort(), make 1480// sure it is async-safe and can handle partially initialized VM. 1481void os::abort(bool dump_core, void* siginfo, void* context) { 1482 os::shutdown(); 1483 if (dump_core) { 1484#ifndef PRODUCT 1485 fdStream out(defaultStream::output_fd()); 1486 out.print_raw("Current thread is "); 1487 char buf[16]; 1488 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id()); 1489 out.print_raw_cr(buf); 1490 out.print_raw_cr("Dumping core ..."); 1491#endif 1492 ::abort(); // dump core 1493 } 1494 1495 ::exit(1); 1496} 1497 1498// Die immediately, no exit hook, no abort hook, no cleanup. 1499void os::die() { 1500 // _exit() on LinuxThreads only kills current thread 1501 ::abort(); 1502} 1503 1504 1505// This method is a copy of JDK's sysGetLastErrorString 1506// from src/solaris/hpi/src/system_md.c 1507 1508size_t os::lasterror(char *buf, size_t len) { 1509 if (errno == 0) return 0; 1510 1511 const char *s = ::strerror(errno); 1512 size_t n = ::strlen(s); 1513 if (n >= len) { 1514 n = len - 1; 1515 } 1516 ::strncpy(buf, s, n); 1517 buf[n] = '\0'; 1518 return n; 1519} 1520 1521intx os::current_thread_id() { return (intx)pthread_self(); } 1522int os::current_process_id() { 1523 1524 // Under the old linux thread library, linux gives each thread 1525 // its own process id. Because of this each thread will return 1526 // a different pid if this method were to return the result 1527 // of getpid(2). Linux provides no api that returns the pid 1528 // of the launcher thread for the vm. This implementation 1529 // returns a unique pid, the pid of the launcher thread 1530 // that starts the vm 'process'. 1531 1532 // Under the NPTL, getpid() returns the same pid as the 1533 // launcher thread rather than a unique pid per thread. 1534 // Use gettid() if you want the old pre NPTL behaviour. 1535 1536 // if you are looking for the result of a call to getpid() that 1537 // returns a unique pid for the calling thread, then look at the 1538 // OSThread::thread_id() method in osThread_linux.hpp file 1539 1540 return (int)(_initial_pid ? _initial_pid : getpid()); 1541} 1542 1543// DLL functions 1544 1545const char* os::dll_file_extension() { return ".so"; } 1546 1547// This must be hard coded because it's the system's temporary 1548// directory not the java application's temp directory, ala java.io.tmpdir. 1549const char* os::get_temp_directory() { return "/tmp"; } 1550 1551static bool file_exists(const char* filename) { 1552 struct stat statbuf; 1553 if (filename == NULL || strlen(filename) == 0) { 1554 return false; 1555 } 1556 return os::stat(filename, &statbuf) == 0; 1557} 1558 1559bool os::dll_build_name(char* buffer, size_t buflen, 1560 const char* pname, const char* fname) { 1561 bool retval = false; 1562 // Copied from libhpi 1563 const size_t pnamelen = pname ? strlen(pname) : 0; 1564 1565 // Return error on buffer overflow. 1566 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) { 1567 return retval; 1568 } 1569 1570 if (pnamelen == 0) { 1571 snprintf(buffer, buflen, "lib%s.so", fname); 1572 retval = true; 1573 } else if (strchr(pname, *os::path_separator()) != NULL) { 1574 int n; 1575 char** pelements = split_path(pname, &n); 1576 if (pelements == NULL) { 1577 return false; 1578 } 1579 for (int i = 0; i < n; i++) { 1580 // Really shouldn't be NULL, but check can't hurt 1581 if (pelements[i] == NULL || strlen(pelements[i]) == 0) { 1582 continue; // skip the empty path values 1583 } 1584 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname); 1585 if (file_exists(buffer)) { 1586 retval = true; 1587 break; 1588 } 1589 } 1590 // release the storage 1591 for (int i = 0; i < n; i++) { 1592 if (pelements[i] != NULL) { 1593 FREE_C_HEAP_ARRAY(char, pelements[i]); 1594 } 1595 } 1596 if (pelements != NULL) { 1597 FREE_C_HEAP_ARRAY(char*, pelements); 1598 } 1599 } else { 1600 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname); 1601 retval = true; 1602 } 1603 return retval; 1604} 1605 1606// check if addr is inside libjvm.so 1607bool os::address_is_in_vm(address addr) { 1608 static address libjvm_base_addr; 1609 Dl_info dlinfo; 1610 1611 if (libjvm_base_addr == NULL) { 1612 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) { 1613 libjvm_base_addr = (address)dlinfo.dli_fbase; 1614 } 1615 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm"); 1616 } 1617 1618 if (dladdr((void *)addr, &dlinfo) != 0) { 1619 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true; 1620 } 1621 1622 return false; 1623} 1624 1625bool os::dll_address_to_function_name(address addr, char *buf, 1626 int buflen, int *offset, 1627 bool demangle) { 1628 // buf is not optional, but offset is optional 1629 assert(buf != NULL, "sanity check"); 1630 1631 Dl_info dlinfo; 1632 1633 if (dladdr((void*)addr, &dlinfo) != 0) { 1634 // see if we have a matching symbol 1635 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) { 1636 if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) { 1637 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname); 1638 } 1639 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr; 1640 return true; 1641 } 1642 // no matching symbol so try for just file info 1643 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) { 1644 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase), 1645 buf, buflen, offset, dlinfo.dli_fname, demangle)) { 1646 return true; 1647 } 1648 } 1649 } 1650 1651 buf[0] = '\0'; 1652 if (offset != NULL) *offset = -1; 1653 return false; 1654} 1655 1656struct _address_to_library_name { 1657 address addr; // input : memory address 1658 size_t buflen; // size of fname 1659 char* fname; // output: library name 1660 address base; // library base addr 1661}; 1662 1663static int address_to_library_name_callback(struct dl_phdr_info *info, 1664 size_t size, void *data) { 1665 int i; 1666 bool found = false; 1667 address libbase = NULL; 1668 struct _address_to_library_name * d = (struct _address_to_library_name *)data; 1669 1670 // iterate through all loadable segments 1671 for (i = 0; i < info->dlpi_phnum; i++) { 1672 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr); 1673 if (info->dlpi_phdr[i].p_type == PT_LOAD) { 1674 // base address of a library is the lowest address of its loaded 1675 // segments. 1676 if (libbase == NULL || libbase > segbase) { 1677 libbase = segbase; 1678 } 1679 // see if 'addr' is within current segment 1680 if (segbase <= d->addr && 1681 d->addr < segbase + info->dlpi_phdr[i].p_memsz) { 1682 found = true; 1683 } 1684 } 1685 } 1686 1687 // dlpi_name is NULL or empty if the ELF file is executable, return 0 1688 // so dll_address_to_library_name() can fall through to use dladdr() which 1689 // can figure out executable name from argv[0]. 1690 if (found && info->dlpi_name && info->dlpi_name[0]) { 1691 d->base = libbase; 1692 if (d->fname) { 1693 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name); 1694 } 1695 return 1; 1696 } 1697 return 0; 1698} 1699 1700bool os::dll_address_to_library_name(address addr, char* buf, 1701 int buflen, int* offset) { 1702 // buf is not optional, but offset is optional 1703 assert(buf != NULL, "sanity check"); 1704 1705 Dl_info dlinfo; 1706 struct _address_to_library_name data; 1707 1708 // There is a bug in old glibc dladdr() implementation that it could resolve 1709 // to wrong library name if the .so file has a base address != NULL. Here 1710 // we iterate through the program headers of all loaded libraries to find 1711 // out which library 'addr' really belongs to. This workaround can be 1712 // removed once the minimum requirement for glibc is moved to 2.3.x. 1713 data.addr = addr; 1714 data.fname = buf; 1715 data.buflen = buflen; 1716 data.base = NULL; 1717 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data); 1718 1719 if (rslt) { 1720 // buf already contains library name 1721 if (offset) *offset = addr - data.base; 1722 return true; 1723 } 1724 if (dladdr((void*)addr, &dlinfo) != 0) { 1725 if (dlinfo.dli_fname != NULL) { 1726 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname); 1727 } 1728 if (dlinfo.dli_fbase != NULL && offset != NULL) { 1729 *offset = addr - (address)dlinfo.dli_fbase; 1730 } 1731 return true; 1732 } 1733 1734 buf[0] = '\0'; 1735 if (offset) *offset = -1; 1736 return false; 1737} 1738 1739// Loads .dll/.so and 1740// in case of error it checks if .dll/.so was built for the 1741// same architecture as Hotspot is running on 1742 1743 1744// Remember the stack's state. The Linux dynamic linker will change 1745// the stack to 'executable' at most once, so we must safepoint only once. 1746bool os::Linux::_stack_is_executable = false; 1747 1748// VM operation that loads a library. This is necessary if stack protection 1749// of the Java stacks can be lost during loading the library. If we 1750// do not stop the Java threads, they can stack overflow before the stacks 1751// are protected again. 1752class VM_LinuxDllLoad: public VM_Operation { 1753 private: 1754 const char *_filename; 1755 char *_ebuf; 1756 int _ebuflen; 1757 void *_lib; 1758 public: 1759 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) : 1760 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {} 1761 VMOp_Type type() const { return VMOp_LinuxDllLoad; } 1762 void doit() { 1763 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen); 1764 os::Linux::_stack_is_executable = true; 1765 } 1766 void* loaded_library() { return _lib; } 1767}; 1768 1769void * os::dll_load(const char *filename, char *ebuf, int ebuflen) { 1770 void * result = NULL; 1771 bool load_attempted = false; 1772 1773 // Check whether the library to load might change execution rights 1774 // of the stack. If they are changed, the protection of the stack 1775 // guard pages will be lost. We need a safepoint to fix this. 1776 // 1777 // See Linux man page execstack(8) for more info. 1778 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) { 1779 ElfFile ef(filename); 1780 if (!ef.specifies_noexecstack()) { 1781 if (!is_init_completed()) { 1782 os::Linux::_stack_is_executable = true; 1783 // This is OK - No Java threads have been created yet, and hence no 1784 // stack guard pages to fix. 1785 // 1786 // This should happen only when you are building JDK7 using a very 1787 // old version of JDK6 (e.g., with JPRT) and running test_gamma. 1788 // 1789 // Dynamic loader will make all stacks executable after 1790 // this function returns, and will not do that again. 1791 assert(Threads::first() == NULL, "no Java threads should exist yet."); 1792 } else { 1793 warning("You have loaded library %s which might have disabled stack guard. " 1794 "The VM will try to fix the stack guard now.\n" 1795 "It's highly recommended that you fix the library with " 1796 "'execstack -c <libfile>', or link it with '-z noexecstack'.", 1797 filename); 1798 1799 assert(Thread::current()->is_Java_thread(), "must be Java thread"); 1800 JavaThread *jt = JavaThread::current(); 1801 if (jt->thread_state() != _thread_in_native) { 1802 // This happens when a compiler thread tries to load a hsdis-<arch>.so file 1803 // that requires ExecStack. Cannot enter safe point. Let's give up. 1804 warning("Unable to fix stack guard. Giving up."); 1805 } else { 1806 if (!LoadExecStackDllInVMThread) { 1807 // This is for the case where the DLL has an static 1808 // constructor function that executes JNI code. We cannot 1809 // load such DLLs in the VMThread. 1810 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1811 } 1812 1813 ThreadInVMfromNative tiv(jt); 1814 debug_only(VMNativeEntryWrapper vew;) 1815 1816 VM_LinuxDllLoad op(filename, ebuf, ebuflen); 1817 VMThread::execute(&op); 1818 if (LoadExecStackDllInVMThread) { 1819 result = op.loaded_library(); 1820 } 1821 load_attempted = true; 1822 } 1823 } 1824 } 1825 } 1826 1827 if (!load_attempted) { 1828 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen); 1829 } 1830 1831 if (result != NULL) { 1832 // Successful loading 1833 return result; 1834 } 1835 1836 Elf32_Ehdr elf_head; 1837 int diag_msg_max_length=ebuflen-strlen(ebuf); 1838 char* diag_msg_buf=ebuf+strlen(ebuf); 1839 1840 if (diag_msg_max_length==0) { 1841 // No more space in ebuf for additional diagnostics message 1842 return NULL; 1843 } 1844 1845 1846 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK); 1847 1848 if (file_descriptor < 0) { 1849 // Can't open library, report dlerror() message 1850 return NULL; 1851 } 1852 1853 bool failed_to_read_elf_head= 1854 (sizeof(elf_head)!= 1855 (::read(file_descriptor, &elf_head,sizeof(elf_head)))); 1856 1857 ::close(file_descriptor); 1858 if (failed_to_read_elf_head) { 1859 // file i/o error - report dlerror() msg 1860 return NULL; 1861 } 1862 1863 typedef struct { 1864 Elf32_Half code; // Actual value as defined in elf.h 1865 Elf32_Half compat_class; // Compatibility of archs at VM's sense 1866 char elf_class; // 32 or 64 bit 1867 char endianess; // MSB or LSB 1868 char* name; // String representation 1869 } arch_t; 1870 1871#ifndef EM_486 1872 #define EM_486 6 /* Intel 80486 */ 1873#endif 1874#ifndef EM_AARCH64 1875 #define EM_AARCH64 183 /* ARM AARCH64 */ 1876#endif 1877 1878 static const arch_t arch_array[]={ 1879 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1880 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"}, 1881 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"}, 1882 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"}, 1883 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1884 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"}, 1885 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"}, 1886 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"}, 1887#if defined(VM_LITTLE_ENDIAN) 1888 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"}, 1889#else 1890 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}, 1891#endif 1892 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"}, 1893 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"}, 1894 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"}, 1895 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"}, 1896 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"}, 1897 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"}, 1898 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}, 1899 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"}, 1900 }; 1901 1902#if (defined IA32) 1903 static Elf32_Half running_arch_code=EM_386; 1904#elif (defined AMD64) 1905 static Elf32_Half running_arch_code=EM_X86_64; 1906#elif (defined IA64) 1907 static Elf32_Half running_arch_code=EM_IA_64; 1908#elif (defined __sparc) && (defined _LP64) 1909 static Elf32_Half running_arch_code=EM_SPARCV9; 1910#elif (defined __sparc) && (!defined _LP64) 1911 static Elf32_Half running_arch_code=EM_SPARC; 1912#elif (defined __powerpc64__) 1913 static Elf32_Half running_arch_code=EM_PPC64; 1914#elif (defined __powerpc__) 1915 static Elf32_Half running_arch_code=EM_PPC; 1916#elif (defined ARM) 1917 static Elf32_Half running_arch_code=EM_ARM; 1918#elif (defined S390) 1919 static Elf32_Half running_arch_code=EM_S390; 1920#elif (defined ALPHA) 1921 static Elf32_Half running_arch_code=EM_ALPHA; 1922#elif (defined MIPSEL) 1923 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE; 1924#elif (defined PARISC) 1925 static Elf32_Half running_arch_code=EM_PARISC; 1926#elif (defined MIPS) 1927 static Elf32_Half running_arch_code=EM_MIPS; 1928#elif (defined M68K) 1929 static Elf32_Half running_arch_code=EM_68K; 1930#elif (defined AARCH64) 1931 static Elf32_Half running_arch_code=EM_AARCH64; 1932#else 1933 #error Method os::dll_load requires that one of following is defined:\ 1934 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64 1935#endif 1936 1937 // Identify compatability class for VM's architecture and library's architecture 1938 // Obtain string descriptions for architectures 1939 1940 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL}; 1941 int running_arch_index=-1; 1942 1943 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) { 1944 if (running_arch_code == arch_array[i].code) { 1945 running_arch_index = i; 1946 } 1947 if (lib_arch.code == arch_array[i].code) { 1948 lib_arch.compat_class = arch_array[i].compat_class; 1949 lib_arch.name = arch_array[i].name; 1950 } 1951 } 1952 1953 assert(running_arch_index != -1, 1954 "Didn't find running architecture code (running_arch_code) in arch_array"); 1955 if (running_arch_index == -1) { 1956 // Even though running architecture detection failed 1957 // we may still continue with reporting dlerror() message 1958 return NULL; 1959 } 1960 1961 if (lib_arch.endianess != arch_array[running_arch_index].endianess) { 1962 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)"); 1963 return NULL; 1964 } 1965 1966#ifndef S390 1967 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) { 1968 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)"); 1969 return NULL; 1970 } 1971#endif // !S390 1972 1973 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) { 1974 if (lib_arch.name!=NULL) { 1975 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1976 " (Possible cause: can't load %s-bit .so on a %s-bit platform)", 1977 lib_arch.name, arch_array[running_arch_index].name); 1978 } else { 1979 ::snprintf(diag_msg_buf, diag_msg_max_length-1, 1980 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)", 1981 lib_arch.code, 1982 arch_array[running_arch_index].name); 1983 } 1984 } 1985 1986 return NULL; 1987} 1988 1989void * os::Linux::dlopen_helper(const char *filename, char *ebuf, 1990 int ebuflen) { 1991 void * result = ::dlopen(filename, RTLD_LAZY); 1992 if (result == NULL) { 1993 ::strncpy(ebuf, ::dlerror(), ebuflen - 1); 1994 ebuf[ebuflen-1] = '\0'; 1995 } 1996 return result; 1997} 1998 1999void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, 2000 int ebuflen) { 2001 void * result = NULL; 2002 if (LoadExecStackDllInVMThread) { 2003 result = dlopen_helper(filename, ebuf, ebuflen); 2004 } 2005 2006 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a 2007 // library that requires an executable stack, or which does not have this 2008 // stack attribute set, dlopen changes the stack attribute to executable. The 2009 // read protection of the guard pages gets lost. 2010 // 2011 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad 2012 // may have been queued at the same time. 2013 2014 if (!_stack_is_executable) { 2015 JavaThread *jt = Threads::first(); 2016 2017 while (jt) { 2018 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized 2019 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions 2020 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(), 2021 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) { 2022 warning("Attempt to reguard stack yellow zone failed."); 2023 } 2024 } 2025 jt = jt->next(); 2026 } 2027 } 2028 2029 return result; 2030} 2031 2032void* os::dll_lookup(void* handle, const char* name) { 2033 void* res = dlsym(handle, name); 2034 return res; 2035} 2036 2037void* os::get_default_process_handle() { 2038 return (void*)::dlopen(NULL, RTLD_LAZY); 2039} 2040 2041static bool _print_ascii_file(const char* filename, outputStream* st) { 2042 int fd = ::open(filename, O_RDONLY); 2043 if (fd == -1) { 2044 return false; 2045 } 2046 2047 char buf[32]; 2048 int bytes; 2049 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) { 2050 st->print_raw(buf, bytes); 2051 } 2052 2053 ::close(fd); 2054 2055 return true; 2056} 2057 2058void os::print_dll_info(outputStream *st) { 2059 st->print_cr("Dynamic libraries:"); 2060 2061 char fname[32]; 2062 pid_t pid = os::Linux::gettid(); 2063 2064 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid); 2065 2066 if (!_print_ascii_file(fname, st)) { 2067 st->print("Can not get library information for pid = %d\n", pid); 2068 } 2069} 2070 2071int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) { 2072 FILE *procmapsFile = NULL; 2073 2074 // Open the procfs maps file for the current process 2075 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) { 2076 // Allocate PATH_MAX for file name plus a reasonable size for other fields. 2077 char line[PATH_MAX + 100]; 2078 2079 // Read line by line from 'file' 2080 while (fgets(line, sizeof(line), procmapsFile) != NULL) { 2081 u8 base, top, offset, inode; 2082 char permissions[5]; 2083 char device[6]; 2084 char name[PATH_MAX + 1]; 2085 2086 // Parse fields from line 2087 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name); 2088 2089 // Filter by device id '00:00' so that we only get file system mapped files. 2090 if (strcmp(device, "00:00") != 0) { 2091 2092 // Call callback with the fields of interest 2093 if(callback(name, (address)base, (address)top, param)) { 2094 // Oops abort, callback aborted 2095 fclose(procmapsFile); 2096 return 1; 2097 } 2098 } 2099 } 2100 fclose(procmapsFile); 2101 } 2102 return 0; 2103} 2104 2105void os::print_os_info_brief(outputStream* st) { 2106 os::Linux::print_distro_info(st); 2107 2108 os::Posix::print_uname_info(st); 2109 2110 os::Linux::print_libversion_info(st); 2111 2112} 2113 2114void os::print_os_info(outputStream* st) { 2115 st->print("OS:"); 2116 2117 os::Linux::print_distro_info(st); 2118 2119 os::Posix::print_uname_info(st); 2120 2121 // Print warning if unsafe chroot environment detected 2122 if (unsafe_chroot_detected) { 2123 st->print("WARNING!! "); 2124 st->print_cr("%s", unstable_chroot_error); 2125 } 2126 2127 os::Linux::print_libversion_info(st); 2128 2129 os::Posix::print_rlimit_info(st); 2130 2131 os::Posix::print_load_average(st); 2132 2133 os::Linux::print_full_memory_info(st); 2134} 2135 2136// Try to identify popular distros. 2137// Most Linux distributions have a /etc/XXX-release file, which contains 2138// the OS version string. Newer Linux distributions have a /etc/lsb-release 2139// file that also contains the OS version string. Some have more than one 2140// /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and 2141// /etc/redhat-release.), so the order is important. 2142// Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have 2143// their own specific XXX-release file as well as a redhat-release file. 2144// Because of this the XXX-release file needs to be searched for before the 2145// redhat-release file. 2146// Since Red Hat has a lsb-release file that is not very descriptive the 2147// search for redhat-release needs to be before lsb-release. 2148// Since the lsb-release file is the new standard it needs to be searched 2149// before the older style release files. 2150// Searching system-release (Red Hat) and os-release (other Linuxes) are a 2151// next to last resort. The os-release file is a new standard that contains 2152// distribution information and the system-release file seems to be an old 2153// standard that has been replaced by the lsb-release and os-release files. 2154// Searching for the debian_version file is the last resort. It contains 2155// an informative string like "6.0.6" or "wheezy/sid". Because of this 2156// "Debian " is printed before the contents of the debian_version file. 2157void os::Linux::print_distro_info(outputStream* st) { 2158 if (!_print_ascii_file("/etc/oracle-release", st) && 2159 !_print_ascii_file("/etc/mandriva-release", st) && 2160 !_print_ascii_file("/etc/mandrake-release", st) && 2161 !_print_ascii_file("/etc/sun-release", st) && 2162 !_print_ascii_file("/etc/redhat-release", st) && 2163 !_print_ascii_file("/etc/lsb-release", st) && 2164 !_print_ascii_file("/etc/SuSE-release", st) && 2165 !_print_ascii_file("/etc/turbolinux-release", st) && 2166 !_print_ascii_file("/etc/gentoo-release", st) && 2167 !_print_ascii_file("/etc/ltib-release", st) && 2168 !_print_ascii_file("/etc/angstrom-version", st) && 2169 !_print_ascii_file("/etc/system-release", st) && 2170 !_print_ascii_file("/etc/os-release", st)) { 2171 2172 if (file_exists("/etc/debian_version")) { 2173 st->print("Debian "); 2174 _print_ascii_file("/etc/debian_version", st); 2175 } else { 2176 st->print("Linux"); 2177 } 2178 } 2179 st->cr(); 2180} 2181 2182void os::Linux::print_libversion_info(outputStream* st) { 2183 // libc, pthread 2184 st->print("libc:"); 2185 st->print("%s ", os::Linux::glibc_version()); 2186 st->print("%s ", os::Linux::libpthread_version()); 2187 if (os::Linux::is_LinuxThreads()) { 2188 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed"); 2189 } 2190 st->cr(); 2191} 2192 2193void os::Linux::print_full_memory_info(outputStream* st) { 2194 st->print("\n/proc/meminfo:\n"); 2195 _print_ascii_file("/proc/meminfo", st); 2196 st->cr(); 2197} 2198 2199void os::print_memory_info(outputStream* st) { 2200 2201 st->print("Memory:"); 2202 st->print(" %dk page", os::vm_page_size()>>10); 2203 2204 // values in struct sysinfo are "unsigned long" 2205 struct sysinfo si; 2206 sysinfo(&si); 2207 2208 st->print(", physical " UINT64_FORMAT "k", 2209 os::physical_memory() >> 10); 2210 st->print("(" UINT64_FORMAT "k free)", 2211 os::available_memory() >> 10); 2212 st->print(", swap " UINT64_FORMAT "k", 2213 ((jlong)si.totalswap * si.mem_unit) >> 10); 2214 st->print("(" UINT64_FORMAT "k free)", 2215 ((jlong)si.freeswap * si.mem_unit) >> 10); 2216 st->cr(); 2217} 2218 2219// Print the first "model name" line and the first "flags" line 2220// that we find and nothing more. We assume "model name" comes 2221// before "flags" so if we find a second "model name", then the 2222// "flags" field is considered missing. 2223static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) { 2224#if defined(IA32) || defined(AMD64) 2225 // Other platforms have less repetitive cpuinfo files 2226 FILE *fp = fopen("/proc/cpuinfo", "r"); 2227 if (fp) { 2228 while (!feof(fp)) { 2229 if (fgets(buf, buflen, fp)) { 2230 // Assume model name comes before flags 2231 bool model_name_printed = false; 2232 if (strstr(buf, "model name") != NULL) { 2233 if (!model_name_printed) { 2234 st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n"); 2235 st->print_raw(buf); 2236 model_name_printed = true; 2237 } else { 2238 // model name printed but not flags? Odd, just return 2239 fclose(fp); 2240 return true; 2241 } 2242 } 2243 // print the flags line too 2244 if (strstr(buf, "flags") != NULL) { 2245 st->print_raw(buf); 2246 fclose(fp); 2247 return true; 2248 } 2249 } 2250 } 2251 fclose(fp); 2252 } 2253#endif // x86 platforms 2254 return false; 2255} 2256 2257void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) { 2258 // Only print the model name if the platform provides this as a summary 2259 if (!print_model_name_and_flags(st, buf, buflen)) { 2260 st->print("\n/proc/cpuinfo:\n"); 2261 if (!_print_ascii_file("/proc/cpuinfo", st)) { 2262 st->print_cr(" <Not Available>"); 2263 } 2264 } 2265} 2266 2267void os::print_siginfo(outputStream* st, void* siginfo) { 2268 const siginfo_t* si = (const siginfo_t*)siginfo; 2269 2270 os::Posix::print_siginfo_brief(st, si); 2271#if INCLUDE_CDS 2272 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) && 2273 UseSharedSpaces) { 2274 FileMapInfo* mapinfo = FileMapInfo::current_info(); 2275 if (mapinfo->is_in_shared_space(si->si_addr)) { 2276 st->print("\n\nError accessing class data sharing archive." \ 2277 " Mapped file inaccessible during execution, " \ 2278 " possible disk/network problem."); 2279 } 2280 } 2281#endif 2282 st->cr(); 2283} 2284 2285 2286static void print_signal_handler(outputStream* st, int sig, 2287 char* buf, size_t buflen); 2288 2289void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) { 2290 st->print_cr("Signal Handlers:"); 2291 print_signal_handler(st, SIGSEGV, buf, buflen); 2292 print_signal_handler(st, SIGBUS , buf, buflen); 2293 print_signal_handler(st, SIGFPE , buf, buflen); 2294 print_signal_handler(st, SIGPIPE, buf, buflen); 2295 print_signal_handler(st, SIGXFSZ, buf, buflen); 2296 print_signal_handler(st, SIGILL , buf, buflen); 2297 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen); 2298 print_signal_handler(st, SR_signum, buf, buflen); 2299 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen); 2300 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen); 2301 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen); 2302 print_signal_handler(st, BREAK_SIGNAL, buf, buflen); 2303#if defined(PPC64) 2304 print_signal_handler(st, SIGTRAP, buf, buflen); 2305#endif 2306} 2307 2308static char saved_jvm_path[MAXPATHLEN] = {0}; 2309 2310// Find the full path to the current module, libjvm.so 2311void os::jvm_path(char *buf, jint buflen) { 2312 // Error checking. 2313 if (buflen < MAXPATHLEN) { 2314 assert(false, "must use a large-enough buffer"); 2315 buf[0] = '\0'; 2316 return; 2317 } 2318 // Lazy resolve the path to current module. 2319 if (saved_jvm_path[0] != 0) { 2320 strcpy(buf, saved_jvm_path); 2321 return; 2322 } 2323 2324 char dli_fname[MAXPATHLEN]; 2325 bool ret = dll_address_to_library_name( 2326 CAST_FROM_FN_PTR(address, os::jvm_path), 2327 dli_fname, sizeof(dli_fname), NULL); 2328 assert(ret, "cannot locate libjvm"); 2329 char *rp = NULL; 2330 if (ret && dli_fname[0] != '\0') { 2331 rp = realpath(dli_fname, buf); 2332 } 2333 if (rp == NULL) { 2334 return; 2335 } 2336 2337 if (Arguments::sun_java_launcher_is_altjvm()) { 2338 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical 2339 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". 2340 // If "/jre/lib/" appears at the right place in the string, then 2341 // assume we are installed in a JDK and we're done. Otherwise, check 2342 // for a JAVA_HOME environment variable and fix up the path so it 2343 // looks like libjvm.so is installed there (append a fake suffix 2344 // hotspot/libjvm.so). 2345 const char *p = buf + strlen(buf) - 1; 2346 for (int count = 0; p > buf && count < 5; ++count) { 2347 for (--p; p > buf && *p != '/'; --p) 2348 /* empty */ ; 2349 } 2350 2351 if (strncmp(p, "/jre/lib/", 9) != 0) { 2352 // Look for JAVA_HOME in the environment. 2353 char* java_home_var = ::getenv("JAVA_HOME"); 2354 if (java_home_var != NULL && java_home_var[0] != 0) { 2355 char* jrelib_p; 2356 int len; 2357 2358 // Check the current module name "libjvm.so". 2359 p = strrchr(buf, '/'); 2360 if (p == NULL) { 2361 return; 2362 } 2363 assert(strstr(p, "/libjvm") == p, "invalid library name"); 2364 2365 rp = realpath(java_home_var, buf); 2366 if (rp == NULL) { 2367 return; 2368 } 2369 2370 // determine if this is a legacy image or modules image 2371 // modules image doesn't have "jre" subdirectory 2372 len = strlen(buf); 2373 assert(len < buflen, "Ran out of buffer room"); 2374 jrelib_p = buf + len; 2375 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch); 2376 if (0 != access(buf, F_OK)) { 2377 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch); 2378 } 2379 2380 if (0 == access(buf, F_OK)) { 2381 // Use current module name "libjvm.so" 2382 len = strlen(buf); 2383 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so"); 2384 } else { 2385 // Go back to path of .so 2386 rp = realpath(dli_fname, buf); 2387 if (rp == NULL) { 2388 return; 2389 } 2390 } 2391 } 2392 } 2393 } 2394 2395 strncpy(saved_jvm_path, buf, MAXPATHLEN); 2396 saved_jvm_path[MAXPATHLEN - 1] = '\0'; 2397} 2398 2399void os::print_jni_name_prefix_on(outputStream* st, int args_size) { 2400 // no prefix required, not even "_" 2401} 2402 2403void os::print_jni_name_suffix_on(outputStream* st, int args_size) { 2404 // no suffix required 2405} 2406 2407//////////////////////////////////////////////////////////////////////////////// 2408// sun.misc.Signal support 2409 2410static volatile jint sigint_count = 0; 2411 2412static void UserHandler(int sig, void *siginfo, void *context) { 2413 // 4511530 - sem_post is serialized and handled by the manager thread. When 2414 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We 2415 // don't want to flood the manager thread with sem_post requests. 2416 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) { 2417 return; 2418 } 2419 2420 // Ctrl-C is pressed during error reporting, likely because the error 2421 // handler fails to abort. Let VM die immediately. 2422 if (sig == SIGINT && is_error_reported()) { 2423 os::die(); 2424 } 2425 2426 os::signal_notify(sig); 2427} 2428 2429void* os::user_handler() { 2430 return CAST_FROM_FN_PTR(void*, UserHandler); 2431} 2432 2433class Semaphore : public StackObj { 2434 public: 2435 Semaphore(); 2436 ~Semaphore(); 2437 void signal(); 2438 void wait(); 2439 bool trywait(); 2440 bool timedwait(unsigned int sec, int nsec); 2441 private: 2442 sem_t _semaphore; 2443}; 2444 2445Semaphore::Semaphore() { 2446 sem_init(&_semaphore, 0, 0); 2447} 2448 2449Semaphore::~Semaphore() { 2450 sem_destroy(&_semaphore); 2451} 2452 2453void Semaphore::signal() { 2454 sem_post(&_semaphore); 2455} 2456 2457void Semaphore::wait() { 2458 sem_wait(&_semaphore); 2459} 2460 2461bool Semaphore::trywait() { 2462 return sem_trywait(&_semaphore) == 0; 2463} 2464 2465bool Semaphore::timedwait(unsigned int sec, int nsec) { 2466 2467 struct timespec ts; 2468 // Semaphore's are always associated with CLOCK_REALTIME 2469 os::Linux::clock_gettime(CLOCK_REALTIME, &ts); 2470 // see unpackTime for discussion on overflow checking 2471 if (sec >= MAX_SECS) { 2472 ts.tv_sec += MAX_SECS; 2473 ts.tv_nsec = 0; 2474 } else { 2475 ts.tv_sec += sec; 2476 ts.tv_nsec += nsec; 2477 if (ts.tv_nsec >= NANOSECS_PER_SEC) { 2478 ts.tv_nsec -= NANOSECS_PER_SEC; 2479 ++ts.tv_sec; // note: this must be <= max_secs 2480 } 2481 } 2482 2483 while (1) { 2484 int result = sem_timedwait(&_semaphore, &ts); 2485 if (result == 0) { 2486 return true; 2487 } else if (errno == EINTR) { 2488 continue; 2489 } else if (errno == ETIMEDOUT) { 2490 return false; 2491 } else { 2492 return false; 2493 } 2494 } 2495} 2496 2497extern "C" { 2498 typedef void (*sa_handler_t)(int); 2499 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *); 2500} 2501 2502void* os::signal(int signal_number, void* handler) { 2503 struct sigaction sigAct, oldSigAct; 2504 2505 sigfillset(&(sigAct.sa_mask)); 2506 sigAct.sa_flags = SA_RESTART|SA_SIGINFO; 2507 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler); 2508 2509 if (sigaction(signal_number, &sigAct, &oldSigAct)) { 2510 // -1 means registration failed 2511 return (void *)-1; 2512 } 2513 2514 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler); 2515} 2516 2517void os::signal_raise(int signal_number) { 2518 ::raise(signal_number); 2519} 2520 2521// The following code is moved from os.cpp for making this 2522// code platform specific, which it is by its very nature. 2523 2524// Will be modified when max signal is changed to be dynamic 2525int os::sigexitnum_pd() { 2526 return NSIG; 2527} 2528 2529// a counter for each possible signal value 2530static volatile jint pending_signals[NSIG+1] = { 0 }; 2531 2532// Linux(POSIX) specific hand shaking semaphore. 2533static sem_t sig_sem; 2534static Semaphore sr_semaphore; 2535 2536void os::signal_init_pd() { 2537 // Initialize signal structures 2538 ::memset((void*)pending_signals, 0, sizeof(pending_signals)); 2539 2540 // Initialize signal semaphore 2541 ::sem_init(&sig_sem, 0, 0); 2542} 2543 2544void os::signal_notify(int sig) { 2545 Atomic::inc(&pending_signals[sig]); 2546 ::sem_post(&sig_sem); 2547} 2548 2549static int check_pending_signals(bool wait) { 2550 Atomic::store(0, &sigint_count); 2551 for (;;) { 2552 for (int i = 0; i < NSIG + 1; i++) { 2553 jint n = pending_signals[i]; 2554 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) { 2555 return i; 2556 } 2557 } 2558 if (!wait) { 2559 return -1; 2560 } 2561 JavaThread *thread = JavaThread::current(); 2562 ThreadBlockInVM tbivm(thread); 2563 2564 bool threadIsSuspended; 2565 do { 2566 thread->set_suspend_equivalent(); 2567 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 2568 ::sem_wait(&sig_sem); 2569 2570 // were we externally suspended while we were waiting? 2571 threadIsSuspended = thread->handle_special_suspend_equivalent_condition(); 2572 if (threadIsSuspended) { 2573 // The semaphore has been incremented, but while we were waiting 2574 // another thread suspended us. We don't want to continue running 2575 // while suspended because that would surprise the thread that 2576 // suspended us. 2577 ::sem_post(&sig_sem); 2578 2579 thread->java_suspend_self(); 2580 } 2581 } while (threadIsSuspended); 2582 } 2583} 2584 2585int os::signal_lookup() { 2586 return check_pending_signals(false); 2587} 2588 2589int os::signal_wait() { 2590 return check_pending_signals(true); 2591} 2592 2593//////////////////////////////////////////////////////////////////////////////// 2594// Virtual Memory 2595 2596int os::vm_page_size() { 2597 // Seems redundant as all get out 2598 assert(os::Linux::page_size() != -1, "must call os::init"); 2599 return os::Linux::page_size(); 2600} 2601 2602// Solaris allocates memory by pages. 2603int os::vm_allocation_granularity() { 2604 assert(os::Linux::page_size() != -1, "must call os::init"); 2605 return os::Linux::page_size(); 2606} 2607 2608// Rationale behind this function: 2609// current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable 2610// mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get 2611// samples for JITted code. Here we create private executable mapping over the code cache 2612// and then we can use standard (well, almost, as mapping can change) way to provide 2613// info for the reporting script by storing timestamp and location of symbol 2614void linux_wrap_code(char* base, size_t size) { 2615 static volatile jint cnt = 0; 2616 2617 if (!UseOprofile) { 2618 return; 2619 } 2620 2621 char buf[PATH_MAX+1]; 2622 int num = Atomic::add(1, &cnt); 2623 2624 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d", 2625 os::get_temp_directory(), os::current_process_id(), num); 2626 unlink(buf); 2627 2628 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU); 2629 2630 if (fd != -1) { 2631 off_t rv = ::lseek(fd, size-2, SEEK_SET); 2632 if (rv != (off_t)-1) { 2633 if (::write(fd, "", 1) == 1) { 2634 mmap(base, size, 2635 PROT_READ|PROT_WRITE|PROT_EXEC, 2636 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0); 2637 } 2638 } 2639 ::close(fd); 2640 unlink(buf); 2641 } 2642} 2643 2644static bool recoverable_mmap_error(int err) { 2645 // See if the error is one we can let the caller handle. This 2646 // list of errno values comes from JBS-6843484. I can't find a 2647 // Linux man page that documents this specific set of errno 2648 // values so while this list currently matches Solaris, it may 2649 // change as we gain experience with this failure mode. 2650 switch (err) { 2651 case EBADF: 2652 case EINVAL: 2653 case ENOTSUP: 2654 // let the caller deal with these errors 2655 return true; 2656 2657 default: 2658 // Any remaining errors on this OS can cause our reserved mapping 2659 // to be lost. That can cause confusion where different data 2660 // structures think they have the same memory mapped. The worst 2661 // scenario is if both the VM and a library think they have the 2662 // same memory mapped. 2663 return false; 2664 } 2665} 2666 2667static void warn_fail_commit_memory(char* addr, size_t size, bool exec, 2668 int err) { 2669 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2670 ", %d) failed; error='%s' (errno=%d)", addr, size, exec, 2671 strerror(err), err); 2672} 2673 2674static void warn_fail_commit_memory(char* addr, size_t size, 2675 size_t alignment_hint, bool exec, 2676 int err) { 2677 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT 2678 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size, 2679 alignment_hint, exec, strerror(err), err); 2680} 2681 2682// NOTE: Linux kernel does not really reserve the pages for us. 2683// All it does is to check if there are enough free pages 2684// left at the time of mmap(). This could be a potential 2685// problem. 2686int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) { 2687 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 2688 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot, 2689 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0); 2690 if (res != (uintptr_t) MAP_FAILED) { 2691 if (UseNUMAInterleaving) { 2692 numa_make_global(addr, size); 2693 } 2694 return 0; 2695 } 2696 2697 int err = errno; // save errno from mmap() call above 2698 2699 if (!recoverable_mmap_error(err)) { 2700 warn_fail_commit_memory(addr, size, exec, err); 2701 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory."); 2702 } 2703 2704 return err; 2705} 2706 2707bool os::pd_commit_memory(char* addr, size_t size, bool exec) { 2708 return os::Linux::commit_memory_impl(addr, size, exec) == 0; 2709} 2710 2711void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec, 2712 const char* mesg) { 2713 assert(mesg != NULL, "mesg must be specified"); 2714 int err = os::Linux::commit_memory_impl(addr, size, exec); 2715 if (err != 0) { 2716 // the caller wants all commit errors to exit with the specified mesg: 2717 warn_fail_commit_memory(addr, size, exec, err); 2718 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2719 } 2720} 2721 2722// Define MAP_HUGETLB here so we can build HotSpot on old systems. 2723#ifndef MAP_HUGETLB 2724 #define MAP_HUGETLB 0x40000 2725#endif 2726 2727// Define MADV_HUGEPAGE here so we can build HotSpot on old systems. 2728#ifndef MADV_HUGEPAGE 2729 #define MADV_HUGEPAGE 14 2730#endif 2731 2732int os::Linux::commit_memory_impl(char* addr, size_t size, 2733 size_t alignment_hint, bool exec) { 2734 int err = os::Linux::commit_memory_impl(addr, size, exec); 2735 if (err == 0) { 2736 realign_memory(addr, size, alignment_hint); 2737 } 2738 return err; 2739} 2740 2741bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint, 2742 bool exec) { 2743 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0; 2744} 2745 2746void os::pd_commit_memory_or_exit(char* addr, size_t size, 2747 size_t alignment_hint, bool exec, 2748 const char* mesg) { 2749 assert(mesg != NULL, "mesg must be specified"); 2750 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec); 2751 if (err != 0) { 2752 // the caller wants all commit errors to exit with the specified mesg: 2753 warn_fail_commit_memory(addr, size, alignment_hint, exec, err); 2754 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg); 2755 } 2756} 2757 2758void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) { 2759 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) { 2760 // We don't check the return value: madvise(MADV_HUGEPAGE) may not 2761 // be supported or the memory may already be backed by huge pages. 2762 ::madvise(addr, bytes, MADV_HUGEPAGE); 2763 } 2764} 2765 2766void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) { 2767 // This method works by doing an mmap over an existing mmaping and effectively discarding 2768 // the existing pages. However it won't work for SHM-based large pages that cannot be 2769 // uncommitted at all. We don't do anything in this case to avoid creating a segment with 2770 // small pages on top of the SHM segment. This method always works for small pages, so we 2771 // allow that in any case. 2772 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) { 2773 commit_memory(addr, bytes, alignment_hint, !ExecMem); 2774 } 2775} 2776 2777void os::numa_make_global(char *addr, size_t bytes) { 2778 Linux::numa_interleave_memory(addr, bytes); 2779} 2780 2781// Define for numa_set_bind_policy(int). Setting the argument to 0 will set the 2782// bind policy to MPOL_PREFERRED for the current thread. 2783#define USE_MPOL_PREFERRED 0 2784 2785void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) { 2786 // To make NUMA and large pages more robust when both enabled, we need to ease 2787 // the requirements on where the memory should be allocated. MPOL_BIND is the 2788 // default policy and it will force memory to be allocated on the specified 2789 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on 2790 // the specified node, but will not force it. Using this policy will prevent 2791 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no 2792 // free large pages. 2793 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED); 2794 Linux::numa_tonode_memory(addr, bytes, lgrp_hint); 2795} 2796 2797bool os::numa_topology_changed() { return false; } 2798 2799size_t os::numa_get_groups_num() { 2800 int max_node = Linux::numa_max_node(); 2801 return max_node > 0 ? max_node + 1 : 1; 2802} 2803 2804int os::numa_get_group_id() { 2805 int cpu_id = Linux::sched_getcpu(); 2806 if (cpu_id != -1) { 2807 int lgrp_id = Linux::get_node_by_cpu(cpu_id); 2808 if (lgrp_id != -1) { 2809 return lgrp_id; 2810 } 2811 } 2812 return 0; 2813} 2814 2815size_t os::numa_get_leaf_groups(int *ids, size_t size) { 2816 for (size_t i = 0; i < size; i++) { 2817 ids[i] = i; 2818 } 2819 return size; 2820} 2821 2822bool os::get_page_info(char *start, page_info* info) { 2823 return false; 2824} 2825 2826char *os::scan_pages(char *start, char* end, page_info* page_expected, 2827 page_info* page_found) { 2828 return end; 2829} 2830 2831 2832int os::Linux::sched_getcpu_syscall(void) { 2833 unsigned int cpu; 2834 int retval = -1; 2835 2836#if defined(IA32) 2837 #ifndef SYS_getcpu 2838 #define SYS_getcpu 318 2839 #endif 2840 retval = syscall(SYS_getcpu, &cpu, NULL, NULL); 2841#elif defined(AMD64) 2842// Unfortunately we have to bring all these macros here from vsyscall.h 2843// to be able to compile on old linuxes. 2844 #define __NR_vgetcpu 2 2845 #define VSYSCALL_START (-10UL << 20) 2846 #define VSYSCALL_SIZE 1024 2847 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr)) 2848 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache); 2849 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu); 2850 retval = vgetcpu(&cpu, NULL, NULL); 2851#endif 2852 2853 return (retval == -1) ? retval : cpu; 2854} 2855 2856// Something to do with the numa-aware allocator needs these symbols 2857extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { } 2858extern "C" JNIEXPORT void numa_error(char *where) { } 2859extern "C" JNIEXPORT int fork1() { return fork(); } 2860 2861 2862// If we are running with libnuma version > 2, then we should 2863// be trying to use symbols with versions 1.1 2864// If we are running with earlier version, which did not have symbol versions, 2865// we should use the base version. 2866void* os::Linux::libnuma_dlsym(void* handle, const char *name) { 2867 void *f = dlvsym(handle, name, "libnuma_1.1"); 2868 if (f == NULL) { 2869 f = dlsym(handle, name); 2870 } 2871 return f; 2872} 2873 2874bool os::Linux::libnuma_init() { 2875 // sched_getcpu() should be in libc. 2876 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2877 dlsym(RTLD_DEFAULT, "sched_getcpu"))); 2878 2879 // If it's not, try a direct syscall. 2880 if (sched_getcpu() == -1) { 2881 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, 2882 (void*)&sched_getcpu_syscall)); 2883 } 2884 2885 if (sched_getcpu() != -1) { // Does it work? 2886 void *handle = dlopen("libnuma.so.1", RTLD_LAZY); 2887 if (handle != NULL) { 2888 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t, 2889 libnuma_dlsym(handle, "numa_node_to_cpus"))); 2890 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t, 2891 libnuma_dlsym(handle, "numa_max_node"))); 2892 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t, 2893 libnuma_dlsym(handle, "numa_available"))); 2894 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t, 2895 libnuma_dlsym(handle, "numa_tonode_memory"))); 2896 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t, 2897 libnuma_dlsym(handle, "numa_interleave_memory"))); 2898 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t, 2899 libnuma_dlsym(handle, "numa_set_bind_policy"))); 2900 2901 2902 if (numa_available() != -1) { 2903 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes")); 2904 // Create a cpu -> node mapping 2905 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true); 2906 rebuild_cpu_to_node_map(); 2907 return true; 2908 } 2909 } 2910 } 2911 return false; 2912} 2913 2914// rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id. 2915// The table is later used in get_node_by_cpu(). 2916void os::Linux::rebuild_cpu_to_node_map() { 2917 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure 2918 // in libnuma (possible values are starting from 16, 2919 // and continuing up with every other power of 2, but less 2920 // than the maximum number of CPUs supported by kernel), and 2921 // is a subject to change (in libnuma version 2 the requirements 2922 // are more reasonable) we'll just hardcode the number they use 2923 // in the library. 2924 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT; 2925 2926 size_t cpu_num = os::active_processor_count(); 2927 size_t cpu_map_size = NCPUS / BitsPerCLong; 2928 size_t cpu_map_valid_size = 2929 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size); 2930 2931 cpu_to_node()->clear(); 2932 cpu_to_node()->at_grow(cpu_num - 1); 2933 size_t node_num = numa_get_groups_num(); 2934 2935 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal); 2936 for (size_t i = 0; i < node_num; i++) { 2937 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) { 2938 for (size_t j = 0; j < cpu_map_valid_size; j++) { 2939 if (cpu_map[j] != 0) { 2940 for (size_t k = 0; k < BitsPerCLong; k++) { 2941 if (cpu_map[j] & (1UL << k)) { 2942 cpu_to_node()->at_put(j * BitsPerCLong + k, i); 2943 } 2944 } 2945 } 2946 } 2947 } 2948 } 2949 FREE_C_HEAP_ARRAY(unsigned long, cpu_map); 2950} 2951 2952int os::Linux::get_node_by_cpu(int cpu_id) { 2953 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) { 2954 return cpu_to_node()->at(cpu_id); 2955 } 2956 return -1; 2957} 2958 2959GrowableArray<int>* os::Linux::_cpu_to_node; 2960os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu; 2961os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus; 2962os::Linux::numa_max_node_func_t os::Linux::_numa_max_node; 2963os::Linux::numa_available_func_t os::Linux::_numa_available; 2964os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory; 2965os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory; 2966os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy; 2967unsigned long* os::Linux::_numa_all_nodes; 2968 2969bool os::pd_uncommit_memory(char* addr, size_t size) { 2970 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE, 2971 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0); 2972 return res != (uintptr_t) MAP_FAILED; 2973} 2974 2975static address get_stack_commited_bottom(address bottom, size_t size) { 2976 address nbot = bottom; 2977 address ntop = bottom + size; 2978 2979 size_t page_sz = os::vm_page_size(); 2980 unsigned pages = size / page_sz; 2981 2982 unsigned char vec[1]; 2983 unsigned imin = 1, imax = pages + 1, imid; 2984 int mincore_return_value = 0; 2985 2986 assert(imin <= imax, "Unexpected page size"); 2987 2988 while (imin < imax) { 2989 imid = (imax + imin) / 2; 2990 nbot = ntop - (imid * page_sz); 2991 2992 // Use a trick with mincore to check whether the page is mapped or not. 2993 // mincore sets vec to 1 if page resides in memory and to 0 if page 2994 // is swapped output but if page we are asking for is unmapped 2995 // it returns -1,ENOMEM 2996 mincore_return_value = mincore(nbot, page_sz, vec); 2997 2998 if (mincore_return_value == -1) { 2999 // Page is not mapped go up 3000 // to find first mapped page 3001 if (errno != EAGAIN) { 3002 assert(errno == ENOMEM, "Unexpected mincore errno"); 3003 imax = imid; 3004 } 3005 } else { 3006 // Page is mapped go down 3007 // to find first not mapped page 3008 imin = imid + 1; 3009 } 3010 } 3011 3012 nbot = nbot + page_sz; 3013 3014 // Adjust stack bottom one page up if last checked page is not mapped 3015 if (mincore_return_value == -1) { 3016 nbot = nbot + page_sz; 3017 } 3018 3019 return nbot; 3020} 3021 3022 3023// Linux uses a growable mapping for the stack, and if the mapping for 3024// the stack guard pages is not removed when we detach a thread the 3025// stack cannot grow beyond the pages where the stack guard was 3026// mapped. If at some point later in the process the stack expands to 3027// that point, the Linux kernel cannot expand the stack any further 3028// because the guard pages are in the way, and a segfault occurs. 3029// 3030// However, it's essential not to split the stack region by unmapping 3031// a region (leaving a hole) that's already part of the stack mapping, 3032// so if the stack mapping has already grown beyond the guard pages at 3033// the time we create them, we have to truncate the stack mapping. 3034// So, we need to know the extent of the stack mapping when 3035// create_stack_guard_pages() is called. 3036 3037// We only need this for stacks that are growable: at the time of 3038// writing thread stacks don't use growable mappings (i.e. those 3039// creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this 3040// only applies to the main thread. 3041 3042// If the (growable) stack mapping already extends beyond the point 3043// where we're going to put our guard pages, truncate the mapping at 3044// that point by munmap()ping it. This ensures that when we later 3045// munmap() the guard pages we don't leave a hole in the stack 3046// mapping. This only affects the main/initial thread 3047 3048bool os::pd_create_stack_guard_pages(char* addr, size_t size) { 3049 if (os::Linux::is_initial_thread()) { 3050 // As we manually grow stack up to bottom inside create_attached_thread(), 3051 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and 3052 // we don't need to do anything special. 3053 // Check it first, before calling heavy function. 3054 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom(); 3055 unsigned char vec[1]; 3056 3057 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) { 3058 // Fallback to slow path on all errors, including EAGAIN 3059 stack_extent = (uintptr_t) get_stack_commited_bottom( 3060 os::Linux::initial_thread_stack_bottom(), 3061 (size_t)addr - stack_extent); 3062 } 3063 3064 if (stack_extent < (uintptr_t)addr) { 3065 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent)); 3066 } 3067 } 3068 3069 return os::commit_memory(addr, size, !ExecMem); 3070} 3071 3072// If this is a growable mapping, remove the guard pages entirely by 3073// munmap()ping them. If not, just call uncommit_memory(). This only 3074// affects the main/initial thread, but guard against future OS changes 3075// It's safe to always unmap guard pages for initial thread because we 3076// always place it right after end of the mapped region 3077 3078bool os::remove_stack_guard_pages(char* addr, size_t size) { 3079 uintptr_t stack_extent, stack_base; 3080 3081 if (os::Linux::is_initial_thread()) { 3082 return ::munmap(addr, size) == 0; 3083 } 3084 3085 return os::uncommit_memory(addr, size); 3086} 3087 3088static address _highest_vm_reserved_address = NULL; 3089 3090// If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory 3091// at 'requested_addr'. If there are existing memory mappings at the same 3092// location, however, they will be overwritten. If 'fixed' is false, 3093// 'requested_addr' is only treated as a hint, the return value may or 3094// may not start from the requested address. Unlike Linux mmap(), this 3095// function returns NULL to indicate failure. 3096static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) { 3097 char * addr; 3098 int flags; 3099 3100 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS; 3101 if (fixed) { 3102 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address"); 3103 flags |= MAP_FIXED; 3104 } 3105 3106 // Map reserved/uncommitted pages PROT_NONE so we fail early if we 3107 // touch an uncommitted page. Otherwise, the read/write might 3108 // succeed if we have enough swap space to back the physical page. 3109 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE, 3110 flags, -1, 0); 3111 3112 if (addr != MAP_FAILED) { 3113 // anon_mmap() should only get called during VM initialization, 3114 // don't need lock (actually we can skip locking even it can be called 3115 // from multiple threads, because _highest_vm_reserved_address is just a 3116 // hint about the upper limit of non-stack memory regions.) 3117 if ((address)addr + bytes > _highest_vm_reserved_address) { 3118 _highest_vm_reserved_address = (address)addr + bytes; 3119 } 3120 } 3121 3122 return addr == MAP_FAILED ? NULL : addr; 3123} 3124 3125// Don't update _highest_vm_reserved_address, because there might be memory 3126// regions above addr + size. If so, releasing a memory region only creates 3127// a hole in the address space, it doesn't help prevent heap-stack collision. 3128// 3129static int anon_munmap(char * addr, size_t size) { 3130 return ::munmap(addr, size) == 0; 3131} 3132 3133char* os::pd_reserve_memory(size_t bytes, char* requested_addr, 3134 size_t alignment_hint) { 3135 return anon_mmap(requested_addr, bytes, (requested_addr != NULL)); 3136} 3137 3138bool os::pd_release_memory(char* addr, size_t size) { 3139 return anon_munmap(addr, size); 3140} 3141 3142static address highest_vm_reserved_address() { 3143 return _highest_vm_reserved_address; 3144} 3145 3146static bool linux_mprotect(char* addr, size_t size, int prot) { 3147 // Linux wants the mprotect address argument to be page aligned. 3148 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size()); 3149 3150 // According to SUSv3, mprotect() should only be used with mappings 3151 // established by mmap(), and mmap() always maps whole pages. Unaligned 3152 // 'addr' likely indicates problem in the VM (e.g. trying to change 3153 // protection of malloc'ed or statically allocated memory). Check the 3154 // caller if you hit this assert. 3155 assert(addr == bottom, "sanity check"); 3156 3157 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size()); 3158 return ::mprotect(bottom, size, prot) == 0; 3159} 3160 3161// Set protections specified 3162bool os::protect_memory(char* addr, size_t bytes, ProtType prot, 3163 bool is_committed) { 3164 unsigned int p = 0; 3165 switch (prot) { 3166 case MEM_PROT_NONE: p = PROT_NONE; break; 3167 case MEM_PROT_READ: p = PROT_READ; break; 3168 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break; 3169 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break; 3170 default: 3171 ShouldNotReachHere(); 3172 } 3173 // is_committed is unused. 3174 return linux_mprotect(addr, bytes, p); 3175} 3176 3177bool os::guard_memory(char* addr, size_t size) { 3178 return linux_mprotect(addr, size, PROT_NONE); 3179} 3180 3181bool os::unguard_memory(char* addr, size_t size) { 3182 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE); 3183} 3184 3185bool os::Linux::transparent_huge_pages_sanity_check(bool warn, 3186 size_t page_size) { 3187 bool result = false; 3188 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE, 3189 MAP_ANONYMOUS|MAP_PRIVATE, 3190 -1, 0); 3191 if (p != MAP_FAILED) { 3192 void *aligned_p = align_ptr_up(p, page_size); 3193 3194 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0; 3195 3196 munmap(p, page_size * 2); 3197 } 3198 3199 if (warn && !result) { 3200 warning("TransparentHugePages is not supported by the operating system."); 3201 } 3202 3203 return result; 3204} 3205 3206bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) { 3207 bool result = false; 3208 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE, 3209 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB, 3210 -1, 0); 3211 3212 if (p != MAP_FAILED) { 3213 // We don't know if this really is a huge page or not. 3214 FILE *fp = fopen("/proc/self/maps", "r"); 3215 if (fp) { 3216 while (!feof(fp)) { 3217 char chars[257]; 3218 long x = 0; 3219 if (fgets(chars, sizeof(chars), fp)) { 3220 if (sscanf(chars, "%lx-%*x", &x) == 1 3221 && x == (long)p) { 3222 if (strstr (chars, "hugepage")) { 3223 result = true; 3224 break; 3225 } 3226 } 3227 } 3228 } 3229 fclose(fp); 3230 } 3231 munmap(p, page_size); 3232 } 3233 3234 if (warn && !result) { 3235 warning("HugeTLBFS is not supported by the operating system."); 3236 } 3237 3238 return result; 3239} 3240 3241// Set the coredump_filter bits to include largepages in core dump (bit 6) 3242// 3243// From the coredump_filter documentation: 3244// 3245// - (bit 0) anonymous private memory 3246// - (bit 1) anonymous shared memory 3247// - (bit 2) file-backed private memory 3248// - (bit 3) file-backed shared memory 3249// - (bit 4) ELF header pages in file-backed private memory areas (it is 3250// effective only if the bit 2 is cleared) 3251// - (bit 5) hugetlb private memory 3252// - (bit 6) hugetlb shared memory 3253// 3254static void set_coredump_filter(void) { 3255 FILE *f; 3256 long cdm; 3257 3258 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) { 3259 return; 3260 } 3261 3262 if (fscanf(f, "%lx", &cdm) != 1) { 3263 fclose(f); 3264 return; 3265 } 3266 3267 rewind(f); 3268 3269 if ((cdm & LARGEPAGES_BIT) == 0) { 3270 cdm |= LARGEPAGES_BIT; 3271 fprintf(f, "%#lx", cdm); 3272 } 3273 3274 fclose(f); 3275} 3276 3277// Large page support 3278 3279static size_t _large_page_size = 0; 3280 3281size_t os::Linux::find_large_page_size() { 3282 size_t large_page_size = 0; 3283 3284 // large_page_size on Linux is used to round up heap size. x86 uses either 3285 // 2M or 4M page, depending on whether PAE (Physical Address Extensions) 3286 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use 3287 // page as large as 256M. 3288 // 3289 // Here we try to figure out page size by parsing /proc/meminfo and looking 3290 // for a line with the following format: 3291 // Hugepagesize: 2048 kB 3292 // 3293 // If we can't determine the value (e.g. /proc is not mounted, or the text 3294 // format has been changed), we'll use the largest page size supported by 3295 // the processor. 3296 3297#ifndef ZERO 3298 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M) 3299 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M); 3300#endif // ZERO 3301 3302 FILE *fp = fopen("/proc/meminfo", "r"); 3303 if (fp) { 3304 while (!feof(fp)) { 3305 int x = 0; 3306 char buf[16]; 3307 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) { 3308 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) { 3309 large_page_size = x * K; 3310 break; 3311 } 3312 } else { 3313 // skip to next line 3314 for (;;) { 3315 int ch = fgetc(fp); 3316 if (ch == EOF || ch == (int)'\n') break; 3317 } 3318 } 3319 } 3320 fclose(fp); 3321 } 3322 3323 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) { 3324 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is " 3325 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size), 3326 proper_unit_for_byte_size(large_page_size)); 3327 } 3328 3329 return large_page_size; 3330} 3331 3332size_t os::Linux::setup_large_page_size() { 3333 _large_page_size = Linux::find_large_page_size(); 3334 const size_t default_page_size = (size_t)Linux::page_size(); 3335 if (_large_page_size > default_page_size) { 3336 _page_sizes[0] = _large_page_size; 3337 _page_sizes[1] = default_page_size; 3338 _page_sizes[2] = 0; 3339 } 3340 3341 return _large_page_size; 3342} 3343 3344bool os::Linux::setup_large_page_type(size_t page_size) { 3345 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && 3346 FLAG_IS_DEFAULT(UseSHM) && 3347 FLAG_IS_DEFAULT(UseTransparentHugePages)) { 3348 3349 // The type of large pages has not been specified by the user. 3350 3351 // Try UseHugeTLBFS and then UseSHM. 3352 UseHugeTLBFS = UseSHM = true; 3353 3354 // Don't try UseTransparentHugePages since there are known 3355 // performance issues with it turned on. This might change in the future. 3356 UseTransparentHugePages = false; 3357 } 3358 3359 if (UseTransparentHugePages) { 3360 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages); 3361 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) { 3362 UseHugeTLBFS = false; 3363 UseSHM = false; 3364 return true; 3365 } 3366 UseTransparentHugePages = false; 3367 } 3368 3369 if (UseHugeTLBFS) { 3370 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS); 3371 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) { 3372 UseSHM = false; 3373 return true; 3374 } 3375 UseHugeTLBFS = false; 3376 } 3377 3378 return UseSHM; 3379} 3380 3381void os::large_page_init() { 3382 if (!UseLargePages && 3383 !UseTransparentHugePages && 3384 !UseHugeTLBFS && 3385 !UseSHM) { 3386 // Not using large pages. 3387 return; 3388 } 3389 3390 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) { 3391 // The user explicitly turned off large pages. 3392 // Ignore the rest of the large pages flags. 3393 UseTransparentHugePages = false; 3394 UseHugeTLBFS = false; 3395 UseSHM = false; 3396 return; 3397 } 3398 3399 size_t large_page_size = Linux::setup_large_page_size(); 3400 UseLargePages = Linux::setup_large_page_type(large_page_size); 3401 3402 set_coredump_filter(); 3403} 3404 3405#ifndef SHM_HUGETLB 3406 #define SHM_HUGETLB 04000 3407#endif 3408 3409char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, 3410 char* req_addr, bool exec) { 3411 // "exec" is passed in but not used. Creating the shared image for 3412 // the code cache doesn't have an SHM_X executable permission to check. 3413 assert(UseLargePages && UseSHM, "only for SHM large pages"); 3414 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3415 3416 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) { 3417 return NULL; // Fallback to small pages. 3418 } 3419 3420 key_t key = IPC_PRIVATE; 3421 char *addr; 3422 3423 bool warn_on_failure = UseLargePages && 3424 (!FLAG_IS_DEFAULT(UseLargePages) || 3425 !FLAG_IS_DEFAULT(UseSHM) || 3426 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3427 char msg[128]; 3428 3429 // Create a large shared memory region to attach to based on size. 3430 // Currently, size is the total size of the heap 3431 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W); 3432 if (shmid == -1) { 3433 // Possible reasons for shmget failure: 3434 // 1. shmmax is too small for Java heap. 3435 // > check shmmax value: cat /proc/sys/kernel/shmmax 3436 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax 3437 // 2. not enough large page memory. 3438 // > check available large pages: cat /proc/meminfo 3439 // > increase amount of large pages: 3440 // echo new_value > /proc/sys/vm/nr_hugepages 3441 // Note 1: different Linux may use different name for this property, 3442 // e.g. on Redhat AS-3 it is "hugetlb_pool". 3443 // Note 2: it's possible there's enough physical memory available but 3444 // they are so fragmented after a long run that they can't 3445 // coalesce into large pages. Try to reserve large pages when 3446 // the system is still "fresh". 3447 if (warn_on_failure) { 3448 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno); 3449 warning("%s", msg); 3450 } 3451 return NULL; 3452 } 3453 3454 // attach to the region 3455 addr = (char*)shmat(shmid, req_addr, 0); 3456 int err = errno; 3457 3458 // Remove shmid. If shmat() is successful, the actual shared memory segment 3459 // will be deleted when it's detached by shmdt() or when the process 3460 // terminates. If shmat() is not successful this will remove the shared 3461 // segment immediately. 3462 shmctl(shmid, IPC_RMID, NULL); 3463 3464 if ((intptr_t)addr == -1) { 3465 if (warn_on_failure) { 3466 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err); 3467 warning("%s", msg); 3468 } 3469 return NULL; 3470 } 3471 3472 return addr; 3473} 3474 3475static void warn_on_large_pages_failure(char* req_addr, size_t bytes, 3476 int error) { 3477 assert(error == ENOMEM, "Only expect to fail if no memory is available"); 3478 3479 bool warn_on_failure = UseLargePages && 3480 (!FLAG_IS_DEFAULT(UseLargePages) || 3481 !FLAG_IS_DEFAULT(UseHugeTLBFS) || 3482 !FLAG_IS_DEFAULT(LargePageSizeInBytes)); 3483 3484 if (warn_on_failure) { 3485 char msg[128]; 3486 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: " 3487 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error); 3488 warning("%s", msg); 3489 } 3490} 3491 3492char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, 3493 char* req_addr, 3494 bool exec) { 3495 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3496 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size"); 3497 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address"); 3498 3499 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3500 char* addr = (char*)::mmap(req_addr, bytes, prot, 3501 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB, 3502 -1, 0); 3503 3504 if (addr == MAP_FAILED) { 3505 warn_on_large_pages_failure(req_addr, bytes, errno); 3506 return NULL; 3507 } 3508 3509 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be"); 3510 3511 return addr; 3512} 3513 3514// Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed(). 3515// Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address 3516// (req_addr != NULL) or with a given alignment. 3517// - bytes shall be a multiple of alignment. 3518// - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3519// - alignment sets the alignment at which memory shall be allocated. 3520// It must be a multiple of allocation granularity. 3521// Returns address of memory or NULL. If req_addr was not NULL, will only return 3522// req_addr or NULL. 3523static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) { 3524 3525 size_t extra_size = bytes; 3526 if (req_addr == NULL && alignment > 0) { 3527 extra_size += alignment; 3528 } 3529 3530 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE, 3531 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 3532 -1, 0); 3533 if (start == MAP_FAILED) { 3534 start = NULL; 3535 } else { 3536 if (req_addr != NULL) { 3537 if (start != req_addr) { 3538 ::munmap(start, extra_size); 3539 start = NULL; 3540 } 3541 } else { 3542 char* const start_aligned = (char*) align_ptr_up(start, alignment); 3543 char* const end_aligned = start_aligned + bytes; 3544 char* const end = start + extra_size; 3545 if (start_aligned > start) { 3546 ::munmap(start, start_aligned - start); 3547 } 3548 if (end_aligned < end) { 3549 ::munmap(end_aligned, end - end_aligned); 3550 } 3551 start = start_aligned; 3552 } 3553 } 3554 return start; 3555 3556} 3557 3558// Reserve memory using mmap(MAP_HUGETLB). 3559// - bytes shall be a multiple of alignment. 3560// - req_addr can be NULL. If not NULL, it must be a multiple of alignment. 3561// - alignment sets the alignment at which memory shall be allocated. 3562// It must be a multiple of allocation granularity. 3563// Returns address of memory or NULL. If req_addr was not NULL, will only return 3564// req_addr or NULL. 3565char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, 3566 size_t alignment, 3567 char* req_addr, 3568 bool exec) { 3569 size_t large_page_size = os::large_page_size(); 3570 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes"); 3571 3572 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3573 assert(is_size_aligned(bytes, alignment), "Must be"); 3574 3575 // First reserve - but not commit - the address range in small pages. 3576 char* const start = anon_mmap_aligned(bytes, alignment, req_addr); 3577 3578 if (start == NULL) { 3579 return NULL; 3580 } 3581 3582 assert(is_ptr_aligned(start, alignment), "Must be"); 3583 3584 char* end = start + bytes; 3585 3586 // Find the regions of the allocated chunk that can be promoted to large pages. 3587 char* lp_start = (char*)align_ptr_up(start, large_page_size); 3588 char* lp_end = (char*)align_ptr_down(end, large_page_size); 3589 3590 size_t lp_bytes = lp_end - lp_start; 3591 3592 assert(is_size_aligned(lp_bytes, large_page_size), "Must be"); 3593 3594 if (lp_bytes == 0) { 3595 // The mapped region doesn't even span the start and the end of a large page. 3596 // Fall back to allocate a non-special area. 3597 ::munmap(start, end - start); 3598 return NULL; 3599 } 3600 3601 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE; 3602 3603 void* result; 3604 3605 // Commit small-paged leading area. 3606 if (start != lp_start) { 3607 result = ::mmap(start, lp_start - start, prot, 3608 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3609 -1, 0); 3610 if (result == MAP_FAILED) { 3611 ::munmap(lp_start, end - lp_start); 3612 return NULL; 3613 } 3614 } 3615 3616 // Commit large-paged area. 3617 result = ::mmap(lp_start, lp_bytes, prot, 3618 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB, 3619 -1, 0); 3620 if (result == MAP_FAILED) { 3621 warn_on_large_pages_failure(lp_start, lp_bytes, errno); 3622 // If the mmap above fails, the large pages region will be unmapped and we 3623 // have regions before and after with small pages. Release these regions. 3624 // 3625 // | mapped | unmapped | mapped | 3626 // ^ ^ ^ ^ 3627 // start lp_start lp_end end 3628 // 3629 ::munmap(start, lp_start - start); 3630 ::munmap(lp_end, end - lp_end); 3631 return NULL; 3632 } 3633 3634 // Commit small-paged trailing area. 3635 if (lp_end != end) { 3636 result = ::mmap(lp_end, end - lp_end, prot, 3637 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, 3638 -1, 0); 3639 if (result == MAP_FAILED) { 3640 ::munmap(start, lp_end - start); 3641 return NULL; 3642 } 3643 } 3644 3645 return start; 3646} 3647 3648char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, 3649 size_t alignment, 3650 char* req_addr, 3651 bool exec) { 3652 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages"); 3653 assert(is_ptr_aligned(req_addr, alignment), "Must be"); 3654 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be"); 3655 assert(is_power_of_2(os::large_page_size()), "Must be"); 3656 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes"); 3657 3658 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) { 3659 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec); 3660 } else { 3661 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec); 3662 } 3663} 3664 3665char* os::reserve_memory_special(size_t bytes, size_t alignment, 3666 char* req_addr, bool exec) { 3667 assert(UseLargePages, "only for large pages"); 3668 3669 char* addr; 3670 if (UseSHM) { 3671 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec); 3672 } else { 3673 assert(UseHugeTLBFS, "must be"); 3674 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec); 3675 } 3676 3677 if (addr != NULL) { 3678 if (UseNUMAInterleaving) { 3679 numa_make_global(addr, bytes); 3680 } 3681 3682 // The memory is committed 3683 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC); 3684 } 3685 3686 return addr; 3687} 3688 3689bool os::Linux::release_memory_special_shm(char* base, size_t bytes) { 3690 // detaching the SHM segment will also delete it, see reserve_memory_special_shm() 3691 return shmdt(base) == 0; 3692} 3693 3694bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) { 3695 return pd_release_memory(base, bytes); 3696} 3697 3698bool os::release_memory_special(char* base, size_t bytes) { 3699 bool res; 3700 if (MemTracker::tracking_level() > NMT_minimal) { 3701 Tracker tkr = MemTracker::get_virtual_memory_release_tracker(); 3702 res = os::Linux::release_memory_special_impl(base, bytes); 3703 if (res) { 3704 tkr.record((address)base, bytes); 3705 } 3706 3707 } else { 3708 res = os::Linux::release_memory_special_impl(base, bytes); 3709 } 3710 return res; 3711} 3712 3713bool os::Linux::release_memory_special_impl(char* base, size_t bytes) { 3714 assert(UseLargePages, "only for large pages"); 3715 bool res; 3716 3717 if (UseSHM) { 3718 res = os::Linux::release_memory_special_shm(base, bytes); 3719 } else { 3720 assert(UseHugeTLBFS, "must be"); 3721 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes); 3722 } 3723 return res; 3724} 3725 3726size_t os::large_page_size() { 3727 return _large_page_size; 3728} 3729 3730// With SysV SHM the entire memory region must be allocated as shared 3731// memory. 3732// HugeTLBFS allows application to commit large page memory on demand. 3733// However, when committing memory with HugeTLBFS fails, the region 3734// that was supposed to be committed will lose the old reservation 3735// and allow other threads to steal that memory region. Because of this 3736// behavior we can't commit HugeTLBFS memory. 3737bool os::can_commit_large_page_memory() { 3738 return UseTransparentHugePages; 3739} 3740 3741bool os::can_execute_large_page_memory() { 3742 return UseTransparentHugePages || UseHugeTLBFS; 3743} 3744 3745// Reserve memory at an arbitrary address, only if that area is 3746// available (and not reserved for something else). 3747 3748char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) { 3749 const int max_tries = 10; 3750 char* base[max_tries]; 3751 size_t size[max_tries]; 3752 const size_t gap = 0x000000; 3753 3754 // Assert only that the size is a multiple of the page size, since 3755 // that's all that mmap requires, and since that's all we really know 3756 // about at this low abstraction level. If we need higher alignment, 3757 // we can either pass an alignment to this method or verify alignment 3758 // in one of the methods further up the call chain. See bug 5044738. 3759 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block"); 3760 3761 // Repeatedly allocate blocks until the block is allocated at the 3762 // right spot. Give up after max_tries. Note that reserve_memory() will 3763 // automatically update _highest_vm_reserved_address if the call is 3764 // successful. The variable tracks the highest memory address every reserved 3765 // by JVM. It is used to detect heap-stack collision if running with 3766 // fixed-stack LinuxThreads. Because here we may attempt to reserve more 3767 // space than needed, it could confuse the collision detecting code. To 3768 // solve the problem, save current _highest_vm_reserved_address and 3769 // calculate the correct value before return. 3770 address old_highest = _highest_vm_reserved_address; 3771 3772 // Linux mmap allows caller to pass an address as hint; give it a try first, 3773 // if kernel honors the hint then we can return immediately. 3774 char * addr = anon_mmap(requested_addr, bytes, false); 3775 if (addr == requested_addr) { 3776 return requested_addr; 3777 } 3778 3779 if (addr != NULL) { 3780 // mmap() is successful but it fails to reserve at the requested address 3781 anon_munmap(addr, bytes); 3782 } 3783 3784 int i; 3785 for (i = 0; i < max_tries; ++i) { 3786 base[i] = reserve_memory(bytes); 3787 3788 if (base[i] != NULL) { 3789 // Is this the block we wanted? 3790 if (base[i] == requested_addr) { 3791 size[i] = bytes; 3792 break; 3793 } 3794 3795 // Does this overlap the block we wanted? Give back the overlapped 3796 // parts and try again. 3797 3798 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i]; 3799 if (top_overlap >= 0 && (size_t)top_overlap < bytes) { 3800 unmap_memory(base[i], top_overlap); 3801 base[i] += top_overlap; 3802 size[i] = bytes - top_overlap; 3803 } else { 3804 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr; 3805 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) { 3806 unmap_memory(requested_addr, bottom_overlap); 3807 size[i] = bytes - bottom_overlap; 3808 } else { 3809 size[i] = bytes; 3810 } 3811 } 3812 } 3813 } 3814 3815 // Give back the unused reserved pieces. 3816 3817 for (int j = 0; j < i; ++j) { 3818 if (base[j] != NULL) { 3819 unmap_memory(base[j], size[j]); 3820 } 3821 } 3822 3823 if (i < max_tries) { 3824 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes); 3825 return requested_addr; 3826 } else { 3827 _highest_vm_reserved_address = old_highest; 3828 return NULL; 3829 } 3830} 3831 3832size_t os::read(int fd, void *buf, unsigned int nBytes) { 3833 return ::read(fd, buf, nBytes); 3834} 3835 3836size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) { 3837 return ::pread(fd, buf, nBytes, offset); 3838} 3839 3840// Short sleep, direct OS call. 3841// 3842// Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee 3843// sched_yield(2) will actually give up the CPU: 3844// 3845// * Alone on this pariticular CPU, keeps running. 3846// * Before the introduction of "skip_buddy" with "compat_yield" disabled 3847// (pre 2.6.39). 3848// 3849// So calling this with 0 is an alternative. 3850// 3851void os::naked_short_sleep(jlong ms) { 3852 struct timespec req; 3853 3854 assert(ms < 1000, "Un-interruptable sleep, short time use only"); 3855 req.tv_sec = 0; 3856 if (ms > 0) { 3857 req.tv_nsec = (ms % 1000) * 1000000; 3858 } else { 3859 req.tv_nsec = 1; 3860 } 3861 3862 nanosleep(&req, NULL); 3863 3864 return; 3865} 3866 3867// Sleep forever; naked call to OS-specific sleep; use with CAUTION 3868void os::infinite_sleep() { 3869 while (true) { // sleep forever ... 3870 ::sleep(100); // ... 100 seconds at a time 3871 } 3872} 3873 3874// Used to convert frequent JVM_Yield() to nops 3875bool os::dont_yield() { 3876 return DontYieldALot; 3877} 3878 3879void os::naked_yield() { 3880 sched_yield(); 3881} 3882 3883//////////////////////////////////////////////////////////////////////////////// 3884// thread priority support 3885 3886// Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER 3887// only supports dynamic priority, static priority must be zero. For real-time 3888// applications, Linux supports SCHED_RR which allows static priority (1-99). 3889// However, for large multi-threaded applications, SCHED_RR is not only slower 3890// than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out 3891// of 5 runs - Sep 2005). 3892// 3893// The following code actually changes the niceness of kernel-thread/LWP. It 3894// has an assumption that setpriority() only modifies one kernel-thread/LWP, 3895// not the entire user process, and user level threads are 1:1 mapped to kernel 3896// threads. It has always been the case, but could change in the future. For 3897// this reason, the code should not be used as default (ThreadPriorityPolicy=0). 3898// It is only used when ThreadPriorityPolicy=1 and requires root privilege. 3899 3900int os::java_to_os_priority[CriticalPriority + 1] = { 3901 19, // 0 Entry should never be used 3902 3903 4, // 1 MinPriority 3904 3, // 2 3905 2, // 3 3906 3907 1, // 4 3908 0, // 5 NormPriority 3909 -1, // 6 3910 3911 -2, // 7 3912 -3, // 8 3913 -4, // 9 NearMaxPriority 3914 3915 -5, // 10 MaxPriority 3916 3917 -5 // 11 CriticalPriority 3918}; 3919 3920static int prio_init() { 3921 if (ThreadPriorityPolicy == 1) { 3922 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1 3923 // if effective uid is not root. Perhaps, a more elegant way of doing 3924 // this is to test CAP_SYS_NICE capability, but that will require libcap.so 3925 if (geteuid() != 0) { 3926 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) { 3927 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux"); 3928 } 3929 ThreadPriorityPolicy = 0; 3930 } 3931 } 3932 if (UseCriticalJavaThreadPriority) { 3933 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority]; 3934 } 3935 return 0; 3936} 3937 3938OSReturn os::set_native_priority(Thread* thread, int newpri) { 3939 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK; 3940 3941 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri); 3942 return (ret == 0) ? OS_OK : OS_ERR; 3943} 3944 3945OSReturn os::get_native_priority(const Thread* const thread, 3946 int *priority_ptr) { 3947 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) { 3948 *priority_ptr = java_to_os_priority[NormPriority]; 3949 return OS_OK; 3950 } 3951 3952 errno = 0; 3953 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id()); 3954 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR); 3955} 3956 3957// Hint to the underlying OS that a task switch would not be good. 3958// Void return because it's a hint and can fail. 3959void os::hint_no_preempt() {} 3960 3961//////////////////////////////////////////////////////////////////////////////// 3962// suspend/resume support 3963 3964// the low-level signal-based suspend/resume support is a remnant from the 3965// old VM-suspension that used to be for java-suspension, safepoints etc, 3966// within hotspot. Now there is a single use-case for this: 3967// - calling get_thread_pc() on the VMThread by the flat-profiler task 3968// that runs in the watcher thread. 3969// The remaining code is greatly simplified from the more general suspension 3970// code that used to be used. 3971// 3972// The protocol is quite simple: 3973// - suspend: 3974// - sends a signal to the target thread 3975// - polls the suspend state of the osthread using a yield loop 3976// - target thread signal handler (SR_handler) sets suspend state 3977// and blocks in sigsuspend until continued 3978// - resume: 3979// - sets target osthread state to continue 3980// - sends signal to end the sigsuspend loop in the SR_handler 3981// 3982// Note that the SR_lock plays no role in this suspend/resume protocol. 3983 3984static void resume_clear_context(OSThread *osthread) { 3985 osthread->set_ucontext(NULL); 3986 osthread->set_siginfo(NULL); 3987} 3988 3989static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, 3990 ucontext_t* context) { 3991 osthread->set_ucontext(context); 3992 osthread->set_siginfo(siginfo); 3993} 3994 3995// Handler function invoked when a thread's execution is suspended or 3996// resumed. We have to be careful that only async-safe functions are 3997// called here (Note: most pthread functions are not async safe and 3998// should be avoided.) 3999// 4000// Note: sigwait() is a more natural fit than sigsuspend() from an 4001// interface point of view, but sigwait() prevents the signal hander 4002// from being run. libpthread would get very confused by not having 4003// its signal handlers run and prevents sigwait()'s use with the 4004// mutex granting granting signal. 4005// 4006// Currently only ever called on the VMThread and JavaThreads (PC sampling) 4007// 4008static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) { 4009 // Save and restore errno to avoid confusing native code with EINTR 4010 // after sigsuspend. 4011 int old_errno = errno; 4012 4013 Thread* thread = Thread::current(); 4014 OSThread* osthread = thread->osthread(); 4015 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread"); 4016 4017 os::SuspendResume::State current = osthread->sr.state(); 4018 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) { 4019 suspend_save_context(osthread, siginfo, context); 4020 4021 // attempt to switch the state, we assume we had a SUSPEND_REQUEST 4022 os::SuspendResume::State state = osthread->sr.suspended(); 4023 if (state == os::SuspendResume::SR_SUSPENDED) { 4024 sigset_t suspend_set; // signals for sigsuspend() 4025 4026 // get current set of blocked signals and unblock resume signal 4027 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set); 4028 sigdelset(&suspend_set, SR_signum); 4029 4030 sr_semaphore.signal(); 4031 // wait here until we are resumed 4032 while (1) { 4033 sigsuspend(&suspend_set); 4034 4035 os::SuspendResume::State result = osthread->sr.running(); 4036 if (result == os::SuspendResume::SR_RUNNING) { 4037 sr_semaphore.signal(); 4038 break; 4039 } 4040 } 4041 4042 } else if (state == os::SuspendResume::SR_RUNNING) { 4043 // request was cancelled, continue 4044 } else { 4045 ShouldNotReachHere(); 4046 } 4047 4048 resume_clear_context(osthread); 4049 } else if (current == os::SuspendResume::SR_RUNNING) { 4050 // request was cancelled, continue 4051 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) { 4052 // ignore 4053 } else { 4054 // ignore 4055 } 4056 4057 errno = old_errno; 4058} 4059 4060 4061static int SR_initialize() { 4062 struct sigaction act; 4063 char *s; 4064 // Get signal number to use for suspend/resume 4065 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) { 4066 int sig = ::strtol(s, 0, 10); 4067 if (sig > 0 || sig < _NSIG) { 4068 SR_signum = sig; 4069 } 4070 } 4071 4072 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS, 4073 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769"); 4074 4075 sigemptyset(&SR_sigset); 4076 sigaddset(&SR_sigset, SR_signum); 4077 4078 // Set up signal handler for suspend/resume 4079 act.sa_flags = SA_RESTART|SA_SIGINFO; 4080 act.sa_handler = (void (*)(int)) SR_handler; 4081 4082 // SR_signum is blocked by default. 4083 // 4528190 - We also need to block pthread restart signal (32 on all 4084 // supported Linux platforms). Note that LinuxThreads need to block 4085 // this signal for all threads to work properly. So we don't have 4086 // to use hard-coded signal number when setting up the mask. 4087 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask); 4088 4089 if (sigaction(SR_signum, &act, 0) == -1) { 4090 return -1; 4091 } 4092 4093 // Save signal flag 4094 os::Linux::set_our_sigflags(SR_signum, act.sa_flags); 4095 return 0; 4096} 4097 4098static int sr_notify(OSThread* osthread) { 4099 int status = pthread_kill(osthread->pthread_id(), SR_signum); 4100 assert_status(status == 0, status, "pthread_kill"); 4101 return status; 4102} 4103 4104// "Randomly" selected value for how long we want to spin 4105// before bailing out on suspending a thread, also how often 4106// we send a signal to a thread we want to resume 4107static const int RANDOMLY_LARGE_INTEGER = 1000000; 4108static const int RANDOMLY_LARGE_INTEGER2 = 100; 4109 4110// returns true on success and false on error - really an error is fatal 4111// but this seems the normal response to library errors 4112static bool do_suspend(OSThread* osthread) { 4113 assert(osthread->sr.is_running(), "thread should be running"); 4114 assert(!sr_semaphore.trywait(), "semaphore has invalid state"); 4115 4116 // mark as suspended and send signal 4117 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) { 4118 // failed to switch, state wasn't running? 4119 ShouldNotReachHere(); 4120 return false; 4121 } 4122 4123 if (sr_notify(osthread) != 0) { 4124 ShouldNotReachHere(); 4125 } 4126 4127 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED 4128 while (true) { 4129 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4130 break; 4131 } else { 4132 // timeout 4133 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend(); 4134 if (cancelled == os::SuspendResume::SR_RUNNING) { 4135 return false; 4136 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) { 4137 // make sure that we consume the signal on the semaphore as well 4138 sr_semaphore.wait(); 4139 break; 4140 } else { 4141 ShouldNotReachHere(); 4142 return false; 4143 } 4144 } 4145 } 4146 4147 guarantee(osthread->sr.is_suspended(), "Must be suspended"); 4148 return true; 4149} 4150 4151static void do_resume(OSThread* osthread) { 4152 assert(osthread->sr.is_suspended(), "thread should be suspended"); 4153 assert(!sr_semaphore.trywait(), "invalid semaphore state"); 4154 4155 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) { 4156 // failed to switch to WAKEUP_REQUEST 4157 ShouldNotReachHere(); 4158 return; 4159 } 4160 4161 while (true) { 4162 if (sr_notify(osthread) == 0) { 4163 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) { 4164 if (osthread->sr.is_running()) { 4165 return; 4166 } 4167 } 4168 } else { 4169 ShouldNotReachHere(); 4170 } 4171 } 4172 4173 guarantee(osthread->sr.is_running(), "Must be running!"); 4174} 4175 4176/////////////////////////////////////////////////////////////////////////////////// 4177// signal handling (except suspend/resume) 4178 4179// This routine may be used by user applications as a "hook" to catch signals. 4180// The user-defined signal handler must pass unrecognized signals to this 4181// routine, and if it returns true (non-zero), then the signal handler must 4182// return immediately. If the flag "abort_if_unrecognized" is true, then this 4183// routine will never retun false (zero), but instead will execute a VM panic 4184// routine kill the process. 4185// 4186// If this routine returns false, it is OK to call it again. This allows 4187// the user-defined signal handler to perform checks either before or after 4188// the VM performs its own checks. Naturally, the user code would be making 4189// a serious error if it tried to handle an exception (such as a null check 4190// or breakpoint) that the VM was generating for its own correct operation. 4191// 4192// This routine may recognize any of the following kinds of signals: 4193// SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1. 4194// It should be consulted by handlers for any of those signals. 4195// 4196// The caller of this routine must pass in the three arguments supplied 4197// to the function referred to in the "sa_sigaction" (not the "sa_handler") 4198// field of the structure passed to sigaction(). This routine assumes that 4199// the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART. 4200// 4201// Note that the VM will print warnings if it detects conflicting signal 4202// handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers". 4203// 4204extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo, 4205 siginfo_t* siginfo, 4206 void* ucontext, 4207 int abort_if_unrecognized); 4208 4209void signalHandler(int sig, siginfo_t* info, void* uc) { 4210 assert(info != NULL && uc != NULL, "it must be old kernel"); 4211 int orig_errno = errno; // Preserve errno value over signal handler. 4212 JVM_handle_linux_signal(sig, info, uc, true); 4213 errno = orig_errno; 4214} 4215 4216 4217// This boolean allows users to forward their own non-matching signals 4218// to JVM_handle_linux_signal, harmlessly. 4219bool os::Linux::signal_handlers_are_installed = false; 4220 4221// For signal-chaining 4222struct sigaction os::Linux::sigact[MAXSIGNUM]; 4223unsigned int os::Linux::sigs = 0; 4224bool os::Linux::libjsig_is_loaded = false; 4225typedef struct sigaction *(*get_signal_t)(int); 4226get_signal_t os::Linux::get_signal_action = NULL; 4227 4228struct sigaction* os::Linux::get_chained_signal_action(int sig) { 4229 struct sigaction *actp = NULL; 4230 4231 if (libjsig_is_loaded) { 4232 // Retrieve the old signal handler from libjsig 4233 actp = (*get_signal_action)(sig); 4234 } 4235 if (actp == NULL) { 4236 // Retrieve the preinstalled signal handler from jvm 4237 actp = get_preinstalled_handler(sig); 4238 } 4239 4240 return actp; 4241} 4242 4243static bool call_chained_handler(struct sigaction *actp, int sig, 4244 siginfo_t *siginfo, void *context) { 4245 // Call the old signal handler 4246 if (actp->sa_handler == SIG_DFL) { 4247 // It's more reasonable to let jvm treat it as an unexpected exception 4248 // instead of taking the default action. 4249 return false; 4250 } else if (actp->sa_handler != SIG_IGN) { 4251 if ((actp->sa_flags & SA_NODEFER) == 0) { 4252 // automaticlly block the signal 4253 sigaddset(&(actp->sa_mask), sig); 4254 } 4255 4256 sa_handler_t hand; 4257 sa_sigaction_t sa; 4258 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0; 4259 // retrieve the chained handler 4260 if (siginfo_flag_set) { 4261 sa = actp->sa_sigaction; 4262 } else { 4263 hand = actp->sa_handler; 4264 } 4265 4266 if ((actp->sa_flags & SA_RESETHAND) != 0) { 4267 actp->sa_handler = SIG_DFL; 4268 } 4269 4270 // try to honor the signal mask 4271 sigset_t oset; 4272 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset); 4273 4274 // call into the chained handler 4275 if (siginfo_flag_set) { 4276 (*sa)(sig, siginfo, context); 4277 } else { 4278 (*hand)(sig); 4279 } 4280 4281 // restore the signal mask 4282 pthread_sigmask(SIG_SETMASK, &oset, 0); 4283 } 4284 // Tell jvm's signal handler the signal is taken care of. 4285 return true; 4286} 4287 4288bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) { 4289 bool chained = false; 4290 // signal-chaining 4291 if (UseSignalChaining) { 4292 struct sigaction *actp = get_chained_signal_action(sig); 4293 if (actp != NULL) { 4294 chained = call_chained_handler(actp, sig, siginfo, context); 4295 } 4296 } 4297 return chained; 4298} 4299 4300struct sigaction* os::Linux::get_preinstalled_handler(int sig) { 4301 if ((((unsigned int)1 << sig) & sigs) != 0) { 4302 return &sigact[sig]; 4303 } 4304 return NULL; 4305} 4306 4307void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) { 4308 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4309 sigact[sig] = oldAct; 4310 sigs |= (unsigned int)1 << sig; 4311} 4312 4313// for diagnostic 4314int os::Linux::sigflags[MAXSIGNUM]; 4315 4316int os::Linux::get_our_sigflags(int sig) { 4317 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4318 return sigflags[sig]; 4319} 4320 4321void os::Linux::set_our_sigflags(int sig, int flags) { 4322 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4323 sigflags[sig] = flags; 4324} 4325 4326void os::Linux::set_signal_handler(int sig, bool set_installed) { 4327 // Check for overwrite. 4328 struct sigaction oldAct; 4329 sigaction(sig, (struct sigaction*)NULL, &oldAct); 4330 4331 void* oldhand = oldAct.sa_sigaction 4332 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4333 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4334 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) && 4335 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) && 4336 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) { 4337 if (AllowUserSignalHandlers || !set_installed) { 4338 // Do not overwrite; user takes responsibility to forward to us. 4339 return; 4340 } else if (UseSignalChaining) { 4341 // save the old handler in jvm 4342 save_preinstalled_handler(sig, oldAct); 4343 // libjsig also interposes the sigaction() call below and saves the 4344 // old sigaction on it own. 4345 } else { 4346 fatal(err_msg("Encountered unexpected pre-existing sigaction handler " 4347 "%#lx for signal %d.", (long)oldhand, sig)); 4348 } 4349 } 4350 4351 struct sigaction sigAct; 4352 sigfillset(&(sigAct.sa_mask)); 4353 sigAct.sa_handler = SIG_DFL; 4354 if (!set_installed) { 4355 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4356 } else { 4357 sigAct.sa_sigaction = signalHandler; 4358 sigAct.sa_flags = SA_SIGINFO|SA_RESTART; 4359 } 4360 // Save flags, which are set by ours 4361 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range"); 4362 sigflags[sig] = sigAct.sa_flags; 4363 4364 int ret = sigaction(sig, &sigAct, &oldAct); 4365 assert(ret == 0, "check"); 4366 4367 void* oldhand2 = oldAct.sa_sigaction 4368 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction) 4369 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler); 4370 assert(oldhand2 == oldhand, "no concurrent signal handler installation"); 4371} 4372 4373// install signal handlers for signals that HotSpot needs to 4374// handle in order to support Java-level exception handling. 4375 4376void os::Linux::install_signal_handlers() { 4377 if (!signal_handlers_are_installed) { 4378 signal_handlers_are_installed = true; 4379 4380 // signal-chaining 4381 typedef void (*signal_setting_t)(); 4382 signal_setting_t begin_signal_setting = NULL; 4383 signal_setting_t end_signal_setting = NULL; 4384 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4385 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting")); 4386 if (begin_signal_setting != NULL) { 4387 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t, 4388 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting")); 4389 get_signal_action = CAST_TO_FN_PTR(get_signal_t, 4390 dlsym(RTLD_DEFAULT, "JVM_get_signal_action")); 4391 libjsig_is_loaded = true; 4392 assert(UseSignalChaining, "should enable signal-chaining"); 4393 } 4394 if (libjsig_is_loaded) { 4395 // Tell libjsig jvm is setting signal handlers 4396 (*begin_signal_setting)(); 4397 } 4398 4399 set_signal_handler(SIGSEGV, true); 4400 set_signal_handler(SIGPIPE, true); 4401 set_signal_handler(SIGBUS, true); 4402 set_signal_handler(SIGILL, true); 4403 set_signal_handler(SIGFPE, true); 4404#if defined(PPC64) 4405 set_signal_handler(SIGTRAP, true); 4406#endif 4407 set_signal_handler(SIGXFSZ, true); 4408 4409 if (libjsig_is_loaded) { 4410 // Tell libjsig jvm finishes setting signal handlers 4411 (*end_signal_setting)(); 4412 } 4413 4414 // We don't activate signal checker if libjsig is in place, we trust ourselves 4415 // and if UserSignalHandler is installed all bets are off. 4416 // Log that signal checking is off only if -verbose:jni is specified. 4417 if (CheckJNICalls) { 4418 if (libjsig_is_loaded) { 4419 if (PrintJNIResolving) { 4420 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled"); 4421 } 4422 check_signals = false; 4423 } 4424 if (AllowUserSignalHandlers) { 4425 if (PrintJNIResolving) { 4426 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled"); 4427 } 4428 check_signals = false; 4429 } 4430 } 4431 } 4432} 4433 4434// This is the fastest way to get thread cpu time on Linux. 4435// Returns cpu time (user+sys) for any thread, not only for current. 4436// POSIX compliant clocks are implemented in the kernels 2.6.16+. 4437// It might work on 2.6.10+ with a special kernel/glibc patch. 4438// For reference, please, see IEEE Std 1003.1-2004: 4439// http://www.unix.org/single_unix_specification 4440 4441jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) { 4442 struct timespec tp; 4443 int rc = os::Linux::clock_gettime(clockid, &tp); 4444 assert(rc == 0, "clock_gettime is expected to return 0 code"); 4445 4446 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec; 4447} 4448 4449///// 4450// glibc on Linux platform uses non-documented flag 4451// to indicate, that some special sort of signal 4452// trampoline is used. 4453// We will never set this flag, and we should 4454// ignore this flag in our diagnostic 4455#ifdef SIGNIFICANT_SIGNAL_MASK 4456 #undef SIGNIFICANT_SIGNAL_MASK 4457#endif 4458#define SIGNIFICANT_SIGNAL_MASK (~0x04000000) 4459 4460static const char* get_signal_handler_name(address handler, 4461 char* buf, int buflen) { 4462 int offset; 4463 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset); 4464 if (found) { 4465 // skip directory names 4466 const char *p1, *p2; 4467 p1 = buf; 4468 size_t len = strlen(os::file_separator()); 4469 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len; 4470 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset); 4471 } else { 4472 jio_snprintf(buf, buflen, PTR_FORMAT, handler); 4473 } 4474 return buf; 4475} 4476 4477static void print_signal_handler(outputStream* st, int sig, 4478 char* buf, size_t buflen) { 4479 struct sigaction sa; 4480 4481 sigaction(sig, NULL, &sa); 4482 4483 // See comment for SIGNIFICANT_SIGNAL_MASK define 4484 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4485 4486 st->print("%s: ", os::exception_name(sig, buf, buflen)); 4487 4488 address handler = (sa.sa_flags & SA_SIGINFO) 4489 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction) 4490 : CAST_FROM_FN_PTR(address, sa.sa_handler); 4491 4492 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) { 4493 st->print("SIG_DFL"); 4494 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) { 4495 st->print("SIG_IGN"); 4496 } else { 4497 st->print("[%s]", get_signal_handler_name(handler, buf, buflen)); 4498 } 4499 4500 st->print(", sa_mask[0]="); 4501 os::Posix::print_signal_set_short(st, &sa.sa_mask); 4502 4503 address rh = VMError::get_resetted_sighandler(sig); 4504 // May be, handler was resetted by VMError? 4505 if (rh != NULL) { 4506 handler = rh; 4507 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK; 4508 } 4509 4510 st->print(", sa_flags="); 4511 os::Posix::print_sa_flags(st, sa.sa_flags); 4512 4513 // Check: is it our handler? 4514 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) || 4515 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) { 4516 // It is our signal handler 4517 // check for flags, reset system-used one! 4518 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) { 4519 st->print( 4520 ", flags was changed from " PTR32_FORMAT ", consider using jsig library", 4521 os::Linux::get_our_sigflags(sig)); 4522 } 4523 } 4524 st->cr(); 4525} 4526 4527 4528#define DO_SIGNAL_CHECK(sig) \ 4529 do { \ 4530 if (!sigismember(&check_signal_done, sig)) { \ 4531 os::Linux::check_signal_handler(sig); \ 4532 } \ 4533 } while (0) 4534 4535// This method is a periodic task to check for misbehaving JNI applications 4536// under CheckJNI, we can add any periodic checks here 4537 4538void os::run_periodic_checks() { 4539 if (check_signals == false) return; 4540 4541 // SEGV and BUS if overridden could potentially prevent 4542 // generation of hs*.log in the event of a crash, debugging 4543 // such a case can be very challenging, so we absolutely 4544 // check the following for a good measure: 4545 DO_SIGNAL_CHECK(SIGSEGV); 4546 DO_SIGNAL_CHECK(SIGILL); 4547 DO_SIGNAL_CHECK(SIGFPE); 4548 DO_SIGNAL_CHECK(SIGBUS); 4549 DO_SIGNAL_CHECK(SIGPIPE); 4550 DO_SIGNAL_CHECK(SIGXFSZ); 4551#if defined(PPC64) 4552 DO_SIGNAL_CHECK(SIGTRAP); 4553#endif 4554 4555 // ReduceSignalUsage allows the user to override these handlers 4556 // see comments at the very top and jvm_solaris.h 4557 if (!ReduceSignalUsage) { 4558 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL); 4559 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL); 4560 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL); 4561 DO_SIGNAL_CHECK(BREAK_SIGNAL); 4562 } 4563 4564 DO_SIGNAL_CHECK(SR_signum); 4565 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL); 4566} 4567 4568typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *); 4569 4570static os_sigaction_t os_sigaction = NULL; 4571 4572void os::Linux::check_signal_handler(int sig) { 4573 char buf[O_BUFLEN]; 4574 address jvmHandler = NULL; 4575 4576 4577 struct sigaction act; 4578 if (os_sigaction == NULL) { 4579 // only trust the default sigaction, in case it has been interposed 4580 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction"); 4581 if (os_sigaction == NULL) return; 4582 } 4583 4584 os_sigaction(sig, (struct sigaction*)NULL, &act); 4585 4586 4587 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK; 4588 4589 address thisHandler = (act.sa_flags & SA_SIGINFO) 4590 ? CAST_FROM_FN_PTR(address, act.sa_sigaction) 4591 : CAST_FROM_FN_PTR(address, act.sa_handler); 4592 4593 4594 switch (sig) { 4595 case SIGSEGV: 4596 case SIGBUS: 4597 case SIGFPE: 4598 case SIGPIPE: 4599 case SIGILL: 4600 case SIGXFSZ: 4601 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler); 4602 break; 4603 4604 case SHUTDOWN1_SIGNAL: 4605 case SHUTDOWN2_SIGNAL: 4606 case SHUTDOWN3_SIGNAL: 4607 case BREAK_SIGNAL: 4608 jvmHandler = (address)user_handler(); 4609 break; 4610 4611 case INTERRUPT_SIGNAL: 4612 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL); 4613 break; 4614 4615 default: 4616 if (sig == SR_signum) { 4617 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler); 4618 } else { 4619 return; 4620 } 4621 break; 4622 } 4623 4624 if (thisHandler != jvmHandler) { 4625 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN)); 4626 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN)); 4627 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN)); 4628 // No need to check this sig any longer 4629 sigaddset(&check_signal_done, sig); 4630 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN 4631 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) { 4632 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell", 4633 exception_name(sig, buf, O_BUFLEN)); 4634 } 4635 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) { 4636 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN)); 4637 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig)); 4638 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags); 4639 // No need to check this sig any longer 4640 sigaddset(&check_signal_done, sig); 4641 } 4642 4643 // Dump all the signal 4644 if (sigismember(&check_signal_done, sig)) { 4645 print_signal_handlers(tty, buf, O_BUFLEN); 4646 } 4647} 4648 4649extern void report_error(char* file_name, int line_no, char* title, 4650 char* format, ...); 4651 4652extern bool signal_name(int signo, char* buf, size_t len); 4653 4654const char* os::exception_name(int exception_code, char* buf, size_t size) { 4655 if (0 < exception_code && exception_code <= SIGRTMAX) { 4656 // signal 4657 if (!signal_name(exception_code, buf, size)) { 4658 jio_snprintf(buf, size, "SIG%d", exception_code); 4659 } 4660 return buf; 4661 } else { 4662 return NULL; 4663 } 4664} 4665 4666// this is called _before_ the most of global arguments have been parsed 4667void os::init(void) { 4668 char dummy; // used to get a guess on initial stack address 4669// first_hrtime = gethrtime(); 4670 4671 // With LinuxThreads the JavaMain thread pid (primordial thread) 4672 // is different than the pid of the java launcher thread. 4673 // So, on Linux, the launcher thread pid is passed to the VM 4674 // via the sun.java.launcher.pid property. 4675 // Use this property instead of getpid() if it was correctly passed. 4676 // See bug 6351349. 4677 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid(); 4678 4679 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid(); 4680 4681 clock_tics_per_sec = sysconf(_SC_CLK_TCK); 4682 4683 init_random(1234567); 4684 4685 ThreadCritical::initialize(); 4686 4687 Linux::set_page_size(sysconf(_SC_PAGESIZE)); 4688 if (Linux::page_size() == -1) { 4689 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)", 4690 strerror(errno))); 4691 } 4692 init_page_sizes((size_t) Linux::page_size()); 4693 4694 Linux::initialize_system_info(); 4695 4696 // main_thread points to the aboriginal thread 4697 Linux::_main_thread = pthread_self(); 4698 4699 Linux::clock_init(); 4700 initial_time_count = javaTimeNanos(); 4701 4702 // pthread_condattr initialization for monotonic clock 4703 int status; 4704 pthread_condattr_t* _condattr = os::Linux::condAttr(); 4705 if ((status = pthread_condattr_init(_condattr)) != 0) { 4706 fatal(err_msg("pthread_condattr_init: %s", strerror(status))); 4707 } 4708 // Only set the clock if CLOCK_MONOTONIC is available 4709 if (os::supports_monotonic_clock()) { 4710 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) { 4711 if (status == EINVAL) { 4712 warning("Unable to use monotonic clock with relative timed-waits" \ 4713 " - changes to the time-of-day clock may have adverse affects"); 4714 } else { 4715 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status))); 4716 } 4717 } 4718 } 4719 // else it defaults to CLOCK_REALTIME 4720 4721 // If the pagesize of the VM is greater than 8K determine the appropriate 4722 // number of initial guard pages. The user can change this with the 4723 // command line arguments, if needed. 4724 if (vm_page_size() > (int)Linux::vm_default_page_size()) { 4725 StackYellowPages = 1; 4726 StackRedPages = 1; 4727 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size(); 4728 } 4729 4730 // retrieve entry point for pthread_setname_np 4731 Linux::_pthread_setname_np = 4732 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np"); 4733 4734} 4735 4736// To install functions for atexit system call 4737extern "C" { 4738 static void perfMemory_exit_helper() { 4739 perfMemory_exit(); 4740 } 4741} 4742 4743// this is called _after_ the global arguments have been parsed 4744jint os::init_2(void) { 4745 Linux::fast_thread_clock_init(); 4746 4747 // Allocate a single page and mark it as readable for safepoint polling 4748 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4749 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page"); 4750 4751 os::set_polling_page(polling_page); 4752 4753#ifndef PRODUCT 4754 if (Verbose && PrintMiscellaneous) { 4755 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", 4756 (intptr_t)polling_page); 4757 } 4758#endif 4759 4760 if (!UseMembar) { 4761 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); 4762 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page"); 4763 os::set_memory_serialize_page(mem_serialize_page); 4764 4765#ifndef PRODUCT 4766 if (Verbose && PrintMiscellaneous) { 4767 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", 4768 (intptr_t)mem_serialize_page); 4769 } 4770#endif 4771 } 4772 4773 // initialize suspend/resume support - must do this before signal_sets_init() 4774 if (SR_initialize() != 0) { 4775 perror("SR_initialize failed"); 4776 return JNI_ERR; 4777 } 4778 4779 Linux::signal_sets_init(); 4780 Linux::install_signal_handlers(); 4781 4782 // Check minimum allowable stack size for thread creation and to initialize 4783 // the java system classes, including StackOverflowError - depends on page 4784 // size. Add a page for compiler2 recursion in main thread. 4785 // Add in 2*BytesPerWord times page size to account for VM stack during 4786 // class initialization depending on 32 or 64 bit VM. 4787 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed, 4788 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() + 4789 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size()); 4790 4791 size_t threadStackSizeInBytes = ThreadStackSize * K; 4792 if (threadStackSizeInBytes != 0 && 4793 threadStackSizeInBytes < os::Linux::min_stack_allowed) { 4794 tty->print_cr("\nThe stack size specified is too small, " 4795 "Specify at least %dk", 4796 os::Linux::min_stack_allowed/ K); 4797 return JNI_ERR; 4798 } 4799 4800 // Make the stack size a multiple of the page size so that 4801 // the yellow/red zones can be guarded. 4802 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes, 4803 vm_page_size())); 4804 4805 Linux::capture_initial_stack(JavaThread::stack_size_at_create()); 4806 4807#if defined(IA32) 4808 workaround_expand_exec_shield_cs_limit(); 4809#endif 4810 4811 Linux::libpthread_init(); 4812 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4813 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n", 4814 Linux::glibc_version(), Linux::libpthread_version(), 4815 Linux::is_floating_stack() ? "floating stack" : "fixed stack"); 4816 } 4817 4818 if (UseNUMA) { 4819 if (!Linux::libnuma_init()) { 4820 UseNUMA = false; 4821 } else { 4822 if ((Linux::numa_max_node() < 1)) { 4823 // There's only one node(they start from 0), disable NUMA. 4824 UseNUMA = false; 4825 } 4826 } 4827 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way 4828 // we can make the adaptive lgrp chunk resizing work. If the user specified 4829 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and 4830 // disable adaptive resizing. 4831 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) { 4832 if (FLAG_IS_DEFAULT(UseNUMA)) { 4833 UseNUMA = false; 4834 } else { 4835 if (FLAG_IS_DEFAULT(UseLargePages) && 4836 FLAG_IS_DEFAULT(UseSHM) && 4837 FLAG_IS_DEFAULT(UseHugeTLBFS)) { 4838 UseLargePages = false; 4839 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) { 4840 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)"); 4841 UseAdaptiveSizePolicy = false; 4842 UseAdaptiveNUMAChunkSizing = false; 4843 } 4844 } 4845 } 4846 if (!UseNUMA && ForceNUMA) { 4847 UseNUMA = true; 4848 } 4849 } 4850 4851 if (MaxFDLimit) { 4852 // set the number of file descriptors to max. print out error 4853 // if getrlimit/setrlimit fails but continue regardless. 4854 struct rlimit nbr_files; 4855 int status = getrlimit(RLIMIT_NOFILE, &nbr_files); 4856 if (status != 0) { 4857 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4858 perror("os::init_2 getrlimit failed"); 4859 } 4860 } else { 4861 nbr_files.rlim_cur = nbr_files.rlim_max; 4862 status = setrlimit(RLIMIT_NOFILE, &nbr_files); 4863 if (status != 0) { 4864 if (PrintMiscellaneous && (Verbose || WizardMode)) { 4865 perror("os::init_2 setrlimit failed"); 4866 } 4867 } 4868 } 4869 } 4870 4871 // Initialize lock used to serialize thread creation (see os::create_thread) 4872 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false)); 4873 4874 // at-exit methods are called in the reverse order of their registration. 4875 // atexit functions are called on return from main or as a result of a 4876 // call to exit(3C). There can be only 32 of these functions registered 4877 // and atexit() does not set errno. 4878 4879 if (PerfAllowAtExitRegistration) { 4880 // only register atexit functions if PerfAllowAtExitRegistration is set. 4881 // atexit functions can be delayed until process exit time, which 4882 // can be problematic for embedded VM situations. Embedded VMs should 4883 // call DestroyJavaVM() to assure that VM resources are released. 4884 4885 // note: perfMemory_exit_helper atexit function may be removed in 4886 // the future if the appropriate cleanup code can be added to the 4887 // VM_Exit VMOperation's doit method. 4888 if (atexit(perfMemory_exit_helper) != 0) { 4889 warning("os::init_2 atexit(perfMemory_exit_helper) failed"); 4890 } 4891 } 4892 4893 // initialize thread priority policy 4894 prio_init(); 4895 4896 return JNI_OK; 4897} 4898 4899// Mark the polling page as unreadable 4900void os::make_polling_page_unreadable(void) { 4901 if (!guard_memory((char*)_polling_page, Linux::page_size())) { 4902 fatal("Could not disable polling page"); 4903 } 4904} 4905 4906// Mark the polling page as readable 4907void os::make_polling_page_readable(void) { 4908 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) { 4909 fatal("Could not enable polling page"); 4910 } 4911} 4912 4913int os::active_processor_count() { 4914 // Linux doesn't yet have a (official) notion of processor sets, 4915 // so just return the number of online processors. 4916 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN); 4917 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check"); 4918 return online_cpus; 4919} 4920 4921void os::set_native_thread_name(const char *name) { 4922 if (Linux::_pthread_setname_np) { 4923 char buf [16]; // according to glibc manpage, 16 chars incl. '/0' 4924 snprintf(buf, sizeof(buf), "%s", name); 4925 buf[sizeof(buf) - 1] = '\0'; 4926 const int rc = Linux::_pthread_setname_np(pthread_self(), buf); 4927 // ERANGE should not happen; all other errors should just be ignored. 4928 assert(rc != ERANGE, "pthread_setname_np failed"); 4929 } 4930} 4931 4932bool os::distribute_processes(uint length, uint* distribution) { 4933 // Not yet implemented. 4934 return false; 4935} 4936 4937bool os::bind_to_processor(uint processor_id) { 4938 // Not yet implemented. 4939 return false; 4940} 4941 4942/// 4943 4944void os::SuspendedThreadTask::internal_do_task() { 4945 if (do_suspend(_thread->osthread())) { 4946 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext()); 4947 do_task(context); 4948 do_resume(_thread->osthread()); 4949 } 4950} 4951 4952class PcFetcher : public os::SuspendedThreadTask { 4953 public: 4954 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {} 4955 ExtendedPC result(); 4956 protected: 4957 void do_task(const os::SuspendedThreadTaskContext& context); 4958 private: 4959 ExtendedPC _epc; 4960}; 4961 4962ExtendedPC PcFetcher::result() { 4963 guarantee(is_done(), "task is not done yet."); 4964 return _epc; 4965} 4966 4967void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) { 4968 Thread* thread = context.thread(); 4969 OSThread* osthread = thread->osthread(); 4970 if (osthread->ucontext() != NULL) { 4971 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext()); 4972 } else { 4973 // NULL context is unexpected, double-check this is the VMThread 4974 guarantee(thread->is_VM_thread(), "can only be called for VMThread"); 4975 } 4976} 4977 4978// Suspends the target using the signal mechanism and then grabs the PC before 4979// resuming the target. Used by the flat-profiler only 4980ExtendedPC os::get_thread_pc(Thread* thread) { 4981 // Make sure that it is called by the watcher for the VMThread 4982 assert(Thread::current()->is_Watcher_thread(), "Must be watcher"); 4983 assert(thread->is_VM_thread(), "Can only be called for VMThread"); 4984 4985 PcFetcher fetcher(thread); 4986 fetcher.run(); 4987 return fetcher.result(); 4988} 4989 4990int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, 4991 pthread_mutex_t *_mutex, 4992 const struct timespec *_abstime) { 4993 if (is_NPTL()) { 4994 return pthread_cond_timedwait(_cond, _mutex, _abstime); 4995 } else { 4996 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control 4997 // word back to default 64bit precision if condvar is signaled. Java 4998 // wants 53bit precision. Save and restore current value. 4999 int fpu = get_fpu_control_word(); 5000 int status = pthread_cond_timedwait(_cond, _mutex, _abstime); 5001 set_fpu_control_word(fpu); 5002 return status; 5003 } 5004} 5005 5006//////////////////////////////////////////////////////////////////////////////// 5007// debug support 5008 5009bool os::find(address addr, outputStream* st) { 5010 Dl_info dlinfo; 5011 memset(&dlinfo, 0, sizeof(dlinfo)); 5012 if (dladdr(addr, &dlinfo) != 0) { 5013 st->print(PTR_FORMAT ": ", addr); 5014 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) { 5015 st->print("%s+%#x", dlinfo.dli_sname, 5016 addr - (intptr_t)dlinfo.dli_saddr); 5017 } else if (dlinfo.dli_fbase != NULL) { 5018 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase); 5019 } else { 5020 st->print("<absolute address>"); 5021 } 5022 if (dlinfo.dli_fname != NULL) { 5023 st->print(" in %s", dlinfo.dli_fname); 5024 } 5025 if (dlinfo.dli_fbase != NULL) { 5026 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase); 5027 } 5028 st->cr(); 5029 5030 if (Verbose) { 5031 // decode some bytes around the PC 5032 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size()); 5033 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size()); 5034 address lowest = (address) dlinfo.dli_sname; 5035 if (!lowest) lowest = (address) dlinfo.dli_fbase; 5036 if (begin < lowest) begin = lowest; 5037 Dl_info dlinfo2; 5038 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr 5039 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) { 5040 end = (address) dlinfo2.dli_saddr; 5041 } 5042 Disassembler::decode(begin, end, st); 5043 } 5044 return true; 5045 } 5046 return false; 5047} 5048 5049//////////////////////////////////////////////////////////////////////////////// 5050// misc 5051 5052// This does not do anything on Linux. This is basically a hook for being 5053// able to use structured exception handling (thread-local exception filters) 5054// on, e.g., Win32. 5055void 5056os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, 5057 JavaCallArguments* args, Thread* thread) { 5058 f(value, method, args, thread); 5059} 5060 5061void os::print_statistics() { 5062} 5063 5064int os::message_box(const char* title, const char* message) { 5065 int i; 5066 fdStream err(defaultStream::error_fd()); 5067 for (i = 0; i < 78; i++) err.print_raw("="); 5068 err.cr(); 5069 err.print_raw_cr(title); 5070 for (i = 0; i < 78; i++) err.print_raw("-"); 5071 err.cr(); 5072 err.print_raw_cr(message); 5073 for (i = 0; i < 78; i++) err.print_raw("="); 5074 err.cr(); 5075 5076 char buf[16]; 5077 // Prevent process from exiting upon "read error" without consuming all CPU 5078 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); } 5079 5080 return buf[0] == 'y' || buf[0] == 'Y'; 5081} 5082 5083int os::stat(const char *path, struct stat *sbuf) { 5084 char pathbuf[MAX_PATH]; 5085 if (strlen(path) > MAX_PATH - 1) { 5086 errno = ENAMETOOLONG; 5087 return -1; 5088 } 5089 os::native_path(strcpy(pathbuf, path)); 5090 return ::stat(pathbuf, sbuf); 5091} 5092 5093bool os::check_heap(bool force) { 5094 return true; 5095} 5096 5097// Is a (classpath) directory empty? 5098bool os::dir_is_empty(const char* path) { 5099 DIR *dir = NULL; 5100 struct dirent *ptr; 5101 5102 dir = opendir(path); 5103 if (dir == NULL) return true; 5104 5105 // Scan the directory 5106 bool result = true; 5107 char buf[sizeof(struct dirent) + MAX_PATH]; 5108 while (result && (ptr = ::readdir(dir)) != NULL) { 5109 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) { 5110 result = false; 5111 } 5112 } 5113 closedir(dir); 5114 return result; 5115} 5116 5117// This code originates from JDK's sysOpen and open64_w 5118// from src/solaris/hpi/src/system_md.c 5119 5120int os::open(const char *path, int oflag, int mode) { 5121 if (strlen(path) > MAX_PATH - 1) { 5122 errno = ENAMETOOLONG; 5123 return -1; 5124 } 5125 5126 // All file descriptors that are opened in the Java process and not 5127 // specifically destined for a subprocess should have the close-on-exec 5128 // flag set. If we don't set it, then careless 3rd party native code 5129 // might fork and exec without closing all appropriate file descriptors 5130 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in 5131 // turn might: 5132 // 5133 // - cause end-of-file to fail to be detected on some file 5134 // descriptors, resulting in mysterious hangs, or 5135 // 5136 // - might cause an fopen in the subprocess to fail on a system 5137 // suffering from bug 1085341. 5138 // 5139 // (Yes, the default setting of the close-on-exec flag is a Unix 5140 // design flaw) 5141 // 5142 // See: 5143 // 1085341: 32-bit stdio routines should support file descriptors >255 5144 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed 5145 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9 5146 // 5147 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open(). 5148 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor 5149 // because it saves a system call and removes a small window where the flag 5150 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored 5151 // and we fall back to using FD_CLOEXEC (see below). 5152#ifdef O_CLOEXEC 5153 oflag |= O_CLOEXEC; 5154#endif 5155 5156 int fd = ::open64(path, oflag, mode); 5157 if (fd == -1) return -1; 5158 5159 //If the open succeeded, the file might still be a directory 5160 { 5161 struct stat64 buf64; 5162 int ret = ::fstat64(fd, &buf64); 5163 int st_mode = buf64.st_mode; 5164 5165 if (ret != -1) { 5166 if ((st_mode & S_IFMT) == S_IFDIR) { 5167 errno = EISDIR; 5168 ::close(fd); 5169 return -1; 5170 } 5171 } else { 5172 ::close(fd); 5173 return -1; 5174 } 5175 } 5176 5177#ifdef FD_CLOEXEC 5178 // Validate that the use of the O_CLOEXEC flag on open above worked. 5179 // With recent kernels, we will perform this check exactly once. 5180 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0; 5181 if (!O_CLOEXEC_is_known_to_work) { 5182 int flags = ::fcntl(fd, F_GETFD); 5183 if (flags != -1) { 5184 if ((flags & FD_CLOEXEC) != 0) 5185 O_CLOEXEC_is_known_to_work = 1; 5186 else 5187 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC); 5188 } 5189 } 5190#endif 5191 5192 return fd; 5193} 5194 5195 5196// create binary file, rewriting existing file if required 5197int os::create_binary_file(const char* path, bool rewrite_existing) { 5198 int oflags = O_WRONLY | O_CREAT; 5199 if (!rewrite_existing) { 5200 oflags |= O_EXCL; 5201 } 5202 return ::open64(path, oflags, S_IREAD | S_IWRITE); 5203} 5204 5205// return current position of file pointer 5206jlong os::current_file_offset(int fd) { 5207 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR); 5208} 5209 5210// move file pointer to the specified offset 5211jlong os::seek_to_file_offset(int fd, jlong offset) { 5212 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET); 5213} 5214 5215// This code originates from JDK's sysAvailable 5216// from src/solaris/hpi/src/native_threads/src/sys_api_td.c 5217 5218int os::available(int fd, jlong *bytes) { 5219 jlong cur, end; 5220 int mode; 5221 struct stat64 buf64; 5222 5223 if (::fstat64(fd, &buf64) >= 0) { 5224 mode = buf64.st_mode; 5225 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) { 5226 // XXX: is the following call interruptible? If so, this might 5227 // need to go through the INTERRUPT_IO() wrapper as for other 5228 // blocking, interruptible calls in this file. 5229 int n; 5230 if (::ioctl(fd, FIONREAD, &n) >= 0) { 5231 *bytes = n; 5232 return 1; 5233 } 5234 } 5235 } 5236 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) { 5237 return 0; 5238 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) { 5239 return 0; 5240 } else if (::lseek64(fd, cur, SEEK_SET) == -1) { 5241 return 0; 5242 } 5243 *bytes = end - cur; 5244 return 1; 5245} 5246 5247// Map a block of memory. 5248char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset, 5249 char *addr, size_t bytes, bool read_only, 5250 bool allow_exec) { 5251 int prot; 5252 int flags = MAP_PRIVATE; 5253 5254 if (read_only) { 5255 prot = PROT_READ; 5256 } else { 5257 prot = PROT_READ | PROT_WRITE; 5258 } 5259 5260 if (allow_exec) { 5261 prot |= PROT_EXEC; 5262 } 5263 5264 if (addr != NULL) { 5265 flags |= MAP_FIXED; 5266 } 5267 5268 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags, 5269 fd, file_offset); 5270 if (mapped_address == MAP_FAILED) { 5271 return NULL; 5272 } 5273 return mapped_address; 5274} 5275 5276 5277// Remap a block of memory. 5278char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset, 5279 char *addr, size_t bytes, bool read_only, 5280 bool allow_exec) { 5281 // same as map_memory() on this OS 5282 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only, 5283 allow_exec); 5284} 5285 5286 5287// Unmap a block of memory. 5288bool os::pd_unmap_memory(char* addr, size_t bytes) { 5289 return munmap(addr, bytes) == 0; 5290} 5291 5292static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time); 5293 5294static clockid_t thread_cpu_clockid(Thread* thread) { 5295 pthread_t tid = thread->osthread()->pthread_id(); 5296 clockid_t clockid; 5297 5298 // Get thread clockid 5299 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid); 5300 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code"); 5301 return clockid; 5302} 5303 5304// current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool) 5305// are used by JVM M&M and JVMTI to get user+sys or user CPU time 5306// of a thread. 5307// 5308// current_thread_cpu_time() and thread_cpu_time(Thread*) returns 5309// the fast estimate available on the platform. 5310 5311jlong os::current_thread_cpu_time() { 5312 if (os::Linux::supports_fast_thread_cpu_time()) { 5313 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5314 } else { 5315 // return user + sys since the cost is the same 5316 return slow_thread_cpu_time(Thread::current(), true /* user + sys */); 5317 } 5318} 5319 5320jlong os::thread_cpu_time(Thread* thread) { 5321 // consistent with what current_thread_cpu_time() returns 5322 if (os::Linux::supports_fast_thread_cpu_time()) { 5323 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5324 } else { 5325 return slow_thread_cpu_time(thread, true /* user + sys */); 5326 } 5327} 5328 5329jlong os::current_thread_cpu_time(bool user_sys_cpu_time) { 5330 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5331 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID); 5332 } else { 5333 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time); 5334 } 5335} 5336 5337jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5338 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) { 5339 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread)); 5340 } else { 5341 return slow_thread_cpu_time(thread, user_sys_cpu_time); 5342 } 5343} 5344 5345// -1 on error. 5346static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) { 5347 pid_t tid = thread->osthread()->thread_id(); 5348 char *s; 5349 char stat[2048]; 5350 int statlen; 5351 char proc_name[64]; 5352 int count; 5353 long sys_time, user_time; 5354 char cdummy; 5355 int idummy; 5356 long ldummy; 5357 FILE *fp; 5358 5359 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid); 5360 fp = fopen(proc_name, "r"); 5361 if (fp == NULL) return -1; 5362 statlen = fread(stat, 1, 2047, fp); 5363 stat[statlen] = '\0'; 5364 fclose(fp); 5365 5366 // Skip pid and the command string. Note that we could be dealing with 5367 // weird command names, e.g. user could decide to rename java launcher 5368 // to "java 1.4.2 :)", then the stat file would look like 5369 // 1234 (java 1.4.2 :)) R ... ... 5370 // We don't really need to know the command string, just find the last 5371 // occurrence of ")" and then start parsing from there. See bug 4726580. 5372 s = strrchr(stat, ')'); 5373 if (s == NULL) return -1; 5374 5375 // Skip blank chars 5376 do { s++; } while (s && isspace(*s)); 5377 5378 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", 5379 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy, 5380 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy, 5381 &user_time, &sys_time); 5382 if (count != 13) return -1; 5383 if (user_sys_cpu_time) { 5384 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec); 5385 } else { 5386 return (jlong)user_time * (1000000000 / clock_tics_per_sec); 5387 } 5388} 5389 5390void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5391 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5392 info_ptr->may_skip_backward = false; // elapsed time not wall time 5393 info_ptr->may_skip_forward = false; // elapsed time not wall time 5394 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5395} 5396 5397void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) { 5398 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits 5399 info_ptr->may_skip_backward = false; // elapsed time not wall time 5400 info_ptr->may_skip_forward = false; // elapsed time not wall time 5401 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned 5402} 5403 5404bool os::is_thread_cpu_time_supported() { 5405 return true; 5406} 5407 5408// System loadavg support. Returns -1 if load average cannot be obtained. 5409// Linux doesn't yet have a (official) notion of processor sets, 5410// so just return the system wide load average. 5411int os::loadavg(double loadavg[], int nelem) { 5412 return ::getloadavg(loadavg, nelem); 5413} 5414 5415void os::pause() { 5416 char filename[MAX_PATH]; 5417 if (PauseAtStartupFile && PauseAtStartupFile[0]) { 5418 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile); 5419 } else { 5420 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id()); 5421 } 5422 5423 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666); 5424 if (fd != -1) { 5425 struct stat buf; 5426 ::close(fd); 5427 while (::stat(filename, &buf) == 0) { 5428 (void)::poll(NULL, 0, 100); 5429 } 5430 } else { 5431 jio_fprintf(stderr, 5432 "Could not open pause file '%s', continuing immediately.\n", filename); 5433 } 5434} 5435 5436 5437// Refer to the comments in os_solaris.cpp park-unpark. The next two 5438// comment paragraphs are worth repeating here: 5439// 5440// Assumption: 5441// Only one parker can exist on an event, which is why we allocate 5442// them per-thread. Multiple unparkers can coexist. 5443// 5444// _Event serves as a restricted-range semaphore. 5445// -1 : thread is blocked, i.e. there is a waiter 5446// 0 : neutral: thread is running or ready, 5447// could have been signaled after a wait started 5448// 1 : signaled - thread is running or ready 5449// 5450// Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can 5451// hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable. 5452// For specifics regarding the bug see GLIBC BUGID 261237 : 5453// http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html. 5454// Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future 5455// will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar 5456// is used. (The simple C test-case provided in the GLIBC bug report manifests the 5457// hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos() 5458// and monitorenter when we're using 1-0 locking. All those operations may result in 5459// calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version 5460// of libpthread avoids the problem, but isn't practical. 5461// 5462// Possible remedies: 5463// 5464// 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work. 5465// This is palliative and probabilistic, however. If the thread is preempted 5466// between the call to compute_abstime() and pthread_cond_timedwait(), more 5467// than the minimum period may have passed, and the abstime may be stale (in the 5468// past) resultin in a hang. Using this technique reduces the odds of a hang 5469// but the JVM is still vulnerable, particularly on heavily loaded systems. 5470// 5471// 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead 5472// of the usual flag-condvar-mutex idiom. The write side of the pipe is set 5473// NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo) 5474// reduces to poll()+read(). This works well, but consumes 2 FDs per extant 5475// thread. 5476// 5477// 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread 5478// that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing 5479// a timeout request to the chron thread and then blocking via pthread_cond_wait(). 5480// This also works well. In fact it avoids kernel-level scalability impediments 5481// on certain platforms that don't handle lots of active pthread_cond_timedwait() 5482// timers in a graceful fashion. 5483// 5484// 4. When the abstime value is in the past it appears that control returns 5485// correctly from pthread_cond_timedwait(), but the condvar is left corrupt. 5486// Subsequent timedwait/wait calls may hang indefinitely. Given that, we 5487// can avoid the problem by reinitializing the condvar -- by cond_destroy() 5488// followed by cond_init() -- after all calls to pthread_cond_timedwait(). 5489// It may be possible to avoid reinitialization by checking the return 5490// value from pthread_cond_timedwait(). In addition to reinitializing the 5491// condvar we must establish the invariant that cond_signal() is only called 5492// within critical sections protected by the adjunct mutex. This prevents 5493// cond_signal() from "seeing" a condvar that's in the midst of being 5494// reinitialized or that is corrupt. Sadly, this invariant obviates the 5495// desirable signal-after-unlock optimization that avoids futile context switching. 5496// 5497// I'm also concerned that some versions of NTPL might allocate an auxilliary 5498// structure when a condvar is used or initialized. cond_destroy() would 5499// release the helper structure. Our reinitialize-after-timedwait fix 5500// put excessive stress on malloc/free and locks protecting the c-heap. 5501// 5502// We currently use (4). See the WorkAroundNTPLTimedWaitHang flag. 5503// It may be possible to refine (4) by checking the kernel and NTPL verisons 5504// and only enabling the work-around for vulnerable environments. 5505 5506// utility to compute the abstime argument to timedwait: 5507// millis is the relative timeout time 5508// abstime will be the absolute timeout time 5509// TODO: replace compute_abstime() with unpackTime() 5510 5511static struct timespec* compute_abstime(timespec* abstime, jlong millis) { 5512 if (millis < 0) millis = 0; 5513 5514 jlong seconds = millis / 1000; 5515 millis %= 1000; 5516 if (seconds > 50000000) { // see man cond_timedwait(3T) 5517 seconds = 50000000; 5518 } 5519 5520 if (os::supports_monotonic_clock()) { 5521 struct timespec now; 5522 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5523 assert_status(status == 0, status, "clock_gettime"); 5524 abstime->tv_sec = now.tv_sec + seconds; 5525 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC; 5526 if (nanos >= NANOSECS_PER_SEC) { 5527 abstime->tv_sec += 1; 5528 nanos -= NANOSECS_PER_SEC; 5529 } 5530 abstime->tv_nsec = nanos; 5531 } else { 5532 struct timeval now; 5533 int status = gettimeofday(&now, NULL); 5534 assert(status == 0, "gettimeofday"); 5535 abstime->tv_sec = now.tv_sec + seconds; 5536 long usec = now.tv_usec + millis * 1000; 5537 if (usec >= 1000000) { 5538 abstime->tv_sec += 1; 5539 usec -= 1000000; 5540 } 5541 abstime->tv_nsec = usec * 1000; 5542 } 5543 return abstime; 5544} 5545 5546void os::PlatformEvent::park() { // AKA "down()" 5547 // Transitions for _Event: 5548 // -1 => -1 : illegal 5549 // 1 => 0 : pass - return immediately 5550 // 0 => -1 : block; then set _Event to 0 before returning 5551 5552 // Invariant: Only the thread associated with the Event/PlatformEvent 5553 // may call park(). 5554 // TODO: assert that _Assoc != NULL or _Assoc == Self 5555 assert(_nParked == 0, "invariant"); 5556 5557 int v; 5558 for (;;) { 5559 v = _Event; 5560 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5561 } 5562 guarantee(v >= 0, "invariant"); 5563 if (v == 0) { 5564 // Do this the hard way by blocking ... 5565 int status = pthread_mutex_lock(_mutex); 5566 assert_status(status == 0, status, "mutex_lock"); 5567 guarantee(_nParked == 0, "invariant"); 5568 ++_nParked; 5569 while (_Event < 0) { 5570 status = pthread_cond_wait(_cond, _mutex); 5571 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ... 5572 // Treat this the same as if the wait was interrupted 5573 if (status == ETIME) { status = EINTR; } 5574 assert_status(status == 0 || status == EINTR, status, "cond_wait"); 5575 } 5576 --_nParked; 5577 5578 _Event = 0; 5579 status = pthread_mutex_unlock(_mutex); 5580 assert_status(status == 0, status, "mutex_unlock"); 5581 // Paranoia to ensure our locked and lock-free paths interact 5582 // correctly with each other. 5583 OrderAccess::fence(); 5584 } 5585 guarantee(_Event >= 0, "invariant"); 5586} 5587 5588int os::PlatformEvent::park(jlong millis) { 5589 // Transitions for _Event: 5590 // -1 => -1 : illegal 5591 // 1 => 0 : pass - return immediately 5592 // 0 => -1 : block; then set _Event to 0 before returning 5593 5594 guarantee(_nParked == 0, "invariant"); 5595 5596 int v; 5597 for (;;) { 5598 v = _Event; 5599 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break; 5600 } 5601 guarantee(v >= 0, "invariant"); 5602 if (v != 0) return OS_OK; 5603 5604 // We do this the hard way, by blocking the thread. 5605 // Consider enforcing a minimum timeout value. 5606 struct timespec abst; 5607 compute_abstime(&abst, millis); 5608 5609 int ret = OS_TIMEOUT; 5610 int status = pthread_mutex_lock(_mutex); 5611 assert_status(status == 0, status, "mutex_lock"); 5612 guarantee(_nParked == 0, "invariant"); 5613 ++_nParked; 5614 5615 // Object.wait(timo) will return because of 5616 // (a) notification 5617 // (b) timeout 5618 // (c) thread.interrupt 5619 // 5620 // Thread.interrupt and object.notify{All} both call Event::set. 5621 // That is, we treat thread.interrupt as a special case of notification. 5622 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false. 5623 // We assume all ETIME returns are valid. 5624 // 5625 // TODO: properly differentiate simultaneous notify+interrupt. 5626 // In that case, we should propagate the notify to another waiter. 5627 5628 while (_Event < 0) { 5629 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst); 5630 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5631 pthread_cond_destroy(_cond); 5632 pthread_cond_init(_cond, os::Linux::condAttr()); 5633 } 5634 assert_status(status == 0 || status == EINTR || 5635 status == ETIME || status == ETIMEDOUT, 5636 status, "cond_timedwait"); 5637 if (!FilterSpuriousWakeups) break; // previous semantics 5638 if (status == ETIME || status == ETIMEDOUT) break; 5639 // We consume and ignore EINTR and spurious wakeups. 5640 } 5641 --_nParked; 5642 if (_Event >= 0) { 5643 ret = OS_OK; 5644 } 5645 _Event = 0; 5646 status = pthread_mutex_unlock(_mutex); 5647 assert_status(status == 0, status, "mutex_unlock"); 5648 assert(_nParked == 0, "invariant"); 5649 // Paranoia to ensure our locked and lock-free paths interact 5650 // correctly with each other. 5651 OrderAccess::fence(); 5652 return ret; 5653} 5654 5655void os::PlatformEvent::unpark() { 5656 // Transitions for _Event: 5657 // 0 => 1 : just return 5658 // 1 => 1 : just return 5659 // -1 => either 0 or 1; must signal target thread 5660 // That is, we can safely transition _Event from -1 to either 5661 // 0 or 1. 5662 // See also: "Semaphores in Plan 9" by Mullender & Cox 5663 // 5664 // Note: Forcing a transition from "-1" to "1" on an unpark() means 5665 // that it will take two back-to-back park() calls for the owning 5666 // thread to block. This has the benefit of forcing a spurious return 5667 // from the first park() call after an unpark() call which will help 5668 // shake out uses of park() and unpark() without condition variables. 5669 5670 if (Atomic::xchg(1, &_Event) >= 0) return; 5671 5672 // Wait for the thread associated with the event to vacate 5673 int status = pthread_mutex_lock(_mutex); 5674 assert_status(status == 0, status, "mutex_lock"); 5675 int AnyWaiters = _nParked; 5676 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant"); 5677 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) { 5678 AnyWaiters = 0; 5679 pthread_cond_signal(_cond); 5680 } 5681 status = pthread_mutex_unlock(_mutex); 5682 assert_status(status == 0, status, "mutex_unlock"); 5683 if (AnyWaiters != 0) { 5684 // Note that we signal() *after* dropping the lock for "immortal" Events. 5685 // This is safe and avoids a common class of futile wakeups. In rare 5686 // circumstances this can cause a thread to return prematurely from 5687 // cond_{timed}wait() but the spurious wakeup is benign and the victim 5688 // will simply re-test the condition and re-park itself. 5689 // This provides particular benefit if the underlying platform does not 5690 // provide wait morphing. 5691 status = pthread_cond_signal(_cond); 5692 assert_status(status == 0, status, "cond_signal"); 5693 } 5694} 5695 5696 5697// JSR166 5698// ------------------------------------------------------- 5699 5700// The solaris and linux implementations of park/unpark are fairly 5701// conservative for now, but can be improved. They currently use a 5702// mutex/condvar pair, plus a a count. 5703// Park decrements count if > 0, else does a condvar wait. Unpark 5704// sets count to 1 and signals condvar. Only one thread ever waits 5705// on the condvar. Contention seen when trying to park implies that someone 5706// is unparking you, so don't wait. And spurious returns are fine, so there 5707// is no need to track notifications. 5708 5709// This code is common to linux and solaris and will be moved to a 5710// common place in dolphin. 5711// 5712// The passed in time value is either a relative time in nanoseconds 5713// or an absolute time in milliseconds. Either way it has to be unpacked 5714// into suitable seconds and nanoseconds components and stored in the 5715// given timespec structure. 5716// Given time is a 64-bit value and the time_t used in the timespec is only 5717// a signed-32-bit value (except on 64-bit Linux) we have to watch for 5718// overflow if times way in the future are given. Further on Solaris versions 5719// prior to 10 there is a restriction (see cond_timedwait) that the specified 5720// number of seconds, in abstime, is less than current_time + 100,000,000. 5721// As it will be 28 years before "now + 100000000" will overflow we can 5722// ignore overflow and just impose a hard-limit on seconds using the value 5723// of "now + 100,000,000". This places a limit on the timeout of about 3.17 5724// years from "now". 5725 5726static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) { 5727 assert(time > 0, "convertTime"); 5728 time_t max_secs = 0; 5729 5730 if (!os::supports_monotonic_clock() || isAbsolute) { 5731 struct timeval now; 5732 int status = gettimeofday(&now, NULL); 5733 assert(status == 0, "gettimeofday"); 5734 5735 max_secs = now.tv_sec + MAX_SECS; 5736 5737 if (isAbsolute) { 5738 jlong secs = time / 1000; 5739 if (secs > max_secs) { 5740 absTime->tv_sec = max_secs; 5741 } else { 5742 absTime->tv_sec = secs; 5743 } 5744 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC; 5745 } else { 5746 jlong secs = time / NANOSECS_PER_SEC; 5747 if (secs >= MAX_SECS) { 5748 absTime->tv_sec = max_secs; 5749 absTime->tv_nsec = 0; 5750 } else { 5751 absTime->tv_sec = now.tv_sec + secs; 5752 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000; 5753 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5754 absTime->tv_nsec -= NANOSECS_PER_SEC; 5755 ++absTime->tv_sec; // note: this must be <= max_secs 5756 } 5757 } 5758 } 5759 } else { 5760 // must be relative using monotonic clock 5761 struct timespec now; 5762 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now); 5763 assert_status(status == 0, status, "clock_gettime"); 5764 max_secs = now.tv_sec + MAX_SECS; 5765 jlong secs = time / NANOSECS_PER_SEC; 5766 if (secs >= MAX_SECS) { 5767 absTime->tv_sec = max_secs; 5768 absTime->tv_nsec = 0; 5769 } else { 5770 absTime->tv_sec = now.tv_sec + secs; 5771 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec; 5772 if (absTime->tv_nsec >= NANOSECS_PER_SEC) { 5773 absTime->tv_nsec -= NANOSECS_PER_SEC; 5774 ++absTime->tv_sec; // note: this must be <= max_secs 5775 } 5776 } 5777 } 5778 assert(absTime->tv_sec >= 0, "tv_sec < 0"); 5779 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs"); 5780 assert(absTime->tv_nsec >= 0, "tv_nsec < 0"); 5781 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec"); 5782} 5783 5784void Parker::park(bool isAbsolute, jlong time) { 5785 // Ideally we'd do something useful while spinning, such 5786 // as calling unpackTime(). 5787 5788 // Optional fast-path check: 5789 // Return immediately if a permit is available. 5790 // We depend on Atomic::xchg() having full barrier semantics 5791 // since we are doing a lock-free update to _counter. 5792 if (Atomic::xchg(0, &_counter) > 0) return; 5793 5794 Thread* thread = Thread::current(); 5795 assert(thread->is_Java_thread(), "Must be JavaThread"); 5796 JavaThread *jt = (JavaThread *)thread; 5797 5798 // Optional optimization -- avoid state transitions if there's an interrupt pending. 5799 // Check interrupt before trying to wait 5800 if (Thread::is_interrupted(thread, false)) { 5801 return; 5802 } 5803 5804 // Next, demultiplex/decode time arguments 5805 timespec absTime; 5806 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all 5807 return; 5808 } 5809 if (time > 0) { 5810 unpackTime(&absTime, isAbsolute, time); 5811 } 5812 5813 5814 // Enter safepoint region 5815 // Beware of deadlocks such as 6317397. 5816 // The per-thread Parker:: mutex is a classic leaf-lock. 5817 // In particular a thread must never block on the Threads_lock while 5818 // holding the Parker:: mutex. If safepoints are pending both the 5819 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock. 5820 ThreadBlockInVM tbivm(jt); 5821 5822 // Don't wait if cannot get lock since interference arises from 5823 // unblocking. Also. check interrupt before trying wait 5824 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) { 5825 return; 5826 } 5827 5828 int status; 5829 if (_counter > 0) { // no wait needed 5830 _counter = 0; 5831 status = pthread_mutex_unlock(_mutex); 5832 assert(status == 0, "invariant"); 5833 // Paranoia to ensure our locked and lock-free paths interact 5834 // correctly with each other and Java-level accesses. 5835 OrderAccess::fence(); 5836 return; 5837 } 5838 5839#ifdef ASSERT 5840 // Don't catch signals while blocked; let the running threads have the signals. 5841 // (This allows a debugger to break into the running thread.) 5842 sigset_t oldsigs; 5843 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals(); 5844 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs); 5845#endif 5846 5847 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */); 5848 jt->set_suspend_equivalent(); 5849 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self() 5850 5851 assert(_cur_index == -1, "invariant"); 5852 if (time == 0) { 5853 _cur_index = REL_INDEX; // arbitrary choice when not timed 5854 status = pthread_cond_wait(&_cond[_cur_index], _mutex); 5855 } else { 5856 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX; 5857 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime); 5858 if (status != 0 && WorkAroundNPTLTimedWaitHang) { 5859 pthread_cond_destroy(&_cond[_cur_index]); 5860 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr()); 5861 } 5862 } 5863 _cur_index = -1; 5864 assert_status(status == 0 || status == EINTR || 5865 status == ETIME || status == ETIMEDOUT, 5866 status, "cond_timedwait"); 5867 5868#ifdef ASSERT 5869 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL); 5870#endif 5871 5872 _counter = 0; 5873 status = pthread_mutex_unlock(_mutex); 5874 assert_status(status == 0, status, "invariant"); 5875 // Paranoia to ensure our locked and lock-free paths interact 5876 // correctly with each other and Java-level accesses. 5877 OrderAccess::fence(); 5878 5879 // If externally suspended while waiting, re-suspend 5880 if (jt->handle_special_suspend_equivalent_condition()) { 5881 jt->java_suspend_self(); 5882 } 5883} 5884 5885void Parker::unpark() { 5886 int status = pthread_mutex_lock(_mutex); 5887 assert(status == 0, "invariant"); 5888 const int s = _counter; 5889 _counter = 1; 5890 if (s < 1) { 5891 // thread might be parked 5892 if (_cur_index != -1) { 5893 // thread is definitely parked 5894 if (WorkAroundNPTLTimedWaitHang) { 5895 status = pthread_cond_signal(&_cond[_cur_index]); 5896 assert(status == 0, "invariant"); 5897 status = pthread_mutex_unlock(_mutex); 5898 assert(status == 0, "invariant"); 5899 } else { 5900 status = pthread_mutex_unlock(_mutex); 5901 assert(status == 0, "invariant"); 5902 status = pthread_cond_signal(&_cond[_cur_index]); 5903 assert(status == 0, "invariant"); 5904 } 5905 } else { 5906 pthread_mutex_unlock(_mutex); 5907 assert(status == 0, "invariant"); 5908 } 5909 } else { 5910 pthread_mutex_unlock(_mutex); 5911 assert(status == 0, "invariant"); 5912 } 5913} 5914 5915 5916extern char** environ; 5917 5918// Run the specified command in a separate process. Return its exit value, 5919// or -1 on failure (e.g. can't fork a new process). 5920// Unlike system(), this function can be called from signal handler. It 5921// doesn't block SIGINT et al. 5922int os::fork_and_exec(char* cmd) { 5923 const char * argv[4] = {"sh", "-c", cmd, NULL}; 5924 5925 pid_t pid = fork(); 5926 5927 if (pid < 0) { 5928 // fork failed 5929 return -1; 5930 5931 } else if (pid == 0) { 5932 // child process 5933 5934 execve("/bin/sh", (char* const*)argv, environ); 5935 5936 // execve failed 5937 _exit(-1); 5938 5939 } else { 5940 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't 5941 // care about the actual exit code, for now. 5942 5943 int status; 5944 5945 // Wait for the child process to exit. This returns immediately if 5946 // the child has already exited. */ 5947 while (waitpid(pid, &status, 0) < 0) { 5948 switch (errno) { 5949 case ECHILD: return 0; 5950 case EINTR: break; 5951 default: return -1; 5952 } 5953 } 5954 5955 if (WIFEXITED(status)) { 5956 // The child exited normally; get its exit code. 5957 return WEXITSTATUS(status); 5958 } else if (WIFSIGNALED(status)) { 5959 // The child exited because of a signal 5960 // The best value to return is 0x80 + signal number, 5961 // because that is what all Unix shells do, and because 5962 // it allows callers to distinguish between process exit and 5963 // process death by signal. 5964 return 0x80 + WTERMSIG(status); 5965 } else { 5966 // Unknown exit code; pass it through 5967 return status; 5968 } 5969 } 5970} 5971 5972// is_headless_jre() 5973// 5974// Test for the existence of xawt/libmawt.so or libawt_xawt.so 5975// in order to report if we are running in a headless jre 5976// 5977// Since JDK8 xawt/libmawt.so was moved into the same directory 5978// as libawt.so, and renamed libawt_xawt.so 5979// 5980bool os::is_headless_jre() { 5981 struct stat statbuf; 5982 char buf[MAXPATHLEN]; 5983 char libmawtpath[MAXPATHLEN]; 5984 const char *xawtstr = "/xawt/libmawt.so"; 5985 const char *new_xawtstr = "/libawt_xawt.so"; 5986 char *p; 5987 5988 // Get path to libjvm.so 5989 os::jvm_path(buf, sizeof(buf)); 5990 5991 // Get rid of libjvm.so 5992 p = strrchr(buf, '/'); 5993 if (p == NULL) { 5994 return false; 5995 } else { 5996 *p = '\0'; 5997 } 5998 5999 // Get rid of client or server 6000 p = strrchr(buf, '/'); 6001 if (p == NULL) { 6002 return false; 6003 } else { 6004 *p = '\0'; 6005 } 6006 6007 // check xawt/libmawt.so 6008 strcpy(libmawtpath, buf); 6009 strcat(libmawtpath, xawtstr); 6010 if (::stat(libmawtpath, &statbuf) == 0) return false; 6011 6012 // check libawt_xawt.so 6013 strcpy(libmawtpath, buf); 6014 strcat(libmawtpath, new_xawtstr); 6015 if (::stat(libmawtpath, &statbuf) == 0) return false; 6016 6017 return true; 6018} 6019 6020// Get the default path to the core file 6021// Returns the length of the string 6022int os::get_core_path(char* buffer, size_t bufferSize) { 6023 /* 6024 * Max length of /proc/sys/kernel/core_pattern is 128 characters. 6025 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt 6026 */ 6027 const int core_pattern_len = 129; 6028 char core_pattern[core_pattern_len] = {0}; 6029 6030 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY); 6031 if (core_pattern_file != -1) { 6032 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len); 6033 ::close(core_pattern_file); 6034 6035 if (ret > 0) { 6036 char *last_char = core_pattern + strlen(core_pattern) - 1; 6037 6038 if (*last_char == '\n') { 6039 *last_char = '\0'; 6040 } 6041 } 6042 } 6043 6044 if (strlen(core_pattern) == 0) { 6045 return -1; 6046 } 6047 6048 char *pid_pos = strstr(core_pattern, "%p"); 6049 int written; 6050 6051 if (core_pattern[0] == '/') { 6052 written = jio_snprintf(buffer, bufferSize, core_pattern); 6053 } else { 6054 char cwd[PATH_MAX]; 6055 6056 const char* p = get_current_directory(cwd, PATH_MAX); 6057 if (p == NULL) { 6058 return -1; 6059 } 6060 6061 if (core_pattern[0] == '|') { 6062 written = jio_snprintf(buffer, bufferSize, 6063 "\"%s\" (or dumping to %s/core.%d)", 6064 &core_pattern[1], p, current_process_id()); 6065 } else { 6066 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern); 6067 } 6068 } 6069 6070 if (written < 0) { 6071 return -1; 6072 } 6073 6074 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) { 6075 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY); 6076 6077 if (core_uses_pid_file != -1) { 6078 char core_uses_pid = 0; 6079 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1); 6080 ::close(core_uses_pid_file); 6081 6082 if (core_uses_pid == '1') { 6083 jio_snprintf(buffer + written, bufferSize - written, 6084 ".%d", current_process_id()); 6085 } 6086 } 6087 } 6088 6089 return strlen(buffer); 6090} 6091 6092/////////////// Unit tests /////////////// 6093 6094#ifndef PRODUCT 6095 6096#define test_log(...) \ 6097 do { \ 6098 if (VerboseInternalVMTests) { \ 6099 tty->print_cr(__VA_ARGS__); \ 6100 tty->flush(); \ 6101 } \ 6102 } while (false) 6103 6104class TestReserveMemorySpecial : AllStatic { 6105 public: 6106 static void small_page_write(void* addr, size_t size) { 6107 size_t page_size = os::vm_page_size(); 6108 6109 char* end = (char*)addr + size; 6110 for (char* p = (char*)addr; p < end; p += page_size) { 6111 *p = 1; 6112 } 6113 } 6114 6115 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) { 6116 if (!UseHugeTLBFS) { 6117 return; 6118 } 6119 6120 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size); 6121 6122 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false); 6123 6124 if (addr != NULL) { 6125 small_page_write(addr, size); 6126 6127 os::Linux::release_memory_special_huge_tlbfs(addr, size); 6128 } 6129 } 6130 6131 static void test_reserve_memory_special_huge_tlbfs_only() { 6132 if (!UseHugeTLBFS) { 6133 return; 6134 } 6135 6136 size_t lp = os::large_page_size(); 6137 6138 for (size_t size = lp; size <= lp * 10; size += lp) { 6139 test_reserve_memory_special_huge_tlbfs_only(size); 6140 } 6141 } 6142 6143 static void test_reserve_memory_special_huge_tlbfs_mixed() { 6144 size_t lp = os::large_page_size(); 6145 size_t ag = os::vm_allocation_granularity(); 6146 6147 // sizes to test 6148 const size_t sizes[] = { 6149 lp, lp + ag, lp + lp / 2, lp * 2, 6150 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2, 6151 lp * 10, lp * 10 + lp / 2 6152 }; 6153 const int num_sizes = sizeof(sizes) / sizeof(size_t); 6154 6155 // For each size/alignment combination, we test three scenarios: 6156 // 1) with req_addr == NULL 6157 // 2) with a non-null req_addr at which we expect to successfully allocate 6158 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we 6159 // expect the allocation to either fail or to ignore req_addr 6160 6161 // Pre-allocate two areas; they shall be as large as the largest allocation 6162 // and aligned to the largest alignment we will be testing. 6163 const size_t mapping_size = sizes[num_sizes - 1] * 2; 6164 char* const mapping1 = (char*) ::mmap(NULL, mapping_size, 6165 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6166 -1, 0); 6167 assert(mapping1 != MAP_FAILED, "should work"); 6168 6169 char* const mapping2 = (char*) ::mmap(NULL, mapping_size, 6170 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE, 6171 -1, 0); 6172 assert(mapping2 != MAP_FAILED, "should work"); 6173 6174 // Unmap the first mapping, but leave the second mapping intact: the first 6175 // mapping will serve as a value for a "good" req_addr (case 2). The second 6176 // mapping, still intact, as "bad" req_addr (case 3). 6177 ::munmap(mapping1, mapping_size); 6178 6179 // Case 1 6180 test_log("%s, req_addr NULL:", __FUNCTION__); 6181 test_log("size align result"); 6182 6183 for (int i = 0; i < num_sizes; i++) { 6184 const size_t size = sizes[i]; 6185 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6186 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false); 6187 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s", 6188 size, alignment, p, (p != NULL ? "" : "(failed)")); 6189 if (p != NULL) { 6190 assert(is_ptr_aligned(p, alignment), "must be"); 6191 small_page_write(p, size); 6192 os::Linux::release_memory_special_huge_tlbfs(p, size); 6193 } 6194 } 6195 } 6196 6197 // Case 2 6198 test_log("%s, req_addr non-NULL:", __FUNCTION__); 6199 test_log("size align req_addr result"); 6200 6201 for (int i = 0; i < num_sizes; i++) { 6202 const size_t size = sizes[i]; 6203 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6204 char* const req_addr = (char*) align_ptr_up(mapping1, alignment); 6205 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6206 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6207 size, alignment, req_addr, p, 6208 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)"))); 6209 if (p != NULL) { 6210 assert(p == req_addr, "must be"); 6211 small_page_write(p, size); 6212 os::Linux::release_memory_special_huge_tlbfs(p, size); 6213 } 6214 } 6215 } 6216 6217 // Case 3 6218 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__); 6219 test_log("size align req_addr result"); 6220 6221 for (int i = 0; i < num_sizes; i++) { 6222 const size_t size = sizes[i]; 6223 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6224 char* const req_addr = (char*) align_ptr_up(mapping2, alignment); 6225 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false); 6226 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s", 6227 size, alignment, req_addr, p, 6228 ((p != NULL ? "" : "(failed)"))); 6229 // as the area around req_addr contains already existing mappings, the API should always 6230 // return NULL (as per contract, it cannot return another address) 6231 assert(p == NULL, "must be"); 6232 } 6233 } 6234 6235 ::munmap(mapping2, mapping_size); 6236 6237 } 6238 6239 static void test_reserve_memory_special_huge_tlbfs() { 6240 if (!UseHugeTLBFS) { 6241 return; 6242 } 6243 6244 test_reserve_memory_special_huge_tlbfs_only(); 6245 test_reserve_memory_special_huge_tlbfs_mixed(); 6246 } 6247 6248 static void test_reserve_memory_special_shm(size_t size, size_t alignment) { 6249 if (!UseSHM) { 6250 return; 6251 } 6252 6253 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment); 6254 6255 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false); 6256 6257 if (addr != NULL) { 6258 assert(is_ptr_aligned(addr, alignment), "Check"); 6259 assert(is_ptr_aligned(addr, os::large_page_size()), "Check"); 6260 6261 small_page_write(addr, size); 6262 6263 os::Linux::release_memory_special_shm(addr, size); 6264 } 6265 } 6266 6267 static void test_reserve_memory_special_shm() { 6268 size_t lp = os::large_page_size(); 6269 size_t ag = os::vm_allocation_granularity(); 6270 6271 for (size_t size = ag; size < lp * 3; size += ag) { 6272 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) { 6273 test_reserve_memory_special_shm(size, alignment); 6274 } 6275 } 6276 } 6277 6278 static void test() { 6279 test_reserve_memory_special_huge_tlbfs(); 6280 test_reserve_memory_special_shm(); 6281 } 6282}; 6283 6284void TestReserveMemorySpecial_test() { 6285 TestReserveMemorySpecial::test(); 6286} 6287 6288#endif 6289